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