Merge branch 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[sfrench/cifs-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 cpumask_var_t __read_mostly     tracing_buffer_mask;
105
106 /*
107  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
108  *
109  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
110  * is set, then ftrace_dump is called. This will output the contents
111  * of the ftrace buffers to the console.  This is very useful for
112  * capturing traces that lead to crashes and outputing it to a
113  * serial console.
114  *
115  * It is default off, but you can enable it with either specifying
116  * "ftrace_dump_on_oops" in the kernel command line, or setting
117  * /proc/sys/kernel/ftrace_dump_on_oops
118  * Set 1 if you want to dump buffers of all CPUs
119  * Set 2 if you want to dump the buffer of the CPU that triggered oops
120  */
121
122 enum ftrace_dump_mode ftrace_dump_on_oops;
123
124 static int tracing_set_tracer(const char *buf);
125
126 #define MAX_TRACER_SIZE         100
127 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
128 static char *default_bootup_tracer;
129
130 static int __init set_cmdline_ftrace(char *str)
131 {
132         strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
133         default_bootup_tracer = bootup_tracer_buf;
134         /* We are using ftrace early, expand it */
135         ring_buffer_expanded = 1;
136         return 1;
137 }
138 __setup("ftrace=", set_cmdline_ftrace);
139
140 static int __init set_ftrace_dump_on_oops(char *str)
141 {
142         if (*str++ != '=' || !*str) {
143                 ftrace_dump_on_oops = DUMP_ALL;
144                 return 1;
145         }
146
147         if (!strcmp("orig_cpu", str)) {
148                 ftrace_dump_on_oops = DUMP_ORIG;
149                 return 1;
150         }
151
152         return 0;
153 }
154 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
155
156 unsigned long long ns2usecs(cycle_t nsec)
157 {
158         nsec += 500;
159         do_div(nsec, 1000);
160         return nsec;
161 }
162
163 /*
164  * The global_trace is the descriptor that holds the tracing
165  * buffers for the live tracing. For each CPU, it contains
166  * a link list of pages that will store trace entries. The
167  * page descriptor of the pages in the memory is used to hold
168  * the link list by linking the lru item in the page descriptor
169  * to each of the pages in the buffer per CPU.
170  *
171  * For each active CPU there is a data field that holds the
172  * pages for the buffer for that CPU. Each CPU has the same number
173  * of pages allocated for its buffer.
174  */
175 static struct trace_array       global_trace;
176
177 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
178
179 int filter_current_check_discard(struct ring_buffer *buffer,
180                                  struct ftrace_event_call *call, void *rec,
181                                  struct ring_buffer_event *event)
182 {
183         return filter_check_discard(call, rec, buffer, event);
184 }
185 EXPORT_SYMBOL_GPL(filter_current_check_discard);
186
187 cycle_t ftrace_now(int cpu)
188 {
189         u64 ts;
190
191         /* Early boot up does not have a buffer yet */
192         if (!global_trace.buffer)
193                 return trace_clock_local();
194
195         ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
196         ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
197
198         return ts;
199 }
200
201 /*
202  * The max_tr is used to snapshot the global_trace when a maximum
203  * latency is reached. Some tracers will use this to store a maximum
204  * trace while it continues examining live traces.
205  *
206  * The buffers for the max_tr are set up the same as the global_trace.
207  * When a snapshot is taken, the link list of the max_tr is swapped
208  * with the link list of the global_trace and the buffers are reset for
209  * the global_trace so the tracing can continue.
210  */
211 static struct trace_array       max_tr;
212
213 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
214
215 /* tracer_enabled is used to toggle activation of a tracer */
216 static int                      tracer_enabled = 1;
217
218 /**
219  * tracing_is_enabled - return tracer_enabled status
220  *
221  * This function is used by other tracers to know the status
222  * of the tracer_enabled flag.  Tracers may use this function
223  * to know if it should enable their features when starting
224  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
225  */
226 int tracing_is_enabled(void)
227 {
228         return tracer_enabled;
229 }
230
231 /*
232  * trace_buf_size is the size in bytes that is allocated
233  * for a buffer. Note, the number of bytes is always rounded
234  * to page size.
235  *
236  * This number is purposely set to a low number of 16384.
237  * If the dump on oops happens, it will be much appreciated
238  * to not have to wait for all that output. Anyway this can be
239  * boot time and run time configurable.
240  */
241 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
242
243 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
244
245 /* trace_types holds a link list of available tracers. */
246 static struct tracer            *trace_types __read_mostly;
247
248 /* current_trace points to the tracer that is currently active */
249 static struct tracer            *current_trace __read_mostly;
250
251 /*
252  * trace_types_lock is used to protect the trace_types list.
253  */
254 static DEFINE_MUTEX(trace_types_lock);
255
256 /*
257  * serialize the access of the ring buffer
258  *
259  * ring buffer serializes readers, but it is low level protection.
260  * The validity of the events (which returns by ring_buffer_peek() ..etc)
261  * are not protected by ring buffer.
262  *
263  * The content of events may become garbage if we allow other process consumes
264  * these events concurrently:
265  *   A) the page of the consumed events may become a normal page
266  *      (not reader page) in ring buffer, and this page will be rewrited
267  *      by events producer.
268  *   B) The page of the consumed events may become a page for splice_read,
269  *      and this page will be returned to system.
270  *
271  * These primitives allow multi process access to different cpu ring buffer
272  * concurrently.
273  *
274  * These primitives don't distinguish read-only and read-consume access.
275  * Multi read-only access are also serialized.
276  */
277
278 #ifdef CONFIG_SMP
279 static DECLARE_RWSEM(all_cpu_access_lock);
280 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
281
282 static inline void trace_access_lock(int cpu)
283 {
284         if (cpu == TRACE_PIPE_ALL_CPU) {
285                 /* gain it for accessing the whole ring buffer. */
286                 down_write(&all_cpu_access_lock);
287         } else {
288                 /* gain it for accessing a cpu ring buffer. */
289
290                 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
291                 down_read(&all_cpu_access_lock);
292
293                 /* Secondly block other access to this @cpu ring buffer. */
294                 mutex_lock(&per_cpu(cpu_access_lock, cpu));
295         }
296 }
297
298 static inline void trace_access_unlock(int cpu)
299 {
300         if (cpu == TRACE_PIPE_ALL_CPU) {
301                 up_write(&all_cpu_access_lock);
302         } else {
303                 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
304                 up_read(&all_cpu_access_lock);
305         }
306 }
307
308 static inline void trace_access_lock_init(void)
309 {
310         int cpu;
311
312         for_each_possible_cpu(cpu)
313                 mutex_init(&per_cpu(cpu_access_lock, cpu));
314 }
315
316 #else
317
318 static DEFINE_MUTEX(access_lock);
319
320 static inline void trace_access_lock(int cpu)
321 {
322         (void)cpu;
323         mutex_lock(&access_lock);
324 }
325
326 static inline void trace_access_unlock(int cpu)
327 {
328         (void)cpu;
329         mutex_unlock(&access_lock);
330 }
331
332 static inline void trace_access_lock_init(void)
333 {
334 }
335
336 #endif
337
338 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
339 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
340
341 /* trace_flags holds trace_options default values */
342 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
343         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
344         TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD;
345
346 static int trace_stop_count;
347 static DEFINE_SPINLOCK(tracing_start_lock);
348
349 /**
350  * trace_wake_up - wake up tasks waiting for trace input
351  *
352  * Simply wakes up any task that is blocked on the trace_wait
353  * queue. These is used with trace_poll for tasks polling the trace.
354  */
355 void trace_wake_up(void)
356 {
357         int cpu;
358
359         if (trace_flags & TRACE_ITER_BLOCK)
360                 return;
361         /*
362          * The runqueue_is_locked() can fail, but this is the best we
363          * have for now:
364          */
365         cpu = get_cpu();
366         if (!runqueue_is_locked(cpu))
367                 wake_up(&trace_wait);
368         put_cpu();
369 }
370
371 static int __init set_buf_size(char *str)
372 {
373         unsigned long buf_size;
374
375         if (!str)
376                 return 0;
377         buf_size = memparse(str, &str);
378         /* nr_entries can not be zero */
379         if (buf_size == 0)
380                 return 0;
381         trace_buf_size = buf_size;
382         return 1;
383 }
384 __setup("trace_buf_size=", set_buf_size);
385
386 static int __init set_tracing_thresh(char *str)
387 {
388         unsigned long threshhold;
389         int ret;
390
391         if (!str)
392                 return 0;
393         ret = strict_strtoul(str, 0, &threshhold);
394         if (ret < 0)
395                 return 0;
396         tracing_thresh = threshhold * 1000;
397         return 1;
398 }
399 __setup("tracing_thresh=", set_tracing_thresh);
400
401 unsigned long nsecs_to_usecs(unsigned long nsecs)
402 {
403         return nsecs / 1000;
404 }
405
406 /* These must match the bit postions in trace_iterator_flags */
407 static const char *trace_options[] = {
408         "print-parent",
409         "sym-offset",
410         "sym-addr",
411         "verbose",
412         "raw",
413         "hex",
414         "bin",
415         "block",
416         "stacktrace",
417         "trace_printk",
418         "ftrace_preempt",
419         "branch",
420         "annotate",
421         "userstacktrace",
422         "sym-userobj",
423         "printk-msg-only",
424         "context-info",
425         "latency-format",
426         "sleep-time",
427         "graph-time",
428         "record-cmd",
429         NULL
430 };
431
432 static struct {
433         u64 (*func)(void);
434         const char *name;
435 } trace_clocks[] = {
436         { trace_clock_local,    "local" },
437         { trace_clock_global,   "global" },
438 };
439
440 int trace_clock_id;
441
442 /*
443  * trace_parser_get_init - gets the buffer for trace parser
444  */
445 int trace_parser_get_init(struct trace_parser *parser, int size)
446 {
447         memset(parser, 0, sizeof(*parser));
448
449         parser->buffer = kmalloc(size, GFP_KERNEL);
450         if (!parser->buffer)
451                 return 1;
452
453         parser->size = size;
454         return 0;
455 }
456
457 /*
458  * trace_parser_put - frees the buffer for trace parser
459  */
460 void trace_parser_put(struct trace_parser *parser)
461 {
462         kfree(parser->buffer);
463 }
464
465 /*
466  * trace_get_user - reads the user input string separated by  space
467  * (matched by isspace(ch))
468  *
469  * For each string found the 'struct trace_parser' is updated,
470  * and the function returns.
471  *
472  * Returns number of bytes read.
473  *
474  * See kernel/trace/trace.h for 'struct trace_parser' details.
475  */
476 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
477         size_t cnt, loff_t *ppos)
478 {
479         char ch;
480         size_t read = 0;
481         ssize_t ret;
482
483         if (!*ppos)
484                 trace_parser_clear(parser);
485
486         ret = get_user(ch, ubuf++);
487         if (ret)
488                 goto out;
489
490         read++;
491         cnt--;
492
493         /*
494          * The parser is not finished with the last write,
495          * continue reading the user input without skipping spaces.
496          */
497         if (!parser->cont) {
498                 /* skip white space */
499                 while (cnt && isspace(ch)) {
500                         ret = get_user(ch, ubuf++);
501                         if (ret)
502                                 goto out;
503                         read++;
504                         cnt--;
505                 }
506
507                 /* only spaces were written */
508                 if (isspace(ch)) {
509                         *ppos += read;
510                         ret = read;
511                         goto out;
512                 }
513
514                 parser->idx = 0;
515         }
516
517         /* read the non-space input */
518         while (cnt && !isspace(ch)) {
519                 if (parser->idx < parser->size - 1)
520                         parser->buffer[parser->idx++] = ch;
521                 else {
522                         ret = -EINVAL;
523                         goto out;
524                 }
525                 ret = get_user(ch, ubuf++);
526                 if (ret)
527                         goto out;
528                 read++;
529                 cnt--;
530         }
531
532         /* We either got finished input or we have to wait for another call. */
533         if (isspace(ch)) {
534                 parser->buffer[parser->idx] = 0;
535                 parser->cont = false;
536         } else {
537                 parser->cont = true;
538                 parser->buffer[parser->idx++] = ch;
539         }
540
541         *ppos += read;
542         ret = read;
543
544 out:
545         return ret;
546 }
547
548 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
549 {
550         int len;
551         int ret;
552
553         if (!cnt)
554                 return 0;
555
556         if (s->len <= s->readpos)
557                 return -EBUSY;
558
559         len = s->len - s->readpos;
560         if (cnt > len)
561                 cnt = len;
562         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
563         if (ret == cnt)
564                 return -EFAULT;
565
566         cnt -= ret;
567
568         s->readpos += cnt;
569         return cnt;
570 }
571
572 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
573 {
574         int len;
575         void *ret;
576
577         if (s->len <= s->readpos)
578                 return -EBUSY;
579
580         len = s->len - s->readpos;
581         if (cnt > len)
582                 cnt = len;
583         ret = memcpy(buf, s->buffer + s->readpos, cnt);
584         if (!ret)
585                 return -EFAULT;
586
587         s->readpos += cnt;
588         return cnt;
589 }
590
591 /*
592  * ftrace_max_lock is used to protect the swapping of buffers
593  * when taking a max snapshot. The buffers themselves are
594  * protected by per_cpu spinlocks. But the action of the swap
595  * needs its own lock.
596  *
597  * This is defined as a arch_spinlock_t in order to help
598  * with performance when lockdep debugging is enabled.
599  *
600  * It is also used in other places outside the update_max_tr
601  * so it needs to be defined outside of the
602  * CONFIG_TRACER_MAX_TRACE.
603  */
604 static arch_spinlock_t ftrace_max_lock =
605         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
606
607 unsigned long __read_mostly     tracing_thresh;
608
609 #ifdef CONFIG_TRACER_MAX_TRACE
610 unsigned long __read_mostly     tracing_max_latency;
611
612 /*
613  * Copy the new maximum trace into the separate maximum-trace
614  * structure. (this way the maximum trace is permanently saved,
615  * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
616  */
617 static void
618 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
619 {
620         struct trace_array_cpu *data = tr->data[cpu];
621         struct trace_array_cpu *max_data;
622
623         max_tr.cpu = cpu;
624         max_tr.time_start = data->preempt_timestamp;
625
626         max_data = max_tr.data[cpu];
627         max_data->saved_latency = tracing_max_latency;
628         max_data->critical_start = data->critical_start;
629         max_data->critical_end = data->critical_end;
630
631         memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
632         max_data->pid = tsk->pid;
633         max_data->uid = task_uid(tsk);
634         max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
635         max_data->policy = tsk->policy;
636         max_data->rt_priority = tsk->rt_priority;
637
638         /* record this tasks comm */
639         tracing_record_cmdline(tsk);
640 }
641
642 /**
643  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
644  * @tr: tracer
645  * @tsk: the task with the latency
646  * @cpu: The cpu that initiated the trace.
647  *
648  * Flip the buffers between the @tr and the max_tr and record information
649  * about which task was the cause of this latency.
650  */
651 void
652 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
653 {
654         struct ring_buffer *buf = tr->buffer;
655
656         if (trace_stop_count)
657                 return;
658
659         WARN_ON_ONCE(!irqs_disabled());
660         if (!current_trace->use_max_tr) {
661                 WARN_ON_ONCE(1);
662                 return;
663         }
664         arch_spin_lock(&ftrace_max_lock);
665
666         tr->buffer = max_tr.buffer;
667         max_tr.buffer = buf;
668
669         __update_max_tr(tr, tsk, cpu);
670         arch_spin_unlock(&ftrace_max_lock);
671 }
672
673 /**
674  * update_max_tr_single - only copy one trace over, and reset the rest
675  * @tr - tracer
676  * @tsk - task with the latency
677  * @cpu - the cpu of the buffer to copy.
678  *
679  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
680  */
681 void
682 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
683 {
684         int ret;
685
686         if (trace_stop_count)
687                 return;
688
689         WARN_ON_ONCE(!irqs_disabled());
690         if (!current_trace->use_max_tr) {
691                 WARN_ON_ONCE(1);
692                 return;
693         }
694
695         arch_spin_lock(&ftrace_max_lock);
696
697         ftrace_disable_cpu();
698
699         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
700
701         if (ret == -EBUSY) {
702                 /*
703                  * We failed to swap the buffer due to a commit taking
704                  * place on this CPU. We fail to record, but we reset
705                  * the max trace buffer (no one writes directly to it)
706                  * and flag that it failed.
707                  */
708                 trace_array_printk(&max_tr, _THIS_IP_,
709                         "Failed to swap buffers due to commit in progress\n");
710         }
711
712         ftrace_enable_cpu();
713
714         WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
715
716         __update_max_tr(tr, tsk, cpu);
717         arch_spin_unlock(&ftrace_max_lock);
718 }
719 #endif /* CONFIG_TRACER_MAX_TRACE */
720
721 /**
722  * register_tracer - register a tracer with the ftrace system.
723  * @type - the plugin for the tracer
724  *
725  * Register a new plugin tracer.
726  */
727 int register_tracer(struct tracer *type)
728 __releases(kernel_lock)
729 __acquires(kernel_lock)
730 {
731         struct tracer *t;
732         int ret = 0;
733
734         if (!type->name) {
735                 pr_info("Tracer must have a name\n");
736                 return -1;
737         }
738
739         if (strlen(type->name) >= MAX_TRACER_SIZE) {
740                 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
741                 return -1;
742         }
743
744         /*
745          * When this gets called we hold the BKL which means that
746          * preemption is disabled. Various trace selftests however
747          * need to disable and enable preemption for successful tests.
748          * So we drop the BKL here and grab it after the tests again.
749          */
750         unlock_kernel();
751         mutex_lock(&trace_types_lock);
752
753         tracing_selftest_running = true;
754
755         for (t = trace_types; t; t = t->next) {
756                 if (strcmp(type->name, t->name) == 0) {
757                         /* already found */
758                         pr_info("Tracer %s already registered\n",
759                                 type->name);
760                         ret = -1;
761                         goto out;
762                 }
763         }
764
765         if (!type->set_flag)
766                 type->set_flag = &dummy_set_flag;
767         if (!type->flags)
768                 type->flags = &dummy_tracer_flags;
769         else
770                 if (!type->flags->opts)
771                         type->flags->opts = dummy_tracer_opt;
772         if (!type->wait_pipe)
773                 type->wait_pipe = default_wait_pipe;
774
775
776 #ifdef CONFIG_FTRACE_STARTUP_TEST
777         if (type->selftest && !tracing_selftest_disabled) {
778                 struct tracer *saved_tracer = current_trace;
779                 struct trace_array *tr = &global_trace;
780
781                 /*
782                  * Run a selftest on this tracer.
783                  * Here we reset the trace buffer, and set the current
784                  * tracer to be this tracer. The tracer can then run some
785                  * internal tracing to verify that everything is in order.
786                  * If we fail, we do not register this tracer.
787                  */
788                 tracing_reset_online_cpus(tr);
789
790                 current_trace = type;
791                 /* the test is responsible for initializing and enabling */
792                 pr_info("Testing tracer %s: ", type->name);
793                 ret = type->selftest(type, tr);
794                 /* the test is responsible for resetting too */
795                 current_trace = saved_tracer;
796                 if (ret) {
797                         printk(KERN_CONT "FAILED!\n");
798                         goto out;
799                 }
800                 /* Only reset on passing, to avoid touching corrupted buffers */
801                 tracing_reset_online_cpus(tr);
802
803                 printk(KERN_CONT "PASSED\n");
804         }
805 #endif
806
807         type->next = trace_types;
808         trace_types = type;
809
810  out:
811         tracing_selftest_running = false;
812         mutex_unlock(&trace_types_lock);
813
814         if (ret || !default_bootup_tracer)
815                 goto out_unlock;
816
817         if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
818                 goto out_unlock;
819
820         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
821         /* Do we want this tracer to start on bootup? */
822         tracing_set_tracer(type->name);
823         default_bootup_tracer = NULL;
824         /* disable other selftests, since this will break it. */
825         tracing_selftest_disabled = 1;
826 #ifdef CONFIG_FTRACE_STARTUP_TEST
827         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
828                type->name);
829 #endif
830
831  out_unlock:
832         lock_kernel();
833         return ret;
834 }
835
836 void unregister_tracer(struct tracer *type)
837 {
838         struct tracer **t;
839
840         mutex_lock(&trace_types_lock);
841         for (t = &trace_types; *t; t = &(*t)->next) {
842                 if (*t == type)
843                         goto found;
844         }
845         pr_info("Tracer %s not registered\n", type->name);
846         goto out;
847
848  found:
849         *t = (*t)->next;
850
851         if (type == current_trace && tracer_enabled) {
852                 tracer_enabled = 0;
853                 tracing_stop();
854                 if (current_trace->stop)
855                         current_trace->stop(&global_trace);
856                 current_trace = &nop_trace;
857         }
858 out:
859         mutex_unlock(&trace_types_lock);
860 }
861
862 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
863 {
864         ftrace_disable_cpu();
865         ring_buffer_reset_cpu(buffer, cpu);
866         ftrace_enable_cpu();
867 }
868
869 void tracing_reset(struct trace_array *tr, int cpu)
870 {
871         struct ring_buffer *buffer = tr->buffer;
872
873         ring_buffer_record_disable(buffer);
874
875         /* Make sure all commits have finished */
876         synchronize_sched();
877         __tracing_reset(buffer, cpu);
878
879         ring_buffer_record_enable(buffer);
880 }
881
882 void tracing_reset_online_cpus(struct trace_array *tr)
883 {
884         struct ring_buffer *buffer = tr->buffer;
885         int cpu;
886
887         ring_buffer_record_disable(buffer);
888
889         /* Make sure all commits have finished */
890         synchronize_sched();
891
892         tr->time_start = ftrace_now(tr->cpu);
893
894         for_each_online_cpu(cpu)
895                 __tracing_reset(buffer, cpu);
896
897         ring_buffer_record_enable(buffer);
898 }
899
900 void tracing_reset_current(int cpu)
901 {
902         tracing_reset(&global_trace, cpu);
903 }
904
905 void tracing_reset_current_online_cpus(void)
906 {
907         tracing_reset_online_cpus(&global_trace);
908 }
909
910 #define SAVED_CMDLINES 128
911 #define NO_CMDLINE_MAP UINT_MAX
912 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
913 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
914 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
915 static int cmdline_idx;
916 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
917
918 /* temporary disable recording */
919 static atomic_t trace_record_cmdline_disabled __read_mostly;
920
921 static void trace_init_cmdlines(void)
922 {
923         memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
924         memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
925         cmdline_idx = 0;
926 }
927
928 int is_tracing_stopped(void)
929 {
930         return trace_stop_count;
931 }
932
933 /**
934  * ftrace_off_permanent - disable all ftrace code permanently
935  *
936  * This should only be called when a serious anomally has
937  * been detected.  This will turn off the function tracing,
938  * ring buffers, and other tracing utilites. It takes no
939  * locks and can be called from any context.
940  */
941 void ftrace_off_permanent(void)
942 {
943         tracing_disabled = 1;
944         ftrace_stop();
945         tracing_off_permanent();
946 }
947
948 /**
949  * tracing_start - quick start of the tracer
950  *
951  * If tracing is enabled but was stopped by tracing_stop,
952  * this will start the tracer back up.
953  */
954 void tracing_start(void)
955 {
956         struct ring_buffer *buffer;
957         unsigned long flags;
958
959         if (tracing_disabled)
960                 return;
961
962         spin_lock_irqsave(&tracing_start_lock, flags);
963         if (--trace_stop_count) {
964                 if (trace_stop_count < 0) {
965                         /* Someone screwed up their debugging */
966                         WARN_ON_ONCE(1);
967                         trace_stop_count = 0;
968                 }
969                 goto out;
970         }
971
972         /* Prevent the buffers from switching */
973         arch_spin_lock(&ftrace_max_lock);
974
975         buffer = global_trace.buffer;
976         if (buffer)
977                 ring_buffer_record_enable(buffer);
978
979         buffer = max_tr.buffer;
980         if (buffer)
981                 ring_buffer_record_enable(buffer);
982
983         arch_spin_unlock(&ftrace_max_lock);
984
985         ftrace_start();
986  out:
987         spin_unlock_irqrestore(&tracing_start_lock, flags);
988 }
989
990 /**
991  * tracing_stop - quick stop of the tracer
992  *
993  * Light weight way to stop tracing. Use in conjunction with
994  * tracing_start.
995  */
996 void tracing_stop(void)
997 {
998         struct ring_buffer *buffer;
999         unsigned long flags;
1000
1001         ftrace_stop();
1002         spin_lock_irqsave(&tracing_start_lock, flags);
1003         if (trace_stop_count++)
1004                 goto out;
1005
1006         /* Prevent the buffers from switching */
1007         arch_spin_lock(&ftrace_max_lock);
1008
1009         buffer = global_trace.buffer;
1010         if (buffer)
1011                 ring_buffer_record_disable(buffer);
1012
1013         buffer = max_tr.buffer;
1014         if (buffer)
1015                 ring_buffer_record_disable(buffer);
1016
1017         arch_spin_unlock(&ftrace_max_lock);
1018
1019  out:
1020         spin_unlock_irqrestore(&tracing_start_lock, flags);
1021 }
1022
1023 void trace_stop_cmdline_recording(void);
1024
1025 static void trace_save_cmdline(struct task_struct *tsk)
1026 {
1027         unsigned pid, idx;
1028
1029         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1030                 return;
1031
1032         /*
1033          * It's not the end of the world if we don't get
1034          * the lock, but we also don't want to spin
1035          * nor do we want to disable interrupts,
1036          * so if we miss here, then better luck next time.
1037          */
1038         if (!arch_spin_trylock(&trace_cmdline_lock))
1039                 return;
1040
1041         idx = map_pid_to_cmdline[tsk->pid];
1042         if (idx == NO_CMDLINE_MAP) {
1043                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1044
1045                 /*
1046                  * Check whether the cmdline buffer at idx has a pid
1047                  * mapped. We are going to overwrite that entry so we
1048                  * need to clear the map_pid_to_cmdline. Otherwise we
1049                  * would read the new comm for the old pid.
1050                  */
1051                 pid = map_cmdline_to_pid[idx];
1052                 if (pid != NO_CMDLINE_MAP)
1053                         map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1054
1055                 map_cmdline_to_pid[idx] = tsk->pid;
1056                 map_pid_to_cmdline[tsk->pid] = idx;
1057
1058                 cmdline_idx = idx;
1059         }
1060
1061         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1062
1063         arch_spin_unlock(&trace_cmdline_lock);
1064 }
1065
1066 void trace_find_cmdline(int pid, char comm[])
1067 {
1068         unsigned map;
1069
1070         if (!pid) {
1071                 strcpy(comm, "<idle>");
1072                 return;
1073         }
1074
1075         if (WARN_ON_ONCE(pid < 0)) {
1076                 strcpy(comm, "<XXX>");
1077                 return;
1078         }
1079
1080         if (pid > PID_MAX_DEFAULT) {
1081                 strcpy(comm, "<...>");
1082                 return;
1083         }
1084
1085         preempt_disable();
1086         arch_spin_lock(&trace_cmdline_lock);
1087         map = map_pid_to_cmdline[pid];
1088         if (map != NO_CMDLINE_MAP)
1089                 strcpy(comm, saved_cmdlines[map]);
1090         else
1091                 strcpy(comm, "<...>");
1092
1093         arch_spin_unlock(&trace_cmdline_lock);
1094         preempt_enable();
1095 }
1096
1097 void tracing_record_cmdline(struct task_struct *tsk)
1098 {
1099         if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1100             !tracing_is_on())
1101                 return;
1102
1103         trace_save_cmdline(tsk);
1104 }
1105
1106 void
1107 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1108                              int pc)
1109 {
1110         struct task_struct *tsk = current;
1111
1112         entry->preempt_count            = pc & 0xff;
1113         entry->pid                      = (tsk) ? tsk->pid : 0;
1114         entry->lock_depth               = (tsk) ? tsk->lock_depth : 0;
1115         entry->flags =
1116 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1117                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1118 #else
1119                 TRACE_FLAG_IRQS_NOSUPPORT |
1120 #endif
1121                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1122                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1123                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1124 }
1125 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1126
1127 struct ring_buffer_event *
1128 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1129                           int type,
1130                           unsigned long len,
1131                           unsigned long flags, int pc)
1132 {
1133         struct ring_buffer_event *event;
1134
1135         event = ring_buffer_lock_reserve(buffer, len);
1136         if (event != NULL) {
1137                 struct trace_entry *ent = ring_buffer_event_data(event);
1138
1139                 tracing_generic_entry_update(ent, flags, pc);
1140                 ent->type = type;
1141         }
1142
1143         return event;
1144 }
1145
1146 static inline void
1147 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1148                              struct ring_buffer_event *event,
1149                              unsigned long flags, int pc,
1150                              int wake)
1151 {
1152         ring_buffer_unlock_commit(buffer, event);
1153
1154         ftrace_trace_stack(buffer, flags, 6, pc);
1155         ftrace_trace_userstack(buffer, flags, pc);
1156
1157         if (wake)
1158                 trace_wake_up();
1159 }
1160
1161 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1162                                 struct ring_buffer_event *event,
1163                                 unsigned long flags, int pc)
1164 {
1165         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1166 }
1167
1168 struct ring_buffer_event *
1169 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1170                                   int type, unsigned long len,
1171                                   unsigned long flags, int pc)
1172 {
1173         *current_rb = global_trace.buffer;
1174         return trace_buffer_lock_reserve(*current_rb,
1175                                          type, len, flags, pc);
1176 }
1177 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1178
1179 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1180                                         struct ring_buffer_event *event,
1181                                         unsigned long flags, int pc)
1182 {
1183         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1184 }
1185 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1186
1187 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1188                                        struct ring_buffer_event *event,
1189                                        unsigned long flags, int pc)
1190 {
1191         __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1192 }
1193 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1194
1195 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1196                                          struct ring_buffer_event *event)
1197 {
1198         ring_buffer_discard_commit(buffer, event);
1199 }
1200 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1201
1202 void
1203 trace_function(struct trace_array *tr,
1204                unsigned long ip, unsigned long parent_ip, unsigned long flags,
1205                int pc)
1206 {
1207         struct ftrace_event_call *call = &event_function;
1208         struct ring_buffer *buffer = tr->buffer;
1209         struct ring_buffer_event *event;
1210         struct ftrace_entry *entry;
1211
1212         /* If we are reading the ring buffer, don't trace */
1213         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1214                 return;
1215
1216         event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1217                                           flags, pc);
1218         if (!event)
1219                 return;
1220         entry   = ring_buffer_event_data(event);
1221         entry->ip                       = ip;
1222         entry->parent_ip                = parent_ip;
1223
1224         if (!filter_check_discard(call, entry, buffer, event))
1225                 ring_buffer_unlock_commit(buffer, event);
1226 }
1227
1228 void
1229 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1230        unsigned long ip, unsigned long parent_ip, unsigned long flags,
1231        int pc)
1232 {
1233         if (likely(!atomic_read(&data->disabled)))
1234                 trace_function(tr, ip, parent_ip, flags, pc);
1235 }
1236
1237 #ifdef CONFIG_STACKTRACE
1238 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1239                                  unsigned long flags,
1240                                  int skip, int pc)
1241 {
1242         struct ftrace_event_call *call = &event_kernel_stack;
1243         struct ring_buffer_event *event;
1244         struct stack_entry *entry;
1245         struct stack_trace trace;
1246
1247         event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1248                                           sizeof(*entry), flags, pc);
1249         if (!event)
1250                 return;
1251         entry   = ring_buffer_event_data(event);
1252         memset(&entry->caller, 0, sizeof(entry->caller));
1253
1254         trace.nr_entries        = 0;
1255         trace.max_entries       = FTRACE_STACK_ENTRIES;
1256         trace.skip              = skip;
1257         trace.entries           = entry->caller;
1258
1259         save_stack_trace(&trace);
1260         if (!filter_check_discard(call, entry, buffer, event))
1261                 ring_buffer_unlock_commit(buffer, event);
1262 }
1263
1264 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1265                         int skip, int pc)
1266 {
1267         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1268                 return;
1269
1270         __ftrace_trace_stack(buffer, flags, skip, pc);
1271 }
1272
1273 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1274                    int pc)
1275 {
1276         __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1277 }
1278
1279 /**
1280  * trace_dump_stack - record a stack back trace in the trace buffer
1281  */
1282 void trace_dump_stack(void)
1283 {
1284         unsigned long flags;
1285
1286         if (tracing_disabled || tracing_selftest_running)
1287                 return;
1288
1289         local_save_flags(flags);
1290
1291         /* skipping 3 traces, seems to get us at the caller of this function */
1292         __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1293 }
1294
1295 void
1296 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1297 {
1298         struct ftrace_event_call *call = &event_user_stack;
1299         struct ring_buffer_event *event;
1300         struct userstack_entry *entry;
1301         struct stack_trace trace;
1302
1303         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1304                 return;
1305
1306         /*
1307          * NMIs can not handle page faults, even with fix ups.
1308          * The save user stack can (and often does) fault.
1309          */
1310         if (unlikely(in_nmi()))
1311                 return;
1312
1313         event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1314                                           sizeof(*entry), flags, pc);
1315         if (!event)
1316                 return;
1317         entry   = ring_buffer_event_data(event);
1318
1319         entry->tgid             = current->tgid;
1320         memset(&entry->caller, 0, sizeof(entry->caller));
1321
1322         trace.nr_entries        = 0;
1323         trace.max_entries       = FTRACE_STACK_ENTRIES;
1324         trace.skip              = 0;
1325         trace.entries           = entry->caller;
1326
1327         save_stack_trace_user(&trace);
1328         if (!filter_check_discard(call, entry, buffer, event))
1329                 ring_buffer_unlock_commit(buffer, event);
1330 }
1331
1332 #ifdef UNUSED
1333 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1334 {
1335         ftrace_trace_userstack(tr, flags, preempt_count());
1336 }
1337 #endif /* UNUSED */
1338
1339 #endif /* CONFIG_STACKTRACE */
1340
1341 /**
1342  * trace_vbprintk - write binary msg to tracing buffer
1343  *
1344  */
1345 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1346 {
1347         static arch_spinlock_t trace_buf_lock =
1348                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1349         static u32 trace_buf[TRACE_BUF_SIZE];
1350
1351         struct ftrace_event_call *call = &event_bprint;
1352         struct ring_buffer_event *event;
1353         struct ring_buffer *buffer;
1354         struct trace_array *tr = &global_trace;
1355         struct trace_array_cpu *data;
1356         struct bprint_entry *entry;
1357         unsigned long flags;
1358         int disable;
1359         int cpu, len = 0, size, pc;
1360
1361         if (unlikely(tracing_selftest_running || tracing_disabled))
1362                 return 0;
1363
1364         /* Don't pollute graph traces with trace_vprintk internals */
1365         pause_graph_tracing();
1366
1367         pc = preempt_count();
1368         preempt_disable_notrace();
1369         cpu = raw_smp_processor_id();
1370         data = tr->data[cpu];
1371
1372         disable = atomic_inc_return(&data->disabled);
1373         if (unlikely(disable != 1))
1374                 goto out;
1375
1376         /* Lockdep uses trace_printk for lock tracing */
1377         local_irq_save(flags);
1378         arch_spin_lock(&trace_buf_lock);
1379         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1380
1381         if (len > TRACE_BUF_SIZE || len < 0)
1382                 goto out_unlock;
1383
1384         size = sizeof(*entry) + sizeof(u32) * len;
1385         buffer = tr->buffer;
1386         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1387                                           flags, pc);
1388         if (!event)
1389                 goto out_unlock;
1390         entry = ring_buffer_event_data(event);
1391         entry->ip                       = ip;
1392         entry->fmt                      = fmt;
1393
1394         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1395         if (!filter_check_discard(call, entry, buffer, event)) {
1396                 ring_buffer_unlock_commit(buffer, event);
1397                 ftrace_trace_stack(buffer, flags, 6, pc);
1398         }
1399
1400 out_unlock:
1401         arch_spin_unlock(&trace_buf_lock);
1402         local_irq_restore(flags);
1403
1404 out:
1405         atomic_dec_return(&data->disabled);
1406         preempt_enable_notrace();
1407         unpause_graph_tracing();
1408
1409         return len;
1410 }
1411 EXPORT_SYMBOL_GPL(trace_vbprintk);
1412
1413 int trace_array_printk(struct trace_array *tr,
1414                        unsigned long ip, const char *fmt, ...)
1415 {
1416         int ret;
1417         va_list ap;
1418
1419         if (!(trace_flags & TRACE_ITER_PRINTK))
1420                 return 0;
1421
1422         va_start(ap, fmt);
1423         ret = trace_array_vprintk(tr, ip, fmt, ap);
1424         va_end(ap);
1425         return ret;
1426 }
1427
1428 int trace_array_vprintk(struct trace_array *tr,
1429                         unsigned long ip, const char *fmt, va_list args)
1430 {
1431         static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1432         static char trace_buf[TRACE_BUF_SIZE];
1433
1434         struct ftrace_event_call *call = &event_print;
1435         struct ring_buffer_event *event;
1436         struct ring_buffer *buffer;
1437         struct trace_array_cpu *data;
1438         int cpu, len = 0, size, pc;
1439         struct print_entry *entry;
1440         unsigned long irq_flags;
1441         int disable;
1442
1443         if (tracing_disabled || tracing_selftest_running)
1444                 return 0;
1445
1446         pc = preempt_count();
1447         preempt_disable_notrace();
1448         cpu = raw_smp_processor_id();
1449         data = tr->data[cpu];
1450
1451         disable = atomic_inc_return(&data->disabled);
1452         if (unlikely(disable != 1))
1453                 goto out;
1454
1455         pause_graph_tracing();
1456         raw_local_irq_save(irq_flags);
1457         arch_spin_lock(&trace_buf_lock);
1458         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1459
1460         size = sizeof(*entry) + len + 1;
1461         buffer = tr->buffer;
1462         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1463                                           irq_flags, pc);
1464         if (!event)
1465                 goto out_unlock;
1466         entry = ring_buffer_event_data(event);
1467         entry->ip = ip;
1468
1469         memcpy(&entry->buf, trace_buf, len);
1470         entry->buf[len] = '\0';
1471         if (!filter_check_discard(call, entry, buffer, event)) {
1472                 ring_buffer_unlock_commit(buffer, event);
1473                 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1474         }
1475
1476  out_unlock:
1477         arch_spin_unlock(&trace_buf_lock);
1478         raw_local_irq_restore(irq_flags);
1479         unpause_graph_tracing();
1480  out:
1481         atomic_dec_return(&data->disabled);
1482         preempt_enable_notrace();
1483
1484         return len;
1485 }
1486
1487 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1488 {
1489         return trace_array_vprintk(&global_trace, ip, fmt, args);
1490 }
1491 EXPORT_SYMBOL_GPL(trace_vprintk);
1492
1493 static void trace_iterator_increment(struct trace_iterator *iter)
1494 {
1495         /* Don't allow ftrace to trace into the ring buffers */
1496         ftrace_disable_cpu();
1497
1498         iter->idx++;
1499         if (iter->buffer_iter[iter->cpu])
1500                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1501
1502         ftrace_enable_cpu();
1503 }
1504
1505 static struct trace_entry *
1506 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1507                 unsigned long *lost_events)
1508 {
1509         struct ring_buffer_event *event;
1510         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1511
1512         /* Don't allow ftrace to trace into the ring buffers */
1513         ftrace_disable_cpu();
1514
1515         if (buf_iter)
1516                 event = ring_buffer_iter_peek(buf_iter, ts);
1517         else
1518                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1519                                          lost_events);
1520
1521         ftrace_enable_cpu();
1522
1523         return event ? ring_buffer_event_data(event) : NULL;
1524 }
1525
1526 static struct trace_entry *
1527 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1528                   unsigned long *missing_events, u64 *ent_ts)
1529 {
1530         struct ring_buffer *buffer = iter->tr->buffer;
1531         struct trace_entry *ent, *next = NULL;
1532         unsigned long lost_events = 0, next_lost = 0;
1533         int cpu_file = iter->cpu_file;
1534         u64 next_ts = 0, ts;
1535         int next_cpu = -1;
1536         int cpu;
1537
1538         /*
1539          * If we are in a per_cpu trace file, don't bother by iterating over
1540          * all cpu and peek directly.
1541          */
1542         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1543                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1544                         return NULL;
1545                 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1546                 if (ent_cpu)
1547                         *ent_cpu = cpu_file;
1548
1549                 return ent;
1550         }
1551
1552         for_each_tracing_cpu(cpu) {
1553
1554                 if (ring_buffer_empty_cpu(buffer, cpu))
1555                         continue;
1556
1557                 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1558
1559                 /*
1560                  * Pick the entry with the smallest timestamp:
1561                  */
1562                 if (ent && (!next || ts < next_ts)) {
1563                         next = ent;
1564                         next_cpu = cpu;
1565                         next_ts = ts;
1566                         next_lost = lost_events;
1567                 }
1568         }
1569
1570         if (ent_cpu)
1571                 *ent_cpu = next_cpu;
1572
1573         if (ent_ts)
1574                 *ent_ts = next_ts;
1575
1576         if (missing_events)
1577                 *missing_events = next_lost;
1578
1579         return next;
1580 }
1581
1582 /* Find the next real entry, without updating the iterator itself */
1583 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1584                                           int *ent_cpu, u64 *ent_ts)
1585 {
1586         return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1587 }
1588
1589 /* Find the next real entry, and increment the iterator to the next entry */
1590 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1591 {
1592         iter->ent = __find_next_entry(iter, &iter->cpu,
1593                                       &iter->lost_events, &iter->ts);
1594
1595         if (iter->ent)
1596                 trace_iterator_increment(iter);
1597
1598         return iter->ent ? iter : NULL;
1599 }
1600
1601 static void trace_consume(struct trace_iterator *iter)
1602 {
1603         /* Don't allow ftrace to trace into the ring buffers */
1604         ftrace_disable_cpu();
1605         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1606                             &iter->lost_events);
1607         ftrace_enable_cpu();
1608 }
1609
1610 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1611 {
1612         struct trace_iterator *iter = m->private;
1613         int i = (int)*pos;
1614         void *ent;
1615
1616         WARN_ON_ONCE(iter->leftover);
1617
1618         (*pos)++;
1619
1620         /* can't go backwards */
1621         if (iter->idx > i)
1622                 return NULL;
1623
1624         if (iter->idx < 0)
1625                 ent = trace_find_next_entry_inc(iter);
1626         else
1627                 ent = iter;
1628
1629         while (ent && iter->idx < i)
1630                 ent = trace_find_next_entry_inc(iter);
1631
1632         iter->pos = *pos;
1633
1634         return ent;
1635 }
1636
1637 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1638 {
1639         struct trace_array *tr = iter->tr;
1640         struct ring_buffer_event *event;
1641         struct ring_buffer_iter *buf_iter;
1642         unsigned long entries = 0;
1643         u64 ts;
1644
1645         tr->data[cpu]->skipped_entries = 0;
1646
1647         if (!iter->buffer_iter[cpu])
1648                 return;
1649
1650         buf_iter = iter->buffer_iter[cpu];
1651         ring_buffer_iter_reset(buf_iter);
1652
1653         /*
1654          * We could have the case with the max latency tracers
1655          * that a reset never took place on a cpu. This is evident
1656          * by the timestamp being before the start of the buffer.
1657          */
1658         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1659                 if (ts >= iter->tr->time_start)
1660                         break;
1661                 entries++;
1662                 ring_buffer_read(buf_iter, NULL);
1663         }
1664
1665         tr->data[cpu]->skipped_entries = entries;
1666 }
1667
1668 /*
1669  * The current tracer is copied to avoid a global locking
1670  * all around.
1671  */
1672 static void *s_start(struct seq_file *m, loff_t *pos)
1673 {
1674         struct trace_iterator *iter = m->private;
1675         static struct tracer *old_tracer;
1676         int cpu_file = iter->cpu_file;
1677         void *p = NULL;
1678         loff_t l = 0;
1679         int cpu;
1680
1681         /* copy the tracer to avoid using a global lock all around */
1682         mutex_lock(&trace_types_lock);
1683         if (unlikely(old_tracer != current_trace && current_trace)) {
1684                 old_tracer = current_trace;
1685                 *iter->trace = *current_trace;
1686         }
1687         mutex_unlock(&trace_types_lock);
1688
1689         atomic_inc(&trace_record_cmdline_disabled);
1690
1691         if (*pos != iter->pos) {
1692                 iter->ent = NULL;
1693                 iter->cpu = 0;
1694                 iter->idx = -1;
1695
1696                 ftrace_disable_cpu();
1697
1698                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1699                         for_each_tracing_cpu(cpu)
1700                                 tracing_iter_reset(iter, cpu);
1701                 } else
1702                         tracing_iter_reset(iter, cpu_file);
1703
1704                 ftrace_enable_cpu();
1705
1706                 iter->leftover = 0;
1707                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1708                         ;
1709
1710         } else {
1711                 /*
1712                  * If we overflowed the seq_file before, then we want
1713                  * to just reuse the trace_seq buffer again.
1714                  */
1715                 if (iter->leftover)
1716                         p = iter;
1717                 else {
1718                         l = *pos - 1;
1719                         p = s_next(m, p, &l);
1720                 }
1721         }
1722
1723         trace_event_read_lock();
1724         trace_access_lock(cpu_file);
1725         return p;
1726 }
1727
1728 static void s_stop(struct seq_file *m, void *p)
1729 {
1730         struct trace_iterator *iter = m->private;
1731
1732         atomic_dec(&trace_record_cmdline_disabled);
1733         trace_access_unlock(iter->cpu_file);
1734         trace_event_read_unlock();
1735 }
1736
1737 static void print_lat_help_header(struct seq_file *m)
1738 {
1739         seq_puts(m, "#                  _------=> CPU#            \n");
1740         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1741         seq_puts(m, "#                | / _----=> need-resched    \n");
1742         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1743         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1744         seq_puts(m, "#                |||| /_--=> lock-depth       \n");
1745         seq_puts(m, "#                |||||/     delay             \n");
1746         seq_puts(m, "#  cmd     pid   |||||| time  |   caller      \n");
1747         seq_puts(m, "#     \\   /      ||||||   \\   |   /           \n");
1748 }
1749
1750 static void print_func_help_header(struct seq_file *m)
1751 {
1752         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1753         seq_puts(m, "#              | |       |          |         |\n");
1754 }
1755
1756
1757 void
1758 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1759 {
1760         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1761         struct trace_array *tr = iter->tr;
1762         struct trace_array_cpu *data = tr->data[tr->cpu];
1763         struct tracer *type = current_trace;
1764         unsigned long entries = 0;
1765         unsigned long total = 0;
1766         unsigned long count;
1767         const char *name = "preemption";
1768         int cpu;
1769
1770         if (type)
1771                 name = type->name;
1772
1773
1774         for_each_tracing_cpu(cpu) {
1775                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1776                 /*
1777                  * If this buffer has skipped entries, then we hold all
1778                  * entries for the trace and we need to ignore the
1779                  * ones before the time stamp.
1780                  */
1781                 if (tr->data[cpu]->skipped_entries) {
1782                         count -= tr->data[cpu]->skipped_entries;
1783                         /* total is the same as the entries */
1784                         total += count;
1785                 } else
1786                         total += count +
1787                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1788                 entries += count;
1789         }
1790
1791         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1792                    name, UTS_RELEASE);
1793         seq_puts(m, "# -----------------------------------"
1794                  "---------------------------------\n");
1795         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1796                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1797                    nsecs_to_usecs(data->saved_latency),
1798                    entries,
1799                    total,
1800                    tr->cpu,
1801 #if defined(CONFIG_PREEMPT_NONE)
1802                    "server",
1803 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1804                    "desktop",
1805 #elif defined(CONFIG_PREEMPT)
1806                    "preempt",
1807 #else
1808                    "unknown",
1809 #endif
1810                    /* These are reserved for later use */
1811                    0, 0, 0, 0);
1812 #ifdef CONFIG_SMP
1813         seq_printf(m, " #P:%d)\n", num_online_cpus());
1814 #else
1815         seq_puts(m, ")\n");
1816 #endif
1817         seq_puts(m, "#    -----------------\n");
1818         seq_printf(m, "#    | task: %.16s-%d "
1819                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1820                    data->comm, data->pid, data->uid, data->nice,
1821                    data->policy, data->rt_priority);
1822         seq_puts(m, "#    -----------------\n");
1823
1824         if (data->critical_start) {
1825                 seq_puts(m, "#  => started at: ");
1826                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1827                 trace_print_seq(m, &iter->seq);
1828                 seq_puts(m, "\n#  => ended at:   ");
1829                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1830                 trace_print_seq(m, &iter->seq);
1831                 seq_puts(m, "\n#\n");
1832         }
1833
1834         seq_puts(m, "#\n");
1835 }
1836
1837 static void test_cpu_buff_start(struct trace_iterator *iter)
1838 {
1839         struct trace_seq *s = &iter->seq;
1840
1841         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1842                 return;
1843
1844         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1845                 return;
1846
1847         if (cpumask_test_cpu(iter->cpu, iter->started))
1848                 return;
1849
1850         if (iter->tr->data[iter->cpu]->skipped_entries)
1851                 return;
1852
1853         cpumask_set_cpu(iter->cpu, iter->started);
1854
1855         /* Don't print started cpu buffer for the first entry of the trace */
1856         if (iter->idx > 1)
1857                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1858                                 iter->cpu);
1859 }
1860
1861 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1862 {
1863         struct trace_seq *s = &iter->seq;
1864         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1865         struct trace_entry *entry;
1866         struct trace_event *event;
1867
1868         entry = iter->ent;
1869
1870         test_cpu_buff_start(iter);
1871
1872         event = ftrace_find_event(entry->type);
1873
1874         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1875                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1876                         if (!trace_print_lat_context(iter))
1877                                 goto partial;
1878                 } else {
1879                         if (!trace_print_context(iter))
1880                                 goto partial;
1881                 }
1882         }
1883
1884         if (event)
1885                 return event->funcs->trace(iter, sym_flags, event);
1886
1887         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1888                 goto partial;
1889
1890         return TRACE_TYPE_HANDLED;
1891 partial:
1892         return TRACE_TYPE_PARTIAL_LINE;
1893 }
1894
1895 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1896 {
1897         struct trace_seq *s = &iter->seq;
1898         struct trace_entry *entry;
1899         struct trace_event *event;
1900
1901         entry = iter->ent;
1902
1903         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1904                 if (!trace_seq_printf(s, "%d %d %llu ",
1905                                       entry->pid, iter->cpu, iter->ts))
1906                         goto partial;
1907         }
1908
1909         event = ftrace_find_event(entry->type);
1910         if (event)
1911                 return event->funcs->raw(iter, 0, event);
1912
1913         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1914                 goto partial;
1915
1916         return TRACE_TYPE_HANDLED;
1917 partial:
1918         return TRACE_TYPE_PARTIAL_LINE;
1919 }
1920
1921 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1922 {
1923         struct trace_seq *s = &iter->seq;
1924         unsigned char newline = '\n';
1925         struct trace_entry *entry;
1926         struct trace_event *event;
1927
1928         entry = iter->ent;
1929
1930         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1931                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1932                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1933                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1934         }
1935
1936         event = ftrace_find_event(entry->type);
1937         if (event) {
1938                 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1939                 if (ret != TRACE_TYPE_HANDLED)
1940                         return ret;
1941         }
1942
1943         SEQ_PUT_FIELD_RET(s, newline);
1944
1945         return TRACE_TYPE_HANDLED;
1946 }
1947
1948 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1949 {
1950         struct trace_seq *s = &iter->seq;
1951         struct trace_entry *entry;
1952         struct trace_event *event;
1953
1954         entry = iter->ent;
1955
1956         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1957                 SEQ_PUT_FIELD_RET(s, entry->pid);
1958                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1959                 SEQ_PUT_FIELD_RET(s, iter->ts);
1960         }
1961
1962         event = ftrace_find_event(entry->type);
1963         return event ? event->funcs->binary(iter, 0, event) :
1964                 TRACE_TYPE_HANDLED;
1965 }
1966
1967 int trace_empty(struct trace_iterator *iter)
1968 {
1969         int cpu;
1970
1971         /* If we are looking at one CPU buffer, only check that one */
1972         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1973                 cpu = iter->cpu_file;
1974                 if (iter->buffer_iter[cpu]) {
1975                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1976                                 return 0;
1977                 } else {
1978                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1979                                 return 0;
1980                 }
1981                 return 1;
1982         }
1983
1984         for_each_tracing_cpu(cpu) {
1985                 if (iter->buffer_iter[cpu]) {
1986                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1987                                 return 0;
1988                 } else {
1989                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1990                                 return 0;
1991                 }
1992         }
1993
1994         return 1;
1995 }
1996
1997 /*  Called with trace_event_read_lock() held. */
1998 enum print_line_t print_trace_line(struct trace_iterator *iter)
1999 {
2000         enum print_line_t ret;
2001
2002         if (iter->lost_events)
2003                 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2004                                  iter->cpu, iter->lost_events);
2005
2006         if (iter->trace && iter->trace->print_line) {
2007                 ret = iter->trace->print_line(iter);
2008                 if (ret != TRACE_TYPE_UNHANDLED)
2009                         return ret;
2010         }
2011
2012         if (iter->ent->type == TRACE_BPRINT &&
2013                         trace_flags & TRACE_ITER_PRINTK &&
2014                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2015                 return trace_print_bprintk_msg_only(iter);
2016
2017         if (iter->ent->type == TRACE_PRINT &&
2018                         trace_flags & TRACE_ITER_PRINTK &&
2019                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2020                 return trace_print_printk_msg_only(iter);
2021
2022         if (trace_flags & TRACE_ITER_BIN)
2023                 return print_bin_fmt(iter);
2024
2025         if (trace_flags & TRACE_ITER_HEX)
2026                 return print_hex_fmt(iter);
2027
2028         if (trace_flags & TRACE_ITER_RAW)
2029                 return print_raw_fmt(iter);
2030
2031         return print_trace_fmt(iter);
2032 }
2033
2034 void trace_default_header(struct seq_file *m)
2035 {
2036         struct trace_iterator *iter = m->private;
2037
2038         if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2039                 /* print nothing if the buffers are empty */
2040                 if (trace_empty(iter))
2041                         return;
2042                 print_trace_header(m, iter);
2043                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2044                         print_lat_help_header(m);
2045         } else {
2046                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2047                         print_func_help_header(m);
2048         }
2049 }
2050
2051 static int s_show(struct seq_file *m, void *v)
2052 {
2053         struct trace_iterator *iter = v;
2054         int ret;
2055
2056         if (iter->ent == NULL) {
2057                 if (iter->tr) {
2058                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2059                         seq_puts(m, "#\n");
2060                 }
2061                 if (iter->trace && iter->trace->print_header)
2062                         iter->trace->print_header(m);
2063                 else
2064                         trace_default_header(m);
2065
2066         } else if (iter->leftover) {
2067                 /*
2068                  * If we filled the seq_file buffer earlier, we
2069                  * want to just show it now.
2070                  */
2071                 ret = trace_print_seq(m, &iter->seq);
2072
2073                 /* ret should this time be zero, but you never know */
2074                 iter->leftover = ret;
2075
2076         } else {
2077                 print_trace_line(iter);
2078                 ret = trace_print_seq(m, &iter->seq);
2079                 /*
2080                  * If we overflow the seq_file buffer, then it will
2081                  * ask us for this data again at start up.
2082                  * Use that instead.
2083                  *  ret is 0 if seq_file write succeeded.
2084                  *        -1 otherwise.
2085                  */
2086                 iter->leftover = ret;
2087         }
2088
2089         return 0;
2090 }
2091
2092 static const struct seq_operations tracer_seq_ops = {
2093         .start          = s_start,
2094         .next           = s_next,
2095         .stop           = s_stop,
2096         .show           = s_show,
2097 };
2098
2099 static struct trace_iterator *
2100 __tracing_open(struct inode *inode, struct file *file)
2101 {
2102         long cpu_file = (long) inode->i_private;
2103         void *fail_ret = ERR_PTR(-ENOMEM);
2104         struct trace_iterator *iter;
2105         struct seq_file *m;
2106         int cpu, ret;
2107
2108         if (tracing_disabled)
2109                 return ERR_PTR(-ENODEV);
2110
2111         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2112         if (!iter)
2113                 return ERR_PTR(-ENOMEM);
2114
2115         /*
2116          * We make a copy of the current tracer to avoid concurrent
2117          * changes on it while we are reading.
2118          */
2119         mutex_lock(&trace_types_lock);
2120         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2121         if (!iter->trace)
2122                 goto fail;
2123
2124         if (current_trace)
2125                 *iter->trace = *current_trace;
2126
2127         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2128                 goto fail;
2129
2130         if (current_trace && current_trace->print_max)
2131                 iter->tr = &max_tr;
2132         else
2133                 iter->tr = &global_trace;
2134         iter->pos = -1;
2135         mutex_init(&iter->mutex);
2136         iter->cpu_file = cpu_file;
2137
2138         /* Notify the tracer early; before we stop tracing. */
2139         if (iter->trace && iter->trace->open)
2140                 iter->trace->open(iter);
2141
2142         /* Annotate start of buffers if we had overruns */
2143         if (ring_buffer_overruns(iter->tr->buffer))
2144                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2145
2146         /* stop the trace while dumping */
2147         tracing_stop();
2148
2149         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2150                 for_each_tracing_cpu(cpu) {
2151                         iter->buffer_iter[cpu] =
2152                                 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2153                 }
2154                 ring_buffer_read_prepare_sync();
2155                 for_each_tracing_cpu(cpu) {
2156                         ring_buffer_read_start(iter->buffer_iter[cpu]);
2157                         tracing_iter_reset(iter, cpu);
2158                 }
2159         } else {
2160                 cpu = iter->cpu_file;
2161                 iter->buffer_iter[cpu] =
2162                         ring_buffer_read_prepare(iter->tr->buffer, cpu);
2163                 ring_buffer_read_prepare_sync();
2164                 ring_buffer_read_start(iter->buffer_iter[cpu]);
2165                 tracing_iter_reset(iter, cpu);
2166         }
2167
2168         ret = seq_open(file, &tracer_seq_ops);
2169         if (ret < 0) {
2170                 fail_ret = ERR_PTR(ret);
2171                 goto fail_buffer;
2172         }
2173
2174         m = file->private_data;
2175         m->private = iter;
2176
2177         mutex_unlock(&trace_types_lock);
2178
2179         return iter;
2180
2181  fail_buffer:
2182         for_each_tracing_cpu(cpu) {
2183                 if (iter->buffer_iter[cpu])
2184                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2185         }
2186         free_cpumask_var(iter->started);
2187         tracing_start();
2188  fail:
2189         mutex_unlock(&trace_types_lock);
2190         kfree(iter->trace);
2191         kfree(iter);
2192
2193         return fail_ret;
2194 }
2195
2196 int tracing_open_generic(struct inode *inode, struct file *filp)
2197 {
2198         if (tracing_disabled)
2199                 return -ENODEV;
2200
2201         filp->private_data = inode->i_private;
2202         return 0;
2203 }
2204
2205 static int tracing_release(struct inode *inode, struct file *file)
2206 {
2207         struct seq_file *m = (struct seq_file *)file->private_data;
2208         struct trace_iterator *iter;
2209         int cpu;
2210
2211         if (!(file->f_mode & FMODE_READ))
2212                 return 0;
2213
2214         iter = m->private;
2215
2216         mutex_lock(&trace_types_lock);
2217         for_each_tracing_cpu(cpu) {
2218                 if (iter->buffer_iter[cpu])
2219                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2220         }
2221
2222         if (iter->trace && iter->trace->close)
2223                 iter->trace->close(iter);
2224
2225         /* reenable tracing if it was previously enabled */
2226         tracing_start();
2227         mutex_unlock(&trace_types_lock);
2228
2229         seq_release(inode, file);
2230         mutex_destroy(&iter->mutex);
2231         free_cpumask_var(iter->started);
2232         kfree(iter->trace);
2233         kfree(iter);
2234         return 0;
2235 }
2236
2237 static int tracing_open(struct inode *inode, struct file *file)
2238 {
2239         struct trace_iterator *iter;
2240         int ret = 0;
2241
2242         /* If this file was open for write, then erase contents */
2243         if ((file->f_mode & FMODE_WRITE) &&
2244             (file->f_flags & O_TRUNC)) {
2245                 long cpu = (long) inode->i_private;
2246
2247                 if (cpu == TRACE_PIPE_ALL_CPU)
2248                         tracing_reset_online_cpus(&global_trace);
2249                 else
2250                         tracing_reset(&global_trace, cpu);
2251         }
2252
2253         if (file->f_mode & FMODE_READ) {
2254                 iter = __tracing_open(inode, file);
2255                 if (IS_ERR(iter))
2256                         ret = PTR_ERR(iter);
2257                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2258                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2259         }
2260         return ret;
2261 }
2262
2263 static void *
2264 t_next(struct seq_file *m, void *v, loff_t *pos)
2265 {
2266         struct tracer *t = v;
2267
2268         (*pos)++;
2269
2270         if (t)
2271                 t = t->next;
2272
2273         return t;
2274 }
2275
2276 static void *t_start(struct seq_file *m, loff_t *pos)
2277 {
2278         struct tracer *t;
2279         loff_t l = 0;
2280
2281         mutex_lock(&trace_types_lock);
2282         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2283                 ;
2284
2285         return t;
2286 }
2287
2288 static void t_stop(struct seq_file *m, void *p)
2289 {
2290         mutex_unlock(&trace_types_lock);
2291 }
2292
2293 static int t_show(struct seq_file *m, void *v)
2294 {
2295         struct tracer *t = v;
2296
2297         if (!t)
2298                 return 0;
2299
2300         seq_printf(m, "%s", t->name);
2301         if (t->next)
2302                 seq_putc(m, ' ');
2303         else
2304                 seq_putc(m, '\n');
2305
2306         return 0;
2307 }
2308
2309 static const struct seq_operations show_traces_seq_ops = {
2310         .start          = t_start,
2311         .next           = t_next,
2312         .stop           = t_stop,
2313         .show           = t_show,
2314 };
2315
2316 static int show_traces_open(struct inode *inode, struct file *file)
2317 {
2318         if (tracing_disabled)
2319                 return -ENODEV;
2320
2321         return seq_open(file, &show_traces_seq_ops);
2322 }
2323
2324 static ssize_t
2325 tracing_write_stub(struct file *filp, const char __user *ubuf,
2326                    size_t count, loff_t *ppos)
2327 {
2328         return count;
2329 }
2330
2331 static const struct file_operations tracing_fops = {
2332         .open           = tracing_open,
2333         .read           = seq_read,
2334         .write          = tracing_write_stub,
2335         .llseek         = seq_lseek,
2336         .release        = tracing_release,
2337 };
2338
2339 static const struct file_operations show_traces_fops = {
2340         .open           = show_traces_open,
2341         .read           = seq_read,
2342         .release        = seq_release,
2343         .llseek         = seq_lseek,
2344 };
2345
2346 /*
2347  * Only trace on a CPU if the bitmask is set:
2348  */
2349 static cpumask_var_t tracing_cpumask;
2350
2351 /*
2352  * The tracer itself will not take this lock, but still we want
2353  * to provide a consistent cpumask to user-space:
2354  */
2355 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2356
2357 /*
2358  * Temporary storage for the character representation of the
2359  * CPU bitmask (and one more byte for the newline):
2360  */
2361 static char mask_str[NR_CPUS + 1];
2362
2363 static ssize_t
2364 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2365                      size_t count, loff_t *ppos)
2366 {
2367         int len;
2368
2369         mutex_lock(&tracing_cpumask_update_lock);
2370
2371         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2372         if (count - len < 2) {
2373                 count = -EINVAL;
2374                 goto out_err;
2375         }
2376         len += sprintf(mask_str + len, "\n");
2377         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2378
2379 out_err:
2380         mutex_unlock(&tracing_cpumask_update_lock);
2381
2382         return count;
2383 }
2384
2385 static ssize_t
2386 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2387                       size_t count, loff_t *ppos)
2388 {
2389         int err, cpu;
2390         cpumask_var_t tracing_cpumask_new;
2391
2392         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2393                 return -ENOMEM;
2394
2395         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2396         if (err)
2397                 goto err_unlock;
2398
2399         mutex_lock(&tracing_cpumask_update_lock);
2400
2401         local_irq_disable();
2402         arch_spin_lock(&ftrace_max_lock);
2403         for_each_tracing_cpu(cpu) {
2404                 /*
2405                  * Increase/decrease the disabled counter if we are
2406                  * about to flip a bit in the cpumask:
2407                  */
2408                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2409                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2410                         atomic_inc(&global_trace.data[cpu]->disabled);
2411                 }
2412                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2413                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2414                         atomic_dec(&global_trace.data[cpu]->disabled);
2415                 }
2416         }
2417         arch_spin_unlock(&ftrace_max_lock);
2418         local_irq_enable();
2419
2420         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2421
2422         mutex_unlock(&tracing_cpumask_update_lock);
2423         free_cpumask_var(tracing_cpumask_new);
2424
2425         return count;
2426
2427 err_unlock:
2428         free_cpumask_var(tracing_cpumask_new);
2429
2430         return err;
2431 }
2432
2433 static const struct file_operations tracing_cpumask_fops = {
2434         .open           = tracing_open_generic,
2435         .read           = tracing_cpumask_read,
2436         .write          = tracing_cpumask_write,
2437         .llseek         = generic_file_llseek,
2438 };
2439
2440 static int tracing_trace_options_show(struct seq_file *m, void *v)
2441 {
2442         struct tracer_opt *trace_opts;
2443         u32 tracer_flags;
2444         int i;
2445
2446         mutex_lock(&trace_types_lock);
2447         tracer_flags = current_trace->flags->val;
2448         trace_opts = current_trace->flags->opts;
2449
2450         for (i = 0; trace_options[i]; i++) {
2451                 if (trace_flags & (1 << i))
2452                         seq_printf(m, "%s\n", trace_options[i]);
2453                 else
2454                         seq_printf(m, "no%s\n", trace_options[i]);
2455         }
2456
2457         for (i = 0; trace_opts[i].name; i++) {
2458                 if (tracer_flags & trace_opts[i].bit)
2459                         seq_printf(m, "%s\n", trace_opts[i].name);
2460                 else
2461                         seq_printf(m, "no%s\n", trace_opts[i].name);
2462         }
2463         mutex_unlock(&trace_types_lock);
2464
2465         return 0;
2466 }
2467
2468 static int __set_tracer_option(struct tracer *trace,
2469                                struct tracer_flags *tracer_flags,
2470                                struct tracer_opt *opts, int neg)
2471 {
2472         int ret;
2473
2474         ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2475         if (ret)
2476                 return ret;
2477
2478         if (neg)
2479                 tracer_flags->val &= ~opts->bit;
2480         else
2481                 tracer_flags->val |= opts->bit;
2482         return 0;
2483 }
2484
2485 /* Try to assign a tracer specific option */
2486 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2487 {
2488         struct tracer_flags *tracer_flags = trace->flags;
2489         struct tracer_opt *opts = NULL;
2490         int i;
2491
2492         for (i = 0; tracer_flags->opts[i].name; i++) {
2493                 opts = &tracer_flags->opts[i];
2494
2495                 if (strcmp(cmp, opts->name) == 0)
2496                         return __set_tracer_option(trace, trace->flags,
2497                                                    opts, neg);
2498         }
2499
2500         return -EINVAL;
2501 }
2502
2503 static void set_tracer_flags(unsigned int mask, int enabled)
2504 {
2505         /* do nothing if flag is already set */
2506         if (!!(trace_flags & mask) == !!enabled)
2507                 return;
2508
2509         if (enabled)
2510                 trace_flags |= mask;
2511         else
2512                 trace_flags &= ~mask;
2513
2514         if (mask == TRACE_ITER_RECORD_CMD)
2515                 trace_event_enable_cmd_record(enabled);
2516 }
2517
2518 static ssize_t
2519 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2520                         size_t cnt, loff_t *ppos)
2521 {
2522         char buf[64];
2523         char *cmp;
2524         int neg = 0;
2525         int ret;
2526         int i;
2527
2528         if (cnt >= sizeof(buf))
2529                 return -EINVAL;
2530
2531         if (copy_from_user(&buf, ubuf, cnt))
2532                 return -EFAULT;
2533
2534         buf[cnt] = 0;
2535         cmp = strstrip(buf);
2536
2537         if (strncmp(cmp, "no", 2) == 0) {
2538                 neg = 1;
2539                 cmp += 2;
2540         }
2541
2542         for (i = 0; trace_options[i]; i++) {
2543                 if (strcmp(cmp, trace_options[i]) == 0) {
2544                         set_tracer_flags(1 << i, !neg);
2545                         break;
2546                 }
2547         }
2548
2549         /* If no option could be set, test the specific tracer options */
2550         if (!trace_options[i]) {
2551                 mutex_lock(&trace_types_lock);
2552                 ret = set_tracer_option(current_trace, cmp, neg);
2553                 mutex_unlock(&trace_types_lock);
2554                 if (ret)
2555                         return ret;
2556         }
2557
2558         *ppos += cnt;
2559
2560         return cnt;
2561 }
2562
2563 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2564 {
2565         if (tracing_disabled)
2566                 return -ENODEV;
2567         return single_open(file, tracing_trace_options_show, NULL);
2568 }
2569
2570 static const struct file_operations tracing_iter_fops = {
2571         .open           = tracing_trace_options_open,
2572         .read           = seq_read,
2573         .llseek         = seq_lseek,
2574         .release        = single_release,
2575         .write          = tracing_trace_options_write,
2576 };
2577
2578 static const char readme_msg[] =
2579         "tracing mini-HOWTO:\n\n"
2580         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2581         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2582         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2583         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2584         "nop\n"
2585         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2586         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2587         "sched_switch\n"
2588         "# cat /sys/kernel/debug/tracing/trace_options\n"
2589         "noprint-parent nosym-offset nosym-addr noverbose\n"
2590         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2591         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2592         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2593         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2594 ;
2595
2596 static ssize_t
2597 tracing_readme_read(struct file *filp, char __user *ubuf,
2598                        size_t cnt, loff_t *ppos)
2599 {
2600         return simple_read_from_buffer(ubuf, cnt, ppos,
2601                                         readme_msg, strlen(readme_msg));
2602 }
2603
2604 static const struct file_operations tracing_readme_fops = {
2605         .open           = tracing_open_generic,
2606         .read           = tracing_readme_read,
2607         .llseek         = generic_file_llseek,
2608 };
2609
2610 static ssize_t
2611 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2612                                 size_t cnt, loff_t *ppos)
2613 {
2614         char *buf_comm;
2615         char *file_buf;
2616         char *buf;
2617         int len = 0;
2618         int pid;
2619         int i;
2620
2621         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2622         if (!file_buf)
2623                 return -ENOMEM;
2624
2625         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2626         if (!buf_comm) {
2627                 kfree(file_buf);
2628                 return -ENOMEM;
2629         }
2630
2631         buf = file_buf;
2632
2633         for (i = 0; i < SAVED_CMDLINES; i++) {
2634                 int r;
2635
2636                 pid = map_cmdline_to_pid[i];
2637                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2638                         continue;
2639
2640                 trace_find_cmdline(pid, buf_comm);
2641                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2642                 buf += r;
2643                 len += r;
2644         }
2645
2646         len = simple_read_from_buffer(ubuf, cnt, ppos,
2647                                       file_buf, len);
2648
2649         kfree(file_buf);
2650         kfree(buf_comm);
2651
2652         return len;
2653 }
2654
2655 static const struct file_operations tracing_saved_cmdlines_fops = {
2656     .open       = tracing_open_generic,
2657     .read       = tracing_saved_cmdlines_read,
2658     .llseek     = generic_file_llseek,
2659 };
2660
2661 static ssize_t
2662 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2663                   size_t cnt, loff_t *ppos)
2664 {
2665         char buf[64];
2666         int r;
2667
2668         r = sprintf(buf, "%u\n", tracer_enabled);
2669         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2670 }
2671
2672 static ssize_t
2673 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2674                    size_t cnt, loff_t *ppos)
2675 {
2676         struct trace_array *tr = filp->private_data;
2677         char buf[64];
2678         unsigned long val;
2679         int ret;
2680
2681         if (cnt >= sizeof(buf))
2682                 return -EINVAL;
2683
2684         if (copy_from_user(&buf, ubuf, cnt))
2685                 return -EFAULT;
2686
2687         buf[cnt] = 0;
2688
2689         ret = strict_strtoul(buf, 10, &val);
2690         if (ret < 0)
2691                 return ret;
2692
2693         val = !!val;
2694
2695         mutex_lock(&trace_types_lock);
2696         if (tracer_enabled ^ val) {
2697                 if (val) {
2698                         tracer_enabled = 1;
2699                         if (current_trace->start)
2700                                 current_trace->start(tr);
2701                         tracing_start();
2702                 } else {
2703                         tracer_enabled = 0;
2704                         tracing_stop();
2705                         if (current_trace->stop)
2706                                 current_trace->stop(tr);
2707                 }
2708         }
2709         mutex_unlock(&trace_types_lock);
2710
2711         *ppos += cnt;
2712
2713         return cnt;
2714 }
2715
2716 static ssize_t
2717 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2718                        size_t cnt, loff_t *ppos)
2719 {
2720         char buf[MAX_TRACER_SIZE+2];
2721         int r;
2722
2723         mutex_lock(&trace_types_lock);
2724         if (current_trace)
2725                 r = sprintf(buf, "%s\n", current_trace->name);
2726         else
2727                 r = sprintf(buf, "\n");
2728         mutex_unlock(&trace_types_lock);
2729
2730         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2731 }
2732
2733 int tracer_init(struct tracer *t, struct trace_array *tr)
2734 {
2735         tracing_reset_online_cpus(tr);
2736         return t->init(tr);
2737 }
2738
2739 static int tracing_resize_ring_buffer(unsigned long size)
2740 {
2741         int ret;
2742
2743         /*
2744          * If kernel or user changes the size of the ring buffer
2745          * we use the size that was given, and we can forget about
2746          * expanding it later.
2747          */
2748         ring_buffer_expanded = 1;
2749
2750         ret = ring_buffer_resize(global_trace.buffer, size);
2751         if (ret < 0)
2752                 return ret;
2753
2754         if (!current_trace->use_max_tr)
2755                 goto out;
2756
2757         ret = ring_buffer_resize(max_tr.buffer, size);
2758         if (ret < 0) {
2759                 int r;
2760
2761                 r = ring_buffer_resize(global_trace.buffer,
2762                                        global_trace.entries);
2763                 if (r < 0) {
2764                         /*
2765                          * AARGH! We are left with different
2766                          * size max buffer!!!!
2767                          * The max buffer is our "snapshot" buffer.
2768                          * When a tracer needs a snapshot (one of the
2769                          * latency tracers), it swaps the max buffer
2770                          * with the saved snap shot. We succeeded to
2771                          * update the size of the main buffer, but failed to
2772                          * update the size of the max buffer. But when we tried
2773                          * to reset the main buffer to the original size, we
2774                          * failed there too. This is very unlikely to
2775                          * happen, but if it does, warn and kill all
2776                          * tracing.
2777                          */
2778                         WARN_ON(1);
2779                         tracing_disabled = 1;
2780                 }
2781                 return ret;
2782         }
2783
2784         max_tr.entries = size;
2785  out:
2786         global_trace.entries = size;
2787
2788         return ret;
2789 }
2790
2791
2792 /**
2793  * tracing_update_buffers - used by tracing facility to expand ring buffers
2794  *
2795  * To save on memory when the tracing is never used on a system with it
2796  * configured in. The ring buffers are set to a minimum size. But once
2797  * a user starts to use the tracing facility, then they need to grow
2798  * to their default size.
2799  *
2800  * This function is to be called when a tracer is about to be used.
2801  */
2802 int tracing_update_buffers(void)
2803 {
2804         int ret = 0;
2805
2806         mutex_lock(&trace_types_lock);
2807         if (!ring_buffer_expanded)
2808                 ret = tracing_resize_ring_buffer(trace_buf_size);
2809         mutex_unlock(&trace_types_lock);
2810
2811         return ret;
2812 }
2813
2814 struct trace_option_dentry;
2815
2816 static struct trace_option_dentry *
2817 create_trace_option_files(struct tracer *tracer);
2818
2819 static void
2820 destroy_trace_option_files(struct trace_option_dentry *topts);
2821
2822 static int tracing_set_tracer(const char *buf)
2823 {
2824         static struct trace_option_dentry *topts;
2825         struct trace_array *tr = &global_trace;
2826         struct tracer *t;
2827         int ret = 0;
2828
2829         mutex_lock(&trace_types_lock);
2830
2831         if (!ring_buffer_expanded) {
2832                 ret = tracing_resize_ring_buffer(trace_buf_size);
2833                 if (ret < 0)
2834                         goto out;
2835                 ret = 0;
2836         }
2837
2838         for (t = trace_types; t; t = t->next) {
2839                 if (strcmp(t->name, buf) == 0)
2840                         break;
2841         }
2842         if (!t) {
2843                 ret = -EINVAL;
2844                 goto out;
2845         }
2846         if (t == current_trace)
2847                 goto out;
2848
2849         trace_branch_disable();
2850         if (current_trace && current_trace->reset)
2851                 current_trace->reset(tr);
2852         if (current_trace && current_trace->use_max_tr) {
2853                 /*
2854                  * We don't free the ring buffer. instead, resize it because
2855                  * The max_tr ring buffer has some state (e.g. ring->clock) and
2856                  * we want preserve it.
2857                  */
2858                 ring_buffer_resize(max_tr.buffer, 1);
2859                 max_tr.entries = 1;
2860         }
2861         destroy_trace_option_files(topts);
2862
2863         current_trace = t;
2864
2865         topts = create_trace_option_files(current_trace);
2866         if (current_trace->use_max_tr) {
2867                 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
2868                 if (ret < 0)
2869                         goto out;
2870                 max_tr.entries = global_trace.entries;
2871         }
2872
2873         if (t->init) {
2874                 ret = tracer_init(t, tr);
2875                 if (ret)
2876                         goto out;
2877         }
2878
2879         trace_branch_enable(tr);
2880  out:
2881         mutex_unlock(&trace_types_lock);
2882
2883         return ret;
2884 }
2885
2886 static ssize_t
2887 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2888                         size_t cnt, loff_t *ppos)
2889 {
2890         char buf[MAX_TRACER_SIZE+1];
2891         int i;
2892         size_t ret;
2893         int err;
2894
2895         ret = cnt;
2896
2897         if (cnt > MAX_TRACER_SIZE)
2898                 cnt = MAX_TRACER_SIZE;
2899
2900         if (copy_from_user(&buf, ubuf, cnt))
2901                 return -EFAULT;
2902
2903         buf[cnt] = 0;
2904
2905         /* strip ending whitespace. */
2906         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2907                 buf[i] = 0;
2908
2909         err = tracing_set_tracer(buf);
2910         if (err)
2911                 return err;
2912
2913         *ppos += ret;
2914
2915         return ret;
2916 }
2917
2918 static ssize_t
2919 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2920                      size_t cnt, loff_t *ppos)
2921 {
2922         unsigned long *ptr = filp->private_data;
2923         char buf[64];
2924         int r;
2925
2926         r = snprintf(buf, sizeof(buf), "%ld\n",
2927                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2928         if (r > sizeof(buf))
2929                 r = sizeof(buf);
2930         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2931 }
2932
2933 static ssize_t
2934 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2935                       size_t cnt, loff_t *ppos)
2936 {
2937         unsigned long *ptr = filp->private_data;
2938         char buf[64];
2939         unsigned long val;
2940         int ret;
2941
2942         if (cnt >= sizeof(buf))
2943                 return -EINVAL;
2944
2945         if (copy_from_user(&buf, ubuf, cnt))
2946                 return -EFAULT;
2947
2948         buf[cnt] = 0;
2949
2950         ret = strict_strtoul(buf, 10, &val);
2951         if (ret < 0)
2952                 return ret;
2953
2954         *ptr = val * 1000;
2955
2956         return cnt;
2957 }
2958
2959 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2960 {
2961         long cpu_file = (long) inode->i_private;
2962         struct trace_iterator *iter;
2963         int ret = 0;
2964
2965         if (tracing_disabled)
2966                 return -ENODEV;
2967
2968         mutex_lock(&trace_types_lock);
2969
2970         /* create a buffer to store the information to pass to userspace */
2971         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2972         if (!iter) {
2973                 ret = -ENOMEM;
2974                 goto out;
2975         }
2976
2977         /*
2978          * We make a copy of the current tracer to avoid concurrent
2979          * changes on it while we are reading.
2980          */
2981         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2982         if (!iter->trace) {
2983                 ret = -ENOMEM;
2984                 goto fail;
2985         }
2986         if (current_trace)
2987                 *iter->trace = *current_trace;
2988
2989         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2990                 ret = -ENOMEM;
2991                 goto fail;
2992         }
2993
2994         /* trace pipe does not show start of buffer */
2995         cpumask_setall(iter->started);
2996
2997         if (trace_flags & TRACE_ITER_LATENCY_FMT)
2998                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2999
3000         iter->cpu_file = cpu_file;
3001         iter->tr = &global_trace;
3002         mutex_init(&iter->mutex);
3003         filp->private_data = iter;
3004
3005         if (iter->trace->pipe_open)
3006                 iter->trace->pipe_open(iter);
3007
3008         nonseekable_open(inode, filp);
3009 out:
3010         mutex_unlock(&trace_types_lock);
3011         return ret;
3012
3013 fail:
3014         kfree(iter->trace);
3015         kfree(iter);
3016         mutex_unlock(&trace_types_lock);
3017         return ret;
3018 }
3019
3020 static int tracing_release_pipe(struct inode *inode, struct file *file)
3021 {
3022         struct trace_iterator *iter = file->private_data;
3023
3024         mutex_lock(&trace_types_lock);
3025
3026         if (iter->trace->pipe_close)
3027                 iter->trace->pipe_close(iter);
3028
3029         mutex_unlock(&trace_types_lock);
3030
3031         free_cpumask_var(iter->started);
3032         mutex_destroy(&iter->mutex);
3033         kfree(iter->trace);
3034         kfree(iter);
3035
3036         return 0;
3037 }
3038
3039 static unsigned int
3040 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3041 {
3042         struct trace_iterator *iter = filp->private_data;
3043
3044         if (trace_flags & TRACE_ITER_BLOCK) {
3045                 /*
3046                  * Always select as readable when in blocking mode
3047                  */
3048                 return POLLIN | POLLRDNORM;
3049         } else {
3050                 if (!trace_empty(iter))
3051                         return POLLIN | POLLRDNORM;
3052                 poll_wait(filp, &trace_wait, poll_table);
3053                 if (!trace_empty(iter))
3054                         return POLLIN | POLLRDNORM;
3055
3056                 return 0;
3057         }
3058 }
3059
3060
3061 void default_wait_pipe(struct trace_iterator *iter)
3062 {
3063         DEFINE_WAIT(wait);
3064
3065         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3066
3067         if (trace_empty(iter))
3068                 schedule();
3069
3070         finish_wait(&trace_wait, &wait);
3071 }
3072
3073 /*
3074  * This is a make-shift waitqueue.
3075  * A tracer might use this callback on some rare cases:
3076  *
3077  *  1) the current tracer might hold the runqueue lock when it wakes up
3078  *     a reader, hence a deadlock (sched, function, and function graph tracers)
3079  *  2) the function tracers, trace all functions, we don't want
3080  *     the overhead of calling wake_up and friends
3081  *     (and tracing them too)
3082  *
3083  *     Anyway, this is really very primitive wakeup.
3084  */
3085 void poll_wait_pipe(struct trace_iterator *iter)
3086 {
3087         set_current_state(TASK_INTERRUPTIBLE);
3088         /* sleep for 100 msecs, and try again. */
3089         schedule_timeout(HZ / 10);
3090 }
3091
3092 /* Must be called with trace_types_lock mutex held. */
3093 static int tracing_wait_pipe(struct file *filp)
3094 {
3095         struct trace_iterator *iter = filp->private_data;
3096
3097         while (trace_empty(iter)) {
3098
3099                 if ((filp->f_flags & O_NONBLOCK)) {
3100                         return -EAGAIN;
3101                 }
3102
3103                 mutex_unlock(&iter->mutex);
3104
3105                 iter->trace->wait_pipe(iter);
3106
3107                 mutex_lock(&iter->mutex);
3108
3109                 if (signal_pending(current))
3110                         return -EINTR;
3111
3112                 /*
3113                  * We block until we read something and tracing is disabled.
3114                  * We still block if tracing is disabled, but we have never
3115                  * read anything. This allows a user to cat this file, and
3116                  * then enable tracing. But after we have read something,
3117                  * we give an EOF when tracing is again disabled.
3118                  *
3119                  * iter->pos will be 0 if we haven't read anything.
3120                  */
3121                 if (!tracer_enabled && iter->pos)
3122                         break;
3123         }
3124
3125         return 1;
3126 }
3127
3128 /*
3129  * Consumer reader.
3130  */
3131 static ssize_t
3132 tracing_read_pipe(struct file *filp, char __user *ubuf,
3133                   size_t cnt, loff_t *ppos)
3134 {
3135         struct trace_iterator *iter = filp->private_data;
3136         static struct tracer *old_tracer;
3137         ssize_t sret;
3138
3139         /* return any leftover data */
3140         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3141         if (sret != -EBUSY)
3142                 return sret;
3143
3144         trace_seq_init(&iter->seq);
3145
3146         /* copy the tracer to avoid using a global lock all around */
3147         mutex_lock(&trace_types_lock);
3148         if (unlikely(old_tracer != current_trace && current_trace)) {
3149                 old_tracer = current_trace;
3150                 *iter->trace = *current_trace;
3151         }
3152         mutex_unlock(&trace_types_lock);
3153
3154         /*
3155          * Avoid more than one consumer on a single file descriptor
3156          * This is just a matter of traces coherency, the ring buffer itself
3157          * is protected.
3158          */
3159         mutex_lock(&iter->mutex);
3160         if (iter->trace->read) {
3161                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3162                 if (sret)
3163                         goto out;
3164         }
3165
3166 waitagain:
3167         sret = tracing_wait_pipe(filp);
3168         if (sret <= 0)
3169                 goto out;
3170
3171         /* stop when tracing is finished */
3172         if (trace_empty(iter)) {
3173                 sret = 0;
3174                 goto out;
3175         }
3176
3177         if (cnt >= PAGE_SIZE)
3178                 cnt = PAGE_SIZE - 1;
3179
3180         /* reset all but tr, trace, and overruns */
3181         memset(&iter->seq, 0,
3182                sizeof(struct trace_iterator) -
3183                offsetof(struct trace_iterator, seq));
3184         iter->pos = -1;
3185
3186         trace_event_read_lock();
3187         trace_access_lock(iter->cpu_file);
3188         while (trace_find_next_entry_inc(iter) != NULL) {
3189                 enum print_line_t ret;
3190                 int len = iter->seq.len;
3191
3192                 ret = print_trace_line(iter);
3193                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3194                         /* don't print partial lines */
3195                         iter->seq.len = len;
3196                         break;
3197                 }
3198                 if (ret != TRACE_TYPE_NO_CONSUME)
3199                         trace_consume(iter);
3200
3201                 if (iter->seq.len >= cnt)
3202                         break;
3203         }
3204         trace_access_unlock(iter->cpu_file);
3205         trace_event_read_unlock();
3206
3207         /* Now copy what we have to the user */
3208         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3209         if (iter->seq.readpos >= iter->seq.len)
3210                 trace_seq_init(&iter->seq);
3211
3212         /*
3213          * If there was nothing to send to user, inspite of consuming trace
3214          * entries, go back to wait for more entries.
3215          */
3216         if (sret == -EBUSY)
3217                 goto waitagain;
3218
3219 out:
3220         mutex_unlock(&iter->mutex);
3221
3222         return sret;
3223 }
3224
3225 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3226                                      struct pipe_buffer *buf)
3227 {
3228         __free_page(buf->page);
3229 }
3230
3231 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3232                                      unsigned int idx)
3233 {
3234         __free_page(spd->pages[idx]);
3235 }
3236
3237 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3238         .can_merge              = 0,
3239         .map                    = generic_pipe_buf_map,
3240         .unmap                  = generic_pipe_buf_unmap,
3241         .confirm                = generic_pipe_buf_confirm,
3242         .release                = tracing_pipe_buf_release,
3243         .steal                  = generic_pipe_buf_steal,
3244         .get                    = generic_pipe_buf_get,
3245 };
3246
3247 static size_t
3248 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3249 {
3250         size_t count;
3251         int ret;
3252
3253         /* Seq buffer is page-sized, exactly what we need. */
3254         for (;;) {
3255                 count = iter->seq.len;
3256                 ret = print_trace_line(iter);
3257                 count = iter->seq.len - count;
3258                 if (rem < count) {
3259                         rem = 0;
3260                         iter->seq.len -= count;
3261                         break;
3262                 }
3263                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3264                         iter->seq.len -= count;
3265                         break;
3266                 }
3267
3268                 if (ret != TRACE_TYPE_NO_CONSUME)
3269                         trace_consume(iter);
3270                 rem -= count;
3271                 if (!trace_find_next_entry_inc(iter))   {
3272                         rem = 0;
3273                         iter->ent = NULL;
3274                         break;
3275                 }
3276         }
3277
3278         return rem;
3279 }
3280
3281 static ssize_t tracing_splice_read_pipe(struct file *filp,
3282                                         loff_t *ppos,
3283                                         struct pipe_inode_info *pipe,
3284                                         size_t len,
3285                                         unsigned int flags)
3286 {
3287         struct page *pages_def[PIPE_DEF_BUFFERS];
3288         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3289         struct trace_iterator *iter = filp->private_data;
3290         struct splice_pipe_desc spd = {
3291                 .pages          = pages_def,
3292                 .partial        = partial_def,
3293                 .nr_pages       = 0, /* This gets updated below. */
3294                 .flags          = flags,
3295                 .ops            = &tracing_pipe_buf_ops,
3296                 .spd_release    = tracing_spd_release_pipe,
3297         };
3298         static struct tracer *old_tracer;
3299         ssize_t ret;
3300         size_t rem;
3301         unsigned int i;
3302
3303         if (splice_grow_spd(pipe, &spd))
3304                 return -ENOMEM;
3305
3306         /* copy the tracer to avoid using a global lock all around */
3307         mutex_lock(&trace_types_lock);
3308         if (unlikely(old_tracer != current_trace && current_trace)) {
3309                 old_tracer = current_trace;
3310                 *iter->trace = *current_trace;
3311         }
3312         mutex_unlock(&trace_types_lock);
3313
3314         mutex_lock(&iter->mutex);
3315
3316         if (iter->trace->splice_read) {
3317                 ret = iter->trace->splice_read(iter, filp,
3318                                                ppos, pipe, len, flags);
3319                 if (ret)
3320                         goto out_err;
3321         }
3322
3323         ret = tracing_wait_pipe(filp);
3324         if (ret <= 0)
3325                 goto out_err;
3326
3327         if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3328                 ret = -EFAULT;
3329                 goto out_err;
3330         }
3331
3332         trace_event_read_lock();
3333         trace_access_lock(iter->cpu_file);
3334
3335         /* Fill as many pages as possible. */
3336         for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3337                 spd.pages[i] = alloc_page(GFP_KERNEL);
3338                 if (!spd.pages[i])
3339                         break;
3340
3341                 rem = tracing_fill_pipe_page(rem, iter);
3342
3343                 /* Copy the data into the page, so we can start over. */
3344                 ret = trace_seq_to_buffer(&iter->seq,
3345                                           page_address(spd.pages[i]),
3346                                           iter->seq.len);
3347                 if (ret < 0) {
3348                         __free_page(spd.pages[i]);
3349                         break;
3350                 }
3351                 spd.partial[i].offset = 0;
3352                 spd.partial[i].len = iter->seq.len;
3353
3354                 trace_seq_init(&iter->seq);
3355         }
3356
3357         trace_access_unlock(iter->cpu_file);
3358         trace_event_read_unlock();
3359         mutex_unlock(&iter->mutex);
3360
3361         spd.nr_pages = i;
3362
3363         ret = splice_to_pipe(pipe, &spd);
3364 out:
3365         splice_shrink_spd(pipe, &spd);
3366         return ret;
3367
3368 out_err:
3369         mutex_unlock(&iter->mutex);
3370         goto out;
3371 }
3372
3373 static ssize_t
3374 tracing_entries_read(struct file *filp, char __user *ubuf,
3375                      size_t cnt, loff_t *ppos)
3376 {
3377         struct trace_array *tr = filp->private_data;
3378         char buf[96];
3379         int r;
3380
3381         mutex_lock(&trace_types_lock);
3382         if (!ring_buffer_expanded)
3383                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3384                             tr->entries >> 10,
3385                             trace_buf_size >> 10);
3386         else
3387                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3388         mutex_unlock(&trace_types_lock);
3389
3390         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3391 }
3392
3393 static ssize_t
3394 tracing_entries_write(struct file *filp, const char __user *ubuf,
3395                       size_t cnt, loff_t *ppos)
3396 {
3397         unsigned long val;
3398         char buf[64];
3399         int ret, cpu;
3400
3401         if (cnt >= sizeof(buf))
3402                 return -EINVAL;
3403
3404         if (copy_from_user(&buf, ubuf, cnt))
3405                 return -EFAULT;
3406
3407         buf[cnt] = 0;
3408
3409         ret = strict_strtoul(buf, 10, &val);
3410         if (ret < 0)
3411                 return ret;
3412
3413         /* must have at least 1 entry */
3414         if (!val)
3415                 return -EINVAL;
3416
3417         mutex_lock(&trace_types_lock);
3418
3419         tracing_stop();
3420
3421         /* disable all cpu buffers */
3422         for_each_tracing_cpu(cpu) {
3423                 if (global_trace.data[cpu])
3424                         atomic_inc(&global_trace.data[cpu]->disabled);
3425                 if (max_tr.data[cpu])
3426                         atomic_inc(&max_tr.data[cpu]->disabled);
3427         }
3428
3429         /* value is in KB */
3430         val <<= 10;
3431
3432         if (val != global_trace.entries) {
3433                 ret = tracing_resize_ring_buffer(val);
3434                 if (ret < 0) {
3435                         cnt = ret;
3436                         goto out;
3437                 }
3438         }
3439
3440         *ppos += cnt;