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