Merge branch 'x86/uaccess' into core/percpu
[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/utsrelease.h>
15 #include <linux/kallsyms.h>
16 #include <linux/seq_file.h>
17 #include <linux/notifier.h>
18 #include <linux/debugfs.h>
19 #include <linux/pagemap.h>
20 #include <linux/hardirq.h>
21 #include <linux/linkage.h>
22 #include <linux/uaccess.h>
23 #include <linux/ftrace.h>
24 #include <linux/module.h>
25 #include <linux/percpu.h>
26 #include <linux/kdebug.h>
27 #include <linux/ctype.h>
28 #include <linux/init.h>
29 #include <linux/poll.h>
30 #include <linux/gfp.h>
31 #include <linux/fs.h>
32 #include <linux/kprobes.h>
33 #include <linux/writeback.h>
34
35 #include <linux/stacktrace.h>
36 #include <linux/ring_buffer.h>
37 #include <linux/irqflags.h>
38
39 #include "trace.h"
40
41 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
42
43 unsigned long __read_mostly     tracing_max_latency;
44 unsigned long __read_mostly     tracing_thresh;
45
46 /*
47  * We need to change this state when a selftest is running.
48  * A selftest will lurk into the ring-buffer to count the
49  * entries inserted during the selftest although some concurrent
50  * insertions into the ring-buffer such as ftrace_printk could occurred
51  * at the same time, giving false positive or negative results.
52  */
53 static bool __read_mostly tracing_selftest_running;
54
55 /* For tracers that don't implement custom flags */
56 static struct tracer_opt dummy_tracer_opt[] = {
57         { }
58 };
59
60 static struct tracer_flags dummy_tracer_flags = {
61         .val = 0,
62         .opts = dummy_tracer_opt
63 };
64
65 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
66 {
67         return 0;
68 }
69
70 /*
71  * Kill all tracing for good (never come back).
72  * It is initialized to 1 but will turn to zero if the initialization
73  * of the tracer is successful. But that is the only place that sets
74  * this back to zero.
75  */
76 int tracing_disabled = 1;
77
78 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
79
80 static inline void ftrace_disable_cpu(void)
81 {
82         preempt_disable();
83         local_inc(&__get_cpu_var(ftrace_cpu_disabled));
84 }
85
86 static inline void ftrace_enable_cpu(void)
87 {
88         local_dec(&__get_cpu_var(ftrace_cpu_disabled));
89         preempt_enable();
90 }
91
92 static cpumask_var_t __read_mostly      tracing_buffer_mask;
93
94 #define for_each_tracing_cpu(cpu)       \
95         for_each_cpu(cpu, tracing_buffer_mask)
96
97 /*
98  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
99  *
100  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
101  * is set, then ftrace_dump is called. This will output the contents
102  * of the ftrace buffers to the console.  This is very useful for
103  * capturing traces that lead to crashes and outputing it to a
104  * serial console.
105  *
106  * It is default off, but you can enable it with either specifying
107  * "ftrace_dump_on_oops" in the kernel command line, or setting
108  * /proc/sys/kernel/ftrace_dump_on_oops to true.
109  */
110 int ftrace_dump_on_oops;
111
112 static int tracing_set_tracer(char *buf);
113
114 static int __init set_ftrace(char *str)
115 {
116         tracing_set_tracer(str);
117         return 1;
118 }
119 __setup("ftrace", set_ftrace);
120
121 static int __init set_ftrace_dump_on_oops(char *str)
122 {
123         ftrace_dump_on_oops = 1;
124         return 1;
125 }
126 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
127
128 long
129 ns2usecs(cycle_t nsec)
130 {
131         nsec += 500;
132         do_div(nsec, 1000);
133         return nsec;
134 }
135
136 cycle_t ftrace_now(int cpu)
137 {
138         u64 ts = ring_buffer_time_stamp(cpu);
139         ring_buffer_normalize_time_stamp(cpu, &ts);
140         return ts;
141 }
142
143 /*
144  * The global_trace is the descriptor that holds the tracing
145  * buffers for the live tracing. For each CPU, it contains
146  * a link list of pages that will store trace entries. The
147  * page descriptor of the pages in the memory is used to hold
148  * the link list by linking the lru item in the page descriptor
149  * to each of the pages in the buffer per CPU.
150  *
151  * For each active CPU there is a data field that holds the
152  * pages for the buffer for that CPU. Each CPU has the same number
153  * of pages allocated for its buffer.
154  */
155 static struct trace_array       global_trace;
156
157 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
158
159 /*
160  * The max_tr is used to snapshot the global_trace when a maximum
161  * latency is reached. Some tracers will use this to store a maximum
162  * trace while it continues examining live traces.
163  *
164  * The buffers for the max_tr are set up the same as the global_trace.
165  * When a snapshot is taken, the link list of the max_tr is swapped
166  * with the link list of the global_trace and the buffers are reset for
167  * the global_trace so the tracing can continue.
168  */
169 static struct trace_array       max_tr;
170
171 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
172
173 /* tracer_enabled is used to toggle activation of a tracer */
174 static int                      tracer_enabled = 1;
175
176 /**
177  * tracing_is_enabled - return tracer_enabled status
178  *
179  * This function is used by other tracers to know the status
180  * of the tracer_enabled flag.  Tracers may use this function
181  * to know if it should enable their features when starting
182  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
183  */
184 int tracing_is_enabled(void)
185 {
186         return tracer_enabled;
187 }
188
189 /* function tracing enabled */
190 int                             ftrace_function_enabled;
191
192 /*
193  * trace_buf_size is the size in bytes that is allocated
194  * for a buffer. Note, the number of bytes is always rounded
195  * to page size.
196  *
197  * This number is purposely set to a low number of 16384.
198  * If the dump on oops happens, it will be much appreciated
199  * to not have to wait for all that output. Anyway this can be
200  * boot time and run time configurable.
201  */
202 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
203
204 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
205
206 /* trace_types holds a link list of available tracers. */
207 static struct tracer            *trace_types __read_mostly;
208
209 /* current_trace points to the tracer that is currently active */
210 static struct tracer            *current_trace __read_mostly;
211
212 /*
213  * max_tracer_type_len is used to simplify the allocating of
214  * buffers to read userspace tracer names. We keep track of
215  * the longest tracer name registered.
216  */
217 static int                      max_tracer_type_len;
218
219 /*
220  * trace_types_lock is used to protect the trace_types list.
221  * This lock is also used to keep user access serialized.
222  * Accesses from userspace will grab this lock while userspace
223  * activities happen inside the kernel.
224  */
225 static DEFINE_MUTEX(trace_types_lock);
226
227 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
228 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
229
230 /* trace_flags holds trace_options default values */
231 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
232         TRACE_ITER_ANNOTATE;
233
234 /**
235  * trace_wake_up - wake up tasks waiting for trace input
236  *
237  * Simply wakes up any task that is blocked on the trace_wait
238  * queue. These is used with trace_poll for tasks polling the trace.
239  */
240 void trace_wake_up(void)
241 {
242         /*
243          * The runqueue_is_locked() can fail, but this is the best we
244          * have for now:
245          */
246         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
247                 wake_up(&trace_wait);
248 }
249
250 static int __init set_buf_size(char *str)
251 {
252         unsigned long buf_size;
253         int ret;
254
255         if (!str)
256                 return 0;
257         ret = strict_strtoul(str, 0, &buf_size);
258         /* nr_entries can not be zero */
259         if (ret < 0 || buf_size == 0)
260                 return 0;
261         trace_buf_size = buf_size;
262         return 1;
263 }
264 __setup("trace_buf_size=", set_buf_size);
265
266 unsigned long nsecs_to_usecs(unsigned long nsecs)
267 {
268         return nsecs / 1000;
269 }
270
271 /* These must match the bit postions in trace_iterator_flags */
272 static const char *trace_options[] = {
273         "print-parent",
274         "sym-offset",
275         "sym-addr",
276         "verbose",
277         "raw",
278         "hex",
279         "bin",
280         "block",
281         "stacktrace",
282         "sched-tree",
283         "ftrace_printk",
284         "ftrace_preempt",
285         "branch",
286         "annotate",
287         "userstacktrace",
288         "sym-userobj",
289         "printk-msg-only",
290         NULL
291 };
292
293 /*
294  * ftrace_max_lock is used to protect the swapping of buffers
295  * when taking a max snapshot. The buffers themselves are
296  * protected by per_cpu spinlocks. But the action of the swap
297  * needs its own lock.
298  *
299  * This is defined as a raw_spinlock_t in order to help
300  * with performance when lockdep debugging is enabled.
301  */
302 static raw_spinlock_t ftrace_max_lock =
303         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
304
305 /*
306  * Copy the new maximum trace into the separate maximum-trace
307  * structure. (this way the maximum trace is permanently saved,
308  * for later retrieval via /debugfs/tracing/latency_trace)
309  */
310 static void
311 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
312 {
313         struct trace_array_cpu *data = tr->data[cpu];
314
315         max_tr.cpu = cpu;
316         max_tr.time_start = data->preempt_timestamp;
317
318         data = max_tr.data[cpu];
319         data->saved_latency = tracing_max_latency;
320
321         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
322         data->pid = tsk->pid;
323         data->uid = task_uid(tsk);
324         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
325         data->policy = tsk->policy;
326         data->rt_priority = tsk->rt_priority;
327
328         /* record this tasks comm */
329         tracing_record_cmdline(current);
330 }
331
332 /**
333  * trace_seq_printf - sequence printing of trace information
334  * @s: trace sequence descriptor
335  * @fmt: printf format string
336  *
337  * The tracer may use either sequence operations or its own
338  * copy to user routines. To simplify formating of a trace
339  * trace_seq_printf is used to store strings into a special
340  * buffer (@s). Then the output may be either used by
341  * the sequencer or pulled into another buffer.
342  */
343 int
344 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
345 {
346         int len = (PAGE_SIZE - 1) - s->len;
347         va_list ap;
348         int ret;
349
350         if (!len)
351                 return 0;
352
353         va_start(ap, fmt);
354         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
355         va_end(ap);
356
357         /* If we can't write it all, don't bother writing anything */
358         if (ret >= len)
359                 return 0;
360
361         s->len += ret;
362
363         return len;
364 }
365
366 /**
367  * trace_seq_puts - trace sequence printing of simple string
368  * @s: trace sequence descriptor
369  * @str: simple string to record
370  *
371  * The tracer may use either the sequence operations or its own
372  * copy to user routines. This function records a simple string
373  * into a special buffer (@s) for later retrieval by a sequencer
374  * or other mechanism.
375  */
376 static int
377 trace_seq_puts(struct trace_seq *s, const char *str)
378 {
379         int len = strlen(str);
380
381         if (len > ((PAGE_SIZE - 1) - s->len))
382                 return 0;
383
384         memcpy(s->buffer + s->len, str, len);
385         s->len += len;
386
387         return len;
388 }
389
390 static int
391 trace_seq_putc(struct trace_seq *s, unsigned char c)
392 {
393         if (s->len >= (PAGE_SIZE - 1))
394                 return 0;
395
396         s->buffer[s->len++] = c;
397
398         return 1;
399 }
400
401 static int
402 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
403 {
404         if (len > ((PAGE_SIZE - 1) - s->len))
405                 return 0;
406
407         memcpy(s->buffer + s->len, mem, len);
408         s->len += len;
409
410         return len;
411 }
412
413 #define MAX_MEMHEX_BYTES        8
414 #define HEX_CHARS               (MAX_MEMHEX_BYTES*2 + 1)
415
416 static int
417 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
418 {
419         unsigned char hex[HEX_CHARS];
420         unsigned char *data = mem;
421         int i, j;
422
423 #ifdef __BIG_ENDIAN
424         for (i = 0, j = 0; i < len; i++) {
425 #else
426         for (i = len-1, j = 0; i >= 0; i--) {
427 #endif
428                 hex[j++] = hex_asc_hi(data[i]);
429                 hex[j++] = hex_asc_lo(data[i]);
430         }
431         hex[j++] = ' ';
432
433         return trace_seq_putmem(s, hex, j);
434 }
435
436 static int
437 trace_seq_path(struct trace_seq *s, struct path *path)
438 {
439         unsigned char *p;
440
441         if (s->len >= (PAGE_SIZE - 1))
442                 return 0;
443         p = d_path(path, s->buffer + s->len, PAGE_SIZE - s->len);
444         if (!IS_ERR(p)) {
445                 p = mangle_path(s->buffer + s->len, p, "\n");
446                 if (p) {
447                         s->len = p - s->buffer;
448                         return 1;
449                 }
450         } else {
451                 s->buffer[s->len++] = '?';
452                 return 1;
453         }
454
455         return 0;
456 }
457
458 static void
459 trace_seq_reset(struct trace_seq *s)
460 {
461         s->len = 0;
462         s->readpos = 0;
463 }
464
465 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
466 {
467         int len;
468         int ret;
469
470         if (s->len <= s->readpos)
471                 return -EBUSY;
472
473         len = s->len - s->readpos;
474         if (cnt > len)
475                 cnt = len;
476         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
477         if (ret)
478                 return -EFAULT;
479
480         s->readpos += len;
481         return cnt;
482 }
483
484 static void
485 trace_print_seq(struct seq_file *m, struct trace_seq *s)
486 {
487         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
488
489         s->buffer[len] = 0;
490         seq_puts(m, s->buffer);
491
492         trace_seq_reset(s);
493 }
494
495 /**
496  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
497  * @tr: tracer
498  * @tsk: the task with the latency
499  * @cpu: The cpu that initiated the trace.
500  *
501  * Flip the buffers between the @tr and the max_tr and record information
502  * about which task was the cause of this latency.
503  */
504 void
505 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
506 {
507         struct ring_buffer *buf = tr->buffer;
508
509         WARN_ON_ONCE(!irqs_disabled());
510         __raw_spin_lock(&ftrace_max_lock);
511
512         tr->buffer = max_tr.buffer;
513         max_tr.buffer = buf;
514
515         ftrace_disable_cpu();
516         ring_buffer_reset(tr->buffer);
517         ftrace_enable_cpu();
518
519         __update_max_tr(tr, tsk, cpu);
520         __raw_spin_unlock(&ftrace_max_lock);
521 }
522
523 /**
524  * update_max_tr_single - only copy one trace over, and reset the rest
525  * @tr - tracer
526  * @tsk - task with the latency
527  * @cpu - the cpu of the buffer to copy.
528  *
529  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
530  */
531 void
532 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
533 {
534         int ret;
535
536         WARN_ON_ONCE(!irqs_disabled());
537         __raw_spin_lock(&ftrace_max_lock);
538
539         ftrace_disable_cpu();
540
541         ring_buffer_reset(max_tr.buffer);
542         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
543
544         ftrace_enable_cpu();
545
546         WARN_ON_ONCE(ret);
547
548         __update_max_tr(tr, tsk, cpu);
549         __raw_spin_unlock(&ftrace_max_lock);
550 }
551
552 /**
553  * register_tracer - register a tracer with the ftrace system.
554  * @type - the plugin for the tracer
555  *
556  * Register a new plugin tracer.
557  */
558 int register_tracer(struct tracer *type)
559 {
560         struct tracer *t;
561         int len;
562         int ret = 0;
563
564         if (!type->name) {
565                 pr_info("Tracer must have a name\n");
566                 return -1;
567         }
568
569         /*
570          * When this gets called we hold the BKL which means that
571          * preemption is disabled. Various trace selftests however
572          * need to disable and enable preemption for successful tests.
573          * So we drop the BKL here and grab it after the tests again.
574          */
575         unlock_kernel();
576         mutex_lock(&trace_types_lock);
577
578         tracing_selftest_running = true;
579
580         for (t = trace_types; t; t = t->next) {
581                 if (strcmp(type->name, t->name) == 0) {
582                         /* already found */
583                         pr_info("Trace %s already registered\n",
584                                 type->name);
585                         ret = -1;
586                         goto out;
587                 }
588         }
589
590         if (!type->set_flag)
591                 type->set_flag = &dummy_set_flag;
592         if (!type->flags)
593                 type->flags = &dummy_tracer_flags;
594         else
595                 if (!type->flags->opts)
596                         type->flags->opts = dummy_tracer_opt;
597
598 #ifdef CONFIG_FTRACE_STARTUP_TEST
599         if (type->selftest) {
600                 struct tracer *saved_tracer = current_trace;
601                 struct trace_array *tr = &global_trace;
602                 int i;
603
604                 /*
605                  * Run a selftest on this tracer.
606                  * Here we reset the trace buffer, and set the current
607                  * tracer to be this tracer. The tracer can then run some
608                  * internal tracing to verify that everything is in order.
609                  * If we fail, we do not register this tracer.
610                  */
611                 for_each_tracing_cpu(i)
612                         tracing_reset(tr, i);
613
614                 current_trace = type;
615                 /* the test is responsible for initializing and enabling */
616                 pr_info("Testing tracer %s: ", type->name);
617                 ret = type->selftest(type, tr);
618                 /* the test is responsible for resetting too */
619                 current_trace = saved_tracer;
620                 if (ret) {
621                         printk(KERN_CONT "FAILED!\n");
622                         goto out;
623                 }
624                 /* Only reset on passing, to avoid touching corrupted buffers */
625                 for_each_tracing_cpu(i)
626                         tracing_reset(tr, i);
627
628                 printk(KERN_CONT "PASSED\n");
629         }
630 #endif
631
632         type->next = trace_types;
633         trace_types = type;
634         len = strlen(type->name);
635         if (len > max_tracer_type_len)
636                 max_tracer_type_len = len;
637
638  out:
639         tracing_selftest_running = false;
640         mutex_unlock(&trace_types_lock);
641         lock_kernel();
642
643         return ret;
644 }
645
646 void unregister_tracer(struct tracer *type)
647 {
648         struct tracer **t;
649         int len;
650
651         mutex_lock(&trace_types_lock);
652         for (t = &trace_types; *t; t = &(*t)->next) {
653                 if (*t == type)
654                         goto found;
655         }
656         pr_info("Trace %s not registered\n", type->name);
657         goto out;
658
659  found:
660         *t = (*t)->next;
661         if (strlen(type->name) != max_tracer_type_len)
662                 goto out;
663
664         max_tracer_type_len = 0;
665         for (t = &trace_types; *t; t = &(*t)->next) {
666                 len = strlen((*t)->name);
667                 if (len > max_tracer_type_len)
668                         max_tracer_type_len = len;
669         }
670  out:
671         mutex_unlock(&trace_types_lock);
672 }
673
674 void tracing_reset(struct trace_array *tr, int cpu)
675 {
676         ftrace_disable_cpu();
677         ring_buffer_reset_cpu(tr->buffer, cpu);
678         ftrace_enable_cpu();
679 }
680
681 void tracing_reset_online_cpus(struct trace_array *tr)
682 {
683         int cpu;
684
685         tr->time_start = ftrace_now(tr->cpu);
686
687         for_each_online_cpu(cpu)
688                 tracing_reset(tr, cpu);
689 }
690
691 #define SAVED_CMDLINES 128
692 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
693 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
694 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
695 static int cmdline_idx;
696 static DEFINE_SPINLOCK(trace_cmdline_lock);
697
698 /* temporary disable recording */
699 atomic_t trace_record_cmdline_disabled __read_mostly;
700
701 static void trace_init_cmdlines(void)
702 {
703         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
704         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
705         cmdline_idx = 0;
706 }
707
708 static int trace_stop_count;
709 static DEFINE_SPINLOCK(tracing_start_lock);
710
711 /**
712  * ftrace_off_permanent - disable all ftrace code permanently
713  *
714  * This should only be called when a serious anomally has
715  * been detected.  This will turn off the function tracing,
716  * ring buffers, and other tracing utilites. It takes no
717  * locks and can be called from any context.
718  */
719 void ftrace_off_permanent(void)
720 {
721         tracing_disabled = 1;
722         ftrace_stop();
723         tracing_off_permanent();
724 }
725
726 /**
727  * tracing_start - quick start of the tracer
728  *
729  * If tracing is enabled but was stopped by tracing_stop,
730  * this will start the tracer back up.
731  */
732 void tracing_start(void)
733 {
734         struct ring_buffer *buffer;
735         unsigned long flags;
736
737         if (tracing_disabled)
738                 return;
739
740         spin_lock_irqsave(&tracing_start_lock, flags);
741         if (--trace_stop_count)
742                 goto out;
743
744         if (trace_stop_count < 0) {
745                 /* Someone screwed up their debugging */
746                 WARN_ON_ONCE(1);
747                 trace_stop_count = 0;
748                 goto out;
749         }
750
751
752         buffer = global_trace.buffer;
753         if (buffer)
754                 ring_buffer_record_enable(buffer);
755
756         buffer = max_tr.buffer;
757         if (buffer)
758                 ring_buffer_record_enable(buffer);
759
760         ftrace_start();
761  out:
762         spin_unlock_irqrestore(&tracing_start_lock, flags);
763 }
764
765 /**
766  * tracing_stop - quick stop of the tracer
767  *
768  * Light weight way to stop tracing. Use in conjunction with
769  * tracing_start.
770  */
771 void tracing_stop(void)
772 {
773         struct ring_buffer *buffer;
774         unsigned long flags;
775
776         ftrace_stop();
777         spin_lock_irqsave(&tracing_start_lock, flags);
778         if (trace_stop_count++)
779                 goto out;
780
781         buffer = global_trace.buffer;
782         if (buffer)
783                 ring_buffer_record_disable(buffer);
784
785         buffer = max_tr.buffer;
786         if (buffer)
787                 ring_buffer_record_disable(buffer);
788
789  out:
790         spin_unlock_irqrestore(&tracing_start_lock, flags);
791 }
792
793 void trace_stop_cmdline_recording(void);
794
795 static void trace_save_cmdline(struct task_struct *tsk)
796 {
797         unsigned map;
798         unsigned idx;
799
800         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
801                 return;
802
803         /*
804          * It's not the end of the world if we don't get
805          * the lock, but we also don't want to spin
806          * nor do we want to disable interrupts,
807          * so if we miss here, then better luck next time.
808          */
809         if (!spin_trylock(&trace_cmdline_lock))
810                 return;
811
812         idx = map_pid_to_cmdline[tsk->pid];
813         if (idx >= SAVED_CMDLINES) {
814                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
815
816                 map = map_cmdline_to_pid[idx];
817                 if (map <= PID_MAX_DEFAULT)
818                         map_pid_to_cmdline[map] = (unsigned)-1;
819
820                 map_pid_to_cmdline[tsk->pid] = idx;
821
822                 cmdline_idx = idx;
823         }
824
825         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
826
827         spin_unlock(&trace_cmdline_lock);
828 }
829
830 char *trace_find_cmdline(int pid)
831 {
832         char *cmdline = "<...>";
833         unsigned map;
834
835         if (!pid)
836                 return "<idle>";
837
838         if (pid > PID_MAX_DEFAULT)
839                 goto out;
840
841         map = map_pid_to_cmdline[pid];
842         if (map >= SAVED_CMDLINES)
843                 goto out;
844
845         cmdline = saved_cmdlines[map];
846
847  out:
848         return cmdline;
849 }
850
851 void tracing_record_cmdline(struct task_struct *tsk)
852 {
853         if (atomic_read(&trace_record_cmdline_disabled))
854                 return;
855
856         trace_save_cmdline(tsk);
857 }
858
859 void
860 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
861                              int pc)
862 {
863         struct task_struct *tsk = current;
864
865         entry->preempt_count            = pc & 0xff;
866         entry->pid                      = (tsk) ? tsk->pid : 0;
867         entry->tgid                     = (tsk) ? tsk->tgid : 0;
868         entry->flags =
869 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
870                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
871 #else
872                 TRACE_FLAG_IRQS_NOSUPPORT |
873 #endif
874                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
875                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
876                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
877 }
878
879 void
880 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
881                unsigned long ip, unsigned long parent_ip, unsigned long flags,
882                int pc)
883 {
884         struct ring_buffer_event *event;
885         struct ftrace_entry *entry;
886         unsigned long irq_flags;
887
888         /* If we are reading the ring buffer, don't trace */
889         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
890                 return;
891
892         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
893                                          &irq_flags);
894         if (!event)
895                 return;
896         entry   = ring_buffer_event_data(event);
897         tracing_generic_entry_update(&entry->ent, flags, pc);
898         entry->ent.type                 = TRACE_FN;
899         entry->ip                       = ip;
900         entry->parent_ip                = parent_ip;
901         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
902 }
903
904 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
905 static void __trace_graph_entry(struct trace_array *tr,
906                                 struct trace_array_cpu *data,
907                                 struct ftrace_graph_ent *trace,
908                                 unsigned long flags,
909                                 int pc)
910 {
911         struct ring_buffer_event *event;
912         struct ftrace_graph_ent_entry *entry;
913         unsigned long irq_flags;
914
915         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
916                 return;
917
918         event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
919                                          &irq_flags);
920         if (!event)
921                 return;
922         entry   = ring_buffer_event_data(event);
923         tracing_generic_entry_update(&entry->ent, flags, pc);
924         entry->ent.type                 = TRACE_GRAPH_ENT;
925         entry->graph_ent                        = *trace;
926         ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
927 }
928
929 static void __trace_graph_return(struct trace_array *tr,
930                                 struct trace_array_cpu *data,
931                                 struct ftrace_graph_ret *trace,
932                                 unsigned long flags,
933                                 int pc)
934 {
935         struct ring_buffer_event *event;
936         struct ftrace_graph_ret_entry *entry;
937         unsigned long irq_flags;
938
939         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
940                 return;
941
942         event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
943                                          &irq_flags);
944         if (!event)
945                 return;
946         entry   = ring_buffer_event_data(event);
947         tracing_generic_entry_update(&entry->ent, flags, pc);
948         entry->ent.type                 = TRACE_GRAPH_RET;
949         entry->ret                              = *trace;
950         ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
951 }
952 #endif
953
954 void
955 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
956        unsigned long ip, unsigned long parent_ip, unsigned long flags,
957        int pc)
958 {
959         if (likely(!atomic_read(&data->disabled)))
960                 trace_function(tr, data, ip, parent_ip, flags, pc);
961 }
962
963 static void ftrace_trace_stack(struct trace_array *tr,
964                                struct trace_array_cpu *data,
965                                unsigned long flags,
966                                int skip, int pc)
967 {
968 #ifdef CONFIG_STACKTRACE
969         struct ring_buffer_event *event;
970         struct stack_entry *entry;
971         struct stack_trace trace;
972         unsigned long irq_flags;
973
974         if (!(trace_flags & TRACE_ITER_STACKTRACE))
975                 return;
976
977         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
978                                          &irq_flags);
979         if (!event)
980                 return;
981         entry   = ring_buffer_event_data(event);
982         tracing_generic_entry_update(&entry->ent, flags, pc);
983         entry->ent.type         = TRACE_STACK;
984
985         memset(&entry->caller, 0, sizeof(entry->caller));
986
987         trace.nr_entries        = 0;
988         trace.max_entries       = FTRACE_STACK_ENTRIES;
989         trace.skip              = skip;
990         trace.entries           = entry->caller;
991
992         save_stack_trace(&trace);
993         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
994 #endif
995 }
996
997 void __trace_stack(struct trace_array *tr,
998                    struct trace_array_cpu *data,
999                    unsigned long flags,
1000                    int skip)
1001 {
1002         ftrace_trace_stack(tr, data, flags, skip, preempt_count());
1003 }
1004
1005 static void ftrace_trace_userstack(struct trace_array *tr,
1006                    struct trace_array_cpu *data,
1007                    unsigned long flags, int pc)
1008 {
1009 #ifdef CONFIG_STACKTRACE
1010         struct ring_buffer_event *event;
1011         struct userstack_entry *entry;
1012         struct stack_trace trace;
1013         unsigned long irq_flags;
1014
1015         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1016                 return;
1017
1018         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1019                                          &irq_flags);
1020         if (!event)
1021                 return;
1022         entry   = ring_buffer_event_data(event);
1023         tracing_generic_entry_update(&entry->ent, flags, pc);
1024         entry->ent.type         = TRACE_USER_STACK;
1025
1026         memset(&entry->caller, 0, sizeof(entry->caller));
1027
1028         trace.nr_entries        = 0;
1029         trace.max_entries       = FTRACE_STACK_ENTRIES;
1030         trace.skip              = 0;
1031         trace.entries           = entry->caller;
1032
1033         save_stack_trace_user(&trace);
1034         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1035 #endif
1036 }
1037
1038 void __trace_userstack(struct trace_array *tr,
1039                    struct trace_array_cpu *data,
1040                    unsigned long flags)
1041 {
1042         ftrace_trace_userstack(tr, data, flags, preempt_count());
1043 }
1044
1045 static void
1046 ftrace_trace_special(void *__tr, void *__data,
1047                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1048                      int pc)
1049 {
1050         struct ring_buffer_event *event;
1051         struct trace_array_cpu *data = __data;
1052         struct trace_array *tr = __tr;
1053         struct special_entry *entry;
1054         unsigned long irq_flags;
1055
1056         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1057                                          &irq_flags);
1058         if (!event)
1059                 return;
1060         entry   = ring_buffer_event_data(event);
1061         tracing_generic_entry_update(&entry->ent, 0, pc);
1062         entry->ent.type                 = TRACE_SPECIAL;
1063         entry->arg1                     = arg1;
1064         entry->arg2                     = arg2;
1065         entry->arg3                     = arg3;
1066         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1067         ftrace_trace_stack(tr, data, irq_flags, 4, pc);
1068         ftrace_trace_userstack(tr, data, irq_flags, pc);
1069
1070         trace_wake_up();
1071 }
1072
1073 void
1074 __trace_special(void *__tr, void *__data,
1075                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1076 {
1077         ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
1078 }
1079
1080 void
1081 tracing_sched_switch_trace(struct trace_array *tr,
1082                            struct trace_array_cpu *data,
1083                            struct task_struct *prev,
1084                            struct task_struct *next,
1085                            unsigned long flags, int pc)
1086 {
1087         struct ring_buffer_event *event;
1088         struct ctx_switch_entry *entry;
1089         unsigned long irq_flags;
1090
1091         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1092                                            &irq_flags);
1093         if (!event)
1094                 return;
1095         entry   = ring_buffer_event_data(event);
1096         tracing_generic_entry_update(&entry->ent, flags, pc);
1097         entry->ent.type                 = TRACE_CTX;
1098         entry->prev_pid                 = prev->pid;
1099         entry->prev_prio                = prev->prio;
1100         entry->prev_state               = prev->state;
1101         entry->next_pid                 = next->pid;
1102         entry->next_prio                = next->prio;
1103         entry->next_state               = next->state;
1104         entry->next_cpu = task_cpu(next);
1105         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1106         ftrace_trace_stack(tr, data, flags, 5, pc);
1107         ftrace_trace_userstack(tr, data, flags, pc);
1108 }
1109
1110 void
1111 tracing_sched_wakeup_trace(struct trace_array *tr,
1112                            struct trace_array_cpu *data,
1113                            struct task_struct *wakee,
1114                            struct task_struct *curr,
1115                            unsigned long flags, int pc)
1116 {
1117         struct ring_buffer_event *event;
1118         struct ctx_switch_entry *entry;
1119         unsigned long irq_flags;
1120
1121         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1122                                            &irq_flags);
1123         if (!event)
1124                 return;
1125         entry   = ring_buffer_event_data(event);
1126         tracing_generic_entry_update(&entry->ent, flags, pc);
1127         entry->ent.type                 = TRACE_WAKE;
1128         entry->prev_pid                 = curr->pid;
1129         entry->prev_prio                = curr->prio;
1130         entry->prev_state               = curr->state;
1131         entry->next_pid                 = wakee->pid;
1132         entry->next_prio                = wakee->prio;
1133         entry->next_state               = wakee->state;
1134         entry->next_cpu                 = task_cpu(wakee);
1135         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1136         ftrace_trace_stack(tr, data, flags, 6, pc);
1137         ftrace_trace_userstack(tr, data, flags, pc);
1138
1139         trace_wake_up();
1140 }
1141
1142 void
1143 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1144 {
1145         struct trace_array *tr = &global_trace;
1146         struct trace_array_cpu *data;
1147         unsigned long flags;
1148         int cpu;
1149         int pc;
1150
1151         if (tracing_disabled)
1152                 return;
1153
1154         pc = preempt_count();
1155         local_irq_save(flags);
1156         cpu = raw_smp_processor_id();
1157         data = tr->data[cpu];
1158
1159         if (likely(atomic_inc_return(&data->disabled) == 1))
1160                 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1161
1162         atomic_dec(&data->disabled);
1163         local_irq_restore(flags);
1164 }
1165
1166 #ifdef CONFIG_FUNCTION_TRACER
1167 static void
1168 function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1169 {
1170         struct trace_array *tr = &global_trace;
1171         struct trace_array_cpu *data;
1172         unsigned long flags;
1173         long disabled;
1174         int cpu, resched;
1175         int pc;
1176
1177         if (unlikely(!ftrace_function_enabled))
1178                 return;
1179
1180         pc = preempt_count();
1181         resched = ftrace_preempt_disable();
1182         local_save_flags(flags);
1183         cpu = raw_smp_processor_id();
1184         data = tr->data[cpu];
1185         disabled = atomic_inc_return(&data->disabled);
1186
1187         if (likely(disabled == 1))
1188                 trace_function(tr, data, ip, parent_ip, flags, pc);
1189
1190         atomic_dec(&data->disabled);
1191         ftrace_preempt_enable(resched);
1192 }
1193
1194 static void
1195 function_trace_call(unsigned long ip, unsigned long parent_ip)
1196 {
1197         struct trace_array *tr = &global_trace;
1198         struct trace_array_cpu *data;
1199         unsigned long flags;
1200         long disabled;
1201         int cpu;
1202         int pc;
1203
1204         if (unlikely(!ftrace_function_enabled))
1205                 return;
1206
1207         /*
1208          * Need to use raw, since this must be called before the
1209          * recursive protection is performed.
1210          */
1211         local_irq_save(flags);
1212         cpu = raw_smp_processor_id();
1213         data = tr->data[cpu];
1214         disabled = atomic_inc_return(&data->disabled);
1215
1216         if (likely(disabled == 1)) {
1217                 pc = preempt_count();
1218                 trace_function(tr, data, ip, parent_ip, flags, pc);
1219         }
1220
1221         atomic_dec(&data->disabled);
1222         local_irq_restore(flags);
1223 }
1224
1225 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1226 int trace_graph_entry(struct ftrace_graph_ent *trace)
1227 {
1228         struct trace_array *tr = &global_trace;
1229         struct trace_array_cpu *data;
1230         unsigned long flags;
1231         long disabled;
1232         int cpu;
1233         int pc;
1234
1235         if (!ftrace_trace_task(current))
1236                 return 0;
1237
1238         if (!ftrace_graph_addr(trace->func))
1239                 return 0;
1240
1241         local_irq_save(flags);
1242         cpu = raw_smp_processor_id();
1243         data = tr->data[cpu];
1244         disabled = atomic_inc_return(&data->disabled);
1245         if (likely(disabled == 1)) {
1246                 pc = preempt_count();
1247                 __trace_graph_entry(tr, data, trace, flags, pc);
1248         }
1249         /* Only do the atomic if it is not already set */
1250         if (!test_tsk_trace_graph(current))
1251                 set_tsk_trace_graph(current);
1252         atomic_dec(&data->disabled);
1253         local_irq_restore(flags);
1254
1255         return 1;
1256 }
1257
1258 void trace_graph_return(struct ftrace_graph_ret *trace)
1259 {
1260         struct trace_array *tr = &global_trace;
1261         struct trace_array_cpu *data;
1262         unsigned long flags;
1263         long disabled;
1264         int cpu;
1265         int pc;
1266
1267         local_irq_save(flags);
1268         cpu = raw_smp_processor_id();
1269         data = tr->data[cpu];
1270         disabled = atomic_inc_return(&data->disabled);
1271         if (likely(disabled == 1)) {
1272                 pc = preempt_count();
1273                 __trace_graph_return(tr, data, trace, flags, pc);
1274         }
1275         if (!trace->depth)
1276                 clear_tsk_trace_graph(current);
1277         atomic_dec(&data->disabled);
1278         local_irq_restore(flags);
1279 }
1280 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1281
1282 static struct ftrace_ops trace_ops __read_mostly =
1283 {
1284         .func = function_trace_call,
1285 };
1286
1287 void tracing_start_function_trace(void)
1288 {
1289         ftrace_function_enabled = 0;
1290
1291         if (trace_flags & TRACE_ITER_PREEMPTONLY)
1292                 trace_ops.func = function_trace_call_preempt_only;
1293         else
1294                 trace_ops.func = function_trace_call;
1295
1296         register_ftrace_function(&trace_ops);
1297         ftrace_function_enabled = 1;
1298 }
1299
1300 void tracing_stop_function_trace(void)
1301 {
1302         ftrace_function_enabled = 0;
1303         unregister_ftrace_function(&trace_ops);
1304 }
1305 #endif
1306
1307 enum trace_file_type {
1308         TRACE_FILE_LAT_FMT      = 1,
1309         TRACE_FILE_ANNOTATE     = 2,
1310 };
1311
1312 static void trace_iterator_increment(struct trace_iterator *iter)
1313 {
1314         /* Don't allow ftrace to trace into the ring buffers */
1315         ftrace_disable_cpu();
1316
1317         iter->idx++;
1318         if (iter->buffer_iter[iter->cpu])
1319                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1320
1321         ftrace_enable_cpu();
1322 }
1323
1324 static struct trace_entry *
1325 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1326 {
1327         struct ring_buffer_event *event;
1328         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1329
1330         /* Don't allow ftrace to trace into the ring buffers */
1331         ftrace_disable_cpu();
1332
1333         if (buf_iter)
1334                 event = ring_buffer_iter_peek(buf_iter, ts);
1335         else
1336                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1337
1338         ftrace_enable_cpu();
1339
1340         return event ? ring_buffer_event_data(event) : NULL;
1341 }
1342
1343 static struct trace_entry *
1344 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1345 {
1346         struct ring_buffer *buffer = iter->tr->buffer;
1347         struct trace_entry *ent, *next = NULL;
1348         u64 next_ts = 0, ts;
1349         int next_cpu = -1;
1350         int cpu;
1351
1352         for_each_tracing_cpu(cpu) {
1353
1354                 if (ring_buffer_empty_cpu(buffer, cpu))
1355                         continue;
1356
1357                 ent = peek_next_entry(iter, cpu, &ts);
1358
1359                 /*
1360                  * Pick the entry with the smallest timestamp:
1361                  */
1362                 if (ent && (!next || ts < next_ts)) {
1363                         next = ent;
1364                         next_cpu = cpu;
1365                         next_ts = ts;
1366                 }
1367         }
1368
1369         if (ent_cpu)
1370                 *ent_cpu = next_cpu;
1371
1372         if (ent_ts)
1373                 *ent_ts = next_ts;
1374
1375         return next;
1376 }
1377
1378 /* Find the next real entry, without updating the iterator itself */
1379 static struct trace_entry *
1380 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1381 {
1382         return __find_next_entry(iter, ent_cpu, ent_ts);
1383 }
1384
1385 /* Find the next real entry, and increment the iterator to the next entry */
1386 static void *find_next_entry_inc(struct trace_iterator *iter)
1387 {
1388         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1389
1390         if (iter->ent)
1391                 trace_iterator_increment(iter);
1392
1393         return iter->ent ? iter : NULL;
1394 }
1395
1396 static void trace_consume(struct trace_iterator *iter)
1397 {
1398         /* Don't allow ftrace to trace into the ring buffers */
1399         ftrace_disable_cpu();
1400         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1401         ftrace_enable_cpu();
1402 }
1403
1404 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1405 {
1406         struct trace_iterator *iter = m->private;
1407         int i = (int)*pos;
1408         void *ent;
1409
1410         (*pos)++;
1411
1412         /* can't go backwards */
1413         if (iter->idx > i)
1414                 return NULL;
1415
1416         if (iter->idx < 0)
1417                 ent = find_next_entry_inc(iter);
1418         else
1419                 ent = iter;
1420
1421         while (ent && iter->idx < i)
1422                 ent = find_next_entry_inc(iter);
1423
1424         iter->pos = *pos;
1425
1426         return ent;
1427 }
1428
1429 static void *s_start(struct seq_file *m, loff_t *pos)
1430 {
1431         struct trace_iterator *iter = m->private;
1432         void *p = NULL;
1433         loff_t l = 0;
1434         int cpu;
1435
1436         mutex_lock(&trace_types_lock);
1437
1438         if (!current_trace || current_trace != iter->trace) {
1439                 mutex_unlock(&trace_types_lock);
1440                 return NULL;
1441         }
1442
1443         atomic_inc(&trace_record_cmdline_disabled);
1444
1445         if (*pos != iter->pos) {
1446                 iter->ent = NULL;
1447                 iter->cpu = 0;
1448                 iter->idx = -1;
1449
1450                 ftrace_disable_cpu();
1451
1452                 for_each_tracing_cpu(cpu) {
1453                         ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1454                 }
1455
1456                 ftrace_enable_cpu();
1457
1458                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1459                         ;
1460
1461         } else {
1462                 l = *pos - 1;
1463                 p = s_next(m, p, &l);
1464         }
1465
1466         return p;
1467 }
1468
1469 static void s_stop(struct seq_file *m, void *p)
1470 {
1471         atomic_dec(&trace_record_cmdline_disabled);
1472         mutex_unlock(&trace_types_lock);
1473 }
1474
1475 #ifdef CONFIG_KRETPROBES
1476 static inline const char *kretprobed(const char *name)
1477 {
1478         static const char tramp_name[] = "kretprobe_trampoline";
1479         int size = sizeof(tramp_name);
1480
1481         if (strncmp(tramp_name, name, size) == 0)
1482                 return "[unknown/kretprobe'd]";
1483         return name;
1484 }
1485 #else
1486 static inline const char *kretprobed(const char *name)
1487 {
1488         return name;
1489 }
1490 #endif /* CONFIG_KRETPROBES */
1491
1492 static int
1493 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1494 {
1495 #ifdef CONFIG_KALLSYMS
1496         char str[KSYM_SYMBOL_LEN];
1497         const char *name;
1498
1499         kallsyms_lookup(address, NULL, NULL, NULL, str);
1500
1501         name = kretprobed(str);
1502
1503         return trace_seq_printf(s, fmt, name);
1504 #endif
1505         return 1;
1506 }
1507
1508 static int
1509 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1510                      unsigned long address)
1511 {
1512 #ifdef CONFIG_KALLSYMS
1513         char str[KSYM_SYMBOL_LEN];
1514         const char *name;
1515
1516         sprint_symbol(str, address);
1517         name = kretprobed(str);
1518
1519         return trace_seq_printf(s, fmt, name);
1520 #endif
1521         return 1;
1522 }
1523
1524 #ifndef CONFIG_64BIT
1525 # define IP_FMT "%08lx"
1526 #else
1527 # define IP_FMT "%016lx"
1528 #endif
1529
1530 int
1531 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1532 {
1533         int ret;
1534
1535         if (!ip)
1536                 return trace_seq_printf(s, "0");
1537
1538         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1539                 ret = seq_print_sym_offset(s, "%s", ip);
1540         else
1541                 ret = seq_print_sym_short(s, "%s", ip);
1542
1543         if (!ret)
1544                 return 0;
1545
1546         if (sym_flags & TRACE_ITER_SYM_ADDR)
1547                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1548         return ret;
1549 }
1550
1551 static inline int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
1552                                     unsigned long ip, unsigned long sym_flags)
1553 {
1554         struct file *file = NULL;
1555         unsigned long vmstart = 0;
1556         int ret = 1;
1557
1558         if (mm) {
1559                 const struct vm_area_struct *vma;
1560
1561                 down_read(&mm->mmap_sem);
1562                 vma = find_vma(mm, ip);
1563                 if (vma) {
1564                         file = vma->vm_file;
1565                         vmstart = vma->vm_start;
1566                 }
1567                 if (file) {
1568                         ret = trace_seq_path(s, &file->f_path);
1569                         if (ret)
1570                                 ret = trace_seq_printf(s, "[+0x%lx]", ip - vmstart);
1571                 }
1572                 up_read(&mm->mmap_sem);
1573         }
1574         if (ret && ((sym_flags & TRACE_ITER_SYM_ADDR) || !file))
1575                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1576         return ret;
1577 }
1578
1579 static int
1580 seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s,
1581                       unsigned long sym_flags)
1582 {
1583         struct mm_struct *mm = NULL;
1584         int ret = 1;
1585         unsigned int i;
1586
1587         if (trace_flags & TRACE_ITER_SYM_USEROBJ) {
1588                 struct task_struct *task;
1589                 /*
1590                  * we do the lookup on the thread group leader,
1591                  * since individual threads might have already quit!
1592                  */
1593                 rcu_read_lock();
1594                 task = find_task_by_vpid(entry->ent.tgid);
1595                 if (task)
1596                         mm = get_task_mm(task);
1597                 rcu_read_unlock();
1598         }
1599
1600         for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1601                 unsigned long ip = entry->caller[i];
1602
1603                 if (ip == ULONG_MAX || !ret)
1604                         break;
1605                 if (i && ret)
1606                         ret = trace_seq_puts(s, " <- ");
1607                 if (!ip) {
1608                         if (ret)
1609                                 ret = trace_seq_puts(s, "??");
1610                         continue;
1611                 }
1612                 if (!ret)
1613                         break;
1614                 if (ret)
1615                         ret = seq_print_user_ip(s, mm, ip, sym_flags);
1616         }
1617
1618         if (mm)
1619                 mmput(mm);
1620         return ret;
1621 }
1622
1623 static void print_lat_help_header(struct seq_file *m)
1624 {
1625         seq_puts(m, "#                  _------=> CPU#            \n");
1626         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1627         seq_puts(m, "#                | / _----=> need-resched    \n");
1628         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1629         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1630         seq_puts(m, "#                |||| /                      \n");
1631         seq_puts(m, "#                |||||     delay             \n");
1632         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1633         seq_puts(m, "#     \\   /      |||||   \\   |   /           \n");
1634 }
1635
1636 static void print_func_help_header(struct seq_file *m)
1637 {
1638         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1639         seq_puts(m, "#              | |       |          |         |\n");
1640 }
1641
1642
1643 static void
1644 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1645 {
1646         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1647         struct trace_array *tr = iter->tr;
1648         struct trace_array_cpu *data = tr->data[tr->cpu];
1649         struct tracer *type = current_trace;
1650         unsigned long total;
1651         unsigned long entries;
1652         const char *name = "preemption";
1653
1654         if (type)
1655                 name = type->name;
1656
1657         entries = ring_buffer_entries(iter->tr->buffer);
1658         total = entries +
1659                 ring_buffer_overruns(iter->tr->buffer);
1660
1661         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1662                    name, UTS_RELEASE);
1663         seq_puts(m, "-----------------------------------"
1664                  "---------------------------------\n");
1665         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1666                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1667                    nsecs_to_usecs(data->saved_latency),
1668                    entries,
1669                    total,
1670                    tr->cpu,
1671 #if defined(CONFIG_PREEMPT_NONE)
1672                    "server",
1673 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1674                    "desktop",
1675 #elif defined(CONFIG_PREEMPT)
1676                    "preempt",
1677 #else
1678                    "unknown",
1679 #endif
1680                    /* These are reserved for later use */
1681                    0, 0, 0, 0);
1682 #ifdef CONFIG_SMP
1683         seq_printf(m, " #P:%d)\n", num_online_cpus());
1684 #else
1685         seq_puts(m, ")\n");
1686 #endif
1687         seq_puts(m, "    -----------------\n");
1688         seq_printf(m, "    | task: %.16s-%d "
1689                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1690                    data->comm, data->pid, data->uid, data->nice,
1691                    data->policy, data->rt_priority);
1692         seq_puts(m, "    -----------------\n");
1693
1694         if (data->critical_start) {
1695                 seq_puts(m, " => started at: ");
1696                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1697                 trace_print_seq(m, &iter->seq);
1698                 seq_puts(m, "\n => ended at:   ");
1699                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1700                 trace_print_seq(m, &iter->seq);
1701                 seq_puts(m, "\n");
1702         }
1703
1704         seq_puts(m, "\n");
1705 }
1706
1707 static void
1708 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1709 {
1710         int hardirq, softirq;
1711         char *comm;
1712
1713         comm = trace_find_cmdline(entry->pid);
1714
1715         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1716         trace_seq_printf(s, "%3d", cpu);
1717         trace_seq_printf(s, "%c%c",
1718                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1719                          (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1720                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1721
1722         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1723         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1724         if (hardirq && softirq) {
1725                 trace_seq_putc(s, 'H');
1726         } else {
1727                 if (hardirq) {
1728                         trace_seq_putc(s, 'h');
1729                 } else {
1730                         if (softirq)
1731                                 trace_seq_putc(s, 's');
1732                         else
1733                                 trace_seq_putc(s, '.');
1734                 }
1735         }
1736
1737         if (entry->preempt_count)
1738                 trace_seq_printf(s, "%x", entry->preempt_count);
1739         else
1740                 trace_seq_puts(s, ".");
1741 }
1742
1743 unsigned long preempt_mark_thresh = 100;
1744
1745 static void
1746 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1747                     unsigned long rel_usecs)
1748 {
1749         trace_seq_printf(s, " %4lldus", abs_usecs);
1750         if (rel_usecs > preempt_mark_thresh)
1751                 trace_seq_puts(s, "!: ");
1752         else if (rel_usecs > 1)
1753                 trace_seq_puts(s, "+: ");
1754         else
1755                 trace_seq_puts(s, " : ");
1756 }
1757
1758 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1759
1760 static int task_state_char(unsigned long state)
1761 {
1762         int bit = state ? __ffs(state) + 1 : 0;
1763
1764         return bit < sizeof(state_to_char) - 1 ? state_to_char[bit] : '?';
1765 }
1766
1767 /*
1768  * The message is supposed to contain an ending newline.
1769  * If the printing stops prematurely, try to add a newline of our own.
1770  */
1771 void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1772 {
1773         struct trace_entry *ent;
1774         struct trace_field_cont *cont;
1775         bool ok = true;
1776
1777         ent = peek_next_entry(iter, iter->cpu, NULL);
1778         if (!ent || ent->type != TRACE_CONT) {
1779                 trace_seq_putc(s, '\n');
1780                 return;
1781         }
1782
1783         do {
1784                 cont = (struct trace_field_cont *)ent;
1785                 if (ok)
1786                         ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1787
1788                 ftrace_disable_cpu();
1789
1790                 if (iter->buffer_iter[iter->cpu])
1791                         ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1792                 else
1793                         ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1794
1795                 ftrace_enable_cpu();
1796
1797                 ent = peek_next_entry(iter, iter->cpu, NULL);
1798         } while (ent && ent->type == TRACE_CONT);
1799
1800         if (!ok)
1801                 trace_seq_putc(s, '\n');
1802 }
1803
1804 static void test_cpu_buff_start(struct trace_iterator *iter)
1805 {
1806         struct trace_seq *s = &iter->seq;
1807
1808         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1809                 return;
1810
1811         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1812                 return;
1813
1814         if (cpumask_test_cpu(iter->cpu, iter->started))
1815                 return;
1816
1817         cpumask_set_cpu(iter->cpu, iter->started);
1818         trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1819 }
1820
1821 static enum print_line_t
1822 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1823 {
1824         struct trace_seq *s = &iter->seq;
1825         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1826         struct trace_entry *next_entry;
1827         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1828         struct trace_entry *entry = iter->ent;
1829         unsigned long abs_usecs;
1830         unsigned long rel_usecs;
1831         u64 next_ts;
1832         char *comm;
1833         int S, T;
1834         int i;
1835
1836         if (entry->type == TRACE_CONT)
1837                 return TRACE_TYPE_HANDLED;
1838
1839         test_cpu_buff_start(iter);
1840
1841         next_entry = find_next_entry(iter, NULL, &next_ts);
1842         if (!next_entry)
1843                 next_ts = iter->ts;
1844         rel_usecs = ns2usecs(next_ts - iter->ts);
1845         abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1846
1847         if (verbose) {
1848                 comm = trace_find_cmdline(entry->pid);
1849                 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1850                                  " %ld.%03ldms (+%ld.%03ldms): ",
1851                                  comm,
1852                                  entry->pid, cpu, entry->flags,
1853                                  entry->preempt_count, trace_idx,
1854                                  ns2usecs(iter->ts),
1855                                  abs_usecs/1000,
1856                                  abs_usecs % 1000, rel_usecs/1000,
1857                                  rel_usecs % 1000);
1858         } else {
1859                 lat_print_generic(s, entry, cpu);
1860                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1861         }
1862         switch (entry->type) {
1863         case TRACE_FN: {
1864                 struct ftrace_entry *field;
1865
1866                 trace_assign_type(field, entry);
1867
1868                 seq_print_ip_sym(s, field->ip, sym_flags);
1869                 trace_seq_puts(s, " (");
1870                 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1871                 trace_seq_puts(s, ")\n");
1872                 break;
1873         }
1874         case TRACE_CTX:
1875         case TRACE_WAKE: {
1876                 struct ctx_switch_entry *field;
1877
1878                 trace_assign_type(field, entry);
1879
1880                 T = task_state_char(field->next_state);
1881                 S = task_state_char(field->prev_state);
1882                 comm = trace_find_cmdline(field->next_pid);
1883                 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1884                                  field->prev_pid,
1885                                  field->prev_prio,
1886                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1887                                  field->next_cpu,
1888                                  field->next_pid,
1889                                  field->next_prio,
1890                                  T, comm);
1891                 break;
1892         }
1893         case TRACE_SPECIAL: {
1894                 struct special_entry *field;
1895
1896                 trace_assign_type(field, entry);
1897
1898                 trace_seq_printf(s, "# %ld %ld %ld\n",
1899                                  field->arg1,
1900                                  field->arg2,
1901                                  field->arg3);
1902                 break;
1903         }
1904         case TRACE_STACK: {
1905                 struct stack_entry *field;
1906
1907                 trace_assign_type(field, entry);
1908
1909                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1910                         if (i)
1911                                 trace_seq_puts(s, " <= ");
1912                         seq_print_ip_sym(s, field->caller[i], sym_flags);
1913                 }
1914                 trace_seq_puts(s, "\n");
1915                 break;
1916         }
1917         case TRACE_PRINT: {
1918                 struct print_entry *field;
1919
1920                 trace_assign_type(field, entry);
1921
1922                 seq_print_ip_sym(s, field->ip, sym_flags);
1923                 trace_seq_printf(s, ": %s", field->buf);
1924                 if (entry->flags & TRACE_FLAG_CONT)
1925                         trace_seq_print_cont(s, iter);
1926                 break;
1927         }
1928         case TRACE_BRANCH: {
1929                 struct trace_branch *field;
1930
1931                 trace_assign_type(field, entry);
1932
1933                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1934                                  field->correct ? "  ok  " : " MISS ",
1935                                  field->func,
1936                                  field->file,
1937                                  field->line);
1938                 break;
1939         }
1940         case TRACE_USER_STACK: {
1941                 struct userstack_entry *field;
1942
1943                 trace_assign_type(field, entry);
1944
1945                 seq_print_userip_objs(field, s, sym_flags);
1946                 trace_seq_putc(s, '\n');
1947                 break;
1948         }
1949         default:
1950                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1951         }
1952         return TRACE_TYPE_HANDLED;
1953 }
1954
1955 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1956 {
1957         struct trace_seq *s = &iter->seq;
1958         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1959         struct trace_entry *entry;
1960         unsigned long usec_rem;
1961         unsigned long long t;
1962         unsigned long secs;
1963         char *comm;
1964         int ret;
1965         int S, T;
1966         int i;
1967
1968         entry = iter->ent;
1969
1970         if (entry->type == TRACE_CONT)
1971                 return TRACE_TYPE_HANDLED;
1972
1973         test_cpu_buff_start(iter);
1974
1975         comm = trace_find_cmdline(iter->ent->pid);
1976
1977         t = ns2usecs(iter->ts);
1978         usec_rem = do_div(t, 1000000ULL);
1979         secs = (unsigned long)t;
1980
1981         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1982         if (!ret)
1983                 return TRACE_TYPE_PARTIAL_LINE;
1984         ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1985         if (!ret)
1986                 return TRACE_TYPE_PARTIAL_LINE;
1987         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1988         if (!ret)
1989                 return TRACE_TYPE_PARTIAL_LINE;
1990
1991         switch (entry->type) {
1992         case TRACE_FN: {
1993                 struct ftrace_entry *field;
1994
1995                 trace_assign_type(field, entry);
1996
1997                 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1998                 if (!ret)
1999                         return TRACE_TYPE_PARTIAL_LINE;
2000                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
2001                                                 field->parent_ip) {
2002                         ret = trace_seq_printf(s, " <-");
2003                         if (!ret)
2004                                 return TRACE_TYPE_PARTIAL_LINE;
2005                         ret = seq_print_ip_sym(s,
2006                                                field->parent_ip,
2007                                                sym_flags);
2008                         if (!ret)
2009                                 return TRACE_TYPE_PARTIAL_LINE;
2010                 }
2011                 ret = trace_seq_printf(s, "\n");
2012                 if (!ret)
2013                         return TRACE_TYPE_PARTIAL_LINE;
2014                 break;
2015         }
2016         case TRACE_CTX:
2017         case TRACE_WAKE: {
2018                 struct ctx_switch_entry *field;
2019
2020                 trace_assign_type(field, entry);
2021
2022                 T = task_state_char(field->next_state);
2023                 S = task_state_char(field->prev_state);
2024                 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
2025                                        field->prev_pid,
2026                                        field->prev_prio,
2027                                        S,
2028                                        entry->type == TRACE_CTX ? "==>" : "  +",
2029                                        field->next_cpu,
2030                                        field->next_pid,
2031                                        field->next_prio,
2032                                        T);
2033                 if (!ret)
2034                         return TRACE_TYPE_PARTIAL_LINE;
2035                 break;
2036         }
2037         case TRACE_SPECIAL: {
2038                 struct special_entry *field;
2039
2040                 trace_assign_type(field, entry);
2041
2042                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2043                                  field->arg1,
2044                                  field->arg2,
2045                                  field->arg3);
2046                 if (!ret)
2047                         return TRACE_TYPE_PARTIAL_LINE;
2048                 break;
2049         }
2050         case TRACE_STACK: {
2051                 struct stack_entry *field;
2052
2053                 trace_assign_type(field, entry);
2054
2055                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
2056                         if (i) {
2057                                 ret = trace_seq_puts(s, " <= ");
2058                                 if (!ret)
2059                                         return TRACE_TYPE_PARTIAL_LINE;
2060                         }
2061                         ret = seq_print_ip_sym(s, field->caller[i],
2062                                                sym_flags);
2063                         if (!ret)
2064                                 return TRACE_TYPE_PARTIAL_LINE;
2065                 }
2066                 ret = trace_seq_puts(s, "\n");
2067                 if (!ret)
2068                         return TRACE_TYPE_PARTIAL_LINE;
2069                 break;
2070         }
2071         case TRACE_PRINT: {
2072                 struct print_entry *field;
2073
2074                 trace_assign_type(field, entry);
2075
2076                 seq_print_ip_sym(s, field->ip, sym_flags);
2077                 trace_seq_printf(s, ": %s", field->buf);
2078                 if (entry->flags & TRACE_FLAG_CONT)
2079                         trace_seq_print_cont(s, iter);
2080                 break;
2081         }
2082         case TRACE_GRAPH_RET: {
2083                 return print_graph_function(iter);
2084         }
2085         case TRACE_GRAPH_ENT: {
2086                 return print_graph_function(iter);
2087         }
2088         case TRACE_BRANCH: {
2089                 struct trace_branch *field;
2090
2091                 trace_assign_type(field, entry);
2092
2093                 trace_seq_printf(s, "[%s] %s:%s:%d\n",
2094                                  field->correct ? "  ok  " : " MISS ",
2095                                  field->func,
2096                                  field->file,
2097                                  field->line);
2098                 break;
2099         }
2100         case TRACE_USER_STACK: {
2101                 struct userstack_entry *field;
2102
2103                 trace_assign_type(field, entry);
2104
2105                 ret = seq_print_userip_objs(field, s, sym_flags);
2106                 if (!ret)
2107                         return TRACE_TYPE_PARTIAL_LINE;
2108                 ret = trace_seq_putc(s, '\n');
2109                 if (!ret)
2110                         return TRACE_TYPE_PARTIAL_LINE;
2111                 break;
2112         }
2113         }
2114         return TRACE_TYPE_HANDLED;
2115 }
2116
2117 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2118 {
2119         struct trace_seq *s = &iter->seq;
2120         struct trace_entry *entry;
2121         int ret;
2122         int S, T;
2123
2124         entry = iter->ent;
2125
2126         if (entry->type == TRACE_CONT)
2127                 return TRACE_TYPE_HANDLED;
2128
2129         ret = trace_seq_printf(s, "%d %d %llu ",
2130                 entry->pid, iter->cpu, iter->ts);
2131         if (!ret)
2132                 return TRACE_TYPE_PARTIAL_LINE;
2133
2134         switch (entry->type) {
2135         case TRACE_FN: {
2136                 struct ftrace_entry *field;
2137
2138                 trace_assign_type(field, entry);
2139
2140                 ret = trace_seq_printf(s, "%x %x\n",
2141                                         field->ip,
2142                                         field->parent_ip);
2143                 if (!ret)
2144                         return TRACE_TYPE_PARTIAL_LINE;
2145                 break;
2146         }
2147         case TRACE_CTX:
2148         case TRACE_WAKE: {
2149                 struct ctx_switch_entry *field;
2150
2151                 trace_assign_type(field, entry);
2152
2153                 T = task_state_char(field->next_state);
2154                 S = entry->type == TRACE_WAKE ? '+' :
2155                         task_state_char(field->prev_state);
2156                 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
2157                                        field->prev_pid,
2158                                        field->prev_prio,
2159                                        S,
2160                                        field->next_cpu,
2161                                        field->next_pid,
2162                                        field->next_prio,
2163                                        T);
2164                 if (!ret)
2165                         return TRACE_TYPE_PARTIAL_LINE;
2166                 break;
2167         }
2168         case TRACE_SPECIAL:
2169         case TRACE_USER_STACK:
2170         case TRACE_STACK: {
2171                 struct special_entry *field;
2172
2173                 trace_assign_type(field, entry);
2174
2175                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
2176                                  field->arg1,
2177                                  field->arg2,
2178                                  field->arg3);
2179                 if (!ret)
2180                         return TRACE_TYPE_PARTIAL_LINE;
2181                 break;
2182         }
2183         case TRACE_PRINT: {
2184                 struct print_entry *field;
2185
2186                 trace_assign_type(field, entry);
2187
2188                 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
2189                 if (entry->flags & TRACE_FLAG_CONT)
2190                         trace_seq_print_cont(s, iter);
2191                 break;
2192         }
2193         }
2194         return TRACE_TYPE_HANDLED;
2195 }
2196
2197 #define SEQ_PUT_FIELD_RET(s, x)                         \
2198 do {                                                    \
2199         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
2200                 return 0;                               \
2201 } while (0)
2202
2203 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
2204 do {                                                    \
2205         BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES);     \
2206         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
2207                 return 0;                               \
2208 } while (0)
2209
2210 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2211 {
2212         struct trace_seq *s = &iter->seq;
2213         unsigned char newline = '\n';
2214         struct trace_entry *entry;
2215         int S, T;
2216
2217         entry = iter->ent;
2218
2219         if (entry->type == TRACE_CONT)
2220                 return TRACE_TYPE_HANDLED;
2221
2222         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2223         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2224         SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2225
2226         switch (entry->type) {
2227         case TRACE_FN: {
2228                 struct ftrace_entry *field;
2229
2230                 trace_assign_type(field, entry);
2231
2232                 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
2233                 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
2234                 break;
2235         }
2236         case TRACE_CTX:
2237         case TRACE_WAKE: {
2238                 struct ctx_switch_entry *field;
2239
2240                 trace_assign_type(field, entry);
2241
2242                 T = task_state_char(field->next_state);
2243                 S = entry->type == TRACE_WAKE ? '+' :
2244                         task_state_char(field->prev_state);
2245                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
2246                 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
2247                 SEQ_PUT_HEX_FIELD_RET(s, S);
2248                 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
2249                 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
2250                 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
2251                 SEQ_PUT_HEX_FIELD_RET(s, T);
2252                 break;
2253         }
2254         case TRACE_SPECIAL:
2255         case TRACE_USER_STACK:
2256         case TRACE_STACK: {
2257                 struct special_entry *field;
2258
2259                 trace_assign_type(field, entry);
2260
2261                 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
2262                 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
2263                 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
2264                 break;
2265         }
2266         }
2267         SEQ_PUT_FIELD_RET(s, newline);
2268
2269         return TRACE_TYPE_HANDLED;
2270 }
2271
2272 static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
2273 {
2274         struct trace_seq *s = &iter->seq;
2275         struct trace_entry *entry = iter->ent;
2276         struct print_entry *field;
2277         int ret;
2278
2279         trace_assign_type(field, entry);
2280
2281         ret = trace_seq_printf(s, field->buf);
2282         if (!ret)
2283                 return TRACE_TYPE_PARTIAL_LINE;
2284
2285         if (entry->flags & TRACE_FLAG_CONT)
2286                 trace_seq_print_cont(s, iter);
2287
2288         return TRACE_TYPE_HANDLED;
2289 }
2290
2291 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2292 {
2293         struct trace_seq *s = &iter->seq;
2294         struct trace_entry *entry;
2295
2296         entry = iter->ent;
2297
2298         if (entry->type == TRACE_CONT)
2299                 return TRACE_TYPE_HANDLED;
2300
2301         SEQ_PUT_FIELD_RET(s, entry->pid);
2302         SEQ_PUT_FIELD_RET(s, entry->cpu);
2303         SEQ_PUT_FIELD_RET(s, iter->ts);
2304
2305         switch (entry->type) {
2306         case TRACE_FN: {
2307                 struct ftrace_entry *field;
2308
2309                 trace_assign_type(field, entry);
2310
2311                 SEQ_PUT_FIELD_RET(s, field->ip);
2312                 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2313                 break;
2314         }
2315         case TRACE_CTX: {
2316                 struct ctx_switch_entry *field;
2317
2318                 trace_assign_type(field, entry);
2319
2320                 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2321                 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2322                 SEQ_PUT_FIELD_RET(s, field->prev_state);
2323                 SEQ_PUT_FIELD_RET(s, field->next_pid);
2324                 SEQ_PUT_FIELD_RET(s, field->next_prio);
2325                 SEQ_PUT_FIELD_RET(s, field->next_state);
2326                 break;
2327         }
2328         case TRACE_SPECIAL:
2329         case TRACE_USER_STACK:
2330         case TRACE_STACK: {
2331                 struct special_entry *field;
2332
2333                 trace_assign_type(field, entry);
2334
2335                 SEQ_PUT_FIELD_RET(s, field->arg1);
2336                 SEQ_PUT_FIELD_RET(s, field->arg2);
2337                 SEQ_PUT_FIELD_RET(s, field->arg3);
2338                 break;
2339         }
2340         }
2341         return 1;
2342 }
2343
2344 static int trace_empty(struct trace_iterator *iter)
2345 {
2346         int cpu;
2347
2348         for_each_tracing_cpu(cpu) {
2349                 if (iter->buffer_iter[cpu]) {
2350                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2351                                 return 0;
2352                 } else {
2353                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2354                                 return 0;
2355                 }
2356         }
2357
2358         return 1;
2359 }
2360
2361 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2362 {
2363         enum print_line_t ret;
2364
2365         if (iter->trace && iter->trace->print_line) {
2366                 ret = iter->trace->print_line(iter);
2367                 if (ret != TRACE_TYPE_UNHANDLED)
2368                         return ret;
2369         }
2370
2371         if (iter->ent->type == TRACE_PRINT &&
2372                         trace_flags & TRACE_ITER_PRINTK &&
2373                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2374                 return print_printk_msg_only(iter);
2375
2376         if (trace_flags & TRACE_ITER_BIN)
2377                 return print_bin_fmt(iter);
2378
2379         if (trace_flags & TRACE_ITER_HEX)
2380                 return print_hex_fmt(iter);
2381
2382         if (trace_flags & TRACE_ITER_RAW)
2383                 return print_raw_fmt(iter);
2384
2385         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2386                 return print_lat_fmt(iter, iter->idx, iter->cpu);
2387
2388         return print_trace_fmt(iter);
2389 }
2390
2391 static int s_show(struct seq_file *m, void *v)
2392 {
2393         struct trace_iterator *iter = v;
2394
2395         if (iter->ent == NULL) {
2396                 if (iter->tr) {
2397                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2398                         seq_puts(m, "#\n");
2399                 }
2400                 if (iter->trace && iter->trace->print_header)
2401                         iter->trace->print_header(m);
2402                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2403                         /* print nothing if the buffers are empty */
2404                         if (trace_empty(iter))
2405                                 return 0;
2406                         print_trace_header(m, iter);
2407                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2408                                 print_lat_help_header(m);
2409                 } else {
2410                         if (!(trace_flags & TRACE_ITER_VERBOSE))
2411                                 print_func_help_header(m);
2412                 }
2413         } else {
2414                 print_trace_line(iter);
2415                 trace_print_seq(m, &iter->seq);
2416         }
2417
2418         return 0;
2419 }
2420
2421 static struct seq_operations tracer_seq_ops = {
2422         .start          = s_start,
2423         .next           = s_next,
2424         .stop           = s_stop,
2425         .show           = s_show,
2426 };
2427
2428 static struct trace_iterator *
2429 __tracing_open(struct inode *inode, struct file *file, int *ret)
2430 {
2431         struct trace_iterator *iter;
2432         struct seq_file *m;
2433         int cpu;
2434
2435         if (tracing_disabled) {
2436                 *ret = -ENODEV;
2437                 return NULL;
2438         }
2439
2440         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2441         if (!iter) {
2442                 *ret = -ENOMEM;
2443                 goto out;
2444         }
2445
2446         mutex_lock(&trace_types_lock);
2447         if (current_trace && current_trace->print_max)
2448                 iter->tr = &max_tr;
2449         else
2450                 iter->tr = inode->i_private;
2451         iter->trace = current_trace;
2452         iter->pos = -1;
2453
2454         /* Notify the tracer early; before we stop tracing. */
2455         if (iter->trace && iter->trace->open)
2456                 iter->trace->open(iter);
2457
2458         /* Annotate start of buffers if we had overruns */
2459         if (ring_buffer_overruns(iter->tr->buffer))
2460                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2461
2462
2463         for_each_tracing_cpu(cpu) {
2464
2465                 iter->buffer_iter[cpu] =
2466                         ring_buffer_read_start(iter->tr->buffer, cpu);
2467
2468                 if (!iter->buffer_iter[cpu])
2469                         goto fail_buffer;
2470         }
2471
2472         /* TODO stop tracer */
2473         *ret = seq_open(file, &tracer_seq_ops);
2474         if (*ret)
2475                 goto fail_buffer;
2476
2477         m = file->private_data;
2478         m->private = iter;
2479
2480         /* stop the trace while dumping */
2481         tracing_stop();
2482
2483         mutex_unlock(&trace_types_lock);
2484
2485  out:
2486         return iter;
2487
2488  fail_buffer:
2489         for_each_tracing_cpu(cpu) {
2490                 if (iter->buffer_iter[cpu])
2491                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2492         }
2493         mutex_unlock(&trace_types_lock);
2494         kfree(iter);
2495
2496         return ERR_PTR(-ENOMEM);
2497 }
2498
2499 int tracing_open_generic(struct inode *inode, struct file *filp)
2500 {
2501         if (tracing_disabled)
2502                 return -ENODEV;
2503
2504         filp->private_data = inode->i_private;
2505         return 0;
2506 }
2507
2508 int tracing_release(struct inode *inode, struct file *file)
2509 {
2510         struct seq_file *m = (struct seq_file *)file->private_data;
2511         struct trace_iterator *iter = m->private;
2512         int cpu;
2513
2514         mutex_lock(&trace_types_lock);
2515         for_each_tracing_cpu(cpu) {
2516                 if (iter->buffer_iter[cpu])
2517                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2518         }
2519
2520         if (iter->trace && iter->trace->close)
2521                 iter->trace->close(iter);
2522
2523         /* reenable tracing if it was previously enabled */
2524         tracing_start();
2525         mutex_unlock(&trace_types_lock);
2526
2527         seq_release(inode, file);
2528         kfree(iter);
2529         return 0;
2530 }
2531
2532 static int tracing_open(struct inode *inode, struct file *file)
2533 {
2534         int ret;
2535
2536         __tracing_open(inode, file, &ret);
2537
2538         return ret;
2539 }
2540
2541 static int tracing_lt_open(struct inode *inode, struct file *file)
2542 {
2543         struct trace_iterator *iter;
2544         int ret;
2545
2546         iter = __tracing_open(inode, file, &ret);
2547
2548         if (!ret)
2549                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2550
2551         return ret;
2552 }
2553
2554
2555 static void *
2556 t_next(struct seq_file *m, void *v, loff_t *pos)
2557 {
2558         struct tracer *t = m->private;
2559
2560         (*pos)++;
2561
2562         if (t)
2563                 t = t->next;
2564
2565         m->private = t;
2566
2567         return t;
2568 }
2569
2570 static void *t_start(struct seq_file *m, loff_t *pos)
2571 {
2572         struct tracer *t = m->private;
2573         loff_t l = 0;
2574
2575         mutex_lock(&trace_types_lock);
2576         for (; t && l < *pos; t = t_next(m, t, &l))
2577                 ;
2578
2579         return t;
2580 }
2581
2582 static void t_stop(struct seq_file *m, void *p)
2583 {
2584         mutex_unlock(&trace_types_lock);
2585 }
2586
2587 static int t_show(struct seq_file *m, void *v)
2588 {
2589         struct tracer *t = v;
2590
2591         if (!t)
2592                 return 0;
2593
2594         seq_printf(m, "%s", t->name);
2595         if (t->next)
2596                 seq_putc(m, ' ');
2597         else
2598                 seq_putc(m, '\n');
2599
2600         return 0;
2601 }
2602
2603 static struct seq_operations show_traces_seq_ops = {
2604         .start          = t_start,
2605         .next           = t_next,
2606         .stop           = t_stop,
2607         .show           = t_show,
2608 };
2609
2610 static int show_traces_open(struct inode *inode, struct file *file)
2611 {
2612         int ret;
2613
2614         if (tracing_disabled)
2615                 return -ENODEV;
2616
2617         ret = seq_open(file, &show_traces_seq_ops);
2618         if (!ret) {
2619                 struct seq_file *m = file->private_data;
2620                 m->private = trace_types;
2621         }
2622
2623         return ret;
2624 }
2625
2626 static struct file_operations tracing_fops = {
2627         .open           = tracing_open,
2628         .read           = seq_read,
2629         .llseek         = seq_lseek,
2630         .release        = tracing_release,
2631 };
2632
2633 static struct file_operations tracing_lt_fops = {
2634         .open           = tracing_lt_open,
2635         .read           = seq_read,
2636         .llseek         = seq_lseek,
2637         .release        = tracing_release,
2638 };
2639
2640 static struct file_operations show_traces_fops = {
2641         .open           = show_traces_open,
2642         .read           = seq_read,
2643         .release        = seq_release,
2644 };
2645
2646 /*
2647  * Only trace on a CPU if the bitmask is set:
2648  */
2649 static cpumask_var_t tracing_cpumask;
2650
2651 /*
2652  * The tracer itself will not take this lock, but still we want
2653  * to provide a consistent cpumask to user-space:
2654  */
2655 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2656
2657 /*
2658  * Temporary storage for the character representation of the
2659  * CPU bitmask (and one more byte for the newline):
2660  */
2661 static char mask_str[NR_CPUS + 1];
2662
2663 static ssize_t
2664 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2665                      size_t count, loff_t *ppos)
2666 {
2667         int len;
2668
2669         mutex_lock(&tracing_cpumask_update_lock);
2670
2671         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2672         if (count - len < 2) {
2673                 count = -EINVAL;
2674                 goto out_err;
2675         }
2676         len += sprintf(mask_str + len, "\n");
2677         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2678
2679 out_err:
2680         mutex_unlock(&tracing_cpumask_update_lock);
2681
2682         return count;
2683 }
2684
2685 static ssize_t
2686 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2687                       size_t count, loff_t *ppos)
2688 {
2689         int err, cpu;
2690         cpumask_var_t tracing_cpumask_new;
2691
2692         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2693                 return -ENOMEM;
2694
2695         mutex_lock(&tracing_cpumask_update_lock);
2696         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2697         if (err)
2698                 goto err_unlock;
2699
2700         local_irq_disable();
2701         __raw_spin_lock(&ftrace_max_lock);
2702         for_each_tracing_cpu(cpu) {
2703                 /*
2704                  * Increase/decrease the disabled counter if we are
2705                  * about to flip a bit in the cpumask:
2706                  */
2707                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2708                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2709                         atomic_inc(&global_trace.data[cpu]->disabled);
2710                 }
2711                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2712                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2713                         atomic_dec(&global_trace.data[cpu]->disabled);
2714                 }
2715         }
2716         __raw_spin_unlock(&ftrace_max_lock);
2717         local_irq_enable();
2718
2719         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2720
2721         mutex_unlock(&tracing_cpumask_update_lock);
2722         free_cpumask_var(tracing_cpumask_new);
2723
2724         return count;
2725
2726 err_unlock:
2727         mutex_unlock(&tracing_cpumask_update_lock);
2728         free_cpumask_var(tracing_cpumask);
2729
2730         return err;
2731 }
2732
2733 static struct file_operations tracing_cpumask_fops = {
2734         .open           = tracing_open_generic,
2735         .read           = tracing_cpumask_read,
2736         .write          = tracing_cpumask_write,
2737 };
2738
2739 static ssize_t
2740 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2741                        size_t cnt, loff_t *ppos)
2742 {
2743         int i;
2744         char *buf;
2745         int r = 0;
2746         int len = 0;
2747         u32 tracer_flags = current_trace->flags->val;
2748         struct tracer_opt *trace_opts = current_trace->flags->opts;
2749
2750
2751         /* calulate max size */
2752         for (i = 0; trace_options[i]; i++) {
2753                 len += strlen(trace_options[i]);
2754                 len += 3; /* "no" and space */
2755         }
2756
2757         /*
2758          * Increase the size with names of options specific
2759          * of the current tracer.
2760          */
2761         for (i = 0; trace_opts[i].name; i++) {
2762                 len += strlen(trace_opts[i].name);
2763                 len += 3; /* "no" and space */
2764         }
2765
2766         /* +2 for \n and \0 */
2767         buf = kmalloc(len + 2, GFP_KERNEL);
2768         if (!buf)
2769                 return -ENOMEM;
2770
2771         for (i = 0; trace_options[i]; i++) {
2772                 if (trace_flags & (1 << i))
2773                         r += sprintf(buf + r, "%s ", trace_options[i]);
2774                 else
2775                         r += sprintf(buf + r, "no%s ", trace_options[i]);
2776         }
2777
2778         for (i = 0; trace_opts[i].name; i++) {
2779                 if (tracer_flags & trace_opts[i].bit)
2780                         r += sprintf(buf + r, "%s ",
2781                                 trace_opts[i].name);
2782                 else
2783                         r += sprintf(buf + r, "no%s ",
2784                                 trace_opts[i].name);
2785         }
2786
2787         r += sprintf(buf + r, "\n");
2788         WARN_ON(r >= len + 2);
2789
2790         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2791
2792         kfree(buf);
2793
2794         return r;
2795 }
2796
2797 /* Try to assign a tracer specific option */
2798 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2799 {
2800         struct tracer_flags *trace_flags = trace->flags;
2801         struct tracer_opt *opts = NULL;
2802         int ret = 0, i = 0;
2803         int len;
2804
2805         for (i = 0; trace_flags->opts[i].name; i++) {
2806                 opts = &trace_flags->opts[i];
2807                 len = strlen(opts->name);
2808
2809                 if (strncmp(cmp, opts->name, len) == 0) {
2810                         ret = trace->set_flag(trace_flags->val,
2811                                 opts->bit, !neg);
2812                         break;
2813                 }
2814         }
2815         /* Not found */
2816         if (!trace_flags->opts[i].name)
2817                 return -EINVAL;
2818
2819         /* Refused to handle */
2820         if (ret)
2821                 return ret;
2822
2823         if (neg)
2824                 trace_flags->val &= ~opts->bit;
2825         else
2826                 trace_flags->val |= opts->bit;
2827
2828         return 0;
2829 }
2830
2831 static ssize_t
2832 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2833                         size_t cnt, loff_t *ppos)
2834 {
2835         char buf[64];
2836         char *cmp = buf;
2837         int neg = 0;
2838         int ret;
2839         int i;
2840
2841         if (cnt >= sizeof(buf))
2842                 return -EINVAL;
2843
2844         if (copy_from_user(&buf, ubuf, cnt))
2845                 return -EFAULT;
2846
2847         buf[cnt] = 0;
2848
2849         if (strncmp(buf, "no", 2) == 0) {
2850                 neg = 1;
2851                 cmp += 2;
2852         }
2853
2854         for (i = 0; trace_options[i]; i++) {
2855                 int len = strlen(trace_options[i]);
2856
2857                 if (strncmp(cmp, trace_options[i], len) == 0) {
2858                         if (neg)
2859                                 trace_flags &= ~(1 << i);
2860                         else
2861                                 trace_flags |= (1 << i);
2862                         break;
2863                 }
2864         }
2865
2866         /* If no option could be set, test the specific tracer options */
2867         if (!trace_options[i]) {
2868                 ret = set_tracer_option(current_trace, cmp, neg);
2869                 if (ret)
2870                         return ret;
2871         }
2872
2873         filp->f_pos += cnt;
2874
2875         return cnt;
2876 }
2877
2878 static struct file_operations tracing_iter_fops = {
2879         .open           = tracing_open_generic,
2880         .read           = tracing_trace_options_read,
2881         .write          = tracing_trace_options_write,
2882 };
2883
2884 static const char readme_msg[] =
2885         "tracing mini-HOWTO:\n\n"
2886         "# mkdir /debug\n"
2887         "# mount -t debugfs nodev /debug\n\n"
2888         "# cat /debug/tracing/available_tracers\n"
2889         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2890         "# cat /debug/tracing/current_tracer\n"
2891         "none\n"
2892         "# echo sched_switch > /debug/tracing/current_tracer\n"
2893         "# cat /debug/tracing/current_tracer\n"
2894         "sched_switch\n"
2895         "# cat /debug/tracing/trace_options\n"
2896         "noprint-parent nosym-offset nosym-addr noverbose\n"
2897         "# echo print-parent > /debug/tracing/trace_options\n"
2898         "# echo 1 > /debug/tracing/tracing_enabled\n"
2899         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2900         "echo 0 > /debug/tracing/tracing_enabled\n"
2901 ;
2902
2903 static ssize_t
2904 tracing_readme_read(struct file *filp, char __user *ubuf,
2905                        size_t cnt, loff_t *ppos)
2906 {
2907         return simple_read_from_buffer(ubuf, cnt, ppos,
2908                                         readme_msg, strlen(readme_msg));
2909 }
2910
2911 static struct file_operations tracing_readme_fops = {
2912         .open           = tracing_open_generic,
2913         .read           = tracing_readme_read,
2914 };
2915
2916 static ssize_t
2917 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2918                   size_t cnt, loff_t *ppos)
2919 {
2920         char buf[64];
2921         int r;
2922
2923         r = sprintf(buf, "%u\n", tracer_enabled);
2924         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2925 }
2926
2927 static ssize_t
2928 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2929                    size_t cnt, loff_t *ppos)
2930 {
2931         struct trace_array *tr = filp->private_data;
2932         char buf[64];
2933         long val;
2934         int ret;
2935
2936         if (cnt >= sizeof(buf))
2937                 return -EINVAL;
2938
2939         if (copy_from_user(&buf, ubuf, cnt))
2940                 return -EFAULT;
2941
2942         buf[cnt] = 0;
2943
2944         ret = strict_strtoul(buf, 10, &val);
2945         if (ret < 0)
2946                 return ret;
2947
2948         val = !!val;
2949
2950         mutex_lock(&trace_types_lock);
2951         if (tracer_enabled ^ val) {
2952                 if (val) {
2953                         tracer_enabled = 1;
2954                         if (current_trace->start)
2955                                 current_trace->start(tr);
2956                         tracing_start();
2957                 } else {
2958                         tracer_enabled = 0;
2959                         tracing_stop();
2960                         if (current_trace->stop)
2961                                 current_trace->stop(tr);
2962                 }
2963         }
2964         mutex_unlock(&trace_types_lock);
2965
2966         filp->f_pos += cnt;
2967
2968         return cnt;
2969 }
2970
2971 static ssize_t
2972 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2973                        size_t cnt, loff_t *ppos)
2974 {
2975         char buf[max_tracer_type_len+2];
2976         int r;
2977
2978         mutex_lock(&trace_types_lock);
2979         if (current_trace)
2980                 r = sprintf(buf, "%s\n", current_trace->name);
2981         else
2982                 r = sprintf(buf, "\n");
2983         mutex_unlock(&trace_types_lock);
2984
2985         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2986 }
2987
2988 static int tracing_set_tracer(char *buf)
2989 {
2990         struct trace_array *tr = &global_trace;
2991         struct tracer *t;
2992         int ret = 0;
2993
2994         mutex_lock(&trace_types_lock);
2995         for (t = trace_types; t; t = t->next) {
2996                 if (strcmp(t->name, buf) == 0)
2997                         break;
2998         }
2999         if (!t) {
3000                 ret = -EINVAL;
3001                 goto out;
3002         }
3003         if (t == current_trace)
3004                 goto out;
3005
3006         trace_branch_disable();
3007         if (current_trace && current_trace->reset)
3008                 current_trace->reset(tr);
3009
3010         current_trace = t;
3011         if (t->init) {
3012                 ret = t->init(tr);
3013                 if (ret)
3014                         goto out;
3015         }
3016
3017         trace_branch_enable(tr);
3018  out:
3019         mutex_unlock(&trace_types_lock);
3020
3021         return ret;
3022 }
3023
3024 static ssize_t
3025 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3026                         size_t cnt, loff_t *ppos)
3027 {
3028         char buf[max_tracer_type_len+1];
3029         int i;
3030         size_t ret;
3031         int err;
3032
3033         ret = cnt;
3034
3035         if (cnt > max_tracer_type_len)
3036                 cnt = max_tracer_type_len;
3037
3038         if (copy_from_user(&buf, ubuf, cnt))
3039                 return -EFAULT;
3040
3041         buf[cnt] = 0;
3042
3043         /* strip ending whitespace. */
3044         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3045                 buf[i] = 0;
3046
3047         err = tracing_set_tracer(buf);
3048         if (err)
3049                 return err;
3050
3051         filp->f_pos += ret;
3052
3053         return ret;
3054 }
3055
3056 static ssize_t
3057 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3058                      size_t cnt, loff_t *ppos)
3059 {
3060         unsigned long *ptr = filp->private_data;
3061         char buf[64];
3062         int r;
3063
3064         r = snprintf(buf, sizeof(buf), "%ld\n",
3065                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3066         if (r > sizeof(buf))
3067                 r = sizeof(buf);
3068         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3069 }
3070
3071 static ssize_t
3072 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3073                       size_t cnt, loff_t *ppos)
3074 {
3075         long *ptr = filp->private_data;
3076         char buf[64];
3077         long val;
3078         int ret;
3079
3080         if (cnt >= sizeof(buf))
3081                 return -EINVAL;
3082
3083         if (copy_from_user(&buf, ubuf, cnt))
3084                 return -EFAULT;
3085
3086         buf[cnt] = 0;
3087
3088         ret = strict_strtoul(buf, 10, &val);
3089         if (ret < 0)
3090                 return ret;
3091
3092         *ptr = val * 1000;
3093
3094         return cnt;
3095 }
3096
3097 static atomic_t tracing_reader;
3098
3099 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3100 {
3101         struct trace_iterator *iter;
3102
3103         if (tracing_disabled)
3104                 return -ENODEV;
3105
3106         /* We only allow for reader of the pipe */
3107         if (atomic_inc_return(&tracing_reader) != 1) {
3108                 atomic_dec(&tracing_reader);
3109                 return -EBUSY;
3110         }
3111
3112         /* create a buffer to store the information to pass to userspace */
3113         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3114         if (!iter)
3115                 return -ENOMEM;
3116
3117         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3118                 kfree(iter);
3119                 return -ENOMEM;
3120         }
3121
3122         mutex_lock(&trace_types_lock);
3123
3124         /* trace pipe does not show start of buffer */
3125         cpumask_setall(iter->started);
3126
3127         iter->tr = &global_trace;
3128         iter->trace = current_trace;
3129         filp->private_data = iter;
3130
3131         if (iter->trace->pipe_open)
3132                 iter->trace->pipe_open(iter);
3133         mutex_unlock(&trace_types_lock);
3134
3135         return 0;
3136 }
3137
3138 static int tracing_release_pipe(struct inode *inode, struct file *file)
3139 {
3140         struct trace_iterator *iter = file->private_data;
3141
3142         free_cpumask_var(iter->started);
3143         kfree(iter);
3144         atomic_dec(&tracing_reader);
3145
3146         return 0;
3147 }
3148
3149 static unsigned int
3150 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3151 {
3152         struct trace_iterator *iter = filp->private_data;
3153
3154         if (trace_flags & TRACE_ITER_BLOCK) {
3155                 /*
3156                  * Always select as readable when in blocking mode
3157                  */
3158                 return POLLIN | POLLRDNORM;
3159         } else {
3160                 if (!trace_empty(iter))
3161                         return POLLIN | POLLRDNORM;
3162                 poll_wait(filp, &trace_wait, poll_table);
3163                 if (!trace_empty(iter))
3164                         return POLLIN | POLLRDNORM;
3165
3166                 return 0;
3167         }
3168 }
3169
3170 /*
3171  * Consumer reader.
3172  */
3173 static ssize_t
3174 tracing_read_pipe(struct file *filp, char __user *ubuf,
3175                   size_t cnt, loff_t *ppos)
3176 {
3177         struct trace_iterator *iter = filp->private_data;
3178         ssize_t sret;
3179
3180         /* return any leftover data */
3181         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3182         if (sret != -EBUSY)
3183                 return sret;
3184
3185         trace_seq_reset(&iter->seq);
3186
3187         mutex_lock(&trace_types_lock);
3188         if (iter->trace->read) {
3189                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3190                 if (sret)
3191                         goto out;
3192         }
3193
3194 waitagain:
3195         sret = 0;
3196         while (trace_empty(iter)) {
3197
3198                 if ((filp->f_flags & O_NONBLOCK)) {
3199                         sret = -EAGAIN;
3200                         goto out;
3201                 }
3202
3203                 /*
3204                  * This is a make-shift waitqueue. The reason we don't use
3205                  * an actual wait queue is because:
3206                  *  1) we only ever have one waiter
3207                  *  2) the tracing, traces all functions, we don't want
3208                  *     the overhead of calling wake_up and friends
3209                  *     (and tracing them too)
3210                  *     Anyway, this is really very primitive wakeup.
3211                  */
3212                 set_current_state(TASK_INTERRUPTIBLE);
3213                 iter->tr->waiter = current;
3214
3215                 mutex_unlock(&trace_types_lock);
3216
3217                 /* sleep for 100 msecs, and try again. */
3218                 schedule_timeout(HZ/10);
3219
3220                 mutex_lock(&trace_types_lock);
3221
3222                 iter->tr->waiter = NULL;
3223
3224                 if (signal_pending(current)) {
3225                         sret = -EINTR;
3226                         goto out;
3227                 }
3228
3229                 if (iter->trace != current_trace)
3230                         goto out;
3231
3232                 /*
3233                  * We block until we read something and tracing is disabled.
3234                  * We still block if tracing is disabled, but we have never
3235                  * read anything. This allows a user to cat this file, and
3236                  * then enable tracing. But after we have read something,
3237                  * we give an EOF when tracing is again disabled.
3238                  *
3239                  * iter->pos will be 0 if we haven't read anything.
3240                  */
3241                 if (!tracer_enabled && iter->pos)
3242                         break;
3243
3244                 continue;
3245         }
3246
3247         /* stop when tracing is finished */
3248         if (trace_empty(iter))
3249                 goto out;
3250
3251         if (cnt >= PAGE_SIZE)
3252                 cnt = PAGE_SIZE - 1;
3253
3254         /* reset all but tr, trace, and overruns */
3255         memset(&iter->seq, 0,
3256                sizeof(struct trace_iterator) -
3257                offsetof(struct trace_iterator, seq));
3258         iter->pos = -1;
3259
3260         while (find_next_entry_inc(iter) != NULL) {
3261                 enum print_line_t ret;
3262                 int len = iter->seq.len;
3263
3264                 ret = print_trace_line(iter);
3265                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3266                         /* don't print partial lines */
3267                         iter->seq.len = len;
3268                         break;
3269                 }
3270
3271                 trace_consume(iter);
3272
3273                 if (iter->seq.len >= cnt)
3274                         break;
3275         }
3276
3277         /* Now copy what we have to the user */
3278         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3279         if (iter->seq.readpos >= iter->seq.len)
3280                 trace_seq_reset(&iter->seq);
3281
3282         /*
3283          * If there was nothing to send to user, inspite of consuming trace
3284          * entries, go back to wait for more entries.
3285          */
3286         if (sret == -EBUSY)
3287                 goto waitagain;
3288
3289 out:
3290         mutex_unlock(&trace_types_lock);
3291
3292         return sret;
3293 }
3294
3295 static ssize_t
3296 tracing_entries_read(struct file *filp, char __user *ubuf,
3297                      size_t cnt, loff_t *ppos)
3298 {
3299         struct trace_array *tr = filp->private_data;
3300         char buf[64];
3301         int r;
3302
3303         r = sprintf(buf, "%lu\n", tr->entries >> 10);
3304         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3305 }
3306
3307 static ssize_t
3308 tracing_entries_write(struct file *filp, const char __user *ubuf,
3309                       size_t cnt, loff_t *ppos)
3310 {
3311         unsigned long val;
3312         char buf[64];
3313         int ret, cpu;
3314
3315         if (cnt >= sizeof(buf))
3316                 return -EINVAL;
3317
3318         if (copy_from_user(&buf, ubuf, cnt))
3319                 return -EFAULT;
3320
3321         buf[cnt] = 0;
3322
3323         ret = strict_strtoul(buf, 10, &val);
3324         if (ret < 0)
3325                 return ret;
3326
3327         /* must have at least 1 entry */
3328         if (!val)
3329                 return -EINVAL;
3330
3331         mutex_lock(&trace_types_lock);
3332
3333         tracing_stop();
3334
3335         /* disable all cpu buffers */
3336         for_each_tracing_cpu(cpu) {
3337                 if (global_trace.data[cpu])
3338                         atomic_inc(&global_trace.data[cpu]->disabled);
3339                 if (max_tr.data[cpu])
3340                         atomic_inc(&max_tr.data[cpu]->disabled);
3341         }
3342
3343         /* value is in KB */
3344         val <<= 10;
3345
3346         if (val != global_trace.entries) {
3347                 ret = ring_buffer_resize(global_trace.buffer, val);
3348                 if (ret < 0) {
3349                         cnt = ret;
3350                         goto out;
3351                 }
3352
3353                 ret = ring_buffer_resize(max_tr.buffer, val);
3354                 if (ret < 0) {
3355                         int r;
3356                         cnt = ret;
3357                         r = ring_buffer_resize(global_trace.buffer,
3358                                                global_trace.entries);
3359                         if (r < 0) {
3360                                 /* AARGH! We are left with different
3361                                  * size max buffer!!!! */
3362                                 WARN_ON(1);
3363                                 tracing_disabled = 1;
3364                         }
3365                         goto out;
3366                 }
3367
3368                 global_trace.entries = val;
3369         }
3370
3371         filp->f_pos += cnt;
3372
3373         /* If check pages failed, return ENOMEM */
3374         if (tracing_disabled)
3375                 cnt = -ENOMEM;
3376  out:
3377         for_each_tracing_cpu(cpu) {
3378                 if (global_trace.data[cpu])
3379                         atomic_dec(&global_trace.data[cpu]->disabled);
3380                 if (max_tr.data[cpu])
3381                         atomic_dec(&max_tr.data[cpu]->disabled);
3382         }
3383
3384         tracing_start();
3385         max_tr.entries = global_trace.entries;
3386         mutex_unlock(&trace_types_lock);
3387
3388         return cnt;
3389 }
3390
3391 static int mark_printk(const char *fmt, ...)
3392 {
3393         int ret;
3394         va_list args;
3395         va_start(args, fmt);
3396         ret = trace_vprintk(0, -1, fmt, args);
3397         va_end(args);
3398         return ret;
3399 }
3400
3401 static ssize_t
3402 tracing_mark_write(struct file *filp, const char __user *ubuf,
3403                                         size_t cnt, loff_t *fpos)
3404 {
3405         char *buf;
3406         char *end;
3407
3408         if (tracing_disabled)
3409                 return -EINVAL;
3410
3411         if (cnt > TRACE_BUF_SIZE)
3412                 cnt = TRACE_BUF_SIZE;
3413
3414         buf = kmalloc(cnt + 1, GFP_KERNEL);
3415         if (buf == NULL)
3416                 return -ENOMEM;
3417
3418         if (copy_from_user(buf, ubuf, cnt)) {
3419                 kfree(buf);
3420                 return -EFAULT;
3421         }
3422
3423         /* Cut from the first nil or newline. */
3424         buf[cnt] = '\0';
3425         end = strchr(buf, '\n');
3426         if (end)
3427                 *end = '\0';
3428
3429         cnt = mark_printk("%s\n", buf);
3430         kfree(buf);
3431         *fpos += cnt;
3432
3433         return cnt;
3434 }
3435
3436 static struct file_operations tracing_max_lat_fops = {
3437         .open           = tracing_open_generic,
3438         .read           = tracing_max_lat_read,
3439         .write          = tracing_max_lat_write,
3440 };
3441
3442 static struct file_operations tracing_ctrl_fops = {
3443         .open           = tracing_open_generic,
3444         .read           = tracing_ctrl_read,
3445         .write          = tracing_ctrl_write,
3446 };
3447
3448 static struct file_operations set_tracer_fops = {
3449         .open           = tracing_open_generic,
3450         .read           = tracing_set_trace_read,
3451         .write          = tracing_set_trace_write,
3452 };
3453
3454 static struct file_operations tracing_pipe_fops = {
3455         .open           = tracing_open_pipe,
3456         .poll           = tracing_poll_pipe,
3457         .read           = tracing_read_pipe,
3458         .release        = tracing_release_pipe,
3459 };
3460
3461 static struct file_operations tracing_entries_fops = {
3462         .open           = tracing_open_generic,
3463         .read           = tracing_entries_read,
3464         .write          = tracing_entries_write,
3465 };
3466
3467 static struct file_operations tracing_mark_fops = {
3468         .open           = tracing_open_generic,
3469         .write          = tracing_mark_write,
3470 };
3471
3472 #ifdef CONFIG_DYNAMIC_FTRACE
3473
3474 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3475 {
3476         return 0;
3477 }
3478
3479 static ssize_t
3480 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3481                   size_t cnt, loff_t *ppos)
3482 {
3483         static char ftrace_dyn_info_buffer[1024];
3484         static DEFINE_MUTEX(dyn_info_mutex);
3485         unsigned long *p = filp->private_data;
3486         char *buf = ftrace_dyn_info_buffer;
3487         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3488         int r;
3489
3490         mutex_lock(&dyn_info_mutex);
3491         r = sprintf(buf, "%ld ", *p);
3492
3493         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3494         buf[r++] = '\n';
3495
3496         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3497
3498         mutex_unlock(&dyn_info_mutex);
3499
3500         return r;
3501 }
3502
3503 static struct file_operations tracing_dyn_info_fops = {
3504         .open           = tracing_open_generic,
3505         .read           = tracing_read_dyn_info,
3506 };
3507 #endif
3508
3509 static struct dentry *d_tracer;
3510
3511 struct dentry *tracing_init_dentry(void)
3512 {
3513         static int once;
3514
3515         if (d_tracer)
3516                 return d_tracer;
3517
3518         d_tracer = debugfs_create_dir("tracing", NULL);
3519
3520         if (!d_tracer && !once) {
3521                 once = 1;
3522                 pr_warning("Could not create debugfs directory 'tracing'\n");
3523                 return NULL;
3524         }
3525
3526         return d_tracer;
3527 }
3528
3529 #ifdef CONFIG_FTRACE_SELFTEST
3530 /* Let selftest have access to static functions in this file */
3531 #include "trace_selftest.c"
3532 #endif
3533
3534 static __init int tracer_init_debugfs(void)
3535 {
3536         struct dentry *d_tracer;
3537         struct dentry *entry;
3538
3539         d_tracer = tracing_init_dentry();
3540
3541         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3542                                     &global_trace, &tracing_ctrl_fops);
3543         if (!entry)
3544                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3545
3546         entry = debugfs_create_file("trace_options", 0644, d_tracer,
3547                                     NULL, &tracing_iter_fops);
3548         if (!entry)
3549                 pr_warning("Could not create debugfs 'trace_options' entry\n");
3550
3551         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3552                                     NULL, &tracing_cpumask_fops);
3553         if (!entry)
3554                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3555
3556         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
3557                                     &global_trace, &tracing_lt_fops);
3558         if (!entry)
3559                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3560
3561         entry = debugfs_create_file("trace", 0444, d_tracer,
3562                                     &global_trace, &tracing_fops);
3563         if (!entry)
3564                 pr_warning("Could not create debugfs 'trace' entry\n");
3565
3566         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3567                                     &global_trace, &show_traces_fops);
3568         if (!entry)
3569                 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3570
3571         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3572                                     &global_trace, &set_tracer_fops);
3573         if (!entry)
3574                 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3575
3576         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3577                                     &tracing_max_latency,
3578                                     &tracing_max_lat_fops);
3579         if (!entry)
3580                 pr_warning("Could not create debugfs "
3581                            "'tracing_max_latency' entry\n");
3582
3583         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3584                                     &tracing_thresh, &tracing_max_lat_fops);
3585         if (!entry)
3586                 pr_warning("Could not create debugfs "
3587                            "'tracing_thresh' entry\n");
3588         entry = debugfs_create_file("README", 0644, d_tracer,
3589                                     NULL, &tracing_readme_fops);
3590         if (!entry)
3591                 pr_warning("Could not create debugfs 'README' entry\n");
3592
3593         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
3594                                     NULL, &tracing_pipe_fops);
3595         if (!entry)
3596                 pr_warning("Could not create debugfs "
3597                            "'trace_pipe' entry\n");
3598
3599         entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3600                                     &global_trace, &tracing_entries_fops);
3601         if (!entry)
3602                 pr_warning("Could not create debugfs "
3603                            "'buffer_size_kb' entry\n");
3604
3605         entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3606                                     NULL, &tracing_mark_fops);
3607         if (!entry)
3608                 pr_warning("Could not create debugfs "
3609                            "'trace_marker' entry\n");
3610
3611 #ifdef CONFIG_DYNAMIC_FTRACE
3612         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3613                                     &ftrace_update_tot_cnt,
3614                                     &tracing_dyn_info_fops);
3615         if (!entry)
3616                 pr_warning("Could not create debugfs "
3617                            "'dyn_ftrace_total_info' entry\n");
3618 #endif
3619 #ifdef CONFIG_SYSPROF_TRACER
3620         init_tracer_sysprof_debugfs(d_tracer);
3621 #endif
3622         return 0;
3623 }
3624
3625 int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
3626 {
3627         static DEFINE_SPINLOCK(trace_buf_lock);
3628         static char trace_buf[TRACE_BUF_SIZE];
3629
3630         struct ring_buffer_event *event;
3631         struct trace_array *tr = &global_trace;
3632         struct trace_array_cpu *data;
3633         int cpu, len = 0, size, pc;
3634         struct print_entry *entry;
3635         unsigned long irq_flags;
3636
3637         if (tracing_disabled || tracing_selftest_running)
3638                 return 0;
3639
3640         pc = preempt_count();
3641         preempt_disable_notrace();
3642         cpu = raw_smp_processor_id();
3643         data = tr->data[cpu];
3644
3645         if (unlikely(atomic_read(&data->disabled)))
3646                 goto out;
3647
3648         pause_graph_tracing();
3649         spin_lock_irqsave(&trace_buf_lock, irq_flags);
3650         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3651
3652         len = min(len, TRACE_BUF_SIZE-1);
3653         trace_buf[len] = 0;
3654
3655         size = sizeof(*entry) + len + 1;
3656         event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3657         if (!event)
3658                 goto out_unlock;
3659         entry = ring_buffer_event_data(event);
3660         tracing_generic_entry_update(&entry->ent, irq_flags, pc);
3661         entry->ent.type                 = TRACE_PRINT;
3662         entry->ip                       = ip;
3663         entry->depth                    = depth;
3664
3665         memcpy(&entry->buf, trace_buf, len);
3666         entry->buf[len] = 0;
3667         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3668
3669  out_unlock:
3670         spin_unlock_irqrestore(&trace_buf_lock, irq_flags);
3671         unpause_graph_tracing();
3672  out:
3673         preempt_enable_notrace();
3674
3675         return len;
3676 }
3677 EXPORT_SYMBOL_GPL(trace_vprintk);
3678
3679 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3680 {
3681         int ret;
3682         va_list ap;
3683
3684         if (!(trace_flags & TRACE_ITER_PRINTK))
3685                 return 0;
3686
3687         va_start(ap, fmt);
3688         ret = trace_vprintk(ip, task_curr_ret_stack(current), fmt, ap);
3689         va_end(ap);
3690         return ret;
3691 }
3692 EXPORT_SYMBOL_GPL(__ftrace_printk);
3693
3694 static int trace_panic_handler(struct notifier_block *this,
3695                                unsigned long event, void *unused)
3696 {
3697         if (ftrace_dump_on_oops)
3698                 ftrace_dump();
3699         return NOTIFY_OK;
3700 }
3701
3702 static struct notifier_block trace_panic_notifier = {
3703         .notifier_call  = trace_panic_handler,
3704         .next           = NULL,
3705         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
3706 };
3707
3708 static int trace_die_handler(struct notifier_block *self,
3709                              unsigned long val,
3710                              void *data)
3711 {
3712         switch (val) {
3713         case DIE_OOPS:
3714                 if (ftrace_dump_on_oops)
3715                         ftrace_dump();
3716                 break;
3717         default:
3718                 break;
3719         }
3720         return NOTIFY_OK;
3721 }
3722
3723 static struct notifier_block trace_die_notifier = {
3724         .notifier_call = trace_die_handler,
3725         .priority = 200
3726 };
3727
3728 /*
3729  * printk is set to max of 1024, we really don't need it that big.
3730  * Nothing should be printing 1000 characters anyway.
3731  */
3732 #define TRACE_MAX_PRINT         1000
3733
3734 /*
3735  * Define here KERN_TRACE so that we have one place to modify
3736  * it if we decide to change what log level the ftrace dump
3737  * should be at.
3738  */
3739 #define KERN_TRACE              KERN_EMERG
3740
3741 static void
3742 trace_printk_seq(struct trace_seq *s)
3743 {
3744         /* Probably should print a warning here. */
3745         if (s->len >= 1000)
3746                 s->len = 1000;
3747
3748         /* should be zero ended, but we are paranoid. */
3749         s->buffer[s->len] = 0;
3750
3751         printk(KERN_TRACE "%s", s->buffer);
3752
3753         trace_seq_reset(s);
3754 }
3755
3756 void ftrace_dump(void)
3757 {
3758         static DEFINE_SPINLOCK(ftrace_dump_lock);
3759         /* use static because iter can be a bit big for the stack */
3760         static struct trace_iterator iter;
3761         static int dump_ran;
3762         unsigned long flags;
3763         int cnt = 0, cpu;
3764
3765         /* only one dump */
3766         spin_lock_irqsave(&ftrace_dump_lock, flags);
3767         if (dump_ran)
3768                 goto out;
3769
3770         dump_ran = 1;
3771
3772         /* No turning back! */
3773         tracing_off();
3774         ftrace_kill();
3775
3776         for_each_tracing_cpu(cpu) {
3777                 atomic_inc(&global_trace.data[cpu]->disabled);
3778         }
3779
3780         /* don't look at user memory in panic mode */
3781         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3782
3783         printk(KERN_TRACE "Dumping ftrace buffer:\n");
3784
3785         iter.tr = &global_trace;
3786         iter.trace = current_trace;
3787
3788         /*
3789          * We need to stop all tracing on all CPUS to read the
3790          * the next buffer. This is a bit expensive, but is
3791          * not done often. We fill all what we can read,
3792          * and then release the locks again.
3793          */
3794
3795         while (!trace_empty(&iter)) {
3796
3797                 if (!cnt)
3798                         printk(KERN_TRACE "---------------------------------\n");
3799
3800                 cnt++;
3801
3802                 /* reset all but tr, trace, and overruns */
3803                 memset(&iter.seq, 0,
3804                        sizeof(struct trace_iterator) -
3805                        offsetof(struct trace_iterator, seq));
3806                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3807                 iter.pos = -1;
3808
3809                 if (find_next_entry_inc(&iter) != NULL) {
3810                         print_trace_line(&iter);
3811                         trace_consume(&iter);
3812                 }
3813
3814                 trace_printk_seq(&iter.seq);
3815         }
3816
3817         if (!cnt)
3818                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
3819         else
3820                 printk(KERN_TRACE "---------------------------------\n");
3821
3822  out:
3823         spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3824 }
3825
3826 __init static int tracer_alloc_buffers(void)
3827 {
3828         struct trace_array_cpu *data;
3829         int i;
3830         int ret = -ENOMEM;
3831
3832         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
3833                 goto out;
3834
3835         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
3836                 goto out_free_buffer_mask;
3837
3838         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
3839         cpumask_copy(tracing_cpumask, cpu_all_mask);
3840
3841         /* TODO: make the number of buffers hot pluggable with CPUS */
3842         global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3843                                                    TRACE_BUFFER_FLAGS);
3844         if (!global_trace.buffer) {
3845                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3846                 WARN_ON(1);
3847                 goto out_free_cpumask;
3848         }
3849         global_trace.entries = ring_buffer_size(global_trace.buffer);
3850
3851
3852 #ifdef CONFIG_TRACER_MAX_TRACE
3853         max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3854                                              TRACE_BUFFER_FLAGS);
3855         if (!max_tr.buffer) {
3856                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3857                 WARN_ON(1);
3858                 ring_buffer_free(global_trace.buffer);
3859                 goto out_free_cpumask;
3860         }
3861         max_tr.entries = ring_buffer_size(max_tr.buffer);
3862         WARN_ON(max_tr.entries != global_trace.entries);
3863 #endif
3864
3865         /* Allocate the first page for all buffers */
3866         for_each_tracing_cpu(i) {
3867                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3868                 max_tr.data[i] = &per_cpu(max_data, i);
3869         }
3870
3871         trace_init_cmdlines();
3872
3873         register_tracer(&nop_trace);
3874 #ifdef CONFIG_BOOT_TRACER
3875         register_tracer(&boot_tracer);
3876         current_trace = &boot_tracer;
3877         current_trace->init(&global_trace);
3878 #else
3879         current_trace = &nop_trace;
3880 #endif
3881
3882         /* All seems OK, enable tracing */
3883         tracing_disabled = 0;
3884
3885         atomic_notifier_chain_register(&panic_notifier_list,
3886                                        &trace_panic_notifier);
3887
3888         register_die_notifier(&trace_die_notifier);
3889         ret = 0;
3890
3891 out_free_cpumask:
3892         free_cpumask_var(tracing_cpumask);
3893 out_free_buffer_mask:
3894         free_cpumask_var(tracing_buffer_mask);
3895 out:
3896         return ret;
3897 }
3898 early_initcall(tracer_alloc_buffers);
3899 fs_initcall(tracer_init_debugfs);