tracing: Remove special traces
[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 /**
1335  * trace_vbprintk - write binary msg to tracing buffer
1336  *
1337  */
1338 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1339 {
1340         static arch_spinlock_t trace_buf_lock =
1341                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1342         static u32 trace_buf[TRACE_BUF_SIZE];
1343
1344         struct ftrace_event_call *call = &event_bprint;
1345         struct ring_buffer_event *event;
1346         struct ring_buffer *buffer;
1347         struct trace_array *tr = &global_trace;
1348         struct trace_array_cpu *data;
1349         struct bprint_entry *entry;
1350         unsigned long flags;
1351         int disable;
1352         int cpu, len = 0, size, pc;
1353
1354         if (unlikely(tracing_selftest_running || tracing_disabled))
1355                 return 0;
1356
1357         /* Don't pollute graph traces with trace_vprintk internals */
1358         pause_graph_tracing();
1359
1360         pc = preempt_count();
1361         preempt_disable_notrace();
1362         cpu = raw_smp_processor_id();
1363         data = tr->data[cpu];
1364
1365         disable = atomic_inc_return(&data->disabled);
1366         if (unlikely(disable != 1))
1367                 goto out;
1368
1369         /* Lockdep uses trace_printk for lock tracing */
1370         local_irq_save(flags);
1371         arch_spin_lock(&trace_buf_lock);
1372         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1373
1374         if (len > TRACE_BUF_SIZE || len < 0)
1375                 goto out_unlock;
1376
1377         size = sizeof(*entry) + sizeof(u32) * len;
1378         buffer = tr->buffer;
1379         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1380                                           flags, pc);
1381         if (!event)
1382                 goto out_unlock;
1383         entry = ring_buffer_event_data(event);
1384         entry->ip                       = ip;
1385         entry->fmt                      = fmt;
1386
1387         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1388         if (!filter_check_discard(call, entry, buffer, event)) {
1389                 ring_buffer_unlock_commit(buffer, event);
1390                 ftrace_trace_stack(buffer, flags, 6, pc);
1391         }
1392
1393 out_unlock:
1394         arch_spin_unlock(&trace_buf_lock);
1395         local_irq_restore(flags);
1396
1397 out:
1398         atomic_dec_return(&data->disabled);
1399         preempt_enable_notrace();
1400         unpause_graph_tracing();
1401
1402         return len;
1403 }
1404 EXPORT_SYMBOL_GPL(trace_vbprintk);
1405
1406 int trace_array_printk(struct trace_array *tr,
1407                        unsigned long ip, const char *fmt, ...)
1408 {
1409         int ret;
1410         va_list ap;
1411
1412         if (!(trace_flags & TRACE_ITER_PRINTK))
1413                 return 0;
1414
1415         va_start(ap, fmt);
1416         ret = trace_array_vprintk(tr, ip, fmt, ap);
1417         va_end(ap);
1418         return ret;
1419 }
1420
1421 int trace_array_vprintk(struct trace_array *tr,
1422                         unsigned long ip, const char *fmt, va_list args)
1423 {
1424         static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1425         static char trace_buf[TRACE_BUF_SIZE];
1426
1427         struct ftrace_event_call *call = &event_print;
1428         struct ring_buffer_event *event;
1429         struct ring_buffer *buffer;
1430         struct trace_array_cpu *data;
1431         int cpu, len = 0, size, pc;
1432         struct print_entry *entry;
1433         unsigned long irq_flags;
1434         int disable;
1435
1436         if (tracing_disabled || tracing_selftest_running)
1437                 return 0;
1438
1439         pc = preempt_count();
1440         preempt_disable_notrace();
1441         cpu = raw_smp_processor_id();
1442         data = tr->data[cpu];
1443
1444         disable = atomic_inc_return(&data->disabled);
1445         if (unlikely(disable != 1))
1446                 goto out;
1447
1448         pause_graph_tracing();
1449         raw_local_irq_save(irq_flags);
1450         arch_spin_lock(&trace_buf_lock);
1451         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1452
1453         size = sizeof(*entry) + len + 1;
1454         buffer = tr->buffer;
1455         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1456                                           irq_flags, pc);
1457         if (!event)
1458                 goto out_unlock;
1459         entry = ring_buffer_event_data(event);
1460         entry->ip = ip;
1461
1462         memcpy(&entry->buf, trace_buf, len);
1463         entry->buf[len] = '\0';
1464         if (!filter_check_discard(call, entry, buffer, event)) {
1465                 ring_buffer_unlock_commit(buffer, event);
1466                 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1467         }
1468
1469  out_unlock:
1470         arch_spin_unlock(&trace_buf_lock);
1471         raw_local_irq_restore(irq_flags);
1472         unpause_graph_tracing();
1473  out:
1474         atomic_dec_return(&data->disabled);
1475         preempt_enable_notrace();
1476
1477         return len;
1478 }
1479
1480 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1481 {
1482         return trace_array_vprintk(&global_trace, ip, fmt, args);
1483 }
1484 EXPORT_SYMBOL_GPL(trace_vprintk);
1485
1486 enum trace_file_type {
1487         TRACE_FILE_LAT_FMT      = 1,
1488         TRACE_FILE_ANNOTATE     = 2,
1489 };
1490
1491 static void trace_iterator_increment(struct trace_iterator *iter)
1492 {
1493         /* Don't allow ftrace to trace into the ring buffers */
1494         ftrace_disable_cpu();
1495
1496         iter->idx++;
1497         if (iter->buffer_iter[iter->cpu])
1498                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1499
1500         ftrace_enable_cpu();
1501 }
1502
1503 static struct trace_entry *
1504 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1505                 unsigned long *lost_events)
1506 {
1507         struct ring_buffer_event *event;
1508         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1509
1510         /* Don't allow ftrace to trace into the ring buffers */
1511         ftrace_disable_cpu();
1512
1513         if (buf_iter)
1514                 event = ring_buffer_iter_peek(buf_iter, ts);
1515         else
1516                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1517                                          lost_events);
1518
1519         ftrace_enable_cpu();
1520
1521         return event ? ring_buffer_event_data(event) : NULL;
1522 }
1523
1524 static struct trace_entry *
1525 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1526                   unsigned long *missing_events, u64 *ent_ts)
1527 {
1528         struct ring_buffer *buffer = iter->tr->buffer;
1529         struct trace_entry *ent, *next = NULL;
1530         unsigned long lost_events = 0, next_lost = 0;
1531         int cpu_file = iter->cpu_file;
1532         u64 next_ts = 0, ts;
1533         int next_cpu = -1;
1534         int cpu;
1535
1536         /*
1537          * If we are in a per_cpu trace file, don't bother by iterating over
1538          * all cpu and peek directly.
1539          */
1540         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1541                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1542                         return NULL;
1543                 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1544                 if (ent_cpu)
1545                         *ent_cpu = cpu_file;
1546
1547                 return ent;
1548         }
1549
1550         for_each_tracing_cpu(cpu) {
1551
1552                 if (ring_buffer_empty_cpu(buffer, cpu))
1553                         continue;
1554
1555                 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1556
1557                 /*
1558                  * Pick the entry with the smallest timestamp:
1559                  */
1560                 if (ent && (!next || ts < next_ts)) {
1561                         next = ent;
1562                         next_cpu = cpu;
1563                         next_ts = ts;
1564                         next_lost = lost_events;
1565                 }
1566         }
1567
1568         if (ent_cpu)
1569                 *ent_cpu = next_cpu;
1570
1571         if (ent_ts)
1572                 *ent_ts = next_ts;
1573
1574         if (missing_events)
1575                 *missing_events = next_lost;
1576
1577         return next;
1578 }
1579
1580 /* Find the next real entry, without updating the iterator itself */
1581 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1582                                           int *ent_cpu, u64 *ent_ts)
1583 {
1584         return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1585 }
1586
1587 /* Find the next real entry, and increment the iterator to the next entry */
1588 static void *find_next_entry_inc(struct trace_iterator *iter)
1589 {
1590         iter->ent = __find_next_entry(iter, &iter->cpu,
1591                                       &iter->lost_events, &iter->ts);
1592
1593         if (iter->ent)
1594                 trace_iterator_increment(iter);
1595
1596         return iter->ent ? iter : NULL;
1597 }
1598
1599 static void trace_consume(struct trace_iterator *iter)
1600 {
1601         /* Don't allow ftrace to trace into the ring buffers */
1602         ftrace_disable_cpu();
1603         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1604                             &iter->lost_events);
1605         ftrace_enable_cpu();
1606 }
1607
1608 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1609 {
1610         struct trace_iterator *iter = m->private;
1611         int i = (int)*pos;
1612         void *ent;
1613
1614         WARN_ON_ONCE(iter->leftover);
1615
1616         (*pos)++;
1617
1618         /* can't go backwards */
1619         if (iter->idx > i)
1620                 return NULL;
1621
1622         if (iter->idx < 0)
1623                 ent = find_next_entry_inc(iter);
1624         else
1625                 ent = iter;
1626
1627         while (ent && iter->idx < i)
1628                 ent = find_next_entry_inc(iter);
1629
1630         iter->pos = *pos;
1631
1632         return ent;
1633 }
1634
1635 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1636 {
1637         struct trace_array *tr = iter->tr;
1638         struct ring_buffer_event *event;
1639         struct ring_buffer_iter *buf_iter;
1640         unsigned long entries = 0;
1641         u64 ts;
1642
1643         tr->data[cpu]->skipped_entries = 0;
1644
1645         if (!iter->buffer_iter[cpu])
1646                 return;
1647
1648         buf_iter = iter->buffer_iter[cpu];
1649         ring_buffer_iter_reset(buf_iter);
1650
1651         /*
1652          * We could have the case with the max latency tracers
1653          * that a reset never took place on a cpu. This is evident
1654          * by the timestamp being before the start of the buffer.
1655          */
1656         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1657                 if (ts >= iter->tr->time_start)
1658                         break;
1659                 entries++;
1660                 ring_buffer_read(buf_iter, NULL);
1661         }
1662
1663         tr->data[cpu]->skipped_entries = entries;
1664 }
1665
1666 /*
1667  * The current tracer is copied to avoid a global locking
1668  * all around.
1669  */
1670 static void *s_start(struct seq_file *m, loff_t *pos)
1671 {
1672         struct trace_iterator *iter = m->private;
1673         static struct tracer *old_tracer;
1674         int cpu_file = iter->cpu_file;
1675         void *p = NULL;
1676         loff_t l = 0;
1677         int cpu;
1678
1679         /* copy the tracer to avoid using a global lock all around */
1680         mutex_lock(&trace_types_lock);
1681         if (unlikely(old_tracer != current_trace && current_trace)) {
1682                 old_tracer = current_trace;
1683                 *iter->trace = *current_trace;
1684         }
1685         mutex_unlock(&trace_types_lock);
1686
1687         atomic_inc(&trace_record_cmdline_disabled);
1688
1689         if (*pos != iter->pos) {
1690                 iter->ent = NULL;
1691                 iter->cpu = 0;
1692                 iter->idx = -1;
1693
1694                 ftrace_disable_cpu();
1695
1696                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1697                         for_each_tracing_cpu(cpu)
1698                                 tracing_iter_reset(iter, cpu);
1699                 } else
1700                         tracing_iter_reset(iter, cpu_file);
1701
1702                 ftrace_enable_cpu();
1703
1704                 iter->leftover = 0;
1705                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1706                         ;
1707
1708         } else {
1709                 /*
1710                  * If we overflowed the seq_file before, then we want
1711                  * to just reuse the trace_seq buffer again.
1712                  */
1713                 if (iter->leftover)
1714                         p = iter;
1715                 else {
1716                         l = *pos - 1;
1717                         p = s_next(m, p, &l);
1718                 }
1719         }
1720
1721         trace_event_read_lock();
1722         trace_access_lock(cpu_file);
1723         return p;
1724 }
1725
1726 static void s_stop(struct seq_file *m, void *p)
1727 {
1728         struct trace_iterator *iter = m->private;
1729
1730         atomic_dec(&trace_record_cmdline_disabled);
1731         trace_access_unlock(iter->cpu_file);
1732         trace_event_read_unlock();
1733 }
1734
1735 static void print_lat_help_header(struct seq_file *m)
1736 {
1737         seq_puts(m, "#                  _------=> CPU#            \n");
1738         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1739         seq_puts(m, "#                | / _----=> need-resched    \n");
1740         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1741         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1742         seq_puts(m, "#                |||| /_--=> lock-depth       \n");
1743         seq_puts(m, "#                |||||/     delay             \n");
1744         seq_puts(m, "#  cmd     pid   |||||| time  |   caller      \n");
1745         seq_puts(m, "#     \\   /      ||||||   \\   |   /           \n");
1746 }
1747
1748 static void print_func_help_header(struct seq_file *m)
1749 {
1750         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1751         seq_puts(m, "#              | |       |          |         |\n");
1752 }
1753
1754
1755 void
1756 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1757 {
1758         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1759         struct trace_array *tr = iter->tr;
1760         struct trace_array_cpu *data = tr->data[tr->cpu];
1761         struct tracer *type = current_trace;
1762         unsigned long entries = 0;
1763         unsigned long total = 0;
1764         unsigned long count;
1765         const char *name = "preemption";
1766         int cpu;
1767
1768         if (type)
1769                 name = type->name;
1770
1771
1772         for_each_tracing_cpu(cpu) {
1773                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1774                 /*
1775                  * If this buffer has skipped entries, then we hold all
1776                  * entries for the trace and we need to ignore the
1777                  * ones before the time stamp.
1778                  */
1779                 if (tr->data[cpu]->skipped_entries) {
1780                         count -= tr->data[cpu]->skipped_entries;
1781                         /* total is the same as the entries */
1782                         total += count;
1783                 } else
1784                         total += count +
1785                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1786                 entries += count;
1787         }
1788
1789         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1790                    name, UTS_RELEASE);
1791         seq_puts(m, "# -----------------------------------"
1792                  "---------------------------------\n");
1793         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1794                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1795                    nsecs_to_usecs(data->saved_latency),
1796                    entries,
1797                    total,
1798                    tr->cpu,
1799 #if defined(CONFIG_PREEMPT_NONE)
1800                    "server",
1801 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1802                    "desktop",
1803 #elif defined(CONFIG_PREEMPT)
1804                    "preempt",
1805 #else
1806                    "unknown",
1807 #endif
1808                    /* These are reserved for later use */
1809                    0, 0, 0, 0);
1810 #ifdef CONFIG_SMP
1811         seq_printf(m, " #P:%d)\n", num_online_cpus());
1812 #else
1813         seq_puts(m, ")\n");
1814 #endif
1815         seq_puts(m, "#    -----------------\n");
1816         seq_printf(m, "#    | task: %.16s-%d "
1817                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1818                    data->comm, data->pid, data->uid, data->nice,
1819                    data->policy, data->rt_priority);
1820         seq_puts(m, "#    -----------------\n");
1821
1822         if (data->critical_start) {
1823                 seq_puts(m, "#  => started at: ");
1824                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1825                 trace_print_seq(m, &iter->seq);
1826                 seq_puts(m, "\n#  => ended at:   ");
1827                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1828                 trace_print_seq(m, &iter->seq);
1829                 seq_puts(m, "\n#\n");
1830         }
1831
1832         seq_puts(m, "#\n");
1833 }
1834
1835 static void test_cpu_buff_start(struct trace_iterator *iter)
1836 {
1837         struct trace_seq *s = &iter->seq;
1838
1839         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1840                 return;
1841
1842         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1843                 return;
1844
1845         if (cpumask_test_cpu(iter->cpu, iter->started))
1846                 return;
1847
1848         if (iter->tr->data[iter->cpu]->skipped_entries)
1849                 return;
1850
1851         cpumask_set_cpu(iter->cpu, iter->started);
1852
1853         /* Don't print started cpu buffer for the first entry of the trace */
1854         if (iter->idx > 1)
1855                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1856                                 iter->cpu);
1857 }
1858
1859 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1860 {
1861         struct trace_seq *s = &iter->seq;
1862         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1863         struct trace_entry *entry;
1864         struct trace_event *event;
1865
1866         entry = iter->ent;
1867
1868         test_cpu_buff_start(iter);
1869
1870         event = ftrace_find_event(entry->type);
1871
1872         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1873                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1874                         if (!trace_print_lat_context(iter))
1875                                 goto partial;
1876                 } else {
1877                         if (!trace_print_context(iter))
1878                                 goto partial;
1879                 }
1880         }
1881
1882         if (event)
1883                 return event->funcs->trace(iter, sym_flags, event);
1884
1885         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1886                 goto partial;
1887
1888         return TRACE_TYPE_HANDLED;
1889 partial:
1890         return TRACE_TYPE_PARTIAL_LINE;
1891 }
1892
1893 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1894 {
1895         struct trace_seq *s = &iter->seq;
1896         struct trace_entry *entry;
1897         struct trace_event *event;
1898
1899         entry = iter->ent;
1900
1901         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1902                 if (!trace_seq_printf(s, "%d %d %llu ",
1903                                       entry->pid, iter->cpu, iter->ts))
1904                         goto partial;
1905         }
1906
1907         event = ftrace_find_event(entry->type);
1908         if (event)
1909                 return event->funcs->raw(iter, 0, event);
1910
1911         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1912                 goto partial;
1913
1914         return TRACE_TYPE_HANDLED;
1915 partial:
1916         return TRACE_TYPE_PARTIAL_LINE;
1917 }
1918
1919 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1920 {
1921         struct trace_seq *s = &iter->seq;
1922         unsigned char newline = '\n';
1923         struct trace_entry *entry;
1924         struct trace_event *event;
1925
1926         entry = iter->ent;
1927
1928         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1929                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1930                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1931                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1932         }
1933
1934         event = ftrace_find_event(entry->type);
1935         if (event) {
1936                 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1937                 if (ret != TRACE_TYPE_HANDLED)
1938                         return ret;
1939         }
1940
1941         SEQ_PUT_FIELD_RET(s, newline);
1942
1943         return TRACE_TYPE_HANDLED;
1944 }
1945
1946 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1947 {
1948         struct trace_seq *s = &iter->seq;
1949         struct trace_entry *entry;
1950         struct trace_event *event;
1951
1952         entry = iter->ent;
1953
1954         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1955                 SEQ_PUT_FIELD_RET(s, entry->pid);
1956                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1957                 SEQ_PUT_FIELD_RET(s, iter->ts);
1958         }
1959
1960         event = ftrace_find_event(entry->type);
1961         return event ? event->funcs->binary(iter, 0, event) :
1962                 TRACE_TYPE_HANDLED;
1963 }
1964
1965 int trace_empty(struct trace_iterator *iter)
1966 {
1967         int cpu;
1968
1969         /* If we are looking at one CPU buffer, only check that one */
1970         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1971                 cpu = iter->cpu_file;
1972                 if (iter->buffer_iter[cpu]) {
1973                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1974                                 return 0;
1975                 } else {
1976                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1977                                 return 0;
1978                 }
1979                 return 1;
1980         }
1981
1982         for_each_tracing_cpu(cpu) {
1983                 if (iter->buffer_iter[cpu]) {
1984                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1985                                 return 0;
1986                 } else {
1987                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1988                                 return 0;
1989                 }
1990         }
1991
1992         return 1;
1993 }
1994
1995 /*  Called with trace_event_read_lock() held. */
1996 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1997 {
1998         enum print_line_t ret;
1999
2000         if (iter->lost_events)
2001                 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2002                                  iter->cpu, iter->lost_events);
2003
2004         if (iter->trace && iter->trace->print_line) {
2005                 ret = iter->trace->print_line(iter);
2006                 if (ret != TRACE_TYPE_UNHANDLED)
2007                         return ret;
2008         }
2009
2010         if (iter->ent->type == TRACE_BPRINT &&
2011                         trace_flags & TRACE_ITER_PRINTK &&
2012                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2013                 return trace_print_bprintk_msg_only(iter);
2014
2015         if (iter->ent->type == TRACE_PRINT &&
2016                         trace_flags & TRACE_ITER_PRINTK &&
2017                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2018                 return trace_print_printk_msg_only(iter);
2019
2020         if (trace_flags & TRACE_ITER_BIN)
2021                 return print_bin_fmt(iter);
2022
2023         if (trace_flags & TRACE_ITER_HEX)
2024                 return print_hex_fmt(iter);
2025
2026         if (trace_flags & TRACE_ITER_RAW)
2027                 return print_raw_fmt(iter);
2028
2029         return print_trace_fmt(iter);
2030 }
2031
2032 void trace_default_header(struct seq_file *m)
2033 {
2034         struct trace_iterator *iter = m->private;
2035
2036         if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2037                 /* print nothing if the buffers are empty */
2038                 if (trace_empty(iter))
2039                         return;
2040                 print_trace_header(m, iter);
2041                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2042                         print_lat_help_header(m);
2043         } else {
2044                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2045                         print_func_help_header(m);
2046         }
2047 }
2048
2049 static int s_show(struct seq_file *m, void *v)
2050 {
2051         struct trace_iterator *iter = v;
2052         int ret;
2053
2054         if (iter->ent == NULL) {
2055                 if (iter->tr) {
2056                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2057                         seq_puts(m, "#\n");
2058                 }
2059                 if (iter->trace && iter->trace->print_header)
2060                         iter->trace->print_header(m);
2061                 else
2062                         trace_default_header(m);
2063
2064         } else if (iter->leftover) {
2065                 /*
2066                  * If we filled the seq_file buffer earlier, we
2067                  * want to just show it now.
2068                  */
2069                 ret = trace_print_seq(m, &iter->seq);
2070
2071                 /* ret should this time be zero, but you never know */
2072                 iter->leftover = ret;
2073
2074         } else {
2075                 print_trace_line(iter);
2076                 ret = trace_print_seq(m, &iter->seq);
2077                 /*
2078                  * If we overflow the seq_file buffer, then it will
2079                  * ask us for this data again at start up.
2080                  * Use that instead.
2081                  *  ret is 0 if seq_file write succeeded.
2082                  *        -1 otherwise.
2083                  */
2084                 iter->leftover = ret;
2085         }
2086
2087         return 0;
2088 }
2089
2090 static const struct seq_operations tracer_seq_ops = {
2091         .start          = s_start,
2092         .next           = s_next,
2093         .stop           = s_stop,
2094         .show           = s_show,
2095 };
2096
2097 static struct trace_iterator *
2098 __tracing_open(struct inode *inode, struct file *file)
2099 {
2100         long cpu_file = (long) inode->i_private;
2101         void *fail_ret = ERR_PTR(-ENOMEM);
2102         struct trace_iterator *iter;
2103         struct seq_file *m;
2104         int cpu, ret;
2105
2106         if (tracing_disabled)
2107                 return ERR_PTR(-ENODEV);
2108
2109         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2110         if (!iter)
2111                 return ERR_PTR(-ENOMEM);
2112
2113         /*
2114          * We make a copy of the current tracer to avoid concurrent
2115          * changes on it while we are reading.
2116          */
2117         mutex_lock(&trace_types_lock);
2118         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2119         if (!iter->trace)
2120                 goto fail;
2121
2122         if (current_trace)
2123                 *iter->trace = *current_trace;
2124
2125         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2126                 goto fail;
2127
2128         if (current_trace && current_trace->print_max)
2129                 iter->tr = &max_tr;
2130         else
2131                 iter->tr = &global_trace;
2132         iter->pos = -1;
2133         mutex_init(&iter->mutex);
2134         iter->cpu_file = cpu_file;
2135
2136         /* Notify the tracer early; before we stop tracing. */
2137         if (iter->trace && iter->trace->open)
2138                 iter->trace->open(iter);
2139
2140         /* Annotate start of buffers if we had overruns */
2141         if (ring_buffer_overruns(iter->tr->buffer))
2142                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2143
2144         /* stop the trace while dumping */
2145         tracing_stop();
2146
2147         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2148                 for_each_tracing_cpu(cpu) {
2149                         iter->buffer_iter[cpu] =
2150                                 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2151                 }
2152                 ring_buffer_read_prepare_sync();
2153                 for_each_tracing_cpu(cpu) {
2154                         ring_buffer_read_start(iter->buffer_iter[cpu]);
2155                         tracing_iter_reset(iter, cpu);
2156                 }
2157         } else {
2158                 cpu = iter->cpu_file;
2159                 iter->buffer_iter[cpu] =
2160                         ring_buffer_read_prepare(iter->tr->buffer, cpu);
2161                 ring_buffer_read_prepare_sync();
2162                 ring_buffer_read_start(iter->buffer_iter[cpu]);
2163                 tracing_iter_reset(iter, cpu);
2164         }
2165
2166         ret = seq_open(file, &tracer_seq_ops);
2167         if (ret < 0) {
2168                 fail_ret = ERR_PTR(ret);
2169                 goto fail_buffer;
2170         }
2171
2172         m = file->private_data;
2173         m->private = iter;
2174
2175         mutex_unlock(&trace_types_lock);
2176
2177         return iter;
2178
2179  fail_buffer:
2180         for_each_tracing_cpu(cpu) {
2181                 if (iter->buffer_iter[cpu])
2182                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2183         }
2184         free_cpumask_var(iter->started);
2185         tracing_start();
2186  fail:
2187         mutex_unlock(&trace_types_lock);
2188         kfree(iter->trace);
2189         kfree(iter);
2190
2191         return fail_ret;
2192 }
2193
2194 int tracing_open_generic(struct inode *inode, struct file *filp)
2195 {
2196         if (tracing_disabled)
2197                 return -ENODEV;
2198
2199         filp->private_data = inode->i_private;
2200         return 0;
2201 }
2202
2203 static int tracing_release(struct inode *inode, struct file *file)
2204 {
2205         struct seq_file *m = (struct seq_file *)file->private_data;
2206         struct trace_iterator *iter;
2207         int cpu;
2208
2209         if (!(file->f_mode & FMODE_READ))
2210                 return 0;
2211
2212         iter = m->private;
2213
2214         mutex_lock(&trace_types_lock);
2215         for_each_tracing_cpu(cpu) {
2216                 if (iter->buffer_iter[cpu])
2217                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2218         }
2219
2220         if (iter->trace && iter->trace->close)
2221                 iter->trace->close(iter);
2222
2223         /* reenable tracing if it was previously enabled */
2224         tracing_start();
2225         mutex_unlock(&trace_types_lock);
2226
2227         seq_release(inode, file);
2228         mutex_destroy(&iter->mutex);
2229         free_cpumask_var(iter->started);
2230         kfree(iter->trace);
2231         kfree(iter);
2232         return 0;
2233 }
2234
2235 static int tracing_open(struct inode *inode, struct file *file)
2236 {
2237         struct trace_iterator *iter;
2238         int ret = 0;
2239
2240         /* If this file was open for write, then erase contents */
2241         if ((file->f_mode & FMODE_WRITE) &&
2242             (file->f_flags & O_TRUNC)) {
2243                 long cpu = (long) inode->i_private;
2244
2245                 if (cpu == TRACE_PIPE_ALL_CPU)
2246                         tracing_reset_online_cpus(&global_trace);
2247                 else
2248                         tracing_reset(&global_trace, cpu);
2249         }
2250
2251         if (file->f_mode & FMODE_READ) {
2252                 iter = __tracing_open(inode, file);
2253                 if (IS_ERR(iter))
2254                         ret = PTR_ERR(iter);
2255                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2256                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2257         }
2258         return ret;
2259 }
2260
2261 static void *
2262 t_next(struct seq_file *m, void *v, loff_t *pos)
2263 {
2264         struct tracer *t = v;
2265
2266         (*pos)++;
2267
2268         if (t)
2269                 t = t->next;
2270
2271         return t;
2272 }
2273
2274 static void *t_start(struct seq_file *m, loff_t *pos)
2275 {
2276         struct tracer *t;
2277         loff_t l = 0;
2278
2279         mutex_lock(&trace_types_lock);
2280         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2281                 ;
2282
2283         return t;
2284 }
2285
2286 static void t_stop(struct seq_file *m, void *p)
2287 {
2288         mutex_unlock(&trace_types_lock);
2289 }
2290
2291 static int t_show(struct seq_file *m, void *v)
2292 {
2293         struct tracer *t = v;
2294
2295         if (!t)
2296                 return 0;
2297
2298         seq_printf(m, "%s", t->name);
2299         if (t->next)
2300                 seq_putc(m, ' ');
2301         else
2302                 seq_putc(m, '\n');
2303
2304         return 0;
2305 }
2306
2307 static const struct seq_operations show_traces_seq_ops = {
2308         .start          = t_start,
2309         .next           = t_next,
2310         .stop           = t_stop,
2311         .show           = t_show,
2312 };
2313
2314 static int show_traces_open(struct inode *inode, struct file *file)
2315 {
2316         if (tracing_disabled)
2317                 return -ENODEV;
2318
2319         return seq_open(file, &show_traces_seq_ops);
2320 }
2321
2322 static ssize_t
2323 tracing_write_stub(struct file *filp, const char __user *ubuf,
2324                    size_t count, loff_t *ppos)
2325 {
2326         return count;
2327 }
2328
2329 static const struct file_operations tracing_fops = {
2330         .open           = tracing_open,
2331         .read           = seq_read,
2332         .write          = tracing_write_stub,
2333         .llseek         = seq_lseek,
2334         .release        = tracing_release,
2335 };
2336
2337 static const struct file_operations show_traces_fops = {
2338         .open           = show_traces_open,
2339         .read           = seq_read,
2340         .release        = seq_release,
2341 };
2342
2343 /*
2344  * Only trace on a CPU if the bitmask is set:
2345  */
2346 static cpumask_var_t tracing_cpumask;
2347
2348 /*
2349  * The tracer itself will not take this lock, but still we want
2350  * to provide a consistent cpumask to user-space:
2351  */
2352 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2353
2354 /*
2355  * Temporary storage for the character representation of the
2356  * CPU bitmask (and one more byte for the newline):
2357  */
2358 static char mask_str[NR_CPUS + 1];
2359
2360 static ssize_t
2361 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2362                      size_t count, loff_t *ppos)
2363 {
2364         int len;
2365
2366         mutex_lock(&tracing_cpumask_update_lock);
2367
2368         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2369         if (count - len < 2) {
2370                 count = -EINVAL;
2371                 goto out_err;
2372         }
2373         len += sprintf(mask_str + len, "\n");
2374         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2375
2376 out_err:
2377         mutex_unlock(&tracing_cpumask_update_lock);
2378
2379         return count;
2380 }
2381
2382 static ssize_t
2383 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2384                       size_t count, loff_t *ppos)
2385 {
2386         int err, cpu;
2387         cpumask_var_t tracing_cpumask_new;
2388
2389         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2390                 return -ENOMEM;
2391
2392         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2393         if (err)
2394                 goto err_unlock;
2395
2396         mutex_lock(&tracing_cpumask_update_lock);
2397
2398         local_irq_disable();
2399         arch_spin_lock(&ftrace_max_lock);
2400         for_each_tracing_cpu(cpu) {
2401                 /*
2402                  * Increase/decrease the disabled counter if we are
2403                  * about to flip a bit in the cpumask:
2404                  */
2405                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2406                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2407                         atomic_inc(&global_trace.data[cpu]->disabled);
2408                 }
2409                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2410                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2411                         atomic_dec(&global_trace.data[cpu]->disabled);
2412                 }
2413         }
2414         arch_spin_unlock(&ftrace_max_lock);
2415         local_irq_enable();
2416
2417         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2418
2419         mutex_unlock(&tracing_cpumask_update_lock);
2420         free_cpumask_var(tracing_cpumask_new);
2421
2422         return count;
2423
2424 err_unlock:
2425         free_cpumask_var(tracing_cpumask_new);
2426
2427         return err;
2428 }
2429
2430 static const struct file_operations tracing_cpumask_fops = {
2431         .open           = tracing_open_generic,
2432         .read           = tracing_cpumask_read,
2433         .write          = tracing_cpumask_write,
2434 };
2435
2436 static int tracing_trace_options_show(struct seq_file *m, void *v)
2437 {
2438         struct tracer_opt *trace_opts;
2439         u32 tracer_flags;
2440         int i;
2441
2442         mutex_lock(&trace_types_lock);
2443         tracer_flags = current_trace->flags->val;
2444         trace_opts = current_trace->flags->opts;
2445
2446         for (i = 0; trace_options[i]; i++) {
2447                 if (trace_flags & (1 << i))
2448                         seq_printf(m, "%s\n", trace_options[i]);
2449                 else
2450                         seq_printf(m, "no%s\n", trace_options[i]);
2451         }
2452
2453         for (i = 0; trace_opts[i].name; i++) {
2454                 if (tracer_flags & trace_opts[i].bit)
2455                         seq_printf(m, "%s\n", trace_opts[i].name);
2456                 else
2457                         seq_printf(m, "no%s\n", trace_opts[i].name);
2458         }
2459         mutex_unlock(&trace_types_lock);
2460
2461         return 0;
2462 }
2463
2464 static int __set_tracer_option(struct tracer *trace,
2465                                struct tracer_flags *tracer_flags,
2466                                struct tracer_opt *opts, int neg)
2467 {
2468         int ret;
2469
2470         ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2471         if (ret)
2472                 return ret;
2473
2474         if (neg)
2475                 tracer_flags->val &= ~opts->bit;
2476         else
2477                 tracer_flags->val |= opts->bit;
2478         return 0;
2479 }
2480
2481 /* Try to assign a tracer specific option */
2482 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2483 {
2484         struct tracer_flags *tracer_flags = trace->flags;
2485         struct tracer_opt *opts = NULL;
2486         int i;
2487
2488         for (i = 0; tracer_flags->opts[i].name; i++) {
2489                 opts = &tracer_flags->opts[i];
2490
2491                 if (strcmp(cmp, opts->name) == 0)
2492                         return __set_tracer_option(trace, trace->flags,
2493                                                    opts, neg);
2494         }
2495
2496         return -EINVAL;
2497 }
2498
2499 static void set_tracer_flags(unsigned int mask, int enabled)
2500 {
2501         /* do nothing if flag is already set */
2502         if (!!(trace_flags & mask) == !!enabled)
2503                 return;
2504
2505         if (enabled)
2506                 trace_flags |= mask;
2507         else
2508                 trace_flags &= ~mask;
2509 }
2510
2511 static ssize_t
2512 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2513                         size_t cnt, loff_t *ppos)
2514 {
2515         char buf[64];
2516         char *cmp;
2517         int neg = 0;
2518         int ret;
2519         int i;
2520
2521         if (cnt >= sizeof(buf))
2522                 return -EINVAL;
2523
2524         if (copy_from_user(&buf, ubuf, cnt))
2525                 return -EFAULT;
2526
2527         buf[cnt] = 0;
2528         cmp = strstrip(buf);
2529
2530         if (strncmp(cmp, "no", 2) == 0) {
2531                 neg = 1;
2532                 cmp += 2;
2533         }
2534
2535         for (i = 0; trace_options[i]; i++) {
2536                 if (strcmp(cmp, trace_options[i]) == 0) {
2537                         set_tracer_flags(1 << i, !neg);
2538                         break;
2539                 }
2540         }
2541
2542         /* If no option could be set, test the specific tracer options */
2543         if (!trace_options[i]) {
2544                 mutex_lock(&trace_types_lock);
2545                 ret = set_tracer_option(current_trace, cmp, neg);
2546                 mutex_unlock(&trace_types_lock);
2547                 if (ret)
2548                         return ret;
2549         }
2550
2551         *ppos += cnt;
2552
2553         return cnt;
2554 }
2555
2556 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2557 {
2558         if (tracing_disabled)
2559                 return -ENODEV;
2560         return single_open(file, tracing_trace_options_show, NULL);
2561 }
2562
2563 static const struct file_operations tracing_iter_fops = {
2564         .open           = tracing_trace_options_open,
2565         .read           = seq_read,
2566         .llseek         = seq_lseek,
2567         .release        = single_release,
2568         .write          = tracing_trace_options_write,
2569 };
2570
2571 static const char readme_msg[] =
2572         "tracing mini-HOWTO:\n\n"
2573         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2574         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2575         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2576         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2577         "nop\n"
2578         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2579         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2580         "sched_switch\n"
2581         "# cat /sys/kernel/debug/tracing/trace_options\n"
2582         "noprint-parent nosym-offset nosym-addr noverbose\n"
2583         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2584         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2585         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2586         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2587 ;
2588
2589 static ssize_t
2590 tracing_readme_read(struct file *filp, char __user *ubuf,
2591                        size_t cnt, loff_t *ppos)
2592 {
2593         return simple_read_from_buffer(ubuf, cnt, ppos,
2594                                         readme_msg, strlen(readme_msg));
2595 }
2596
2597 static const struct file_operations tracing_readme_fops = {
2598         .open           = tracing_open_generic,
2599         .read           = tracing_readme_read,
2600 };
2601
2602 static ssize_t
2603 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2604                                 size_t cnt, loff_t *ppos)
2605 {
2606         char *buf_comm;
2607         char *file_buf;
2608         char *buf;
2609         int len = 0;
2610         int pid;
2611         int i;
2612
2613         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2614         if (!file_buf)
2615                 return -ENOMEM;
2616
2617         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2618         if (!buf_comm) {
2619                 kfree(file_buf);
2620                 return -ENOMEM;
2621         }
2622
2623         buf = file_buf;
2624
2625         for (i = 0; i < SAVED_CMDLINES; i++) {
2626                 int r;
2627
2628                 pid = map_cmdline_to_pid[i];
2629                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2630                         continue;
2631
2632                 trace_find_cmdline(pid, buf_comm);
2633                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2634                 buf += r;
2635                 len += r;
2636         }
2637
2638         len = simple_read_from_buffer(ubuf, cnt, ppos,
2639                                       file_buf, len);
2640
2641         kfree(file_buf);
2642         kfree(buf_comm);
2643
2644         return len;
2645 }
2646
2647 static const struct file_operations tracing_saved_cmdlines_fops = {
2648     .open       = tracing_open_generic,
2649     .read       = tracing_saved_cmdlines_read,
2650 };
2651
2652 static ssize_t
2653 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2654                   size_t cnt, loff_t *ppos)
2655 {
2656         char buf[64];
2657         int r;
2658
2659         r = sprintf(buf, "%u\n", tracer_enabled);
2660         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2661 }
2662
2663 static ssize_t
2664 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2665                    size_t cnt, loff_t *ppos)
2666 {
2667         struct trace_array *tr = filp->private_data;
2668         char buf[64];
2669         unsigned long val;
2670         int ret;
2671
2672         if (cnt >= sizeof(buf))
2673                 return -EINVAL;
2674
2675         if (copy_from_user(&buf, ubuf, cnt))
2676                 return -EFAULT;
2677
2678         buf[cnt] = 0;
2679
2680         ret = strict_strtoul(buf, 10, &val);
2681         if (ret < 0)
2682                 return ret;
2683
2684         val = !!val;
2685
2686         mutex_lock(&trace_types_lock);
2687         if (tracer_enabled ^ val) {
2688                 if (val) {
2689                         tracer_enabled = 1;
2690                         if (current_trace->start)
2691                                 current_trace->start(tr);
2692                         tracing_start();
2693                 } else {
2694                         tracer_enabled = 0;
2695                         tracing_stop();
2696                         if (current_trace->stop)
2697                                 current_trace->stop(tr);
2698                 }
2699         }
2700         mutex_unlock(&trace_types_lock);
2701
2702         *ppos += cnt;
2703
2704         return cnt;
2705 }
2706
2707 static ssize_t
2708 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2709                        size_t cnt, loff_t *ppos)
2710 {
2711         char buf[MAX_TRACER_SIZE+2];
2712         int r;
2713
2714         mutex_lock(&trace_types_lock);
2715         if (current_trace)
2716                 r = sprintf(buf, "%s\n", current_trace->name);
2717         else
2718                 r = sprintf(buf, "\n");
2719         mutex_unlock(&trace_types_lock);
2720
2721         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2722 }
2723
2724 int tracer_init(struct tracer *t, struct trace_array *tr)
2725 {
2726         tracing_reset_online_cpus(tr);
2727         return t->init(tr);
2728 }
2729
2730 static int tracing_resize_ring_buffer(unsigned long size)
2731 {
2732         int ret;
2733
2734         /*
2735          * If kernel or user changes the size of the ring buffer
2736          * we use the size that was given, and we can forget about
2737          * expanding it later.
2738          */
2739         ring_buffer_expanded = 1;
2740
2741         ret = ring_buffer_resize(global_trace.buffer, size);
2742         if (ret < 0)
2743                 return ret;
2744
2745         ret = ring_buffer_resize(max_tr.buffer, size);
2746         if (ret < 0) {
2747                 int r;
2748
2749                 r = ring_buffer_resize(global_trace.buffer,
2750                                        global_trace.entries);
2751                 if (r < 0) {
2752                         /*
2753                          * AARGH! We are left with different
2754                          * size max buffer!!!!
2755                          * The max buffer is our "snapshot" buffer.
2756                          * When a tracer needs a snapshot (one of the
2757                          * latency tracers), it swaps the max buffer
2758                          * with the saved snap shot. We succeeded to
2759                          * update the size of the main buffer, but failed to
2760                          * update the size of the max buffer. But when we tried
2761                          * to reset the main buffer to the original size, we
2762                          * failed there too. This is very unlikely to
2763                          * happen, but if it does, warn and kill all
2764                          * tracing.
2765                          */
2766                         WARN_ON(1);
2767                         tracing_disabled = 1;
2768                 }
2769                 return ret;
2770         }
2771
2772         global_trace.entries = size;
2773
2774         return ret;
2775 }
2776
2777 /**
2778  * tracing_update_buffers - used by tracing facility to expand ring buffers
2779  *
2780  * To save on memory when the tracing is never used on a system with it
2781  * configured in. The ring buffers are set to a minimum size. But once
2782  * a user starts to use the tracing facility, then they need to grow
2783  * to their default size.
2784  *
2785  * This function is to be called when a tracer is about to be used.
2786  */
2787 int tracing_update_buffers(void)
2788 {
2789         int ret = 0;
2790
2791         mutex_lock(&trace_types_lock);
2792         if (!ring_buffer_expanded)
2793                 ret = tracing_resize_ring_buffer(trace_buf_size);
2794         mutex_unlock(&trace_types_lock);
2795
2796         return ret;
2797 }
2798
2799 struct trace_option_dentry;
2800
2801 static struct trace_option_dentry *
2802 create_trace_option_files(struct tracer *tracer);
2803
2804 static void
2805 destroy_trace_option_files(struct trace_option_dentry *topts);
2806
2807 static int tracing_set_tracer(const char *buf)
2808 {
2809         static struct trace_option_dentry *topts;
2810         struct trace_array *tr = &global_trace;
2811         struct tracer *t;
2812         int ret = 0;
2813
2814         mutex_lock(&trace_types_lock);
2815
2816         if (!ring_buffer_expanded) {
2817                 ret = tracing_resize_ring_buffer(trace_buf_size);
2818                 if (ret < 0)
2819                         goto out;
2820                 ret = 0;
2821         }
2822
2823         for (t = trace_types; t; t = t->next) {
2824                 if (strcmp(t->name, buf) == 0)
2825                         break;
2826         }
2827         if (!t) {
2828                 ret = -EINVAL;
2829                 goto out;
2830         }
2831         if (t == current_trace)
2832                 goto out;
2833
2834         trace_branch_disable();
2835         if (current_trace && current_trace->reset)
2836                 current_trace->reset(tr);
2837
2838         destroy_trace_option_files(topts);
2839
2840         current_trace = t;
2841
2842         topts = create_trace_option_files(current_trace);
2843
2844         if (t->init) {
2845                 ret = tracer_init(t, tr);
2846                 if (ret)
2847                         goto out;
2848         }
2849
2850         trace_branch_enable(tr);
2851  out:
2852         mutex_unlock(&trace_types_lock);
2853
2854         return ret;
2855 }
2856
2857 static ssize_t
2858 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2859                         size_t cnt, loff_t *ppos)
2860 {
2861         char buf[MAX_TRACER_SIZE+1];
2862         int i;
2863         size_t ret;
2864         int err;
2865
2866         ret = cnt;
2867
2868         if (cnt > MAX_TRACER_SIZE)
2869                 cnt = MAX_TRACER_SIZE;
2870
2871         if (copy_from_user(&buf, ubuf, cnt))
2872                 return -EFAULT;
2873
2874         buf[cnt] = 0;
2875
2876         /* strip ending whitespace. */
2877         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2878                 buf[i] = 0;
2879
2880         err = tracing_set_tracer(buf);
2881         if (err)
2882                 return err;
2883
2884         *ppos += ret;
2885
2886         return ret;
2887 }
2888
2889 static ssize_t
2890 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2891                      size_t cnt, loff_t *ppos)
2892 {
2893         unsigned long *ptr = filp->private_data;
2894         char buf[64];
2895         int r;
2896
2897         r = snprintf(buf, sizeof(buf), "%ld\n",
2898                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2899         if (r > sizeof(buf))
2900                 r = sizeof(buf);
2901         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2902 }
2903
2904 static ssize_t
2905 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2906                       size_t cnt, loff_t *ppos)
2907 {
2908         unsigned long *ptr = filp->private_data;
2909         char buf[64];
2910         unsigned long val;
2911         int ret;
2912
2913         if (cnt >= sizeof(buf))
2914                 return -EINVAL;
2915
2916         if (copy_from_user(&buf, ubuf, cnt))
2917                 return -EFAULT;
2918
2919         buf[cnt] = 0;
2920
2921         ret = strict_strtoul(buf, 10, &val);
2922         if (ret < 0)
2923                 return ret;
2924
2925         *ptr = val * 1000;
2926
2927         return cnt;
2928 }
2929
2930 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2931 {
2932         long cpu_file = (long) inode->i_private;
2933         struct trace_iterator *iter;
2934         int ret = 0;
2935
2936         if (tracing_disabled)
2937                 return -ENODEV;
2938
2939         mutex_lock(&trace_types_lock);
2940
2941         /* create a buffer to store the information to pass to userspace */
2942         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2943         if (!iter) {
2944                 ret = -ENOMEM;
2945                 goto out;
2946         }
2947
2948         /*
2949          * We make a copy of the current tracer to avoid concurrent
2950          * changes on it while we are reading.
2951          */
2952         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2953         if (!iter->trace) {
2954                 ret = -ENOMEM;
2955                 goto fail;
2956         }
2957         if (current_trace)
2958                 *iter->trace = *current_trace;
2959
2960         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2961                 ret = -ENOMEM;
2962                 goto fail;
2963         }
2964
2965         /* trace pipe does not show start of buffer */
2966         cpumask_setall(iter->started);
2967
2968         if (trace_flags & TRACE_ITER_LATENCY_FMT)
2969                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2970
2971         iter->cpu_file = cpu_file;
2972         iter->tr = &global_trace;
2973         mutex_init(&iter->mutex);
2974         filp->private_data = iter;
2975
2976         if (iter->trace->pipe_open)
2977                 iter->trace->pipe_open(iter);
2978
2979 out:
2980         mutex_unlock(&trace_types_lock);
2981         return ret;
2982
2983 fail:
2984         kfree(iter->trace);
2985         kfree(iter);
2986         mutex_unlock(&trace_types_lock);
2987         return ret;
2988 }
2989
2990 static int tracing_release_pipe(struct inode *inode, struct file *file)
2991 {
2992         struct trace_iterator *iter = file->private_data;
2993
2994         mutex_lock(&trace_types_lock);
2995
2996         if (iter->trace->pipe_close)
2997                 iter->trace->pipe_close(iter);
2998
2999         mutex_unlock(&trace_types_lock);
3000
3001         free_cpumask_var(iter->started);
3002         mutex_destroy(&iter->mutex);
3003         kfree(iter->trace);
3004         kfree(iter);
3005
3006         return 0;
3007 }
3008
3009 static unsigned int
3010 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3011 {
3012         struct trace_iterator *iter = filp->private_data;
3013
3014         if (trace_flags & TRACE_ITER_BLOCK) {
3015                 /*
3016                  * Always select as readable when in blocking mode
3017                  */
3018                 return POLLIN | POLLRDNORM;
3019         } else {
3020                 if (!trace_empty(iter))
3021                         return POLLIN | POLLRDNORM;
3022                 poll_wait(filp, &trace_wait, poll_table);
3023                 if (!trace_empty(iter))
3024                         return POLLIN | POLLRDNORM;
3025
3026                 return 0;
3027         }
3028 }
3029
3030
3031 void default_wait_pipe(struct trace_iterator *iter)
3032 {
3033         DEFINE_WAIT(wait);
3034
3035         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3036
3037         if (trace_empty(iter))
3038                 schedule();
3039
3040         finish_wait(&trace_wait, &wait);
3041 }
3042
3043 /*
3044  * This is a make-shift waitqueue.
3045  * A tracer might use this callback on some rare cases:
3046  *
3047  *  1) the current tracer might hold the runqueue lock when it wakes up
3048  *     a reader, hence a deadlock (sched, function, and function graph tracers)
3049  *  2) the function tracers, trace all functions, we don't want
3050  *     the overhead of calling wake_up and friends
3051  *     (and tracing them too)
3052  *
3053  *     Anyway, this is really very primitive wakeup.
3054  */
3055 void poll_wait_pipe(struct trace_iterator *iter)
3056 {
3057         set_current_state(TASK_INTERRUPTIBLE);
3058         /* sleep for 100 msecs, and try again. */
3059         schedule_timeout(HZ / 10);
3060 }
3061
3062 /* Must be called with trace_types_lock mutex held. */
3063 static int tracing_wait_pipe(struct file *filp)
3064 {
3065         struct trace_iterator *iter = filp->private_data;
3066
3067         while (trace_empty(iter)) {
3068
3069                 if ((filp->f_flags & O_NONBLOCK)) {
3070                         return -EAGAIN;
3071                 }
3072
3073                 mutex_unlock(&iter->mutex);
3074
3075                 iter->trace->wait_pipe(iter);
3076
3077                 mutex_lock(&iter->mutex);
3078
3079                 if (signal_pending(current))
3080                         return -EINTR;
3081
3082                 /*
3083                  * We block until we read something and tracing is disabled.
3084                  * We still block if tracing is disabled, but we have never
3085                  * read anything. This allows a user to cat this file, and
3086                  * then enable tracing. But after we have read something,
3087                  * we give an EOF when tracing is again disabled.
3088                  *
3089                  * iter->pos will be 0 if we haven't read anything.
3090                  */
3091                 if (!tracer_enabled && iter->pos)
3092                         break;
3093         }
3094
3095         return 1;
3096 }
3097
3098 /*
3099  * Consumer reader.
3100  */
3101 static ssize_t
3102 tracing_read_pipe(struct file *filp, char __user *ubuf,
3103                   size_t cnt, loff_t *ppos)
3104 {
3105         struct trace_iterator *iter = filp->private_data;
3106         static struct tracer *old_tracer;
3107         ssize_t sret;
3108
3109         /* return any leftover data */
3110         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3111         if (sret != -EBUSY)
3112                 return sret;
3113
3114         trace_seq_init(&iter->seq);
3115
3116         /* copy the tracer to avoid using a global lock all around */
3117         mutex_lock(&trace_types_lock);
3118         if (unlikely(old_tracer != current_trace && current_trace)) {
3119                 old_tracer = current_trace;
3120                 *iter->trace = *current_trace;
3121         }
3122         mutex_unlock(&trace_types_lock);
3123
3124         /*
3125          * Avoid more than one consumer on a single file descriptor
3126          * This is just a matter of traces coherency, the ring buffer itself
3127          * is protected.
3128          */
3129         mutex_lock(&iter->mutex);
3130         if (iter->trace->read) {
3131                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3132                 if (sret)
3133                         goto out;
3134         }
3135
3136 waitagain:
3137         sret = tracing_wait_pipe(filp);
3138         if (sret <= 0)
3139                 goto out;
3140
3141         /* stop when tracing is finished */
3142         if (trace_empty(iter)) {
3143                 sret = 0;
3144                 goto out;
3145         }
3146
3147         if (cnt >= PAGE_SIZE)
3148                 cnt = PAGE_SIZE - 1;
3149
3150         /* reset all but tr, trace, and overruns */
3151         memset(&iter->seq, 0,
3152                sizeof(struct trace_iterator) -
3153                offsetof(struct trace_iterator, seq));
3154         iter->pos = -1;
3155
3156         trace_event_read_lock();
3157         trace_access_lock(iter->cpu_file);
3158         while (find_next_entry_inc(iter) != NULL) {
3159                 enum print_line_t ret;
3160                 int len = iter->seq.len;
3161
3162                 ret = print_trace_line(iter);
3163                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3164                         /* don't print partial lines */
3165                         iter->seq.len = len;
3166                         break;
3167                 }
3168                 if (ret != TRACE_TYPE_NO_CONSUME)
3169                         trace_consume(iter);
3170
3171                 if (iter->seq.len >= cnt)
3172                         break;
3173         }
3174         trace_access_unlock(iter->cpu_file);
3175         trace_event_read_unlock();
3176
3177         /* Now copy what we have to the user */
3178         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3179         if (iter->seq.readpos >= iter->seq.len)
3180                 trace_seq_init(&iter->seq);
3181
3182         /*
3183          * If there was nothing to send to user, inspite of consuming trace
3184          * entries, go back to wait for more entries.
3185          */
3186         if (sret == -EBUSY)
3187                 goto waitagain;
3188
3189 out:
3190         mutex_unlock(&iter->mutex);
3191
3192         return sret;
3193 }
3194
3195 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3196                                      struct pipe_buffer *buf)
3197 {
3198         __free_page(buf->page);
3199 }
3200
3201 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3202                                      unsigned int idx)
3203 {
3204         __free_page(spd->pages[idx]);
3205 }
3206
3207 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3208         .can_merge              = 0,
3209         .map                    = generic_pipe_buf_map,
3210         .unmap                  = generic_pipe_buf_unmap,
3211         .confirm                = generic_pipe_buf_confirm,
3212         .release                = tracing_pipe_buf_release,
3213         .steal                  = generic_pipe_buf_steal,
3214         .get                    = generic_pipe_buf_get,
3215 };
3216
3217 static size_t
3218 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3219 {
3220         size_t count;
3221         int ret;
3222
3223         /* Seq buffer is page-sized, exactly what we need. */
3224         for (;;) {
3225                 count = iter->seq.len;
3226                 ret = print_trace_line(iter);
3227                 count = iter->seq.len - count;
3228                 if (rem < count) {
3229                         rem = 0;
3230                         iter->seq.len -= count;
3231                         break;
3232                 }
3233                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3234                         iter->seq.len -= count;
3235                         break;
3236                 }
3237
3238                 if (ret != TRACE_TYPE_NO_CONSUME)
3239                         trace_consume(iter);
3240                 rem -= count;
3241                 if (!find_next_entry_inc(iter)) {
3242                         rem = 0;
3243                         iter->ent = NULL;
3244                         break;
3245                 }
3246         }
3247
3248         return rem;
3249 }
3250
3251 static ssize_t tracing_splice_read_pipe(struct file *filp,
3252                                         loff_t *ppos,
3253                                         struct pipe_inode_info *pipe,
3254                                         size_t len,
3255                                         unsigned int flags)
3256 {
3257         struct page *pages_def[PIPE_DEF_BUFFERS];
3258         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3259         struct trace_iterator *iter = filp->private_data;
3260         struct splice_pipe_desc spd = {
3261                 .pages          = pages_def,
3262                 .partial        = partial_def,
3263                 .nr_pages       = 0, /* This gets updated below. */
3264                 .flags          = flags,
3265                 .ops            = &tracing_pipe_buf_ops,
3266                 .spd_release    = tracing_spd_release_pipe,
3267         };
3268         static struct tracer *old_tracer;
3269         ssize_t ret;
3270         size_t rem;
3271         unsigned int i;
3272
3273         if (splice_grow_spd(pipe, &spd))
3274                 return -ENOMEM;
3275
3276         /* copy the tracer to avoid using a global lock all around */
3277         mutex_lock(&trace_types_lock);
3278         if (unlikely(old_tracer != current_trace && current_trace)) {
3279                 old_tracer = current_trace;
3280                 *iter->trace = *current_trace;
3281         }
3282         mutex_unlock(&trace_types_lock);
3283
3284         mutex_lock(&iter->mutex);
3285
3286         if (iter->trace->splice_read) {
3287                 ret = iter->trace->splice_read(iter, filp,
3288                                                ppos, pipe, len, flags);
3289                 if (ret)
3290                         goto out_err;
3291         }
3292
3293         ret = tracing_wait_pipe(filp);
3294         if (ret <= 0)
3295                 goto out_err;
3296
3297         if (!iter->ent && !find_next_entry_inc(iter)) {
3298                 ret = -EFAULT;
3299                 goto out_err;
3300         }
3301
3302         trace_event_read_lock();
3303         trace_access_lock(iter->cpu_file);
3304
3305         /* Fill as many pages as possible. */
3306         for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3307                 spd.pages[i] = alloc_page(GFP_KERNEL);
3308                 if (!spd.pages[i])
3309                         break;
3310
3311                 rem = tracing_fill_pipe_page(rem, iter);
3312
3313                 /* Copy the data into the page, so we can start over. */
3314                 ret = trace_seq_to_buffer(&iter->seq,
3315                                           page_address(spd.pages[i]),
3316                                           iter->seq.len);
3317                 if (ret < 0) {
3318                         __free_page(spd.pages[i]);
3319                         break;
3320                 }
3321                 spd.partial[i].offset = 0;
3322                 spd.partial[i].len = iter->seq.len;
3323
3324                 trace_seq_init(&iter->seq);
3325         }
3326
3327         trace_access_unlock(iter->cpu_file);
3328         trace_event_read_unlock();
3329         mutex_unlock(&iter->mutex);
3330
3331         spd.nr_pages = i;
3332
3333         ret = splice_to_pipe(pipe, &spd);
3334 out:
3335         splice_shrink_spd(pipe, &spd);
3336         return ret;
3337
3338 out_err:
3339         mutex_unlock(&iter->mutex);
3340         goto out;
3341 }
3342
3343 static ssize_t
3344 tracing_entries_read(struct file *filp, char __user *ubuf,
3345                      size_t cnt, loff_t *ppos)
3346 {
3347         struct trace_array *tr = filp->private_data;
3348         char buf[96];
3349         int r;
3350
3351         mutex_lock(&trace_types_lock);
3352         if (!ring_buffer_expanded)
3353                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3354                             tr->entries >> 10,
3355                             trace_buf_size >> 10);
3356         else
3357                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3358         mutex_unlock(&trace_types_lock);
3359
3360         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3361 }
3362
3363 static ssize_t
3364 tracing_entries_write(struct file *filp, const char __user *ubuf,
3365                       size_t cnt, loff_t *ppos)
3366 {
3367         unsigned long val;
3368         char buf[64];
3369         int ret, cpu;
3370
3371         if (cnt >= sizeof(buf))
3372                 return -EINVAL;
3373
3374         if (copy_from_user(&buf, ubuf, cnt))
3375                 return -EFAULT;
3376
3377         buf[cnt] = 0;
3378
3379         ret = strict_strtoul(buf, 10, &val);
3380         if (ret < 0)
3381                 return ret;
3382
3383         /* must have at least 1 entry */
3384         if (!val)
3385                 return -EINVAL;
3386
3387         mutex_lock(&trace_types_lock);
3388
3389         tracing_stop();
3390
3391         /* disable all cpu buffers */
3392         for_each_tracing_cpu(cpu) {
3393                 if (global_trace.data[cpu])
3394                         atomic_inc(&global_trace.data[cpu]->disabled);
3395                 if (max_tr.data[cpu])
3396                         atomic_inc(&max_tr.data[cpu]->disabled);
3397         }
3398
3399         /* value is in KB */
3400         val <<= 10;
3401
3402         if (val != global_trace.entries) {
3403                 ret = tracing_resize_ring_buffer(val);
3404                 if (ret < 0) {
3405                         cnt = ret;
3406                         goto out;
3407                 }
3408         }
3409
3410         *ppos += cnt;
3411
3412         /* If check pages failed, return ENOMEM */
3413         if (tracing_disabled)
3414                 cnt = -ENOMEM;
3415  out:
3416         for_each_tracing_cpu(cpu) {
3417                 if (global_trace.data[cpu])
3418                         atomic_dec(&global_trace.data[cpu]->disabled);
3419                 if (max_tr.data[cpu])
3420                         atomic_dec(&max_tr.data[cpu]->disabled);
3421         }
3422
3423         tracing_start();
3424         max_tr.entries = global_trace.entries;
3425         mutex_unlock(&trace_types_lock);
3426
3427         return cnt;
3428 }
3429
3430 static int mark_printk(const char *fmt, ...)
3431 {
3432         int ret;
3433         va_list args;
3434         va_start(args, fmt);
3435         ret = trace_vprintk(0, fmt, args);
3436         va_end(args);
3437         return ret;
3438 }
3439
3440 static ssize_t
3441 tracing_mark_write(struct file *filp, const char __user *ubuf,
3442                                         size_t cnt, loff_t *fpos)
3443 {
3444         char *buf;
3445
3446         if (tracing_disabled)
3447                 return -EINVAL;
3448
3449         if (cnt > TRACE_BUF_SIZE)
3450                 cnt = TRACE_BUF_SIZE;
3451
3452         buf = kmalloc(cnt + 2, GFP_KERNEL);
3453         if (buf == NULL)
3454                 return -ENOMEM;
3455
3456         if (copy_from_user(buf, ubuf, cnt)) {
3457                 kfree(buf);
3458                 return -EFAULT;
3459         }
3460         if (buf[cnt-1] != '\n') {
3461                 buf[cnt] = '\n';
3462                 buf[cnt+1] = '\0';
3463         } else
3464                 buf[cnt] = '\0';
3465
3466         cnt = mark_printk("%s", buf);
3467         kfree(buf);
3468         *fpos += cnt;
3469
3470         return cnt;
3471 }
3472
3473 static int tracing_clock_show(struct seq_file *m, void *v)
3474 {
3475         int i;
3476
3477         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3478                 seq_printf(m,
3479                         "%s%s%s%s", i ? " " : "",
3480                         i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3481                         i == trace_clock_id ? "]" : "");
3482         seq_putc(m, '\n');
3483
3484         return 0;
3485 }
3486
3487 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3488                                    size_t cnt, loff_t *fpos)
3489 {
3490         char buf[64];
3491         const char *clockstr;
3492         int i;
3493
3494         if (cnt >= sizeof(buf))
3495                 return -EINVAL;
3496
3497         if (copy_from_user(&buf, ubuf, cnt))
3498                 return -EFAULT;
3499
3500         buf[cnt] = 0;
3501
3502         clockstr = strstrip(buf);
3503
3504         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3505                 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3506                         break;
3507         }
3508         if (i == ARRAY_SIZE(trace_clocks))
3509                 return -EINVAL;
3510
3511         trace_clock_id = i;
3512
3513         mutex_lock(&trace_types_lock);
3514
3515         ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3516         if (max_tr.buffer)
3517                 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3518
3519         mutex_unlock(&trace_types_lock);
3520
3521         *fpos += cnt;
3522
3523         return cnt;
3524 }
3525
3526 static int tracing_clock_open(struct inode *inode, struct file *file)
3527 {
3528         if (tracing_disabled)
3529                 return -ENODEV;
3530         return single_open(file, tracing_clock_show, NULL);
3531 }
3532
3533 static const struct file_operations tracing_max_lat_fops = {
3534         .open           = tracing_open_generic,
3535         .read           = tracing_max_lat_read,
3536         .write          = tracing_max_lat_write,
3537 };
3538
3539 static const struct file_operations tracing_ctrl_fops = {
3540         .open           = tracing_open_generic,
3541         .read           = tracing_ctrl_read,
3542         .write          = tracing_ctrl_write,
3543 };
3544
3545 static const struct file_operations set_tracer_fops = {
3546         .open           = tracing_open_generic,
3547         .read           = tracing_set_trace_read,
3548         .write          = tracing_set_trace_write,
3549 };
3550
3551 static const struct file_operations tracing_pipe_fops = {
3552         .open           = tracing_open_pipe,
3553         .poll           = tracing_poll_pipe,
3554         .read           = tracing_read_pipe,
3555         .splice_read    = tracing_splice_read_pipe,
3556         .release        = tracing_release_pipe,
3557 };
3558
3559 static const struct file_operations tracing_entries_fops = {
3560         .open           = tracing_open_generic,
3561         .read           = tracing_entries_read,
3562         .write          = tracing_entries_write,
3563 };
3564
3565 static const struct file_operations tracing_mark_fops = {
3566         .open           = tracing_open_generic,
3567         .write          = tracing_mark_write,
3568 };
3569
3570 static const struct file_operations trace_clock_fops = {
3571         .open           = tracing_clock_open,
3572         .read           = seq_read,
3573         .llseek         = seq_lseek,
3574         .release        = single_release,
3575         .write          = tracing_clock_write,
3576 };
3577
3578 struct ftrace_buffer_info {
3579         struct trace_array      *tr;
3580         void                    *spare;
3581         int                     cpu;
3582         unsigned int            read;
3583 };
3584
3585 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3586 {
3587         int cpu = (int)(long)inode->i_private;
3588         struct ftrace_buffer_info *info;
3589
3590         if (tracing_disabled)
3591                 return -ENODEV;
3592
3593         info = kzalloc(sizeof(*info), GFP_KERNEL);
3594         if (!info)
3595                 return -ENOMEM;
3596
3597         info->tr        = &global_trace;
3598         info->cpu       = cpu;
3599         info->spare     = NULL;
3600         /* Force reading ring buffer for first read */
3601         info->read      = (unsigned int)-1;
3602
3603         filp->private_data = info;
3604
3605         return nonseekable_open(inode, filp);
3606 }
3607
3608 static ssize_t
3609 tracing_buffers_read(struct file *filp, char __user *ubuf,
3610                      size_t count, loff_t *ppos)
3611 {
3612         struct ftrace_buffer_info *info = filp->private_data;
3613         ssize_t ret;
3614         size_t size;
3615
3616         if (!count)
3617                 return 0;
3618
3619         if (!info->spare)
3620                 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3621         if (!info->spare)
3622                 return -ENOMEM;
3623
3624         /* Do we have previous read data to read? */
3625         if (info->read < PAGE_SIZE)
3626                 goto read;
3627
3628         info->read = 0;
3629
3630         trace_access_lock(info->cpu);
3631         ret = ring_buffer_read_page(info->tr->buffer,
3632                                     &info->spare,
3633                                     count,
3634                                     info->cpu, 0);
3635         trace_access_unlock(info->cpu);
3636         if (ret < 0)
3637                 return 0;
3638
3639 read:
3640         size = PAGE_SIZE - info->read;
3641         if (size > count)
3642                 size = count;
3643
3644         ret = copy_to_user(ubuf, info->spare + info->read, size);
3645         if (ret == size)
3646                 return -EFAULT;
3647         size -= ret;
3648
3649         *ppos += size;
3650         info->read += size;
3651
3652         return size;
3653 }
3654
3655 static int tracing_buffers_release(struct inode *inode, struct file *file)
3656 {
3657         struct ftrace_buffer_info *info = file->private_data;
3658
3659         if (info->spare)
3660                 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3661         kfree(info);
3662
3663         return 0;
3664 }
3665
3666 struct buffer_ref {
3667         struct ring_buffer      *buffer;
3668         void                    *page;
3669         int                     ref;
3670 };
3671
3672 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3673                                     struct pipe_buffer *buf)
3674 {
3675         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3676
3677         if (--ref->ref)
3678                 return;
3679
3680         ring_buffer_free_read_page(ref->buffer, ref->page);
3681         kfree(ref);
3682         buf->private = 0;
3683 }
3684
3685 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3686                                  struct pipe_buffer *buf)
3687 {
3688         return 1;
3689 }
3690
3691 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3692                                 struct pipe_buffer *buf)
3693 {
3694         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3695
3696         ref->ref++;
3697 }
3698
3699 /* Pipe buffer operations for a buffer. */
3700 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3701         .can_merge              = 0,
3702         .map                    = generic_pipe_buf_map,
3703         .unmap                  = generic_pipe_buf_unmap,
3704         .confirm                = generic_pipe_buf_confirm,
3705         .release                = buffer_pipe_buf_release,
3706         .steal                  = buffer_pipe_buf_steal,
3707         .get                    = buffer_pipe_buf_get,
3708 };
3709
3710 /*
3711  * Callback from splice_to_pipe(), if we need to release some pages
3712  * at the end of the spd in case we error'ed out in filling the pipe.
3713  */
3714 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3715 {
3716         struct buffer_ref *ref =
3717                 (struct buffer_ref *)spd->partial[i].private;
3718
3719         if (--ref->ref)
3720                 return;
3721
3722         ring_buffer_free_read_page(ref->buffer, ref->page);
3723         kfree(ref);
3724         spd->partial[i].private = 0;
3725 }
3726
3727 static ssize_t
3728 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3729                             struct pipe_inode_info *pipe, size_t len,
3730                             unsigned int flags)
3731 {
3732         struct ftrace_buffer_info *info = file->private_data;
3733         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3734         struct page *pages_def[PIPE_DEF_BUFFERS];
3735         struct splice_pipe_desc spd = {
3736                 .pages          = pages_def,
3737                 .partial        = partial_def,
3738                 .flags          = flags,
3739                 .ops            = &buffer_pipe_buf_ops,
3740                 .spd_release    = buffer_spd_release,
3741         };
3742         struct buffer_ref *ref;
3743         int entries, size, i;
3744         size_t ret;
3745
3746         if (splice_grow_spd(pipe, &spd))
3747                 return -ENOMEM;
3748
3749         if (*ppos & (PAGE_SIZE - 1)) {
3750                 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3751                 ret = -EINVAL;
3752                 goto out;
3753         }
3754
3755         if (len & (PAGE_SIZE - 1)) {
3756                 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3757                 if (len < PAGE_SIZE) {
3758                         ret = -EINVAL;
3759                         goto out;
3760                 }
3761                 len &= PAGE_MASK;
3762         }
3763
3764         trace_access_lock(info->cpu);
3765         entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3766
3767         for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3768                 struct page *page;
3769                 int r;
3770
3771                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3772                 if (!ref)
3773                         break;
3774
3775                 ref->ref = 1;
3776                 ref->buffer = info->tr->buffer;
3777                 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3778                 if (!ref->page) {
3779                         kfree(ref);
3780                         break;
3781                 }
3782
3783                 r = ring_buffer_read_page(ref->buffer, &ref->page,
3784                                           len, info->cpu, 1);
3785                 if (r < 0) {
3786                         ring_buffer_free_read_page(ref->buffer,
3787                                                    ref->page);
3788                         kfree(ref);
3789                         break;
3790                 }
3791
3792                 /*
3793                  * zero out any left over data, this is going to
3794                  * user land.
3795                  */
3796                 size = ring_buffer_page_len(ref->page);
3797                 if (size < PAGE_SIZE)
3798                         memset(ref->page + size, 0, PAGE_SIZE - size);
3799
3800                 page = virt_to_page(ref->page);
3801
3802                 spd.pages[i] = page;
3803                 spd.partial[i].len = PAGE_SIZE;
3804                 spd.partial[i].offset = 0;
3805                 spd.partial[i].private = (unsigned long)ref;
3806                 spd.nr_pages++;
3807                 *ppos += PAGE_SIZE;
3808
3809                 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3810         }
3811
3812         trace_access_unlock(info->cpu);
3813         spd.nr_pages = i;
3814
3815         /* did we read anything? */
3816         if (!spd.nr_pages) {
3817                 if (flags & SPLICE_F_NONBLOCK)
3818                         ret = -EAGAIN;
3819                 else
3820                         ret = 0;
3821                 /* TODO: block */
3822                 goto out;
3823         }
3824
3825         ret = splice_to_pipe(pipe, &spd);
3826         splice_shrink_spd(pipe, &spd);
3827 out:
3828         return ret;
3829 }
3830
3831 static const struct file_operations tracing_buffers_fops = {
3832         .open           = tracing_buffers_open,
3833         .read           = tracing_buffers_read,
3834         .release        = tracing_buffers_release,
3835         .splice_read    = tracing_buffers_splice_read,
3836         .llseek         = no_llseek,
3837 };
3838
3839 static ssize_t
3840 tracing_stats_read(struct file *filp, char __user *ubuf,
3841                    size_t count, loff_t *ppos)
3842 {
3843         unsigned long cpu = (unsigned long)filp->private_data;
3844         struct trace_array *tr = &global_trace;
3845         struct trace_seq *s;
3846         unsigned long cnt;
3847
3848         s = kmalloc(sizeof(*s), GFP_KERNEL);
3849         if (!s)
3850                 return -ENOMEM;
3851
3852         trace_seq_init(s);
3853
3854         cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3855         trace_seq_printf(s, "entries: %ld\n", cnt);
3856
3857         cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3858         trace_seq_printf(s, "overrun: %ld\n", cnt);
3859
3860         cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3861         trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3862
3863         count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3864
3865         kfree(s);
3866
3867         return count;
3868 }
3869
3870 static const struct file_operations tracing_stats_fops = {
3871         .open           = tracing_open_generic,
3872         .read           = tracing_stats_read,
3873 };
3874
3875 #ifdef CONFIG_DYNAMIC_FTRACE
3876
3877 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3878 {
3879         return 0;
3880 }
3881
3882 static ssize_t
3883 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3884                   size_t cnt, loff_t *ppos)
3885 {
3886         static char ftrace_dyn_info_buffer[1024];
3887         static DEFINE_MUTEX(dyn_info_mutex);
3888         unsigned long *p = filp->private_data;
3889         char *buf = ftrace_dyn_info_buffer;
3890         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3891         int r;
3892
3893         mutex_lock(&dyn_info_mutex);
3894         r = sprintf(buf, "%ld ", *p);
3895
3896         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3897         buf[r++] = '\n';
3898
3899         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3900
3901         mutex_unlock(&dyn_info_mutex);
3902
3903         return r;
3904 }
3905
3906 static const struct file_operations tracing_dyn_info_fops = {
3907         .open           = tracing_open_generic,
3908         .read           = tracing_read_dyn_info,
3909 };
3910 #endif
3911
3912 static struct dentry *d_tracer;
3913
3914 struct dentry *tracing_init_dentry(void)
3915 {
3916         static int once;
3917
3918         if (d_tracer)
3919                 return d_tracer;
3920
3921         if (!debugfs_initialized())
3922                 return NULL;
3923
3924         d_tracer = debugfs_create_dir("tracing", NULL);
3925
3926         if (!d_tracer && !once) {
3927                 once = 1;
3928                 pr_warning("Could not create debugfs directory 'tracing'\n");
3929                 return NULL;
3930         }
3931
3932         return d_tracer;
3933 }
3934
3935 static struct dentry *d_percpu;
3936
3937 struct dentry *tracing_dentry_percpu(void)
3938 {
3939         static int once;
3940         struct dentry *d_tracer;
3941
3942         if (d_percpu)
3943                 return d_percpu;
3944
3945         d_tracer = tracing_init_dentry();
3946
3947         if (!d_tracer)
3948                 return NULL;
3949
3950         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3951
3952         if (!d_percpu && !once) {
3953                 once = 1;
3954                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3955                 return NULL;
3956         }
3957
3958         return d_percpu;
3959 }
3960
3961 static void tracing_init_debugfs_percpu(long cpu)
3962 {
3963         struct dentry *d_percpu = tracing_dentry_percpu();
3964         struct dentry *d_cpu;
3965         /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3966         char cpu_dir[7];
3967
3968         if (cpu > 999 || cpu < 0)
3969                 return;
3970
3971         sprintf(cpu_dir, "cpu%ld", cpu);
3972         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3973         if (!d_cpu) {
3974                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3975                 return;
3976         }
3977
3978         /* per cpu trace_pipe */
3979         trace_create_file("trace_pipe", 0444, d_cpu,
3980                         (void *) cpu, &tracing_pipe_fops);
3981
3982         /* per cpu trace */
3983         trace_create_file("trace", 0644, d_cpu,
3984                         (void *) cpu, &tracing_fops);
3985
3986         trace_create_file("trace_pipe_raw", 0444, d_cpu,
3987                         (void *) cpu, &tracing_buffers_fops);
3988
3989         trace_create_file("stats", 0444, d_cpu,
3990                         (void *) cpu, &tracing_stats_fops);
3991 }
3992
3993 #ifdef CONFIG_FTRACE_SELFTEST
3994 /* Let selftest have access to static functions in this file */
3995 #include "trace_selftest.c"
3996 #endif
3997
3998 struct trace_option_dentry {
3999         struct tracer_opt               *opt;
4000         struct tracer_flags             *flags;
4001         struct dentry                   *entry;
4002 };
4003
4004 static ssize_t
4005 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4006                         loff_t *ppos)
4007 {
4008         struct trace_option_dentry *topt = filp->private_data;
4009         char *buf;
4010
4011         if (topt->flags->val & topt->opt->bit)
4012                 buf = "1\n";
4013         else
4014                 buf = "0\n";
4015
4016         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4017 }
4018
4019 static ssize_t
4020 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4021                          loff_t *ppos)
4022 {
4023         struct trace_option_dentry *topt = filp->private_data;
4024         unsigned long val;
4025         char buf[64];
4026         int ret;
4027
4028         if (cnt >= sizeof(buf))
4029                 return -EINVAL;
4030
4031         if (copy_from_user(&buf, ubuf, cnt))
4032                 return -EFAULT;
4033
4034         buf[cnt] = 0;
4035
4036         ret = strict_strtoul(buf, 10, &val);
4037         if (ret < 0)
4038                 return ret;
4039
4040         if (val != 0 && val != 1)
4041                 return -EINVAL;
4042
4043         if (!!(topt->flags->val & topt->opt->bit) != val) {
4044                 mutex_lock(&trace_types_lock);
4045                 ret = __set_tracer_option(current_trace, topt->flags,
4046                                           topt->opt, !val);
4047                 mutex_unlock(&trace_types_lock);
4048                 if (ret)
4049                         return ret;
4050         }
4051
4052         *ppos += cnt;
4053
4054         return cnt;
4055 }
4056
4057
4058 static const struct file_operations trace_options_fops = {
4059         .open = tracing_open_generic,
4060         .read = trace_options_read,
4061         .write = trace_options_write,
4062 };
4063
4064 static ssize_t
4065 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4066                         loff_t *ppos)
4067 {
4068         long index = (long)filp->private_data;
4069         char *buf;
4070
4071         if (trace_flags & (1 << index))
4072                 buf = "1\n";
4073         else
4074                 buf = "0\n";
4075
4076         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4077 }
4078
4079 static ssize_t
4080 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4081                          loff_t *ppos)
4082 {
4083         long index = (long)filp->private_data;
4084         char buf[64];
4085         unsigned long val;
4086         int ret;
4087
4088         if (cnt >= sizeof(buf))
4089                 return -EINVAL;
4090
4091         if (copy_from_user(&buf, ubuf, cnt))
4092                 return -EFAULT;
4093
4094         buf[cnt] = 0;
4095
4096         ret = strict_strtoul(buf, 10, &val);
4097         if (ret < 0)
4098                 return ret;
4099
4100         if (val != 0 && val != 1)
4101                 return -EINVAL;
4102         set_tracer_flags(1 << index, val);
4103
4104         *ppos += cnt;
4105
4106         return cnt;
4107 }
4108
4109 static const struct file_operations trace_options_core_fops = {
4110         .open = tracing_open_generic,
4111         .read = trace_options_core_read,
4112         .write = trace_options_core_write,
4113 };
4114
4115 struct dentry *trace_create_file(const char *name,
4116                                  mode_t mode,
4117                                  struct dentry *parent,
4118                                  void *data,
4119                                  const struct file_operations *fops)
4120 {
4121         struct dentry *ret;
4122
4123         ret = debugfs_create_file(name, mode, parent, data, fops);
4124         if (!ret)
4125                 pr_warning("Could not create debugfs '%s' entry\n", name);
4126
4127         return ret;
4128 }
4129
4130
4131 static struct dentry *trace_options_init_dentry(void)
4132 {
4133         struct dentry *d_tracer;
4134         static struct dentry *t_options;
4135
4136         if (t_options)
4137                 return t_options;
4138
4139         d_tracer = tracing_init_dentry();
4140         if (!d_tracer)
4141                 return NULL;
4142
4143         t_options = debugfs_create_dir("options", d_tracer);
4144         if (!t_options) {
4145                 pr_warning("Could not create debugfs directory 'options'\n");
4146                 return NULL;
4147         }
4148
4149         return t_options;
4150 }
4151
4152 static void
4153 create_trace_option_file(struct trace_option_dentry *topt,
4154                          struct tracer_flags *flags,
4155                          struct tracer_opt *opt)
4156 {
4157         struct dentry *t_options;
4158
4159         t_options = trace_options_init_dentry();
4160         if (!t_options)
4161                 return;
4162
4163         topt->flags = flags;
4164         topt->opt = opt;
4165
4166         topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4167                                     &trace_options_fops);
4168
4169 }
4170
4171 static struct trace_option_dentry *
4172 create_trace_option_files(struct tracer *tracer)
4173 {
4174         struct trace_option_dentry *topts;
4175         struct tracer_flags *flags;
4176         struct tracer_opt *opts;
4177         int cnt;
4178
4179         if (!tracer)
4180                 return NULL;
4181
4182         flags = tracer->flags;
4183
4184         if (!flags || !flags->opts)
4185                 return NULL;
4186
4187         opts = flags->opts;
4188
4189         for (cnt = 0; opts[cnt].name; cnt++)
4190                 ;
4191
4192         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4193         if (!topts)
4194                 return NULL;
4195
4196         for (cnt = 0; opts[cnt].name; cnt++)
4197                 create_trace_option_file(&topts[cnt], flags,
4198                                          &opts[cnt]);
4199
4200         return topts;
4201 }
4202
4203 static void
4204 destroy_trace_option_files(struct trace_option_dentry *topts)
4205 {
4206         int cnt;
4207
4208         if (!topts)
4209                 return;
4210
4211         for (cnt = 0; topts[cnt].opt; cnt++) {
4212                 if (topts[cnt].entry)
4213                         debugfs_remove(topts[cnt].entry);
4214         }
4215
4216         kfree(topts);
4217 }
4218
4219 static struct dentry *
4220 create_trace_option_core_file(const char *option, long index)
4221 {
4222         struct dentry *t_options;
4223
4224         t_options = trace_options_init_dentry();
4225         if (!t_options)
4226                 return NULL;
4227
4228         return trace_create_file(option, 0644, t_options, (void *)index,
4229                                     &trace_options_core_fops);
4230 }
4231
4232 static __init void create_trace_options_dir(void)
4233 {
4234         struct dentry *t_options;
4235         int i;
4236
4237         t_options = trace_options_init_dentry();
4238         if (!t_options)
4239                 return;
4240
4241         for (i = 0; trace_options[i]; i++)
4242                 create_trace_option_core_file(trace_options[i], i);
4243 }
4244
4245 static __init int tracer_init_debugfs(void)
4246 {
4247         struct dentry *d_tracer;
4248         int cpu;
4249
4250         trace_access_lock_init();
4251
4252         d_tracer = tracing_init_dentry();
4253
4254         trace_create_file("tracing_enabled", 0644, d_tracer,
4255                         &global_trace, &tracing_ctrl_fops);
4256
4257         trace_create_file("trace_options", 0644, d_tracer,
4258                         NULL, &tracing_iter_fops);
4259
4260         trace_create_file("tracing_cpumask", 0644, d_tracer,
4261                         NULL, &tracing_cpumask_fops);
4262
4263         trace_create_file("trace", 0644, d_tracer,
4264                         (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4265
4266         trace_create_file("available_tracers", 0444, d_tracer,
4267                         &global_trace, &show_traces_fops);
4268
4269         trace_create_file("current_tracer", 0644, d_tracer,
4270                         &global_trace, &set_tracer_fops);
4271
4272 #ifdef CONFIG_TRACER_MAX_TRACE
4273         trace_create_file("tracing_max_latency", 0644, d_tracer,
4274                         &tracing_max_latency, &tracing_max_lat_fops);
4275 #endif
4276
4277         trace_create_file("tracing_thresh", 0644, d_tracer,
4278                         &tracing_thresh, &tracing_max_lat_fops);
4279
4280         trace_create_file("README", 0444, d_tracer,
4281                         NULL, &tracing_readme_fops);
4282
4283         trace_create_file("trace_pipe", 0444, d_tracer,
4284                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4285
4286         trace_create_file("buffer_size_kb", 0644, d_tracer,
4287                         &global_trace, &tracing_entries_fops);
4288
4289         trace_create_file("trace_marker", 0220, d_tracer,
4290                         NULL, &tracing_mark_fops);
4291
4292         trace_create_file("saved_cmdlines", 0444, d_tracer,
4293                         NULL, &tracing_saved_cmdlines_fops);
4294
4295         trace_create_file("trace_clock", 0644, d_tracer, NULL,
4296                           &trace_clock_fops);
4297
4298 #ifdef CONFIG_DYNAMIC_FTRACE
4299         trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4300                         &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4301 #endif
4302
4303         create_trace_options_dir();
4304
4305         for_each_tracing_cpu(cpu)
4306                 tracing_init_debugfs_percpu(cpu);
4307
4308         return 0;
4309 }
4310
4311 static int trace_panic_handler(struct notifier_block *this,
4312                                unsigned long event, void *unused)
4313 {
4314         if (ftrace_dump_on_oops)
4315                 ftrace_dump(ftrace_dump_on_oops);
4316         return NOTIFY_OK;
4317 }
4318
4319 static struct notifier_block trace_panic_notifier = {
4320         .notifier_call  = trace_panic_handler,
4321         .next           = NULL,
4322         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
4323 };
4324
4325 static int trace_die_handler(struct notifier_block *self,
4326                              unsigned long val,
4327                              void *data)
4328 {
4329         switch (val) {
4330         case DIE_OOPS:
4331                 if (ftrace_dump_on_oops)
4332                         ftrace_dump(ftrace_dump_on_oops);
4333                 break;
4334         default:
4335                 break;
4336         }
4337         return NOTIFY_OK;
4338 }
4339
4340 static struct notifier_block trace_die_notifier = {
4341         .notifier_call = trace_die_handler,
4342         .priority = 200
4343 };
4344
4345 /*
4346  * printk is set to max of 1024, we really don't need it that big.
4347  * Nothing should be printing 1000 characters anyway.
4348  */
4349 #define TRACE_MAX_PRINT         1000
4350
4351 /*
4352  * Define here KERN_TRACE so that we have one place to modify
4353  * it if we decide to change what log level the ftrace dump
4354  * should be at.
4355  */
4356 #define KERN_TRACE              KERN_EMERG
4357
4358 static void
4359 trace_printk_seq(struct trace_seq *s)
4360 {
4361         /* Probably should print a warning here. */
4362         if (s->len >= 1000)
4363                 s->len = 1000;
4364
4365         /* should be zero ended, but we are paranoid. */
4366         s->buffer[s->len] = 0;
4367
4368         printk(KERN_TRACE "%s", s->buffer);
4369
4370         trace_seq_init(s);
4371 }
4372
4373 static void
4374 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4375 {
4376         static arch_spinlock_t ftrace_dump_lock =
4377                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4378         /* use static because iter can be a bit big for the stack */
4379         static struct trace_iterator iter;
4380         unsigned int old_userobj;
4381         static int dump_ran;
4382         unsigned long flags;
4383         int cnt = 0, cpu;
4384
4385         /* only one dump */
4386         local_irq_save(flags);
4387         arch_spin_lock(&ftrace_dump_lock);
4388         if (dump_ran)
4389                 goto out;
4390
4391         dump_ran = 1;
4392
4393         tracing_off();
4394
4395         if (disable_tracing)
4396                 ftrace_kill();
4397
4398         for_each_tracing_cpu(cpu) {
4399                 atomic_inc(&global_trace.data[cpu]->disabled);
4400         }
4401
4402         old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4403
4404         /* don't look at user memory in panic mode */
4405         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4406
4407         /* Simulate the iterator */
4408         iter.tr = &global_trace;
4409         iter.trace = current_trace;
4410
4411         switch (oops_dump_mode) {
4412         case DUMP_ALL:
4413                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4414                 break;
4415         case DUMP_ORIG:
4416                 iter.cpu_file = raw_smp_processor_id();
4417                 break;
4418         case DUMP_NONE:
4419                 goto out_enable;
4420         default:
4421                 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4422                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4423         }
4424
4425         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4426
4427         /*
4428          * We need to stop all tracing on all CPUS to read the
4429          * the next buffer. This is a bit expensive, but is
4430          * not done often. We fill all what we can read,
4431          * and then release the locks again.
4432          */
4433
4434         while (!trace_empty(&iter)) {
4435
4436                 if (!cnt)
4437                         printk(KERN_TRACE "---------------------------------\n");
4438
4439                 cnt++;
4440
4441                 /* reset all but tr, trace, and overruns */
4442                 memset(&iter.seq, 0,
4443                        sizeof(struct trace_iterator) -
4444                        offsetof(struct trace_iterator, seq));
4445                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4446                 iter.pos = -1;
4447
4448                 if (find_next_entry_inc(&iter) != NULL) {
4449                         int ret;
4450
4451                         ret = print_trace_line(&iter);
4452                         if (ret != TRACE_TYPE_NO_CONSUME)
4453                                 trace_consume(&iter);
4454                 }
4455
4456                 trace_printk_seq(&iter.seq);
4457         }
4458
4459         if (!cnt)
4460                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4461         else
4462                 printk(KERN_TRACE "---------------------------------\n");
4463
4464  out_enable:
4465         /* Re-enable tracing if requested */
4466         if (!disable_tracing) {
4467                 trace_flags |= old_userobj;
4468
4469                 for_each_tracing_cpu(cpu) {
4470                         atomic_dec(&global_trace.data[cpu]->disabled);
4471                 }
4472                 tracing_on();
4473         }
4474
4475  out:
4476         arch_spin_unlock(&ftrace_dump_lock);
4477         local_irq_restore(flags);
4478 }
4479
4480 /* By default: disable tracing after the dump */
4481 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4482 {
4483         __ftrace_dump(true, oops_dump_mode);
4484 }
4485
4486 __init static int tracer_alloc_buffers(void)
4487 {
4488         int ring_buf_size;
4489         int i;
4490         int ret = -ENOMEM;
4491
4492         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4493                 goto out;
4494
4495         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4496                 goto out_free_buffer_mask;
4497
4498         /* To save memory, keep the ring buffer size to its minimum */
4499         if (ring_buffer_expanded)
4500                 ring_buf_size = trace_buf_size;
4501         else
4502                 ring_buf_size = 1;
4503
4504         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4505         cpumask_copy(tracing_cpumask, cpu_all_mask);
4506
4507         /* TODO: make the number of buffers hot pluggable with CPUS */
4508         global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4509                                                    TRACE_BUFFER_FLAGS);
4510         if (!global_trace.buffer) {
4511                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4512                 WARN_ON(1);
4513                 goto out_free_cpumask;
4514         }
4515         global_trace.entries = ring_buffer_size(global_trace.buffer);
4516
4517
4518 #ifdef CONFIG_TRACER_MAX_TRACE
4519         max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4520                                              TRACE_BUFFER_FLAGS);
4521         if (!max_tr.buffer) {
4522                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4523                 WARN_ON(1);
4524                 ring_buffer_free(global_trace.buffer);
4525                 goto out_free_cpumask;
4526         }
4527         max_tr.entries = ring_buffer_size(max_tr.buffer);
4528         WARN_ON(max_tr.entries != global_trace.entries);
4529 #endif
4530
4531         /* Allocate the first page for all buffers */
4532         for_each_tracing_cpu(i) {
4533                 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4534                 max_tr.data[i] = &per_cpu(max_tr_data, i);
4535         }
4536
4537         trace_init_cmdlines();
4538
4539         register_tracer(&nop_trace);
4540         current_trace = &nop_trace;
4541         /* All seems OK, enable tracing */
4542         tracing_disabled = 0;
4543
4544         atomic_notifier_chain_register(&panic_notifier_list,
4545                                        &trace_panic_notifier);
4546
4547         register_die_notifier(&trace_die_notifier);
4548
4549         return 0;
4550
4551 out_free_cpumask:
4552         free_cpumask_var(tracing_cpumask);
4553 out_free_buffer_mask:
4554         free_cpumask_var(tracing_buffer_mask);
4555 out:
4556         return ret;
4557 }
4558
4559 __init static int clear_boot_tracer(void)
4560 {
4561         /*
4562          * The default tracer at boot buffer is an init section.
4563          * This function is called in lateinit. If we did not
4564          * find the boot tracer, then clear it out, to prevent
4565          * later registration from accessing the buffer that is
4566          * about to be freed.
4567          */
4568         if (!default_bootup_tracer)
4569                 return 0;
4570
4571         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4572                default_bootup_tracer);
4573         default_bootup_tracer = NULL;
4574
4575         return 0;
4576 }
4577
4578 early_initcall(tracer_alloc_buffers);
4579 fs_initcall(tracer_init_debugfs);
4580 late_initcall(clear_boot_tracer);