Merge branch 'bkl/core' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic...
[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         mutex_lock(&trace_types_lock);
745
746         tracing_selftest_running = true;
747
748         for (t = trace_types; t; t = t->next) {
749                 if (strcmp(type->name, t->name) == 0) {
750                         /* already found */
751                         pr_info("Tracer %s already registered\n",
752                                 type->name);
753                         ret = -1;
754                         goto out;
755                 }
756         }
757
758         if (!type->set_flag)
759                 type->set_flag = &dummy_set_flag;
760         if (!type->flags)
761                 type->flags = &dummy_tracer_flags;
762         else
763                 if (!type->flags->opts)
764                         type->flags->opts = dummy_tracer_opt;
765         if (!type->wait_pipe)
766                 type->wait_pipe = default_wait_pipe;
767
768
769 #ifdef CONFIG_FTRACE_STARTUP_TEST
770         if (type->selftest && !tracing_selftest_disabled) {
771                 struct tracer *saved_tracer = current_trace;
772                 struct trace_array *tr = &global_trace;
773
774                 /*
775                  * Run a selftest on this tracer.
776                  * Here we reset the trace buffer, and set the current
777                  * tracer to be this tracer. The tracer can then run some
778                  * internal tracing to verify that everything is in order.
779                  * If we fail, we do not register this tracer.
780                  */
781                 tracing_reset_online_cpus(tr);
782
783                 current_trace = type;
784                 /* the test is responsible for initializing and enabling */
785                 pr_info("Testing tracer %s: ", type->name);
786                 ret = type->selftest(type, tr);
787                 /* the test is responsible for resetting too */
788                 current_trace = saved_tracer;
789                 if (ret) {
790                         printk(KERN_CONT "FAILED!\n");
791                         goto out;
792                 }
793                 /* Only reset on passing, to avoid touching corrupted buffers */
794                 tracing_reset_online_cpus(tr);
795
796                 printk(KERN_CONT "PASSED\n");
797         }
798 #endif
799
800         type->next = trace_types;
801         trace_types = type;
802
803  out:
804         tracing_selftest_running = false;
805         mutex_unlock(&trace_types_lock);
806
807         if (ret || !default_bootup_tracer)
808                 goto out_unlock;
809
810         if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
811                 goto out_unlock;
812
813         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
814         /* Do we want this tracer to start on bootup? */
815         tracing_set_tracer(type->name);
816         default_bootup_tracer = NULL;
817         /* disable other selftests, since this will break it. */
818         tracing_selftest_disabled = 1;
819 #ifdef CONFIG_FTRACE_STARTUP_TEST
820         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
821                type->name);
822 #endif
823
824  out_unlock:
825         return ret;
826 }
827
828 void unregister_tracer(struct tracer *type)
829 {
830         struct tracer **t;
831
832         mutex_lock(&trace_types_lock);
833         for (t = &trace_types; *t; t = &(*t)->next) {
834                 if (*t == type)
835                         goto found;
836         }
837         pr_info("Tracer %s not registered\n", type->name);
838         goto out;
839
840  found:
841         *t = (*t)->next;
842
843         if (type == current_trace && tracer_enabled) {
844                 tracer_enabled = 0;
845                 tracing_stop();
846                 if (current_trace->stop)
847                         current_trace->stop(&global_trace);
848                 current_trace = &nop_trace;
849         }
850 out:
851         mutex_unlock(&trace_types_lock);
852 }
853
854 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
855 {
856         ftrace_disable_cpu();
857         ring_buffer_reset_cpu(buffer, cpu);
858         ftrace_enable_cpu();
859 }
860
861 void tracing_reset(struct trace_array *tr, int cpu)
862 {
863         struct ring_buffer *buffer = tr->buffer;
864
865         ring_buffer_record_disable(buffer);
866
867         /* Make sure all commits have finished */
868         synchronize_sched();
869         __tracing_reset(buffer, cpu);
870
871         ring_buffer_record_enable(buffer);
872 }
873
874 void tracing_reset_online_cpus(struct trace_array *tr)
875 {
876         struct ring_buffer *buffer = tr->buffer;
877         int cpu;
878
879         ring_buffer_record_disable(buffer);
880
881         /* Make sure all commits have finished */
882         synchronize_sched();
883
884         tr->time_start = ftrace_now(tr->cpu);
885
886         for_each_online_cpu(cpu)
887                 __tracing_reset(buffer, cpu);
888
889         ring_buffer_record_enable(buffer);
890 }
891
892 void tracing_reset_current(int cpu)
893 {
894         tracing_reset(&global_trace, cpu);
895 }
896
897 void tracing_reset_current_online_cpus(void)
898 {
899         tracing_reset_online_cpus(&global_trace);
900 }
901
902 #define SAVED_CMDLINES 128
903 #define NO_CMDLINE_MAP UINT_MAX
904 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
905 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
906 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
907 static int cmdline_idx;
908 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
909
910 /* temporary disable recording */
911 static atomic_t trace_record_cmdline_disabled __read_mostly;
912
913 static void trace_init_cmdlines(void)
914 {
915         memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
916         memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
917         cmdline_idx = 0;
918 }
919
920 int is_tracing_stopped(void)
921 {
922         return trace_stop_count;
923 }
924
925 /**
926  * ftrace_off_permanent - disable all ftrace code permanently
927  *
928  * This should only be called when a serious anomally has
929  * been detected.  This will turn off the function tracing,
930  * ring buffers, and other tracing utilites. It takes no
931  * locks and can be called from any context.
932  */
933 void ftrace_off_permanent(void)
934 {
935         tracing_disabled = 1;
936         ftrace_stop();
937         tracing_off_permanent();
938 }
939
940 /**
941  * tracing_start - quick start of the tracer
942  *
943  * If tracing is enabled but was stopped by tracing_stop,
944  * this will start the tracer back up.
945  */
946 void tracing_start(void)
947 {
948         struct ring_buffer *buffer;
949         unsigned long flags;
950
951         if (tracing_disabled)
952                 return;
953
954         spin_lock_irqsave(&tracing_start_lock, flags);
955         if (--trace_stop_count) {
956                 if (trace_stop_count < 0) {
957                         /* Someone screwed up their debugging */
958                         WARN_ON_ONCE(1);
959                         trace_stop_count = 0;
960                 }
961                 goto out;
962         }
963
964         /* Prevent the buffers from switching */
965         arch_spin_lock(&ftrace_max_lock);
966
967         buffer = global_trace.buffer;
968         if (buffer)
969                 ring_buffer_record_enable(buffer);
970
971         buffer = max_tr.buffer;
972         if (buffer)
973                 ring_buffer_record_enable(buffer);
974
975         arch_spin_unlock(&ftrace_max_lock);
976
977         ftrace_start();
978  out:
979         spin_unlock_irqrestore(&tracing_start_lock, flags);
980 }
981
982 /**
983  * tracing_stop - quick stop of the tracer
984  *
985  * Light weight way to stop tracing. Use in conjunction with
986  * tracing_start.
987  */
988 void tracing_stop(void)
989 {
990         struct ring_buffer *buffer;
991         unsigned long flags;
992
993         ftrace_stop();
994         spin_lock_irqsave(&tracing_start_lock, flags);
995         if (trace_stop_count++)
996                 goto out;
997
998         /* Prevent the buffers from switching */
999         arch_spin_lock(&ftrace_max_lock);
1000
1001         buffer = global_trace.buffer;
1002         if (buffer)
1003                 ring_buffer_record_disable(buffer);
1004
1005         buffer = max_tr.buffer;
1006         if (buffer)
1007                 ring_buffer_record_disable(buffer);
1008
1009         arch_spin_unlock(&ftrace_max_lock);
1010
1011  out:
1012         spin_unlock_irqrestore(&tracing_start_lock, flags);
1013 }
1014
1015 void trace_stop_cmdline_recording(void);
1016
1017 static void trace_save_cmdline(struct task_struct *tsk)
1018 {
1019         unsigned pid, idx;
1020
1021         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1022                 return;
1023
1024         /*
1025          * It's not the end of the world if we don't get
1026          * the lock, but we also don't want to spin
1027          * nor do we want to disable interrupts,
1028          * so if we miss here, then better luck next time.
1029          */
1030         if (!arch_spin_trylock(&trace_cmdline_lock))
1031                 return;
1032
1033         idx = map_pid_to_cmdline[tsk->pid];
1034         if (idx == NO_CMDLINE_MAP) {
1035                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1036
1037                 /*
1038                  * Check whether the cmdline buffer at idx has a pid
1039                  * mapped. We are going to overwrite that entry so we
1040                  * need to clear the map_pid_to_cmdline. Otherwise we
1041                  * would read the new comm for the old pid.
1042                  */
1043                 pid = map_cmdline_to_pid[idx];
1044                 if (pid != NO_CMDLINE_MAP)
1045                         map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1046
1047                 map_cmdline_to_pid[idx] = tsk->pid;
1048                 map_pid_to_cmdline[tsk->pid] = idx;
1049
1050                 cmdline_idx = idx;
1051         }
1052
1053         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1054
1055         arch_spin_unlock(&trace_cmdline_lock);
1056 }
1057
1058 void trace_find_cmdline(int pid, char comm[])
1059 {
1060         unsigned map;
1061
1062         if (!pid) {
1063                 strcpy(comm, "<idle>");
1064                 return;
1065         }
1066
1067         if (WARN_ON_ONCE(pid < 0)) {
1068                 strcpy(comm, "<XXX>");
1069                 return;
1070         }
1071
1072         if (pid > PID_MAX_DEFAULT) {
1073                 strcpy(comm, "<...>");
1074                 return;
1075         }
1076
1077         preempt_disable();
1078         arch_spin_lock(&trace_cmdline_lock);
1079         map = map_pid_to_cmdline[pid];
1080         if (map != NO_CMDLINE_MAP)
1081                 strcpy(comm, saved_cmdlines[map]);
1082         else
1083                 strcpy(comm, "<...>");
1084
1085         arch_spin_unlock(&trace_cmdline_lock);
1086         preempt_enable();
1087 }
1088
1089 void tracing_record_cmdline(struct task_struct *tsk)
1090 {
1091         if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1092             !tracing_is_on())
1093                 return;
1094
1095         trace_save_cmdline(tsk);
1096 }
1097
1098 void
1099 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1100                              int pc)
1101 {
1102         struct task_struct *tsk = current;
1103
1104         entry->preempt_count            = pc & 0xff;
1105         entry->pid                      = (tsk) ? tsk->pid : 0;
1106         entry->lock_depth               = (tsk) ? tsk->lock_depth : 0;
1107         entry->flags =
1108 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1109                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1110 #else
1111                 TRACE_FLAG_IRQS_NOSUPPORT |
1112 #endif
1113                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1114                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1115                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1116 }
1117 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1118
1119 struct ring_buffer_event *
1120 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1121                           int type,
1122                           unsigned long len,
1123                           unsigned long flags, int pc)
1124 {
1125         struct ring_buffer_event *event;
1126
1127         event = ring_buffer_lock_reserve(buffer, len);
1128         if (event != NULL) {
1129                 struct trace_entry *ent = ring_buffer_event_data(event);
1130
1131                 tracing_generic_entry_update(ent, flags, pc);
1132                 ent->type = type;
1133         }
1134
1135         return event;
1136 }
1137
1138 static inline void
1139 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1140                              struct ring_buffer_event *event,
1141                              unsigned long flags, int pc,
1142                              int wake)
1143 {
1144         ring_buffer_unlock_commit(buffer, event);
1145
1146         ftrace_trace_stack(buffer, flags, 6, pc);
1147         ftrace_trace_userstack(buffer, flags, pc);
1148
1149         if (wake)
1150                 trace_wake_up();
1151 }
1152
1153 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1154                                 struct ring_buffer_event *event,
1155                                 unsigned long flags, int pc)
1156 {
1157         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1158 }
1159
1160 struct ring_buffer_event *
1161 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1162                                   int type, unsigned long len,
1163                                   unsigned long flags, int pc)
1164 {
1165         *current_rb = global_trace.buffer;
1166         return trace_buffer_lock_reserve(*current_rb,
1167                                          type, len, flags, pc);
1168 }
1169 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1170
1171 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1172                                         struct ring_buffer_event *event,
1173                                         unsigned long flags, int pc)
1174 {
1175         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1176 }
1177 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1178
1179 void trace_nowake_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, 0);
1184 }
1185 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1186
1187 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1188                                          struct ring_buffer_event *event)
1189 {
1190         ring_buffer_discard_commit(buffer, event);
1191 }
1192 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1193
1194 void
1195 trace_function(struct trace_array *tr,
1196                unsigned long ip, unsigned long parent_ip, unsigned long flags,
1197                int pc)
1198 {
1199         struct ftrace_event_call *call = &event_function;
1200         struct ring_buffer *buffer = tr->buffer;
1201         struct ring_buffer_event *event;
1202         struct ftrace_entry *entry;
1203
1204         /* If we are reading the ring buffer, don't trace */
1205         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1206                 return;
1207
1208         event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1209                                           flags, pc);
1210         if (!event)
1211                 return;
1212         entry   = ring_buffer_event_data(event);
1213         entry->ip                       = ip;
1214         entry->parent_ip                = parent_ip;
1215
1216         if (!filter_check_discard(call, entry, buffer, event))
1217                 ring_buffer_unlock_commit(buffer, event);
1218 }
1219
1220 void
1221 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1222        unsigned long ip, unsigned long parent_ip, unsigned long flags,
1223        int pc)
1224 {
1225         if (likely(!atomic_read(&data->disabled)))
1226                 trace_function(tr, ip, parent_ip, flags, pc);
1227 }
1228
1229 #ifdef CONFIG_STACKTRACE
1230 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1231                                  unsigned long flags,
1232                                  int skip, int pc)
1233 {
1234         struct ftrace_event_call *call = &event_kernel_stack;
1235         struct ring_buffer_event *event;
1236         struct stack_entry *entry;
1237         struct stack_trace trace;
1238
1239         event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1240                                           sizeof(*entry), flags, pc);
1241         if (!event)
1242                 return;
1243         entry   = ring_buffer_event_data(event);
1244         memset(&entry->caller, 0, sizeof(entry->caller));
1245
1246         trace.nr_entries        = 0;
1247         trace.max_entries       = FTRACE_STACK_ENTRIES;
1248         trace.skip              = skip;
1249         trace.entries           = entry->caller;
1250
1251         save_stack_trace(&trace);
1252         if (!filter_check_discard(call, entry, buffer, event))
1253                 ring_buffer_unlock_commit(buffer, event);
1254 }
1255
1256 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1257                         int skip, int pc)
1258 {
1259         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1260                 return;
1261
1262         __ftrace_trace_stack(buffer, flags, skip, pc);
1263 }
1264
1265 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1266                    int pc)
1267 {
1268         __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1269 }
1270
1271 /**
1272  * trace_dump_stack - record a stack back trace in the trace buffer
1273  */
1274 void trace_dump_stack(void)
1275 {
1276         unsigned long flags;
1277
1278         if (tracing_disabled || tracing_selftest_running)
1279                 return;
1280
1281         local_save_flags(flags);
1282
1283         /* skipping 3 traces, seems to get us at the caller of this function */
1284         __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1285 }
1286
1287 void
1288 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1289 {
1290         struct ftrace_event_call *call = &event_user_stack;
1291         struct ring_buffer_event *event;
1292         struct userstack_entry *entry;
1293         struct stack_trace trace;
1294
1295         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1296                 return;
1297
1298         /*
1299          * NMIs can not handle page faults, even with fix ups.
1300          * The save user stack can (and often does) fault.
1301          */
1302         if (unlikely(in_nmi()))
1303                 return;
1304
1305         event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1306                                           sizeof(*entry), flags, pc);
1307         if (!event)
1308                 return;
1309         entry   = ring_buffer_event_data(event);
1310
1311         entry->tgid             = current->tgid;
1312         memset(&entry->caller, 0, sizeof(entry->caller));
1313
1314         trace.nr_entries        = 0;
1315         trace.max_entries       = FTRACE_STACK_ENTRIES;
1316         trace.skip              = 0;
1317         trace.entries           = entry->caller;
1318
1319         save_stack_trace_user(&trace);
1320         if (!filter_check_discard(call, entry, buffer, event))
1321                 ring_buffer_unlock_commit(buffer, event);
1322 }
1323
1324 #ifdef UNUSED
1325 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1326 {
1327         ftrace_trace_userstack(tr, flags, preempt_count());
1328 }
1329 #endif /* UNUSED */
1330
1331 #endif /* CONFIG_STACKTRACE */
1332
1333 /**
1334  * trace_vbprintk - write binary msg to tracing buffer
1335  *
1336  */
1337 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1338 {
1339         static arch_spinlock_t trace_buf_lock =
1340                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1341         static u32 trace_buf[TRACE_BUF_SIZE];
1342
1343         struct ftrace_event_call *call = &event_bprint;
1344         struct ring_buffer_event *event;
1345         struct ring_buffer *buffer;
1346         struct trace_array *tr = &global_trace;
1347         struct trace_array_cpu *data;
1348         struct bprint_entry *entry;
1349         unsigned long flags;
1350         int disable;
1351         int cpu, len = 0, size, pc;
1352
1353         if (unlikely(tracing_selftest_running || tracing_disabled))
1354                 return 0;
1355
1356         /* Don't pollute graph traces with trace_vprintk internals */
1357         pause_graph_tracing();
1358
1359         pc = preempt_count();
1360         preempt_disable_notrace();
1361         cpu = raw_smp_processor_id();
1362         data = tr->data[cpu];
1363
1364         disable = atomic_inc_return(&data->disabled);
1365         if (unlikely(disable != 1))
1366                 goto out;
1367
1368         /* Lockdep uses trace_printk for lock tracing */
1369         local_irq_save(flags);
1370         arch_spin_lock(&trace_buf_lock);
1371         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1372
1373         if (len > TRACE_BUF_SIZE || len < 0)
1374                 goto out_unlock;
1375
1376         size = sizeof(*entry) + sizeof(u32) * len;
1377         buffer = tr->buffer;
1378         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1379                                           flags, pc);
1380         if (!event)
1381                 goto out_unlock;
1382         entry = ring_buffer_event_data(event);
1383         entry->ip                       = ip;
1384         entry->fmt                      = fmt;
1385
1386         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1387         if (!filter_check_discard(call, entry, buffer, event)) {
1388                 ring_buffer_unlock_commit(buffer, event);
1389                 ftrace_trace_stack(buffer, flags, 6, pc);
1390         }
1391
1392 out_unlock:
1393         arch_spin_unlock(&trace_buf_lock);
1394         local_irq_restore(flags);
1395
1396 out:
1397         atomic_dec_return(&data->disabled);
1398         preempt_enable_notrace();
1399         unpause_graph_tracing();
1400
1401         return len;
1402 }
1403 EXPORT_SYMBOL_GPL(trace_vbprintk);
1404
1405 int trace_array_printk(struct trace_array *tr,
1406                        unsigned long ip, const char *fmt, ...)
1407 {
1408         int ret;
1409         va_list ap;
1410
1411         if (!(trace_flags & TRACE_ITER_PRINTK))
1412                 return 0;
1413
1414         va_start(ap, fmt);
1415         ret = trace_array_vprintk(tr, ip, fmt, ap);
1416         va_end(ap);
1417         return ret;
1418 }
1419
1420 int trace_array_vprintk(struct trace_array *tr,
1421                         unsigned long ip, const char *fmt, va_list args)
1422 {
1423         static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1424         static char trace_buf[TRACE_BUF_SIZE];
1425
1426         struct ftrace_event_call *call = &event_print;
1427         struct ring_buffer_event *event;
1428         struct ring_buffer *buffer;
1429         struct trace_array_cpu *data;
1430         int cpu, len = 0, size, pc;
1431         struct print_entry *entry;
1432         unsigned long irq_flags;
1433         int disable;
1434
1435         if (tracing_disabled || tracing_selftest_running)
1436                 return 0;
1437
1438         pc = preempt_count();
1439         preempt_disable_notrace();
1440         cpu = raw_smp_processor_id();
1441         data = tr->data[cpu];
1442
1443         disable = atomic_inc_return(&data->disabled);
1444         if (unlikely(disable != 1))
1445                 goto out;
1446
1447         pause_graph_tracing();
1448         raw_local_irq_save(irq_flags);
1449         arch_spin_lock(&trace_buf_lock);
1450         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1451
1452         size = sizeof(*entry) + len + 1;
1453         buffer = tr->buffer;
1454         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1455                                           irq_flags, pc);
1456         if (!event)
1457                 goto out_unlock;
1458         entry = ring_buffer_event_data(event);
1459         entry->ip = ip;
1460
1461         memcpy(&entry->buf, trace_buf, len);
1462         entry->buf[len] = '\0';
1463         if (!filter_check_discard(call, entry, buffer, event)) {
1464                 ring_buffer_unlock_commit(buffer, event);
1465                 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1466         }
1467
1468  out_unlock:
1469         arch_spin_unlock(&trace_buf_lock);
1470         raw_local_irq_restore(irq_flags);
1471         unpause_graph_tracing();
1472  out:
1473         atomic_dec_return(&data->disabled);
1474         preempt_enable_notrace();
1475
1476         return len;
1477 }
1478
1479 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1480 {
1481         return trace_array_vprintk(&global_trace, ip, fmt, args);
1482 }
1483 EXPORT_SYMBOL_GPL(trace_vprintk);
1484
1485 static void trace_iterator_increment(struct trace_iterator *iter)
1486 {
1487         /* Don't allow ftrace to trace into the ring buffers */
1488         ftrace_disable_cpu();
1489
1490         iter->idx++;
1491         if (iter->buffer_iter[iter->cpu])
1492                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1493
1494         ftrace_enable_cpu();
1495 }
1496
1497 static struct trace_entry *
1498 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1499                 unsigned long *lost_events)
1500 {
1501         struct ring_buffer_event *event;
1502         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1503
1504         /* Don't allow ftrace to trace into the ring buffers */
1505         ftrace_disable_cpu();
1506
1507         if (buf_iter)
1508                 event = ring_buffer_iter_peek(buf_iter, ts);
1509         else
1510                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1511                                          lost_events);
1512
1513         ftrace_enable_cpu();
1514
1515         return event ? ring_buffer_event_data(event) : NULL;
1516 }
1517
1518 static struct trace_entry *
1519 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1520                   unsigned long *missing_events, u64 *ent_ts)
1521 {
1522         struct ring_buffer *buffer = iter->tr->buffer;
1523         struct trace_entry *ent, *next = NULL;
1524         unsigned long lost_events = 0, next_lost = 0;
1525         int cpu_file = iter->cpu_file;
1526         u64 next_ts = 0, ts;
1527         int next_cpu = -1;
1528         int cpu;
1529
1530         /*
1531          * If we are in a per_cpu trace file, don't bother by iterating over
1532          * all cpu and peek directly.
1533          */
1534         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1535                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1536                         return NULL;
1537                 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1538                 if (ent_cpu)
1539                         *ent_cpu = cpu_file;
1540
1541                 return ent;
1542         }
1543
1544         for_each_tracing_cpu(cpu) {
1545
1546                 if (ring_buffer_empty_cpu(buffer, cpu))
1547                         continue;
1548
1549                 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1550
1551                 /*
1552                  * Pick the entry with the smallest timestamp:
1553                  */
1554                 if (ent && (!next || ts < next_ts)) {
1555                         next = ent;
1556                         next_cpu = cpu;
1557                         next_ts = ts;
1558                         next_lost = lost_events;
1559                 }
1560         }
1561
1562         if (ent_cpu)
1563                 *ent_cpu = next_cpu;
1564
1565         if (ent_ts)
1566                 *ent_ts = next_ts;
1567
1568         if (missing_events)
1569                 *missing_events = next_lost;
1570
1571         return next;
1572 }
1573
1574 /* Find the next real entry, without updating the iterator itself */
1575 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1576                                           int *ent_cpu, u64 *ent_ts)
1577 {
1578         return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1579 }
1580
1581 /* Find the next real entry, and increment the iterator to the next entry */
1582 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1583 {
1584         iter->ent = __find_next_entry(iter, &iter->cpu,
1585                                       &iter->lost_events, &iter->ts);
1586
1587         if (iter->ent)
1588                 trace_iterator_increment(iter);
1589
1590         return iter->ent ? iter : NULL;
1591 }
1592
1593 static void trace_consume(struct trace_iterator *iter)
1594 {
1595         /* Don't allow ftrace to trace into the ring buffers */
1596         ftrace_disable_cpu();
1597         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1598                             &iter->lost_events);
1599         ftrace_enable_cpu();
1600 }
1601
1602 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1603 {
1604         struct trace_iterator *iter = m->private;
1605         int i = (int)*pos;
1606         void *ent;
1607
1608         WARN_ON_ONCE(iter->leftover);
1609
1610         (*pos)++;
1611
1612         /* can't go backwards */
1613         if (iter->idx > i)
1614                 return NULL;
1615
1616         if (iter->idx < 0)
1617                 ent = trace_find_next_entry_inc(iter);
1618         else
1619                 ent = iter;
1620
1621         while (ent && iter->idx < i)
1622                 ent = trace_find_next_entry_inc(iter);
1623
1624         iter->pos = *pos;
1625
1626         return ent;
1627 }
1628
1629 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1630 {
1631         struct trace_array *tr = iter->tr;
1632         struct ring_buffer_event *event;
1633         struct ring_buffer_iter *buf_iter;
1634         unsigned long entries = 0;
1635         u64 ts;
1636
1637         tr->data[cpu]->skipped_entries = 0;
1638
1639         if (!iter->buffer_iter[cpu])
1640                 return;
1641
1642         buf_iter = iter->buffer_iter[cpu];
1643         ring_buffer_iter_reset(buf_iter);
1644
1645         /*
1646          * We could have the case with the max latency tracers
1647          * that a reset never took place on a cpu. This is evident
1648          * by the timestamp being before the start of the buffer.
1649          */
1650         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1651                 if (ts >= iter->tr->time_start)
1652                         break;
1653                 entries++;
1654                 ring_buffer_read(buf_iter, NULL);
1655         }
1656
1657         tr->data[cpu]->skipped_entries = entries;
1658 }
1659
1660 /*
1661  * The current tracer is copied to avoid a global locking
1662  * all around.
1663  */
1664 static void *s_start(struct seq_file *m, loff_t *pos)
1665 {
1666         struct trace_iterator *iter = m->private;
1667         static struct tracer *old_tracer;
1668         int cpu_file = iter->cpu_file;
1669         void *p = NULL;
1670         loff_t l = 0;
1671         int cpu;
1672
1673         /* copy the tracer to avoid using a global lock all around */
1674         mutex_lock(&trace_types_lock);
1675         if (unlikely(old_tracer != current_trace && current_trace)) {
1676                 old_tracer = current_trace;
1677                 *iter->trace = *current_trace;
1678         }
1679         mutex_unlock(&trace_types_lock);
1680
1681         atomic_inc(&trace_record_cmdline_disabled);
1682
1683         if (*pos != iter->pos) {
1684                 iter->ent = NULL;
1685                 iter->cpu = 0;
1686                 iter->idx = -1;
1687
1688                 ftrace_disable_cpu();
1689
1690                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1691                         for_each_tracing_cpu(cpu)
1692                                 tracing_iter_reset(iter, cpu);
1693                 } else
1694                         tracing_iter_reset(iter, cpu_file);
1695
1696                 ftrace_enable_cpu();
1697
1698                 iter->leftover = 0;
1699                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1700                         ;
1701
1702         } else {
1703                 /*
1704                  * If we overflowed the seq_file before, then we want
1705                  * to just reuse the trace_seq buffer again.
1706                  */
1707                 if (iter->leftover)
1708                         p = iter;
1709                 else {
1710                         l = *pos - 1;
1711                         p = s_next(m, p, &l);
1712                 }
1713         }
1714
1715         trace_event_read_lock();
1716         trace_access_lock(cpu_file);
1717         return p;
1718 }
1719
1720 static void s_stop(struct seq_file *m, void *p)
1721 {
1722         struct trace_iterator *iter = m->private;
1723
1724         atomic_dec(&trace_record_cmdline_disabled);
1725         trace_access_unlock(iter->cpu_file);
1726         trace_event_read_unlock();
1727 }
1728
1729 static void print_lat_help_header(struct seq_file *m)
1730 {
1731         seq_puts(m, "#                  _------=> CPU#            \n");
1732         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1733         seq_puts(m, "#                | / _----=> need-resched    \n");
1734         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1735         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1736         seq_puts(m, "#                |||| /_--=> lock-depth       \n");
1737         seq_puts(m, "#                |||||/     delay             \n");
1738         seq_puts(m, "#  cmd     pid   |||||| time  |   caller      \n");
1739         seq_puts(m, "#     \\   /      ||||||   \\   |   /           \n");
1740 }
1741
1742 static void print_func_help_header(struct seq_file *m)
1743 {
1744         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1745         seq_puts(m, "#              | |       |          |         |\n");
1746 }
1747
1748
1749 void
1750 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1751 {
1752         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1753         struct trace_array *tr = iter->tr;
1754         struct trace_array_cpu *data = tr->data[tr->cpu];
1755         struct tracer *type = current_trace;
1756         unsigned long entries = 0;
1757         unsigned long total = 0;
1758         unsigned long count;
1759         const char *name = "preemption";
1760         int cpu;
1761
1762         if (type)
1763                 name = type->name;
1764
1765
1766         for_each_tracing_cpu(cpu) {
1767                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1768                 /*
1769                  * If this buffer has skipped entries, then we hold all
1770                  * entries for the trace and we need to ignore the
1771                  * ones before the time stamp.
1772                  */
1773                 if (tr->data[cpu]->skipped_entries) {
1774                         count -= tr->data[cpu]->skipped_entries;
1775                         /* total is the same as the entries */
1776                         total += count;
1777                 } else
1778                         total += count +
1779                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1780                 entries += count;
1781         }
1782
1783         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1784                    name, UTS_RELEASE);
1785         seq_puts(m, "# -----------------------------------"
1786                  "---------------------------------\n");
1787         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1788                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1789                    nsecs_to_usecs(data->saved_latency),
1790                    entries,
1791                    total,
1792                    tr->cpu,
1793 #if defined(CONFIG_PREEMPT_NONE)
1794                    "server",
1795 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1796                    "desktop",
1797 #elif defined(CONFIG_PREEMPT)
1798                    "preempt",
1799 #else
1800                    "unknown",
1801 #endif
1802                    /* These are reserved for later use */
1803                    0, 0, 0, 0);
1804 #ifdef CONFIG_SMP
1805         seq_printf(m, " #P:%d)\n", num_online_cpus());
1806 #else
1807         seq_puts(m, ")\n");
1808 #endif
1809         seq_puts(m, "#    -----------------\n");
1810         seq_printf(m, "#    | task: %.16s-%d "
1811                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1812                    data->comm, data->pid, data->uid, data->nice,
1813                    data->policy, data->rt_priority);
1814         seq_puts(m, "#    -----------------\n");
1815
1816         if (data->critical_start) {
1817                 seq_puts(m, "#  => started at: ");
1818                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1819                 trace_print_seq(m, &iter->seq);
1820                 seq_puts(m, "\n#  => ended at:   ");
1821                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1822                 trace_print_seq(m, &iter->seq);
1823                 seq_puts(m, "\n#\n");
1824         }
1825
1826         seq_puts(m, "#\n");
1827 }
1828
1829 static void test_cpu_buff_start(struct trace_iterator *iter)
1830 {
1831         struct trace_seq *s = &iter->seq;
1832
1833         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1834                 return;
1835
1836         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1837                 return;
1838
1839         if (cpumask_test_cpu(iter->cpu, iter->started))
1840                 return;
1841
1842         if (iter->tr->data[iter->cpu]->skipped_entries)
1843                 return;
1844
1845         cpumask_set_cpu(iter->cpu, iter->started);
1846
1847         /* Don't print started cpu buffer for the first entry of the trace */
1848         if (iter->idx > 1)
1849                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1850                                 iter->cpu);
1851 }
1852
1853 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1854 {
1855         struct trace_seq *s = &iter->seq;
1856         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1857         struct trace_entry *entry;
1858         struct trace_event *event;
1859
1860         entry = iter->ent;
1861
1862         test_cpu_buff_start(iter);
1863
1864         event = ftrace_find_event(entry->type);
1865
1866         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1867                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1868                         if (!trace_print_lat_context(iter))
1869                                 goto partial;
1870                 } else {
1871                         if (!trace_print_context(iter))
1872                                 goto partial;
1873                 }
1874         }
1875
1876         if (event)
1877                 return event->funcs->trace(iter, sym_flags, event);
1878
1879         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1880                 goto partial;
1881
1882         return TRACE_TYPE_HANDLED;
1883 partial:
1884         return TRACE_TYPE_PARTIAL_LINE;
1885 }
1886
1887 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1888 {
1889         struct trace_seq *s = &iter->seq;
1890         struct trace_entry *entry;
1891         struct trace_event *event;
1892
1893         entry = iter->ent;
1894
1895         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1896                 if (!trace_seq_printf(s, "%d %d %llu ",
1897                                       entry->pid, iter->cpu, iter->ts))
1898                         goto partial;
1899         }
1900
1901         event = ftrace_find_event(entry->type);
1902         if (event)
1903                 return event->funcs->raw(iter, 0, event);
1904
1905         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1906                 goto partial;
1907
1908         return TRACE_TYPE_HANDLED;
1909 partial:
1910         return TRACE_TYPE_PARTIAL_LINE;
1911 }
1912
1913 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1914 {
1915         struct trace_seq *s = &iter->seq;
1916         unsigned char newline = '\n';
1917         struct trace_entry *entry;
1918         struct trace_event *event;
1919
1920         entry = iter->ent;
1921
1922         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1923                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1924                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1925                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1926         }
1927
1928         event = ftrace_find_event(entry->type);
1929         if (event) {
1930                 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1931                 if (ret != TRACE_TYPE_HANDLED)
1932                         return ret;
1933         }
1934
1935         SEQ_PUT_FIELD_RET(s, newline);
1936
1937         return TRACE_TYPE_HANDLED;
1938 }
1939
1940 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1941 {
1942         struct trace_seq *s = &iter->seq;
1943         struct trace_entry *entry;
1944         struct trace_event *event;
1945
1946         entry = iter->ent;
1947
1948         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1949                 SEQ_PUT_FIELD_RET(s, entry->pid);
1950                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1951                 SEQ_PUT_FIELD_RET(s, iter->ts);
1952         }
1953
1954         event = ftrace_find_event(entry->type);
1955         return event ? event->funcs->binary(iter, 0, event) :
1956                 TRACE_TYPE_HANDLED;
1957 }
1958
1959 int trace_empty(struct trace_iterator *iter)
1960 {
1961         int cpu;
1962
1963         /* If we are looking at one CPU buffer, only check that one */
1964         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1965                 cpu = iter->cpu_file;
1966                 if (iter->buffer_iter[cpu]) {
1967                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1968                                 return 0;
1969                 } else {
1970                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1971                                 return 0;
1972                 }
1973                 return 1;
1974         }
1975
1976         for_each_tracing_cpu(cpu) {
1977                 if (iter->buffer_iter[cpu]) {
1978                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1979                                 return 0;
1980                 } else {
1981                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1982                                 return 0;
1983                 }
1984         }
1985
1986         return 1;
1987 }
1988
1989 /*  Called with trace_event_read_lock() held. */
1990 enum print_line_t print_trace_line(struct trace_iterator *iter)
1991 {
1992         enum print_line_t ret;
1993
1994         if (iter->lost_events)
1995                 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
1996                                  iter->cpu, iter->lost_events);
1997
1998         if (iter->trace && iter->trace->print_line) {
1999                 ret = iter->trace->print_line(iter);
2000                 if (ret != TRACE_TYPE_UNHANDLED)
2001                         return ret;
2002         }
2003
2004         if (iter->ent->type == TRACE_BPRINT &&
2005                         trace_flags & TRACE_ITER_PRINTK &&
2006                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2007                 return trace_print_bprintk_msg_only(iter);
2008
2009         if (iter->ent->type == TRACE_PRINT &&
2010                         trace_flags & TRACE_ITER_PRINTK &&
2011                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2012                 return trace_print_printk_msg_only(iter);
2013
2014         if (trace_flags & TRACE_ITER_BIN)
2015                 return print_bin_fmt(iter);
2016
2017         if (trace_flags & TRACE_ITER_HEX)
2018                 return print_hex_fmt(iter);
2019
2020         if (trace_flags & TRACE_ITER_RAW)
2021                 return print_raw_fmt(iter);
2022
2023         return print_trace_fmt(iter);
2024 }
2025
2026 void trace_default_header(struct seq_file *m)
2027 {
2028         struct trace_iterator *iter = m->private;
2029
2030         if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2031                 /* print nothing if the buffers are empty */
2032                 if (trace_empty(iter))
2033                         return;
2034                 print_trace_header(m, iter);
2035                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2036                         print_lat_help_header(m);
2037         } else {
2038                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2039                         print_func_help_header(m);
2040         }
2041 }
2042
2043 static int s_show(struct seq_file *m, void *v)
2044 {
2045         struct trace_iterator *iter = v;
2046         int ret;
2047
2048         if (iter->ent == NULL) {
2049                 if (iter->tr) {
2050                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2051                         seq_puts(m, "#\n");
2052                 }
2053                 if (iter->trace && iter->trace->print_header)
2054                         iter->trace->print_header(m);
2055                 else
2056                         trace_default_header(m);
2057
2058         } else if (iter->leftover) {
2059                 /*
2060                  * If we filled the seq_file buffer earlier, we
2061                  * want to just show it now.
2062                  */
2063                 ret = trace_print_seq(m, &iter->seq);
2064
2065                 /* ret should this time be zero, but you never know */
2066                 iter->leftover = ret;
2067
2068         } else {
2069                 print_trace_line(iter);
2070                 ret = trace_print_seq(m, &iter->seq);
2071                 /*
2072                  * If we overflow the seq_file buffer, then it will
2073                  * ask us for this data again at start up.
2074                  * Use that instead.
2075                  *  ret is 0 if seq_file write succeeded.
2076                  *        -1 otherwise.
2077                  */
2078                 iter->leftover = ret;
2079         }
2080
2081         return 0;
2082 }
2083
2084 static const struct seq_operations tracer_seq_ops = {
2085         .start          = s_start,
2086         .next           = s_next,
2087         .stop           = s_stop,
2088         .show           = s_show,
2089 };
2090
2091 static struct trace_iterator *
2092 __tracing_open(struct inode *inode, struct file *file)
2093 {
2094         long cpu_file = (long) inode->i_private;
2095         void *fail_ret = ERR_PTR(-ENOMEM);
2096         struct trace_iterator *iter;
2097         struct seq_file *m;
2098         int cpu, ret;
2099
2100         if (tracing_disabled)
2101                 return ERR_PTR(-ENODEV);
2102
2103         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2104         if (!iter)
2105                 return ERR_PTR(-ENOMEM);
2106
2107         /*
2108          * We make a copy of the current tracer to avoid concurrent
2109          * changes on it while we are reading.
2110          */
2111         mutex_lock(&trace_types_lock);
2112         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2113         if (!iter->trace)
2114                 goto fail;
2115
2116         if (current_trace)
2117                 *iter->trace = *current_trace;
2118
2119         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2120                 goto fail;
2121
2122         if (current_trace && current_trace->print_max)
2123                 iter->tr = &max_tr;
2124         else
2125                 iter->tr = &global_trace;
2126         iter->pos = -1;
2127         mutex_init(&iter->mutex);
2128         iter->cpu_file = cpu_file;
2129
2130         /* Notify the tracer early; before we stop tracing. */
2131         if (iter->trace && iter->trace->open)
2132                 iter->trace->open(iter);
2133
2134         /* Annotate start of buffers if we had overruns */
2135         if (ring_buffer_overruns(iter->tr->buffer))
2136                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2137
2138         /* stop the trace while dumping */
2139         tracing_stop();
2140
2141         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2142                 for_each_tracing_cpu(cpu) {
2143                         iter->buffer_iter[cpu] =
2144                                 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2145                 }
2146                 ring_buffer_read_prepare_sync();
2147                 for_each_tracing_cpu(cpu) {
2148                         ring_buffer_read_start(iter->buffer_iter[cpu]);
2149                         tracing_iter_reset(iter, cpu);
2150                 }
2151         } else {
2152                 cpu = iter->cpu_file;
2153                 iter->buffer_iter[cpu] =
2154                         ring_buffer_read_prepare(iter->tr->buffer, cpu);
2155                 ring_buffer_read_prepare_sync();
2156                 ring_buffer_read_start(iter->buffer_iter[cpu]);
2157                 tracing_iter_reset(iter, cpu);
2158         }
2159
2160         ret = seq_open(file, &tracer_seq_ops);
2161         if (ret < 0) {
2162                 fail_ret = ERR_PTR(ret);
2163                 goto fail_buffer;
2164         }
2165
2166         m = file->private_data;
2167         m->private = iter;
2168
2169         mutex_unlock(&trace_types_lock);
2170
2171         return iter;
2172
2173  fail_buffer:
2174         for_each_tracing_cpu(cpu) {
2175                 if (iter->buffer_iter[cpu])
2176                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2177         }
2178         free_cpumask_var(iter->started);
2179         tracing_start();
2180  fail:
2181         mutex_unlock(&trace_types_lock);
2182         kfree(iter->trace);
2183         kfree(iter);
2184
2185         return fail_ret;
2186 }
2187
2188 int tracing_open_generic(struct inode *inode, struct file *filp)
2189 {
2190         if (tracing_disabled)
2191                 return -ENODEV;
2192
2193         filp->private_data = inode->i_private;
2194         return 0;
2195 }
2196
2197 static int tracing_release(struct inode *inode, struct file *file)
2198 {
2199         struct seq_file *m = (struct seq_file *)file->private_data;
2200         struct trace_iterator *iter;
2201         int cpu;
2202
2203         if (!(file->f_mode & FMODE_READ))
2204                 return 0;
2205
2206         iter = m->private;
2207
2208         mutex_lock(&trace_types_lock);
2209         for_each_tracing_cpu(cpu) {
2210                 if (iter->buffer_iter[cpu])
2211                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2212         }
2213
2214         if (iter->trace && iter->trace->close)
2215                 iter->trace->close(iter);
2216
2217         /* reenable tracing if it was previously enabled */
2218         tracing_start();
2219         mutex_unlock(&trace_types_lock);
2220
2221         seq_release(inode, file);
2222         mutex_destroy(&iter->mutex);
2223         free_cpumask_var(iter->started);
2224         kfree(iter->trace);
2225         kfree(iter);
2226         return 0;
2227 }
2228
2229 static int tracing_open(struct inode *inode, struct file *file)
2230 {
2231         struct trace_iterator *iter;
2232         int ret = 0;
2233
2234         /* If this file was open for write, then erase contents */
2235         if ((file->f_mode & FMODE_WRITE) &&
2236             (file->f_flags & O_TRUNC)) {
2237                 long cpu = (long) inode->i_private;
2238
2239                 if (cpu == TRACE_PIPE_ALL_CPU)
2240                         tracing_reset_online_cpus(&global_trace);
2241                 else
2242                         tracing_reset(&global_trace, cpu);
2243         }
2244
2245         if (file->f_mode & FMODE_READ) {
2246                 iter = __tracing_open(inode, file);
2247                 if (IS_ERR(iter))
2248                         ret = PTR_ERR(iter);
2249                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2250                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2251         }
2252         return ret;
2253 }
2254
2255 static void *
2256 t_next(struct seq_file *m, void *v, loff_t *pos)
2257 {
2258         struct tracer *t = v;
2259
2260         (*pos)++;
2261
2262         if (t)
2263                 t = t->next;
2264
2265         return t;
2266 }
2267
2268 static void *t_start(struct seq_file *m, loff_t *pos)
2269 {
2270         struct tracer *t;
2271         loff_t l = 0;
2272
2273         mutex_lock(&trace_types_lock);
2274         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2275                 ;
2276
2277         return t;
2278 }
2279
2280 static void t_stop(struct seq_file *m, void *p)
2281 {
2282         mutex_unlock(&trace_types_lock);
2283 }
2284
2285 static int t_show(struct seq_file *m, void *v)
2286 {
2287         struct tracer *t = v;
2288
2289         if (!t)
2290                 return 0;
2291
2292         seq_printf(m, "%s", t->name);
2293         if (t->next)
2294                 seq_putc(m, ' ');
2295         else
2296                 seq_putc(m, '\n');
2297
2298         return 0;
2299 }
2300
2301 static const struct seq_operations show_traces_seq_ops = {
2302         .start          = t_start,
2303         .next           = t_next,
2304         .stop           = t_stop,
2305         .show           = t_show,
2306 };
2307
2308 static int show_traces_open(struct inode *inode, struct file *file)
2309 {
2310         if (tracing_disabled)
2311                 return -ENODEV;
2312
2313         return seq_open(file, &show_traces_seq_ops);
2314 }
2315
2316 static ssize_t
2317 tracing_write_stub(struct file *filp, const char __user *ubuf,
2318                    size_t count, loff_t *ppos)
2319 {
2320         return count;
2321 }
2322
2323 static const struct file_operations tracing_fops = {
2324         .open           = tracing_open,
2325         .read           = seq_read,
2326         .write          = tracing_write_stub,
2327         .llseek         = seq_lseek,
2328         .release        = tracing_release,
2329 };
2330
2331 static const struct file_operations show_traces_fops = {
2332         .open           = show_traces_open,
2333         .read           = seq_read,
2334         .release        = seq_release,
2335         .llseek         = seq_lseek,
2336 };
2337
2338 /*
2339  * Only trace on a CPU if the bitmask is set:
2340  */
2341 static cpumask_var_t tracing_cpumask;
2342
2343 /*
2344  * The tracer itself will not take this lock, but still we want
2345  * to provide a consistent cpumask to user-space:
2346  */
2347 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2348
2349 /*
2350  * Temporary storage for the character representation of the
2351  * CPU bitmask (and one more byte for the newline):
2352  */
2353 static char mask_str[NR_CPUS + 1];
2354
2355 static ssize_t
2356 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2357                      size_t count, loff_t *ppos)
2358 {
2359         int len;
2360
2361         mutex_lock(&tracing_cpumask_update_lock);
2362
2363         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2364         if (count - len < 2) {
2365                 count = -EINVAL;
2366                 goto out_err;
2367         }
2368         len += sprintf(mask_str + len, "\n");
2369         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2370
2371 out_err:
2372         mutex_unlock(&tracing_cpumask_update_lock);
2373
2374         return count;
2375 }
2376
2377 static ssize_t
2378 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2379                       size_t count, loff_t *ppos)
2380 {
2381         int err, cpu;
2382         cpumask_var_t tracing_cpumask_new;
2383
2384         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2385                 return -ENOMEM;
2386
2387         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2388         if (err)
2389                 goto err_unlock;
2390
2391         mutex_lock(&tracing_cpumask_update_lock);
2392
2393         local_irq_disable();
2394         arch_spin_lock(&ftrace_max_lock);
2395         for_each_tracing_cpu(cpu) {
2396                 /*
2397                  * Increase/decrease the disabled counter if we are
2398                  * about to flip a bit in the cpumask:
2399                  */
2400                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2401                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2402                         atomic_inc(&global_trace.data[cpu]->disabled);
2403                 }
2404                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2405                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2406                         atomic_dec(&global_trace.data[cpu]->disabled);
2407                 }
2408         }
2409         arch_spin_unlock(&ftrace_max_lock);
2410         local_irq_enable();
2411
2412         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2413
2414         mutex_unlock(&tracing_cpumask_update_lock);
2415         free_cpumask_var(tracing_cpumask_new);
2416
2417         return count;
2418
2419 err_unlock:
2420         free_cpumask_var(tracing_cpumask_new);
2421
2422         return err;
2423 }
2424
2425 static const struct file_operations tracing_cpumask_fops = {
2426         .open           = tracing_open_generic,
2427         .read           = tracing_cpumask_read,
2428         .write          = tracing_cpumask_write,
2429         .llseek         = generic_file_llseek,
2430 };
2431
2432 static int tracing_trace_options_show(struct seq_file *m, void *v)
2433 {
2434         struct tracer_opt *trace_opts;
2435         u32 tracer_flags;
2436         int i;
2437
2438         mutex_lock(&trace_types_lock);
2439         tracer_flags = current_trace->flags->val;
2440         trace_opts = current_trace->flags->opts;
2441
2442         for (i = 0; trace_options[i]; i++) {
2443                 if (trace_flags & (1 << i))
2444                         seq_printf(m, "%s\n", trace_options[i]);
2445                 else
2446                         seq_printf(m, "no%s\n", trace_options[i]);
2447         }
2448
2449         for (i = 0; trace_opts[i].name; i++) {
2450                 if (tracer_flags & trace_opts[i].bit)
2451                         seq_printf(m, "%s\n", trace_opts[i].name);
2452                 else
2453                         seq_printf(m, "no%s\n", trace_opts[i].name);
2454         }
2455         mutex_unlock(&trace_types_lock);
2456
2457         return 0;
2458 }
2459
2460 static int __set_tracer_option(struct tracer *trace,
2461                                struct tracer_flags *tracer_flags,
2462                                struct tracer_opt *opts, int neg)
2463 {
2464         int ret;
2465
2466         ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2467         if (ret)
2468                 return ret;
2469
2470         if (neg)
2471                 tracer_flags->val &= ~opts->bit;
2472         else
2473                 tracer_flags->val |= opts->bit;
2474         return 0;
2475 }
2476
2477 /* Try to assign a tracer specific option */
2478 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2479 {
2480         struct tracer_flags *tracer_flags = trace->flags;
2481         struct tracer_opt *opts = NULL;
2482         int i;
2483
2484         for (i = 0; tracer_flags->opts[i].name; i++) {
2485                 opts = &tracer_flags->opts[i];
2486
2487                 if (strcmp(cmp, opts->name) == 0)
2488                         return __set_tracer_option(trace, trace->flags,
2489                                                    opts, neg);
2490         }
2491
2492         return -EINVAL;
2493 }
2494
2495 static void set_tracer_flags(unsigned int mask, int enabled)
2496 {
2497         /* do nothing if flag is already set */
2498         if (!!(trace_flags & mask) == !!enabled)
2499                 return;
2500
2501         if (enabled)
2502                 trace_flags |= mask;
2503         else
2504                 trace_flags &= ~mask;
2505
2506         if (mask == TRACE_ITER_RECORD_CMD)
2507                 trace_event_enable_cmd_record(enabled);
2508 }
2509
2510 static ssize_t
2511 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2512                         size_t cnt, loff_t *ppos)
2513 {
2514         char buf[64];
2515         char *cmp;
2516         int neg = 0;
2517         int ret;
2518         int i;
2519
2520         if (cnt >= sizeof(buf))
2521                 return -EINVAL;
2522
2523         if (copy_from_user(&buf, ubuf, cnt))
2524                 return -EFAULT;
2525
2526         buf[cnt] = 0;
2527         cmp = strstrip(buf);
2528
2529         if (strncmp(cmp, "no", 2) == 0) {
2530                 neg = 1;
2531                 cmp += 2;
2532         }
2533
2534         for (i = 0; trace_options[i]; i++) {
2535                 if (strcmp(cmp, trace_options[i]) == 0) {
2536                         set_tracer_flags(1 << i, !neg);
2537                         break;
2538                 }
2539         }
2540
2541         /* If no option could be set, test the specific tracer options */
2542         if (!trace_options[i]) {
2543                 mutex_lock(&trace_types_lock);
2544                 ret = set_tracer_option(current_trace, cmp, neg);
2545                 mutex_unlock(&trace_types_lock);
2546                 if (ret)
2547                         return ret;
2548         }
2549
2550         *ppos += cnt;
2551
2552         return cnt;
2553 }
2554
2555 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2556 {
2557         if (tracing_disabled)
2558                 return -ENODEV;
2559         return single_open(file, tracing_trace_options_show, NULL);
2560 }
2561
2562 static const struct file_operations tracing_iter_fops = {
2563         .open           = tracing_trace_options_open,
2564         .read           = seq_read,
2565         .llseek         = seq_lseek,
2566         .release        = single_release,
2567         .write          = tracing_trace_options_write,
2568 };
2569
2570 static const char readme_msg[] =
2571         "tracing mini-HOWTO:\n\n"
2572         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2573         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2574         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2575         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2576         "nop\n"
2577         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2578         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2579         "sched_switch\n"
2580         "# cat /sys/kernel/debug/tracing/trace_options\n"
2581         "noprint-parent nosym-offset nosym-addr noverbose\n"
2582         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2583         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2584         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2585         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2586 ;
2587
2588 static ssize_t
2589 tracing_readme_read(struct file *filp, char __user *ubuf,
2590                        size_t cnt, loff_t *ppos)
2591 {
2592         return simple_read_from_buffer(ubuf, cnt, ppos,
2593                                         readme_msg, strlen(readme_msg));
2594 }
2595
2596 static const struct file_operations tracing_readme_fops = {
2597         .open           = tracing_open_generic,
2598         .read           = tracing_readme_read,
2599         .llseek         = generic_file_llseek,
2600 };
2601
2602 static ssize_t
2603 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2604                                 size_t cnt, loff_t *ppos)
2605 {
2606         char *buf_comm;
2607         char *file_buf;
2608         char *buf;
2609         int len = 0;
2610         int pid;
2611         int i;
2612
2613         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2614         if (!file_buf)
2615                 return -ENOMEM;
2616
2617         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2618         if (!buf_comm) {
2619                 kfree(file_buf);
2620                 return -ENOMEM;
2621         }
2622
2623         buf = file_buf;
2624
2625         for (i = 0; i < SAVED_CMDLINES; i++) {
2626                 int r;
2627
2628                 pid = map_cmdline_to_pid[i];
2629                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2630                         continue;
2631
2632                 trace_find_cmdline(pid, buf_comm);
2633                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2634                 buf += r;
2635                 len += r;
2636         }
2637
2638         len = simple_read_from_buffer(ubuf, cnt, ppos,
2639                                       file_buf, len);
2640
2641         kfree(file_buf);
2642         kfree(buf_comm);
2643
2644         return len;
2645 }
2646
2647 static const struct file_operations tracing_saved_cmdlines_fops = {
2648     .open       = tracing_open_generic,
2649     .read       = tracing_saved_cmdlines_read,
2650     .llseek     = generic_file_llseek,
2651 };
2652
2653 static ssize_t
2654 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2655                   size_t cnt, loff_t *ppos)
2656 {
2657         char buf[64];
2658         int r;
2659
2660         r = sprintf(buf, "%u\n", tracer_enabled);
2661         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2662 }
2663
2664 static ssize_t
2665 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2666                    size_t cnt, loff_t *ppos)
2667 {
2668         struct trace_array *tr = filp->private_data;
2669         char buf[64];
2670         unsigned long val;
2671         int ret;
2672
2673         if (cnt >= sizeof(buf))
2674                 return -EINVAL;
2675
2676         if (copy_from_user(&buf, ubuf, cnt))
2677                 return -EFAULT;
2678
2679         buf[cnt] = 0;
2680
2681         ret = strict_strtoul(buf, 10, &val);
2682         if (ret < 0)
2683                 return ret;
2684
2685         val = !!val;
2686
2687         mutex_lock(&trace_types_lock);
2688         if (tracer_enabled ^ val) {
2689                 if (val) {
2690                         tracer_enabled = 1;
2691                         if (current_trace->start)
2692                                 current_trace->start(tr);
2693                         tracing_start();
2694                 } else {
2695                         tracer_enabled = 0;
2696                         tracing_stop();
2697                         if (current_trace->stop)
2698                                 current_trace->stop(tr);
2699                 }
2700         }
2701         mutex_unlock(&trace_types_lock);
2702
2703         *ppos += cnt;
2704
2705         return cnt;
2706 }
2707
2708 static ssize_t
2709 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2710                        size_t cnt, loff_t *ppos)
2711 {
2712         char buf[MAX_TRACER_SIZE+2];
2713         int r;
2714
2715         mutex_lock(&trace_types_lock);
2716         if (current_trace)
2717                 r = sprintf(buf, "%s\n", current_trace->name);
2718         else
2719                 r = sprintf(buf, "\n");
2720         mutex_unlock(&trace_types_lock);
2721
2722         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2723 }
2724
2725 int tracer_init(struct tracer *t, struct trace_array *tr)
2726 {
2727         tracing_reset_online_cpus(tr);
2728         return t->init(tr);
2729 }
2730
2731 static int tracing_resize_ring_buffer(unsigned long size)
2732 {
2733         int ret;
2734
2735         /*
2736          * If kernel or user changes the size of the ring buffer
2737          * we use the size that was given, and we can forget about
2738          * expanding it later.
2739          */
2740         ring_buffer_expanded = 1;
2741
2742         ret = ring_buffer_resize(global_trace.buffer, size);
2743         if (ret < 0)
2744                 return ret;
2745
2746         if (!current_trace->use_max_tr)
2747                 goto out;
2748
2749         ret = ring_buffer_resize(max_tr.buffer, size);
2750         if (ret < 0) {
2751                 int r;
2752
2753                 r = ring_buffer_resize(global_trace.buffer,
2754                                        global_trace.entries);
2755                 if (r < 0) {
2756                         /*
2757                          * AARGH! We are left with different
2758                          * size max buffer!!!!
2759                          * The max buffer is our "snapshot" buffer.
2760                          * When a tracer needs a snapshot (one of the
2761                          * latency tracers), it swaps the max buffer
2762                          * with the saved snap shot. We succeeded to
2763                          * update the size of the main buffer, but failed to
2764                          * update the size of the max buffer. But when we tried
2765                          * to reset the main buffer to the original size, we
2766                          * failed there too. This is very unlikely to
2767                          * happen, but if it does, warn and kill all
2768                          * tracing.
2769                          */
2770                         WARN_ON(1);
2771                         tracing_disabled = 1;
2772                 }
2773                 return ret;
2774         }
2775
2776         max_tr.entries = size;
2777  out:
2778         global_trace.entries = size;
2779
2780         return ret;
2781 }
2782
2783
2784 /**
2785  * tracing_update_buffers - used by tracing facility to expand ring buffers
2786  *
2787  * To save on memory when the tracing is never used on a system with it
2788  * configured in. The ring buffers are set to a minimum size. But once
2789  * a user starts to use the tracing facility, then they need to grow
2790  * to their default size.
2791  *
2792  * This function is to be called when a tracer is about to be used.
2793  */
2794 int tracing_update_buffers(void)
2795 {
2796         int ret = 0;
2797
2798         mutex_lock(&trace_types_lock);
2799         if (!ring_buffer_expanded)
2800                 ret = tracing_resize_ring_buffer(trace_buf_size);
2801         mutex_unlock(&trace_types_lock);
2802
2803         return ret;
2804 }
2805
2806 struct trace_option_dentry;
2807
2808 static struct trace_option_dentry *
2809 create_trace_option_files(struct tracer *tracer);
2810
2811 static void
2812 destroy_trace_option_files(struct trace_option_dentry *topts);
2813
2814 static int tracing_set_tracer(const char *buf)
2815 {
2816         static struct trace_option_dentry *topts;
2817         struct trace_array *tr = &global_trace;
2818         struct tracer *t;
2819         int ret = 0;
2820
2821         mutex_lock(&trace_types_lock);
2822
2823         if (!ring_buffer_expanded) {
2824                 ret = tracing_resize_ring_buffer(trace_buf_size);
2825                 if (ret < 0)
2826                         goto out;
2827                 ret = 0;
2828         }
2829
2830         for (t = trace_types; t; t = t->next) {
2831                 if (strcmp(t->name, buf) == 0)
2832                         break;
2833         }
2834         if (!t) {
2835                 ret = -EINVAL;
2836                 goto out;
2837         }
2838         if (t == current_trace)
2839                 goto out;
2840
2841         trace_branch_disable();
2842         if (current_trace && current_trace->reset)
2843                 current_trace->reset(tr);
2844         if (current_trace && current_trace->use_max_tr) {
2845                 /*
2846                  * We don't free the ring buffer. instead, resize it because
2847                  * The max_tr ring buffer has some state (e.g. ring->clock) and
2848                  * we want preserve it.
2849                  */
2850                 ring_buffer_resize(max_tr.buffer, 1);
2851                 max_tr.entries = 1;
2852         }
2853         destroy_trace_option_files(topts);
2854
2855         current_trace = t;
2856
2857         topts = create_trace_option_files(current_trace);
2858         if (current_trace->use_max_tr) {
2859                 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
2860                 if (ret < 0)
2861                         goto out;
2862                 max_tr.entries = global_trace.entries;
2863         }
2864
2865         if (t->init) {
2866                 ret = tracer_init(t, tr);
2867                 if (ret)
2868                         goto out;
2869         }
2870
2871         trace_branch_enable(tr);
2872  out:
2873         mutex_unlock(&trace_types_lock);
2874
2875         return ret;
2876 }
2877
2878 static ssize_t
2879 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2880                         size_t cnt, loff_t *ppos)
2881 {
2882         char buf[MAX_TRACER_SIZE+1];
2883         int i;
2884         size_t ret;
2885         int err;
2886
2887         ret = cnt;
2888
2889         if (cnt > MAX_TRACER_SIZE)
2890                 cnt = MAX_TRACER_SIZE;
2891
2892         if (copy_from_user(&buf, ubuf, cnt))
2893                 return -EFAULT;
2894
2895         buf[cnt] = 0;
2896
2897         /* strip ending whitespace. */
2898         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2899                 buf[i] = 0;
2900
2901         err = tracing_set_tracer(buf);
2902         if (err)
2903                 return err;
2904
2905         *ppos += ret;
2906
2907         return ret;
2908 }
2909
2910 static ssize_t
2911 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2912                      size_t cnt, loff_t *ppos)
2913 {
2914         unsigned long *ptr = filp->private_data;
2915         char buf[64];
2916         int r;
2917
2918         r = snprintf(buf, sizeof(buf), "%ld\n",
2919                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2920         if (r > sizeof(buf))
2921                 r = sizeof(buf);
2922         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2923 }
2924
2925 static ssize_t
2926 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2927                       size_t cnt, loff_t *ppos)
2928 {
2929         unsigned long *ptr = filp->private_data;
2930         char buf[64];
2931         unsigned long val;
2932         int ret;
2933
2934         if (cnt >= sizeof(buf))
2935                 return -EINVAL;
2936
2937         if (copy_from_user(&buf, ubuf, cnt))
2938                 return -EFAULT;
2939
2940         buf[cnt] = 0;
2941
2942         ret = strict_strtoul(buf, 10, &val);
2943         if (ret < 0)
2944                 return ret;
2945
2946         *ptr = val * 1000;
2947
2948         return cnt;
2949 }
2950
2951 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2952 {
2953         long cpu_file = (long) inode->i_private;
2954         struct trace_iterator *iter;
2955         int ret = 0;
2956
2957         if (tracing_disabled)
2958                 return -ENODEV;
2959
2960         mutex_lock(&trace_types_lock);
2961
2962         /* create a buffer to store the information to pass to userspace */
2963         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2964         if (!iter) {
2965                 ret = -ENOMEM;
2966                 goto out;
2967         }
2968
2969         /*
2970          * We make a copy of the current tracer to avoid concurrent
2971          * changes on it while we are reading.
2972          */
2973         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2974         if (!iter->trace) {
2975                 ret = -ENOMEM;
2976                 goto fail;
2977         }
2978         if (current_trace)
2979                 *iter->trace = *current_trace;
2980
2981         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2982                 ret = -ENOMEM;
2983                 goto fail;
2984         }
2985
2986         /* trace pipe does not show start of buffer */
2987         cpumask_setall(iter->started);
2988
2989         if (trace_flags & TRACE_ITER_LATENCY_FMT)
2990                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2991
2992         iter->cpu_file = cpu_file;
2993         iter->tr = &global_trace;
2994         mutex_init(&iter->mutex);
2995         filp->private_data = iter;
2996
2997         if (iter->trace->pipe_open)
2998                 iter->trace->pipe_open(iter);
2999
3000         nonseekable_open(inode, filp);
3001 out:
3002         mutex_unlock(&trace_types_lock);
3003         return ret;
3004
3005 fail:
3006         kfree(iter->trace);
3007         kfree(iter);
3008         mutex_unlock(&trace_types_lock);
3009         return ret;
3010 }
3011
3012 static int tracing_release_pipe(struct inode *inode, struct file *file)
3013 {
3014         struct trace_iterator *iter = file->private_data;
3015
3016         mutex_lock(&trace_types_lock);
3017
3018         if (iter->trace->pipe_close)
3019                 iter->trace->pipe_close(iter);
3020
3021         mutex_unlock(&trace_types_lock);
3022
3023         free_cpumask_var(iter->started);
3024         mutex_destroy(&iter->mutex);
3025         kfree(iter->trace);
3026         kfree(iter);
3027
3028         return 0;
3029 }
3030
3031 static unsigned int
3032 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3033 {
3034         struct trace_iterator *iter = filp->private_data;
3035
3036         if (trace_flags & TRACE_ITER_BLOCK) {
3037                 /*
3038                  * Always select as readable when in blocking mode
3039                  */
3040                 return POLLIN | POLLRDNORM;
3041         } else {
3042                 if (!trace_empty(iter))
3043                         return POLLIN | POLLRDNORM;
3044                 poll_wait(filp, &trace_wait, poll_table);
3045                 if (!trace_empty(iter))
3046                         return POLLIN | POLLRDNORM;
3047
3048                 return 0;
3049         }
3050 }
3051
3052
3053 void default_wait_pipe(struct trace_iterator *iter)
3054 {
3055         DEFINE_WAIT(wait);
3056
3057         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3058
3059         if (trace_empty(iter))
3060                 schedule();
3061
3062         finish_wait(&trace_wait, &wait);
3063 }
3064
3065 /*
3066  * This is a make-shift waitqueue.
3067  * A tracer might use this callback on some rare cases:
3068  *
3069  *  1) the current tracer might hold the runqueue lock when it wakes up
3070  *     a reader, hence a deadlock (sched, function, and function graph tracers)
3071  *  2) the function tracers, trace all functions, we don't want
3072  *     the overhead of calling wake_up and friends
3073  *     (and tracing them too)
3074  *
3075  *     Anyway, this is really very primitive wakeup.
3076  */
3077 void poll_wait_pipe(struct trace_iterator *iter)
3078 {
3079         set_current_state(TASK_INTERRUPTIBLE);
3080         /* sleep for 100 msecs, and try again. */
3081         schedule_timeout(HZ / 10);
3082 }
3083
3084 /* Must be called with trace_types_lock mutex held. */
3085 static int tracing_wait_pipe(struct file *filp)
3086 {
3087         struct trace_iterator *iter = filp->private_data;
3088
3089         while (trace_empty(iter)) {
3090
3091                 if ((filp->f_flags & O_NONBLOCK)) {
3092                         return -EAGAIN;
3093                 }
3094
3095                 mutex_unlock(&iter->mutex);
3096
3097                 iter->trace->wait_pipe(iter);
3098
3099                 mutex_lock(&iter->mutex);
3100
3101                 if (signal_pending(current))
3102                         return -EINTR;
3103
3104                 /*
3105                  * We block until we read something and tracing is disabled.
3106                  * We still block if tracing is disabled, but we have never
3107                  * read anything. This allows a user to cat this file, and
3108                  * then enable tracing. But after we have read something,
3109                  * we give an EOF when tracing is again disabled.
3110                  *
3111                  * iter->pos will be 0 if we haven't read anything.
3112                  */
3113                 if (!tracer_enabled && iter->pos)
3114                         break;
3115         }
3116
3117         return 1;
3118 }
3119
3120 /*
3121  * Consumer reader.
3122  */
3123 static ssize_t
3124 tracing_read_pipe(struct file *filp, char __user *ubuf,
3125                   size_t cnt, loff_t *ppos)
3126 {
3127         struct trace_iterator *iter = filp->private_data;
3128         static struct tracer *old_tracer;
3129         ssize_t sret;
3130
3131         /* return any leftover data */
3132         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3133         if (sret != -EBUSY)
3134                 return sret;
3135
3136         trace_seq_init(&iter->seq);
3137
3138         /* copy the tracer to avoid using a global lock all around */
3139         mutex_lock(&trace_types_lock);
3140         if (unlikely(old_tracer != current_trace && current_trace)) {
3141                 old_tracer = current_trace;
3142                 *iter->trace = *current_trace;
3143         }
3144         mutex_unlock(&trace_types_lock);
3145
3146         /*
3147          * Avoid more than one consumer on a single file descriptor
3148          * This is just a matter of traces coherency, the ring buffer itself
3149          * is protected.
3150          */
3151         mutex_lock(&iter->mutex);
3152         if (iter->trace->read) {
3153                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3154                 if (sret)
3155                         goto out;
3156         }
3157
3158 waitagain:
3159         sret = tracing_wait_pipe(filp);
3160         if (sret <= 0)
3161                 goto out;
3162
3163         /* stop when tracing is finished */
3164         if (trace_empty(iter)) {
3165                 sret = 0;
3166                 goto out;
3167         }
3168
3169         if (cnt >= PAGE_SIZE)
3170                 cnt = PAGE_SIZE - 1;
3171
3172         /* reset all but tr, trace, and overruns */
3173         memset(&iter->seq, 0,
3174                sizeof(struct trace_iterator) -
3175                offsetof(struct trace_iterator, seq));
3176         iter->pos = -1;
3177
3178         trace_event_read_lock();
3179         trace_access_lock(iter->cpu_file);
3180         while (trace_find_next_entry_inc(iter) != NULL) {
3181                 enum print_line_t ret;
3182                 int len = iter->seq.len;
3183
3184                 ret = print_trace_line(iter);
3185                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3186                         /* don't print partial lines */
3187                         iter->seq.len = len;
3188                         break;
3189                 }
3190                 if (ret != TRACE_TYPE_NO_CONSUME)
3191                         trace_consume(iter);
3192
3193                 if (iter->seq.len >= cnt)
3194                         break;
3195         }
3196         trace_access_unlock(iter->cpu_file);
3197         trace_event_read_unlock();
3198
3199         /* Now copy what we have to the user */
3200         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3201         if (iter->seq.readpos >= iter->seq.len)
3202                 trace_seq_init(&iter->seq);
3203
3204         /*
3205          * If there was nothing to send to user, inspite of consuming trace
3206          * entries, go back to wait for more entries.
3207          */
3208         if (sret == -EBUSY)
3209                 goto waitagain;
3210
3211 out:
3212         mutex_unlock(&iter->mutex);
3213
3214         return sret;
3215 }
3216
3217 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3218                                      struct pipe_buffer *buf)
3219 {
3220         __free_page(buf->page);
3221 }
3222
3223 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3224                                      unsigned int idx)
3225 {
3226         __free_page(spd->pages[idx]);
3227 }
3228
3229 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3230         .can_merge              = 0,
3231         .map                    = generic_pipe_buf_map,
3232         .unmap                  = generic_pipe_buf_unmap,
3233         .confirm                = generic_pipe_buf_confirm,
3234         .release                = tracing_pipe_buf_release,
3235         .steal                  = generic_pipe_buf_steal,
3236         .get                    = generic_pipe_buf_get,
3237 };
3238
3239 static size_t
3240 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3241 {
3242         size_t count;
3243         int ret;
3244
3245         /* Seq buffer is page-sized, exactly what we need. */
3246         for (;;) {
3247                 count = iter->seq.len;
3248                 ret = print_trace_line(iter);
3249                 count = iter->seq.len - count;
3250                 if (rem < count) {
3251                         rem = 0;
3252                         iter->seq.len -= count;
3253                         break;
3254                 }
3255                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3256                         iter->seq.len -= count;
3257                         break;
3258                 }
3259
3260                 if (ret != TRACE_TYPE_NO_CONSUME)
3261                         trace_consume(iter);
3262                 rem -= count;
3263                 if (!trace_find_next_entry_inc(iter))   {
3264                         rem = 0;
3265                         iter->ent = NULL;
3266                         break;
3267                 }
3268         }
3269
3270         return rem;
3271 }
3272
3273 static ssize_t tracing_splice_read_pipe(struct file *filp,
3274                                         loff_t *ppos,
3275                                         struct pipe_inode_info *pipe,
3276                                         size_t len,
3277                                         unsigned int flags)
3278 {
3279         struct page *pages_def[PIPE_DEF_BUFFERS];
3280         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3281         struct trace_iterator *iter = filp->private_data;
3282         struct splice_pipe_desc spd = {
3283                 .pages          = pages_def,
3284                 .partial        = partial_def,
3285                 .nr_pages       = 0, /* This gets updated below. */
3286                 .flags          = flags,
3287                 .ops            = &tracing_pipe_buf_ops,
3288                 .spd_release    = tracing_spd_release_pipe,
3289         };
3290         static struct tracer *old_tracer;
3291         ssize_t ret;
3292         size_t rem;
3293         unsigned int i;
3294
3295         if (splice_grow_spd(pipe, &spd))
3296                 return -ENOMEM;
3297
3298         /* copy the tracer to avoid using a global lock all around */
3299         mutex_lock(&trace_types_lock);
3300         if (unlikely(old_tracer != current_trace && current_trace)) {
3301                 old_tracer = current_trace;
3302                 *iter->trace = *current_trace;
3303         }
3304         mutex_unlock(&trace_types_lock);
3305
3306         mutex_lock(&iter->mutex);
3307
3308         if (iter->trace->splice_read) {
3309                 ret = iter->trace->splice_read(iter, filp,
3310                                                ppos, pipe, len, flags);
3311                 if (ret)
3312                         goto out_err;
3313         }
3314
3315         ret = tracing_wait_pipe(filp);
3316         if (ret <= 0)
3317                 goto out_err;
3318
3319         if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3320                 ret = -EFAULT;
3321                 goto out_err;
3322         }
3323
3324         trace_event_read_lock();
3325         trace_access_lock(iter->cpu_file);
3326
3327         /* Fill as many pages as possible. */
3328         for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3329                 spd.pages[i] = alloc_page(GFP_KERNEL);
3330                 if (!spd.pages[i])
3331                         break;
3332
3333                 rem = tracing_fill_pipe_page(rem, iter);
3334
3335                 /* Copy the data into the page, so we can start over. */
3336                 ret = trace_seq_to_buffer(&iter->seq,
3337                                           page_address(spd.pages[i]),
3338                                           iter->seq.len);
3339                 if (ret < 0) {
3340                         __free_page(spd.pages[i]);
3341                         break;
3342                 }
3343                 spd.partial[i].offset = 0;
3344                 spd.partial[i].len = iter->seq.len;
3345
3346                 trace_seq_init(&iter->seq);
3347         }
3348
3349         trace_access_unlock(iter->cpu_file);
3350         trace_event_read_unlock();
3351         mutex_unlock(&iter->mutex);
3352
3353         spd.nr_pages = i;
3354
3355         ret = splice_to_pipe(pipe, &spd);
3356 out:
3357         splice_shrink_spd(pipe, &spd);
3358         return ret;
3359
3360 out_err:
3361         mutex_unlock(&iter->mutex);
3362         goto out;
3363 }
3364
3365 static ssize_t
3366 tracing_entries_read(struct file *filp, char __user *ubuf,
3367                      size_t cnt, loff_t *ppos)
3368 {
3369         struct trace_array *tr = filp->private_data;
3370         char buf[96];
3371         int r;
3372
3373         mutex_lock(&trace_types_lock);
3374         if (!ring_buffer_expanded)
3375                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3376                             tr->entries >> 10,
3377                             trace_buf_size >> 10);
3378         else
3379                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3380         mutex_unlock(&trace_types_lock);
3381
3382         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3383 }
3384
3385 static ssize_t
3386 tracing_entries_write(struct file *filp, const char __user *ubuf,
3387                       size_t cnt, loff_t *ppos)
3388 {
3389         unsigned long val;
3390         char buf[64];
3391         int ret, cpu;
3392
3393         if (cnt >= sizeof(buf))
3394                 return -EINVAL;
3395
3396         if (copy_from_user(&buf, ubuf, cnt))
3397                 return -EFAULT;
3398
3399         buf[cnt] = 0;
3400
3401         ret = strict_strtoul(buf, 10, &val);
3402         if (ret < 0)
3403                 return ret;
3404
3405         /* must have at least 1 entry */
3406         if (!val)
3407                 return -EINVAL;
3408
3409         mutex_lock(&trace_types_lock);
3410
3411         tracing_stop();
3412
3413         /* disable all cpu buffers */
3414         for_each_tracing_cpu(cpu) {
3415                 if (global_trace.data[cpu])
3416                         atomic_inc(&global_trace.data[cpu]->disabled);
3417                 if (max_tr.data[cpu])
3418                         atomic_inc(&max_tr.data[cpu]->disabled);
3419         }
3420
3421         /* value is in KB */
3422         val <<= 10;
3423
3424         if (val != global_trace.entries) {
3425                 ret = tracing_resize_ring_buffer(val);
3426                 if (ret < 0) {
3427                         cnt = ret;
3428                         goto out;
3429                 }
3430         }
3431
3432         *ppos += cnt;
3433
3434         /* If check pages failed, return ENOMEM */
3435         if (tracing_disabled)
3436                 cnt = -ENOMEM;
3437  out:
3438         for_each_tracing_cpu(cpu) {
3439                 if (global_trace.data[cpu])
3440                         atomic_dec(&global_trace.data[cpu]->disabled);
3441