]> git.karo-electronics.de Git - mv-sheeva.git/blob - kernel/trace/trace.c
tracing: add global-clock option to provide cross CPU clock to traces
[mv-sheeva.git] / kernel / trace / trace.c
1 /*
2  * ring buffer based function tracer
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally taken from the RT patch by:
8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code from the latency_tracer, that is:
11  *  Copyright (C) 2004-2006 Ingo Molnar
12  *  Copyright (C) 2004 William Lee Irwin III
13  */
14 #include <linux/ring_buffer.h>
15 #include <linux/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/ctype.h>
34 #include <linux/init.h>
35 #include <linux/poll.h>
36 #include <linux/gfp.h>
37 #include <linux/fs.h>
38
39 #include "trace.h"
40 #include "trace_output.h"
41
42 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
43
44 unsigned long __read_mostly     tracing_max_latency;
45 unsigned long __read_mostly     tracing_thresh;
46
47 /*
48  * On boot up, the ring buffer is set to the minimum size, so that
49  * we do not waste memory on systems that are not using tracing.
50  */
51 static int ring_buffer_expanded;
52
53 /*
54  * We need to change this state when a selftest is running.
55  * A selftest will lurk into the ring-buffer to count the
56  * entries inserted during the selftest although some concurrent
57  * insertions into the ring-buffer such as trace_printk could occurred
58  * at the same time, giving false positive or negative results.
59  */
60 static bool __read_mostly tracing_selftest_running;
61
62 /*
63  * If a tracer is running, we do not want to run SELFTEST.
64  */
65 static bool __read_mostly tracing_selftest_disabled;
66
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt[] = {
69         { }
70 };
71
72 static struct tracer_flags dummy_tracer_flags = {
73         .val = 0,
74         .opts = dummy_tracer_opt
75 };
76
77 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
78 {
79         return 0;
80 }
81
82 /*
83  * Kill all tracing for good (never come back).
84  * It is initialized to 1 but will turn to zero if the initialization
85  * of the tracer is successful. But that is the only place that sets
86  * this back to zero.
87  */
88 static int tracing_disabled = 1;
89
90 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
91
92 static inline void ftrace_disable_cpu(void)
93 {
94         preempt_disable();
95         local_inc(&__get_cpu_var(ftrace_cpu_disabled));
96 }
97
98 static inline void ftrace_enable_cpu(void)
99 {
100         local_dec(&__get_cpu_var(ftrace_cpu_disabled));
101         preempt_enable();
102 }
103
104 static cpumask_var_t __read_mostly      tracing_buffer_mask;
105
106 /* Define which cpu buffers are currently read in trace_pipe */
107 static cpumask_var_t                    tracing_reader_cpumask;
108
109 #define for_each_tracing_cpu(cpu)       \
110         for_each_cpu(cpu, tracing_buffer_mask)
111
112 /*
113  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
114  *
115  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
116  * is set, then ftrace_dump is called. This will output the contents
117  * of the ftrace buffers to the console.  This is very useful for
118  * capturing traces that lead to crashes and outputing it to a
119  * serial console.
120  *
121  * It is default off, but you can enable it with either specifying
122  * "ftrace_dump_on_oops" in the kernel command line, or setting
123  * /proc/sys/kernel/ftrace_dump_on_oops to true.
124  */
125 int ftrace_dump_on_oops;
126
127 static int tracing_set_tracer(const char *buf);
128
129 #define BOOTUP_TRACER_SIZE              100
130 static char bootup_tracer_buf[BOOTUP_TRACER_SIZE] __initdata;
131 static char *default_bootup_tracer;
132
133 static int __init set_ftrace(char *str)
134 {
135         strncpy(bootup_tracer_buf, str, BOOTUP_TRACER_SIZE);
136         default_bootup_tracer = bootup_tracer_buf;
137         /* We are using ftrace early, expand it */
138         ring_buffer_expanded = 1;
139         return 1;
140 }
141 __setup("ftrace=", set_ftrace);
142
143 static int __init set_ftrace_dump_on_oops(char *str)
144 {
145         ftrace_dump_on_oops = 1;
146         return 1;
147 }
148 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
149
150 long
151 ns2usecs(cycle_t nsec)
152 {
153         nsec += 500;
154         do_div(nsec, 1000);
155         return nsec;
156 }
157
158 /*
159  * The global_trace is the descriptor that holds the tracing
160  * buffers for the live tracing. For each CPU, it contains
161  * a link list of pages that will store trace entries. The
162  * page descriptor of the pages in the memory is used to hold
163  * the link list by linking the lru item in the page descriptor
164  * to each of the pages in the buffer per CPU.
165  *
166  * For each active CPU there is a data field that holds the
167  * pages for the buffer for that CPU. Each CPU has the same number
168  * of pages allocated for its buffer.
169  */
170 static struct trace_array       global_trace;
171
172 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
173
174 cycle_t ftrace_now(int cpu)
175 {
176         u64 ts;
177
178         /* Early boot up does not have a buffer yet */
179         if (!global_trace.buffer)
180                 return trace_clock_local();
181
182         ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
183         ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
184
185         return ts;
186 }
187
188 /*
189  * The max_tr is used to snapshot the global_trace when a maximum
190  * latency is reached. Some tracers will use this to store a maximum
191  * trace while it continues examining live traces.
192  *
193  * The buffers for the max_tr are set up the same as the global_trace.
194  * When a snapshot is taken, the link list of the max_tr is swapped
195  * with the link list of the global_trace and the buffers are reset for
196  * the global_trace so the tracing can continue.
197  */
198 static struct trace_array       max_tr;
199
200 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
201
202 /* tracer_enabled is used to toggle activation of a tracer */
203 static int                      tracer_enabled = 1;
204
205 /**
206  * tracing_is_enabled - return tracer_enabled status
207  *
208  * This function is used by other tracers to know the status
209  * of the tracer_enabled flag.  Tracers may use this function
210  * to know if it should enable their features when starting
211  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
212  */
213 int tracing_is_enabled(void)
214 {
215         return tracer_enabled;
216 }
217
218 /*
219  * trace_buf_size is the size in bytes that is allocated
220  * for a buffer. Note, the number of bytes is always rounded
221  * to page size.
222  *
223  * This number is purposely set to a low number of 16384.
224  * If the dump on oops happens, it will be much appreciated
225  * to not have to wait for all that output. Anyway this can be
226  * boot time and run time configurable.
227  */
228 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
229
230 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
231
232 /* trace_types holds a link list of available tracers. */
233 static struct tracer            *trace_types __read_mostly;
234
235 /* current_trace points to the tracer that is currently active */
236 static struct tracer            *current_trace __read_mostly;
237
238 /*
239  * max_tracer_type_len is used to simplify the allocating of
240  * buffers to read userspace tracer names. We keep track of
241  * the longest tracer name registered.
242  */
243 static int                      max_tracer_type_len;
244
245 /*
246  * trace_types_lock is used to protect the trace_types list.
247  * This lock is also used to keep user access serialized.
248  * Accesses from userspace will grab this lock while userspace
249  * activities happen inside the kernel.
250  */
251 static DEFINE_MUTEX(trace_types_lock);
252
253 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
254 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
255
256 /* trace_flags holds trace_options default values */
257 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
258         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO;
259
260 /**
261  * trace_wake_up - wake up tasks waiting for trace input
262  *
263  * Simply wakes up any task that is blocked on the trace_wait
264  * queue. These is used with trace_poll for tasks polling the trace.
265  */
266 void trace_wake_up(void)
267 {
268         /*
269          * The runqueue_is_locked() can fail, but this is the best we
270          * have for now:
271          */
272         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
273                 wake_up(&trace_wait);
274 }
275
276 static int __init set_buf_size(char *str)
277 {
278         unsigned long buf_size;
279         int ret;
280
281         if (!str)
282                 return 0;
283         ret = strict_strtoul(str, 0, &buf_size);
284         /* nr_entries can not be zero */
285         if (ret < 0 || buf_size == 0)
286                 return 0;
287         trace_buf_size = buf_size;
288         return 1;
289 }
290 __setup("trace_buf_size=", set_buf_size);
291
292 unsigned long nsecs_to_usecs(unsigned long nsecs)
293 {
294         return nsecs / 1000;
295 }
296
297 /* These must match the bit postions in trace_iterator_flags */
298 static const char *trace_options[] = {
299         "print-parent",
300         "sym-offset",
301         "sym-addr",
302         "verbose",
303         "raw",
304         "hex",
305         "bin",
306         "block",
307         "stacktrace",
308         "sched-tree",
309         "trace_printk",
310         "ftrace_preempt",
311         "branch",
312         "annotate",
313         "userstacktrace",
314         "sym-userobj",
315         "printk-msg-only",
316         "context-info",
317         "latency-format",
318         "global-clock",
319         NULL
320 };
321
322 /*
323  * ftrace_max_lock is used to protect the swapping of buffers
324  * when taking a max snapshot. The buffers themselves are
325  * protected by per_cpu spinlocks. But the action of the swap
326  * needs its own lock.
327  *
328  * This is defined as a raw_spinlock_t in order to help
329  * with performance when lockdep debugging is enabled.
330  */
331 static raw_spinlock_t ftrace_max_lock =
332         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
333
334 /*
335  * Copy the new maximum trace into the separate maximum-trace
336  * structure. (this way the maximum trace is permanently saved,
337  * for later retrieval via /debugfs/tracing/latency_trace)
338  */
339 static void
340 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
341 {
342         struct trace_array_cpu *data = tr->data[cpu];
343
344         max_tr.cpu = cpu;
345         max_tr.time_start = data->preempt_timestamp;
346
347         data = max_tr.data[cpu];
348         data->saved_latency = tracing_max_latency;
349
350         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
351         data->pid = tsk->pid;
352         data->uid = task_uid(tsk);
353         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
354         data->policy = tsk->policy;
355         data->rt_priority = tsk->rt_priority;
356
357         /* record this tasks comm */
358         tracing_record_cmdline(tsk);
359 }
360
361 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
362 {
363         int len;
364         int ret;
365
366         if (!cnt)
367                 return 0;
368
369         if (s->len <= s->readpos)
370                 return -EBUSY;
371
372         len = s->len - s->readpos;
373         if (cnt > len)
374                 cnt = len;
375         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
376         if (ret == cnt)
377                 return -EFAULT;
378
379         cnt -= ret;
380
381         s->readpos += cnt;
382         return cnt;
383 }
384
385 ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
386 {
387         int len;
388         void *ret;
389
390         if (s->len <= s->readpos)
391                 return -EBUSY;
392
393         len = s->len - s->readpos;
394         if (cnt > len)
395                 cnt = len;
396         ret = memcpy(buf, s->buffer + s->readpos, cnt);
397         if (!ret)
398                 return -EFAULT;
399
400         s->readpos += cnt;
401         return cnt;
402 }
403
404 static void
405 trace_print_seq(struct seq_file *m, struct trace_seq *s)
406 {
407         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
408
409         s->buffer[len] = 0;
410         seq_puts(m, s->buffer);
411
412         trace_seq_init(s);
413 }
414
415 /**
416  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
417  * @tr: tracer
418  * @tsk: the task with the latency
419  * @cpu: The cpu that initiated the trace.
420  *
421  * Flip the buffers between the @tr and the max_tr and record information
422  * about which task was the cause of this latency.
423  */
424 void
425 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
426 {
427         struct ring_buffer *buf = tr->buffer;
428
429         WARN_ON_ONCE(!irqs_disabled());
430         __raw_spin_lock(&ftrace_max_lock);
431
432         tr->buffer = max_tr.buffer;
433         max_tr.buffer = buf;
434
435         ftrace_disable_cpu();
436         ring_buffer_reset(tr->buffer);
437         ftrace_enable_cpu();
438
439         __update_max_tr(tr, tsk, cpu);
440         __raw_spin_unlock(&ftrace_max_lock);
441 }
442
443 /**
444  * update_max_tr_single - only copy one trace over, and reset the rest
445  * @tr - tracer
446  * @tsk - task with the latency
447  * @cpu - the cpu of the buffer to copy.
448  *
449  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
450  */
451 void
452 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
453 {
454         int ret;
455
456         WARN_ON_ONCE(!irqs_disabled());
457         __raw_spin_lock(&ftrace_max_lock);
458
459         ftrace_disable_cpu();
460
461         ring_buffer_reset(max_tr.buffer);
462         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
463
464         ftrace_enable_cpu();
465
466         WARN_ON_ONCE(ret && ret != -EAGAIN);
467
468         __update_max_tr(tr, tsk, cpu);
469         __raw_spin_unlock(&ftrace_max_lock);
470 }
471
472 /**
473  * register_tracer - register a tracer with the ftrace system.
474  * @type - the plugin for the tracer
475  *
476  * Register a new plugin tracer.
477  */
478 int register_tracer(struct tracer *type)
479 __releases(kernel_lock)
480 __acquires(kernel_lock)
481 {
482         struct tracer *t;
483         int len;
484         int ret = 0;
485
486         if (!type->name) {
487                 pr_info("Tracer must have a name\n");
488                 return -1;
489         }
490
491         /*
492          * When this gets called we hold the BKL which means that
493          * preemption is disabled. Various trace selftests however
494          * need to disable and enable preemption for successful tests.
495          * So we drop the BKL here and grab it after the tests again.
496          */
497         unlock_kernel();
498         mutex_lock(&trace_types_lock);
499
500         tracing_selftest_running = true;
501
502         for (t = trace_types; t; t = t->next) {
503                 if (strcmp(type->name, t->name) == 0) {
504                         /* already found */
505                         pr_info("Trace %s already registered\n",
506                                 type->name);
507                         ret = -1;
508                         goto out;
509                 }
510         }
511
512         if (!type->set_flag)
513                 type->set_flag = &dummy_set_flag;
514         if (!type->flags)
515                 type->flags = &dummy_tracer_flags;
516         else
517                 if (!type->flags->opts)
518                         type->flags->opts = dummy_tracer_opt;
519         if (!type->wait_pipe)
520                 type->wait_pipe = default_wait_pipe;
521
522
523 #ifdef CONFIG_FTRACE_STARTUP_TEST
524         if (type->selftest && !tracing_selftest_disabled) {
525                 struct tracer *saved_tracer = current_trace;
526                 struct trace_array *tr = &global_trace;
527                 int i;
528
529                 /*
530                  * Run a selftest on this tracer.
531                  * Here we reset the trace buffer, and set the current
532                  * tracer to be this tracer. The tracer can then run some
533                  * internal tracing to verify that everything is in order.
534                  * If we fail, we do not register this tracer.
535                  */
536                 for_each_tracing_cpu(i)
537                         tracing_reset(tr, i);
538
539                 current_trace = type;
540                 /* the test is responsible for initializing and enabling */
541                 pr_info("Testing tracer %s: ", type->name);
542                 ret = type->selftest(type, tr);
543                 /* the test is responsible for resetting too */
544                 current_trace = saved_tracer;
545                 if (ret) {
546                         printk(KERN_CONT "FAILED!\n");
547                         goto out;
548                 }
549                 /* Only reset on passing, to avoid touching corrupted buffers */
550                 for_each_tracing_cpu(i)
551                         tracing_reset(tr, i);
552
553                 printk(KERN_CONT "PASSED\n");
554         }
555 #endif
556
557         type->next = trace_types;
558         trace_types = type;
559         len = strlen(type->name);
560         if (len > max_tracer_type_len)
561                 max_tracer_type_len = len;
562
563  out:
564         tracing_selftest_running = false;
565         mutex_unlock(&trace_types_lock);
566
567         if (ret || !default_bootup_tracer)
568                 goto out_unlock;
569
570         if (strncmp(default_bootup_tracer, type->name, BOOTUP_TRACER_SIZE))
571                 goto out_unlock;
572
573         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
574         /* Do we want this tracer to start on bootup? */
575         tracing_set_tracer(type->name);
576         default_bootup_tracer = NULL;
577         /* disable other selftests, since this will break it. */
578         tracing_selftest_disabled = 1;
579 #ifdef CONFIG_FTRACE_STARTUP_TEST
580         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
581                type->name);
582 #endif
583
584  out_unlock:
585         lock_kernel();
586         return ret;
587 }
588
589 void unregister_tracer(struct tracer *type)
590 {
591         struct tracer **t;
592         int len;
593
594         mutex_lock(&trace_types_lock);
595         for (t = &trace_types; *t; t = &(*t)->next) {
596                 if (*t == type)
597                         goto found;
598         }
599         pr_info("Trace %s not registered\n", type->name);
600         goto out;
601
602  found:
603         *t = (*t)->next;
604
605         if (type == current_trace && tracer_enabled) {
606                 tracer_enabled = 0;
607                 tracing_stop();
608                 if (current_trace->stop)
609                         current_trace->stop(&global_trace);
610                 current_trace = &nop_trace;
611         }
612
613         if (strlen(type->name) != max_tracer_type_len)
614                 goto out;
615
616         max_tracer_type_len = 0;
617         for (t = &trace_types; *t; t = &(*t)->next) {
618                 len = strlen((*t)->name);
619                 if (len > max_tracer_type_len)
620                         max_tracer_type_len = len;
621         }
622  out:
623         mutex_unlock(&trace_types_lock);
624 }
625
626 void tracing_reset(struct trace_array *tr, int cpu)
627 {
628         ftrace_disable_cpu();
629         ring_buffer_reset_cpu(tr->buffer, cpu);
630         ftrace_enable_cpu();
631 }
632
633 void tracing_reset_online_cpus(struct trace_array *tr)
634 {
635         int cpu;
636
637         tr->time_start = ftrace_now(tr->cpu);
638
639         for_each_online_cpu(cpu)
640                 tracing_reset(tr, cpu);
641 }
642
643 #define SAVED_CMDLINES 128
644 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
645 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
646 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
647 static int cmdline_idx;
648 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
649
650 /* temporary disable recording */
651 static atomic_t trace_record_cmdline_disabled __read_mostly;
652
653 static void trace_init_cmdlines(void)
654 {
655         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
656         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
657         cmdline_idx = 0;
658 }
659
660 static int trace_stop_count;
661 static DEFINE_SPINLOCK(tracing_start_lock);
662
663 /**
664  * ftrace_off_permanent - disable all ftrace code permanently
665  *
666  * This should only be called when a serious anomally has
667  * been detected.  This will turn off the function tracing,
668  * ring buffers, and other tracing utilites. It takes no
669  * locks and can be called from any context.
670  */
671 void ftrace_off_permanent(void)
672 {
673         tracing_disabled = 1;
674         ftrace_stop();
675         tracing_off_permanent();
676 }
677
678 /**
679  * tracing_start - quick start of the tracer
680  *
681  * If tracing is enabled but was stopped by tracing_stop,
682  * this will start the tracer back up.
683  */
684 void tracing_start(void)
685 {
686         struct ring_buffer *buffer;
687         unsigned long flags;
688
689         if (tracing_disabled)
690                 return;
691
692         spin_lock_irqsave(&tracing_start_lock, flags);
693         if (--trace_stop_count) {
694                 if (trace_stop_count < 0) {
695                         /* Someone screwed up their debugging */
696                         WARN_ON_ONCE(1);
697                         trace_stop_count = 0;
698                 }
699                 goto out;
700         }
701
702
703         buffer = global_trace.buffer;
704         if (buffer)
705                 ring_buffer_record_enable(buffer);
706
707         buffer = max_tr.buffer;
708         if (buffer)
709                 ring_buffer_record_enable(buffer);
710
711         ftrace_start();
712  out:
713         spin_unlock_irqrestore(&tracing_start_lock, flags);
714 }
715
716 /**
717  * tracing_stop - quick stop of the tracer
718  *
719  * Light weight way to stop tracing. Use in conjunction with
720  * tracing_start.
721  */
722 void tracing_stop(void)
723 {
724         struct ring_buffer *buffer;
725         unsigned long flags;
726
727         ftrace_stop();
728         spin_lock_irqsave(&tracing_start_lock, flags);
729         if (trace_stop_count++)
730                 goto out;
731
732         buffer = global_trace.buffer;
733         if (buffer)
734                 ring_buffer_record_disable(buffer);
735
736         buffer = max_tr.buffer;
737         if (buffer)
738                 ring_buffer_record_disable(buffer);
739
740  out:
741         spin_unlock_irqrestore(&tracing_start_lock, flags);
742 }
743
744 void trace_stop_cmdline_recording(void);
745
746 static void trace_save_cmdline(struct task_struct *tsk)
747 {
748         unsigned map;
749         unsigned idx;
750
751         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
752                 return;
753
754         /*
755          * It's not the end of the world if we don't get
756          * the lock, but we also don't want to spin
757          * nor do we want to disable interrupts,
758          * so if we miss here, then better luck next time.
759          */
760         if (!__raw_spin_trylock(&trace_cmdline_lock))
761                 return;
762
763         idx = map_pid_to_cmdline[tsk->pid];
764         if (idx >= SAVED_CMDLINES) {
765                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
766
767                 map = map_cmdline_to_pid[idx];
768                 if (map <= PID_MAX_DEFAULT)
769                         map_pid_to_cmdline[map] = (unsigned)-1;
770
771                 map_pid_to_cmdline[tsk->pid] = idx;
772
773                 cmdline_idx = idx;
774         }
775
776         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
777
778         __raw_spin_unlock(&trace_cmdline_lock);
779 }
780
781 void trace_find_cmdline(int pid, char comm[])
782 {
783         unsigned map;
784
785         if (!pid) {
786                 strcpy(comm, "<idle>");
787                 return;
788         }
789
790         if (pid > PID_MAX_DEFAULT) {
791                 strcpy(comm, "<...>");
792                 return;
793         }
794
795         __raw_spin_lock(&trace_cmdline_lock);
796         map = map_pid_to_cmdline[pid];
797         if (map >= SAVED_CMDLINES)
798                 goto out;
799
800         strcpy(comm, saved_cmdlines[map]);
801
802  out:
803         __raw_spin_unlock(&trace_cmdline_lock);
804 }
805
806 void tracing_record_cmdline(struct task_struct *tsk)
807 {
808         if (atomic_read(&trace_record_cmdline_disabled) || !tracing_is_on())
809                 return;
810
811         trace_save_cmdline(tsk);
812 }
813
814 void
815 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
816                              int pc)
817 {
818         struct task_struct *tsk = current;
819
820         entry->preempt_count            = pc & 0xff;
821         entry->pid                      = (tsk) ? tsk->pid : 0;
822         entry->tgid                     = (tsk) ? tsk->tgid : 0;
823         entry->flags =
824 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
825                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
826 #else
827                 TRACE_FLAG_IRQS_NOSUPPORT |
828 #endif
829                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
830                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
831                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
832 }
833
834 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
835                                                     unsigned char type,
836                                                     unsigned long len,
837                                                     unsigned long flags, int pc)
838 {
839         struct ring_buffer_event *event;
840
841         event = ring_buffer_lock_reserve(tr->buffer, len);
842         if (event != NULL) {
843                 struct trace_entry *ent = ring_buffer_event_data(event);
844
845                 tracing_generic_entry_update(ent, flags, pc);
846                 ent->type = type;
847         }
848
849         return event;
850 }
851 static void ftrace_trace_stack(struct trace_array *tr,
852                                unsigned long flags, int skip, int pc);
853 static void ftrace_trace_userstack(struct trace_array *tr,
854                                    unsigned long flags, int pc);
855
856 void trace_buffer_unlock_commit(struct trace_array *tr,
857                                 struct ring_buffer_event *event,
858                                 unsigned long flags, int pc)
859 {
860         ring_buffer_unlock_commit(tr->buffer, event);
861
862         ftrace_trace_stack(tr, flags, 6, pc);
863         ftrace_trace_userstack(tr, flags, pc);
864         trace_wake_up();
865 }
866
867 struct ring_buffer_event *
868 trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
869                                   unsigned long flags, int pc)
870 {
871         return trace_buffer_lock_reserve(&global_trace,
872                                          type, len, flags, pc);
873 }
874
875 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
876                                         unsigned long flags, int pc)
877 {
878         return trace_buffer_unlock_commit(&global_trace, event, flags, pc);
879 }
880
881 void
882 trace_function(struct trace_array *tr,
883                unsigned long ip, unsigned long parent_ip, unsigned long flags,
884                int pc)
885 {
886         struct ring_buffer_event *event;
887         struct ftrace_entry *entry;
888
889         /* If we are reading the ring buffer, don't trace */
890         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
891                 return;
892
893         event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
894                                           flags, pc);
895         if (!event)
896                 return;
897         entry   = ring_buffer_event_data(event);
898         entry->ip                       = ip;
899         entry->parent_ip                = parent_ip;
900         ring_buffer_unlock_commit(tr->buffer, event);
901 }
902
903 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
904 static void __trace_graph_entry(struct trace_array *tr,
905                                 struct ftrace_graph_ent *trace,
906                                 unsigned long flags,
907                                 int pc)
908 {
909         struct ring_buffer_event *event;
910         struct ftrace_graph_ent_entry *entry;
911
912         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
913                 return;
914
915         event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
916                                           sizeof(*entry), flags, pc);
917         if (!event)
918                 return;
919         entry   = ring_buffer_event_data(event);
920         entry->graph_ent                        = *trace;
921         ring_buffer_unlock_commit(global_trace.buffer, event);
922 }
923
924 static void __trace_graph_return(struct trace_array *tr,
925                                 struct ftrace_graph_ret *trace,
926                                 unsigned long flags,
927                                 int pc)
928 {
929         struct ring_buffer_event *event;
930         struct ftrace_graph_ret_entry *entry;
931
932         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
933                 return;
934
935         event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
936                                           sizeof(*entry), flags, pc);
937         if (!event)
938                 return;
939         entry   = ring_buffer_event_data(event);
940         entry->ret                              = *trace;
941         ring_buffer_unlock_commit(global_trace.buffer, event);
942 }
943 #endif
944
945 void
946 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
947        unsigned long ip, unsigned long parent_ip, unsigned long flags,
948        int pc)
949 {
950         if (likely(!atomic_read(&data->disabled)))
951                 trace_function(tr, ip, parent_ip, flags, pc);
952 }
953
954 static void __ftrace_trace_stack(struct trace_array *tr,
955                                  unsigned long flags,
956                                  int skip, int pc)
957 {
958 #ifdef CONFIG_STACKTRACE
959         struct ring_buffer_event *event;
960         struct stack_entry *entry;
961         struct stack_trace trace;
962
963         event = trace_buffer_lock_reserve(tr, TRACE_STACK,
964                                           sizeof(*entry), flags, pc);
965         if (!event)
966                 return;
967         entry   = ring_buffer_event_data(event);
968         memset(&entry->caller, 0, sizeof(entry->caller));
969
970         trace.nr_entries        = 0;
971         trace.max_entries       = FTRACE_STACK_ENTRIES;
972         trace.skip              = skip;
973         trace.entries           = entry->caller;
974
975         save_stack_trace(&trace);
976         ring_buffer_unlock_commit(tr->buffer, event);
977 #endif
978 }
979
980 static void ftrace_trace_stack(struct trace_array *tr,
981                                unsigned long flags,
982                                int skip, int pc)
983 {
984         if (!(trace_flags & TRACE_ITER_STACKTRACE))
985                 return;
986
987         __ftrace_trace_stack(tr, flags, skip, pc);
988 }
989
990 void __trace_stack(struct trace_array *tr,
991                    unsigned long flags,
992                    int skip, int pc)
993 {
994         __ftrace_trace_stack(tr, flags, skip, pc);
995 }
996
997 static void ftrace_trace_userstack(struct trace_array *tr,
998                                    unsigned long flags, int pc)
999 {
1000 #ifdef CONFIG_STACKTRACE
1001         struct ring_buffer_event *event;
1002         struct userstack_entry *entry;
1003         struct stack_trace trace;
1004
1005         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1006                 return;
1007
1008         event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
1009                                           sizeof(*entry), flags, pc);
1010         if (!event)
1011                 return;
1012         entry   = ring_buffer_event_data(event);
1013
1014         memset(&entry->caller, 0, sizeof(entry->caller));
1015
1016         trace.nr_entries        = 0;
1017         trace.max_entries       = FTRACE_STACK_ENTRIES;
1018         trace.skip              = 0;
1019         trace.entries           = entry->caller;
1020
1021         save_stack_trace_user(&trace);
1022         ring_buffer_unlock_commit(tr->buffer, event);
1023 #endif
1024 }
1025
1026 #ifdef UNUSED
1027 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1028 {
1029         ftrace_trace_userstack(tr, flags, preempt_count());
1030 }
1031 #endif /* UNUSED */
1032
1033 static void
1034 ftrace_trace_special(void *__tr,
1035                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1036                      int pc)
1037 {
1038         struct ring_buffer_event *event;
1039         struct trace_array *tr = __tr;
1040         struct special_entry *entry;
1041
1042         event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1043                                           sizeof(*entry), 0, pc);
1044         if (!event)
1045                 return;
1046         entry   = ring_buffer_event_data(event);
1047         entry->arg1                     = arg1;
1048         entry->arg2                     = arg2;
1049         entry->arg3                     = arg3;
1050         trace_buffer_unlock_commit(tr, event, 0, pc);
1051 }
1052
1053 void
1054 __trace_special(void *__tr, void *__data,
1055                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1056 {
1057         ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1058 }
1059
1060 void
1061 tracing_sched_switch_trace(struct trace_array *tr,
1062                            struct task_struct *prev,
1063                            struct task_struct *next,
1064                            unsigned long flags, int pc)
1065 {
1066         struct ring_buffer_event *event;
1067         struct ctx_switch_entry *entry;
1068
1069         event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1070                                           sizeof(*entry), flags, pc);
1071         if (!event)
1072                 return;
1073         entry   = ring_buffer_event_data(event);
1074         entry->prev_pid                 = prev->pid;
1075         entry->prev_prio                = prev->prio;
1076         entry->prev_state               = prev->state;
1077         entry->next_pid                 = next->pid;
1078         entry->next_prio                = next->prio;
1079         entry->next_state               = next->state;
1080         entry->next_cpu = task_cpu(next);
1081         trace_buffer_unlock_commit(tr, event, flags, pc);
1082 }
1083
1084 void
1085 tracing_sched_wakeup_trace(struct trace_array *tr,
1086                            struct task_struct *wakee,
1087                            struct task_struct *curr,
1088                            unsigned long flags, int pc)
1089 {
1090         struct ring_buffer_event *event;
1091         struct ctx_switch_entry *entry;
1092
1093         event = trace_buffer_lock_reserve(tr, TRACE_WAKE,
1094                                           sizeof(*entry), flags, pc);
1095         if (!event)
1096                 return;
1097         entry   = ring_buffer_event_data(event);
1098         entry->prev_pid                 = curr->pid;
1099         entry->prev_prio                = curr->prio;
1100         entry->prev_state               = curr->state;
1101         entry->next_pid                 = wakee->pid;
1102         entry->next_prio                = wakee->prio;
1103         entry->next_state               = wakee->state;
1104         entry->next_cpu                 = task_cpu(wakee);
1105
1106         ring_buffer_unlock_commit(tr->buffer, event);
1107         ftrace_trace_stack(tr, flags, 6, pc);
1108         ftrace_trace_userstack(tr, flags, pc);
1109 }
1110
1111 void
1112 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1113 {
1114         struct trace_array *tr = &global_trace;
1115         struct trace_array_cpu *data;
1116         unsigned long flags;
1117         int cpu;
1118         int pc;
1119
1120         if (tracing_disabled)
1121                 return;
1122
1123         pc = preempt_count();
1124         local_irq_save(flags);
1125         cpu = raw_smp_processor_id();
1126         data = tr->data[cpu];
1127
1128         if (likely(atomic_inc_return(&data->disabled) == 1))
1129                 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1130
1131         atomic_dec(&data->disabled);
1132         local_irq_restore(flags);
1133 }
1134
1135 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1136 int trace_graph_entry(struct ftrace_graph_ent *trace)
1137 {
1138         struct trace_array *tr = &global_trace;
1139         struct trace_array_cpu *data;
1140         unsigned long flags;
1141         long disabled;
1142         int cpu;
1143         int pc;
1144
1145         if (!ftrace_trace_task(current))
1146                 return 0;
1147
1148         if (!ftrace_graph_addr(trace->func))
1149                 return 0;
1150
1151         local_irq_save(flags);
1152         cpu = raw_smp_processor_id();
1153         data = tr->data[cpu];
1154         disabled = atomic_inc_return(&data->disabled);
1155         if (likely(disabled == 1)) {
1156                 pc = preempt_count();
1157                 __trace_graph_entry(tr, trace, flags, pc);
1158         }
1159         /* Only do the atomic if it is not already set */
1160         if (!test_tsk_trace_graph(current))
1161                 set_tsk_trace_graph(current);
1162         atomic_dec(&data->disabled);
1163         local_irq_restore(flags);
1164
1165         return 1;
1166 }
1167
1168 void trace_graph_return(struct ftrace_graph_ret *trace)
1169 {
1170         struct trace_array *tr = &global_trace;
1171         struct trace_array_cpu *data;
1172         unsigned long flags;
1173         long disabled;
1174         int cpu;
1175         int pc;
1176
1177         local_irq_save(flags);
1178         cpu = raw_smp_processor_id();
1179         data = tr->data[cpu];
1180         disabled = atomic_inc_return(&data->disabled);
1181         if (likely(disabled == 1)) {
1182                 pc = preempt_count();
1183                 __trace_graph_return(tr, trace, flags, pc);
1184         }
1185         if (!trace->depth)
1186                 clear_tsk_trace_graph(current);
1187         atomic_dec(&data->disabled);
1188         local_irq_restore(flags);
1189 }
1190 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1191
1192
1193 /**
1194  * trace_vbprintk - write binary msg to tracing buffer
1195  *
1196  */
1197 int trace_vbprintk(unsigned long ip, int depth, const char *fmt, va_list args)
1198 {
1199         static raw_spinlock_t trace_buf_lock =
1200                 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1201         static u32 trace_buf[TRACE_BUF_SIZE];
1202
1203         struct ring_buffer_event *event;
1204         struct trace_array *tr = &global_trace;
1205         struct trace_array_cpu *data;
1206         struct bprint_entry *entry;
1207         unsigned long flags;
1208         int resched;
1209         int cpu, len = 0, size, pc;
1210
1211         if (unlikely(tracing_selftest_running || tracing_disabled))
1212                 return 0;
1213
1214         /* Don't pollute graph traces with trace_vprintk internals */
1215         pause_graph_tracing();
1216
1217         pc = preempt_count();
1218         resched = ftrace_preempt_disable();
1219         cpu = raw_smp_processor_id();
1220         data = tr->data[cpu];
1221
1222         if (unlikely(atomic_read(&data->disabled)))
1223                 goto out;
1224
1225         /* Lockdep uses trace_printk for lock tracing */
1226         local_irq_save(flags);
1227         __raw_spin_lock(&trace_buf_lock);
1228         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1229
1230         if (len > TRACE_BUF_SIZE || len < 0)
1231                 goto out_unlock;
1232
1233         size = sizeof(*entry) + sizeof(u32) * len;
1234         event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc);
1235         if (!event)
1236                 goto out_unlock;
1237         entry = ring_buffer_event_data(event);
1238         entry->ip                       = ip;
1239         entry->depth                    = depth;
1240         entry->fmt                      = fmt;
1241
1242         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1243         ring_buffer_unlock_commit(tr->buffer, event);
1244
1245 out_unlock:
1246         __raw_spin_unlock(&trace_buf_lock);
1247         local_irq_restore(flags);
1248
1249 out:
1250         ftrace_preempt_enable(resched);
1251         unpause_graph_tracing();
1252
1253         return len;
1254 }
1255 EXPORT_SYMBOL_GPL(trace_vbprintk);
1256
1257 int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
1258 {
1259         static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1260         static char trace_buf[TRACE_BUF_SIZE];
1261
1262         struct ring_buffer_event *event;
1263         struct trace_array *tr = &global_trace;
1264         struct trace_array_cpu *data;
1265         int cpu, len = 0, size, pc;
1266         struct print_entry *entry;
1267         unsigned long irq_flags;
1268
1269         if (tracing_disabled || tracing_selftest_running)
1270                 return 0;
1271
1272         pc = preempt_count();
1273         preempt_disable_notrace();
1274         cpu = raw_smp_processor_id();
1275         data = tr->data[cpu];
1276
1277         if (unlikely(atomic_read(&data->disabled)))
1278                 goto out;
1279
1280         pause_graph_tracing();
1281         raw_local_irq_save(irq_flags);
1282         __raw_spin_lock(&trace_buf_lock);
1283         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1284
1285         len = min(len, TRACE_BUF_SIZE-1);
1286         trace_buf[len] = 0;
1287
1288         size = sizeof(*entry) + len + 1;
1289         event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
1290         if (!event)
1291                 goto out_unlock;
1292         entry = ring_buffer_event_data(event);
1293         entry->ip                       = ip;
1294         entry->depth                    = depth;
1295
1296         memcpy(&entry->buf, trace_buf, len);
1297         entry->buf[len] = 0;
1298         ring_buffer_unlock_commit(tr->buffer, event);
1299
1300  out_unlock:
1301         __raw_spin_unlock(&trace_buf_lock);
1302         raw_local_irq_restore(irq_flags);
1303         unpause_graph_tracing();
1304  out:
1305         preempt_enable_notrace();
1306
1307         return len;
1308 }
1309 EXPORT_SYMBOL_GPL(trace_vprintk);
1310
1311 enum trace_file_type {
1312         TRACE_FILE_LAT_FMT      = 1,
1313         TRACE_FILE_ANNOTATE     = 2,
1314 };
1315
1316 static void trace_iterator_increment(struct trace_iterator *iter)
1317 {
1318         /* Don't allow ftrace to trace into the ring buffers */
1319         ftrace_disable_cpu();
1320
1321         iter->idx++;
1322         if (iter->buffer_iter[iter->cpu])
1323                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1324
1325         ftrace_enable_cpu();
1326 }
1327
1328 static struct trace_entry *
1329 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1330 {
1331         struct ring_buffer_event *event;
1332         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1333
1334         /* Don't allow ftrace to trace into the ring buffers */
1335         ftrace_disable_cpu();
1336
1337         if (buf_iter)
1338                 event = ring_buffer_iter_peek(buf_iter, ts);
1339         else
1340                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1341
1342         ftrace_enable_cpu();
1343
1344         return event ? ring_buffer_event_data(event) : NULL;
1345 }
1346
1347 static struct trace_entry *
1348 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1349 {
1350         struct ring_buffer *buffer = iter->tr->buffer;
1351         struct trace_entry *ent, *next = NULL;
1352         int cpu_file = iter->cpu_file;
1353         u64 next_ts = 0, ts;
1354         int next_cpu = -1;
1355         int cpu;
1356
1357         /*
1358          * If we are in a per_cpu trace file, don't bother by iterating over
1359          * all cpu and peek directly.
1360          */
1361         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1362                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1363                         return NULL;
1364                 ent = peek_next_entry(iter, cpu_file, ent_ts);
1365                 if (ent_cpu)
1366                         *ent_cpu = cpu_file;
1367
1368                 return ent;
1369         }
1370
1371         for_each_tracing_cpu(cpu) {
1372
1373                 if (ring_buffer_empty_cpu(buffer, cpu))
1374                         continue;
1375
1376                 ent = peek_next_entry(iter, cpu, &ts);
1377
1378                 /*
1379                  * Pick the entry with the smallest timestamp:
1380                  */
1381                 if (ent && (!next || ts < next_ts)) {
1382                         next = ent;
1383                         next_cpu = cpu;
1384                         next_ts = ts;
1385                 }
1386         }
1387
1388         if (ent_cpu)
1389                 *ent_cpu = next_cpu;
1390
1391         if (ent_ts)
1392                 *ent_ts = next_ts;
1393
1394         return next;
1395 }
1396
1397 /* Find the next real entry, without updating the iterator itself */
1398 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1399                                           int *ent_cpu, u64 *ent_ts)
1400 {
1401         return __find_next_entry(iter, ent_cpu, ent_ts);
1402 }
1403
1404 /* Find the next real entry, and increment the iterator to the next entry */
1405 static void *find_next_entry_inc(struct trace_iterator *iter)
1406 {
1407         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1408
1409         if (iter->ent)
1410                 trace_iterator_increment(iter);
1411
1412         return iter->ent ? iter : NULL;
1413 }
1414
1415 static void trace_consume(struct trace_iterator *iter)
1416 {
1417         /* Don't allow ftrace to trace into the ring buffers */
1418         ftrace_disable_cpu();
1419         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1420         ftrace_enable_cpu();
1421 }
1422
1423 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1424 {
1425         struct trace_iterator *iter = m->private;
1426         int i = (int)*pos;
1427         void *ent;
1428
1429         (*pos)++;
1430
1431         /* can't go backwards */
1432         if (iter->idx > i)
1433                 return NULL;
1434
1435         if (iter->idx < 0)
1436                 ent = find_next_entry_inc(iter);
1437         else
1438                 ent = iter;
1439
1440         while (ent && iter->idx < i)
1441                 ent = find_next_entry_inc(iter);
1442
1443         iter->pos = *pos;
1444
1445         return ent;
1446 }
1447
1448 /*
1449  * No necessary locking here. The worst thing which can
1450  * happen is loosing events consumed at the same time
1451  * by a trace_pipe reader.
1452  * Other than that, we don't risk to crash the ring buffer
1453  * because it serializes the readers.
1454  *
1455  * The current tracer is copied to avoid a global locking
1456  * all around.
1457  */
1458 static void *s_start(struct seq_file *m, loff_t *pos)
1459 {
1460         struct trace_iterator *iter = m->private;
1461         static struct tracer *old_tracer;
1462         int cpu_file = iter->cpu_file;
1463         void *p = NULL;
1464         loff_t l = 0;
1465         int cpu;
1466
1467         /* copy the tracer to avoid using a global lock all around */
1468         mutex_lock(&trace_types_lock);
1469         if (unlikely(old_tracer != current_trace && current_trace)) {
1470                 old_tracer = current_trace;
1471                 *iter->trace = *current_trace;
1472         }
1473         mutex_unlock(&trace_types_lock);
1474
1475         atomic_inc(&trace_record_cmdline_disabled);
1476
1477         if (*pos != iter->pos) {
1478                 iter->ent = NULL;
1479                 iter->cpu = 0;
1480                 iter->idx = -1;
1481
1482                 ftrace_disable_cpu();
1483
1484                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1485                         for_each_tracing_cpu(cpu)
1486                                 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1487                 } else
1488                         ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1489
1490
1491                 ftrace_enable_cpu();
1492
1493                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1494                         ;
1495
1496         } else {
1497                 l = *pos - 1;
1498                 p = s_next(m, p, &l);
1499         }
1500
1501         return p;
1502 }
1503
1504 static void s_stop(struct seq_file *m, void *p)
1505 {
1506         atomic_dec(&trace_record_cmdline_disabled);
1507 }
1508
1509 static void print_lat_help_header(struct seq_file *m)
1510 {
1511         seq_puts(m, "#                  _------=> CPU#            \n");
1512         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1513         seq_puts(m, "#                | / _----=> need-resched    \n");
1514         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1515         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1516         seq_puts(m, "#                |||| /                      \n");
1517         seq_puts(m, "#                |||||     delay             \n");
1518         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1519         seq_puts(m, "#     \\   /      |||||   \\   |   /           \n");
1520 }
1521
1522 static void print_func_help_header(struct seq_file *m)
1523 {
1524         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1525         seq_puts(m, "#              | |       |          |         |\n");
1526 }
1527
1528
1529 static void
1530 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1531 {
1532         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1533         struct trace_array *tr = iter->tr;
1534         struct trace_array_cpu *data = tr->data[tr->cpu];
1535         struct tracer *type = current_trace;
1536         unsigned long total;
1537         unsigned long entries;
1538         const char *name = "preemption";
1539
1540         if (type)
1541                 name = type->name;
1542
1543         entries = ring_buffer_entries(iter->tr->buffer);
1544         total = entries +
1545                 ring_buffer_overruns(iter->tr->buffer);
1546
1547         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1548                    name, UTS_RELEASE);
1549         seq_puts(m, "# -----------------------------------"
1550                  "---------------------------------\n");
1551         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1552                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1553                    nsecs_to_usecs(data->saved_latency),
1554                    entries,
1555                    total,
1556                    tr->cpu,
1557 #if defined(CONFIG_PREEMPT_NONE)
1558                    "server",
1559 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1560                    "desktop",
1561 #elif defined(CONFIG_PREEMPT)
1562                    "preempt",
1563 #else
1564                    "unknown",
1565 #endif
1566                    /* These are reserved for later use */
1567                    0, 0, 0, 0);
1568 #ifdef CONFIG_SMP
1569         seq_printf(m, " #P:%d)\n", num_online_cpus());
1570 #else
1571         seq_puts(m, ")\n");
1572 #endif
1573         seq_puts(m, "#    -----------------\n");
1574         seq_printf(m, "#    | task: %.16s-%d "
1575                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1576                    data->comm, data->pid, data->uid, data->nice,
1577                    data->policy, data->rt_priority);
1578         seq_puts(m, "#    -----------------\n");
1579
1580         if (data->critical_start) {
1581                 seq_puts(m, "#  => started at: ");
1582                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1583                 trace_print_seq(m, &iter->seq);
1584                 seq_puts(m, "\n#  => ended at:   ");
1585                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1586                 trace_print_seq(m, &iter->seq);
1587                 seq_puts(m, "#\n");
1588         }
1589
1590         seq_puts(m, "#\n");
1591 }
1592
1593 static void test_cpu_buff_start(struct trace_iterator *iter)
1594 {
1595         struct trace_seq *s = &iter->seq;
1596
1597         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1598                 return;
1599
1600         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1601                 return;
1602
1603         if (cpumask_test_cpu(iter->cpu, iter->started))
1604                 return;
1605
1606         cpumask_set_cpu(iter->cpu, iter->started);
1607         trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1608 }
1609
1610 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1611 {
1612         struct trace_seq *s = &iter->seq;
1613         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1614         struct trace_entry *entry;
1615         struct trace_event *event;
1616
1617         entry = iter->ent;
1618
1619         test_cpu_buff_start(iter);
1620
1621         event = ftrace_find_event(entry->type);
1622
1623         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1624                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1625                         if (!trace_print_lat_context(iter))
1626                                 goto partial;
1627                 } else {
1628                         if (!trace_print_context(iter))
1629                                 goto partial;
1630                 }
1631         }
1632
1633         if (event)
1634                 return event->trace(iter, sym_flags);
1635
1636         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1637                 goto partial;
1638
1639         return TRACE_TYPE_HANDLED;
1640 partial:
1641         return TRACE_TYPE_PARTIAL_LINE;
1642 }
1643
1644 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1645 {
1646         struct trace_seq *s = &iter->seq;
1647         struct trace_entry *entry;
1648         struct trace_event *event;
1649
1650         entry = iter->ent;
1651
1652         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1653                 if (!trace_seq_printf(s, "%d %d %llu ",
1654                                       entry->pid, iter->cpu, iter->ts))
1655                         goto partial;
1656         }
1657
1658         event = ftrace_find_event(entry->type);
1659         if (event)
1660                 return event->raw(iter, 0);
1661
1662         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1663                 goto partial;
1664
1665         return TRACE_TYPE_HANDLED;
1666 partial:
1667         return TRACE_TYPE_PARTIAL_LINE;
1668 }
1669
1670 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1671 {
1672         struct trace_seq *s = &iter->seq;
1673         unsigned char newline = '\n';
1674         struct trace_entry *entry;
1675         struct trace_event *event;
1676
1677         entry = iter->ent;
1678
1679         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1680                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1681                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1682                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1683         }
1684
1685         event = ftrace_find_event(entry->type);
1686         if (event) {
1687                 enum print_line_t ret = event->hex(iter, 0);
1688                 if (ret != TRACE_TYPE_HANDLED)
1689                         return ret;
1690         }
1691
1692         SEQ_PUT_FIELD_RET(s, newline);
1693
1694         return TRACE_TYPE_HANDLED;
1695 }
1696
1697 static enum print_line_t print_bprintk_msg_only(struct trace_iterator *iter)
1698 {
1699         struct trace_seq *s = &iter->seq;
1700         struct trace_entry *entry = iter->ent;
1701         struct bprint_entry *field;
1702         int ret;
1703
1704         trace_assign_type(field, entry);
1705
1706         ret = trace_seq_bprintf(s, field->fmt, field->buf);
1707         if (!ret)
1708                 return TRACE_TYPE_PARTIAL_LINE;
1709
1710         return TRACE_TYPE_HANDLED;
1711 }
1712
1713 static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
1714 {
1715         struct trace_seq *s = &iter->seq;
1716         struct trace_entry *entry = iter->ent;
1717         struct print_entry *field;
1718         int ret;
1719
1720         trace_assign_type(field, entry);
1721
1722         ret = trace_seq_printf(s, "%s", field->buf);
1723         if (!ret)
1724                 return TRACE_TYPE_PARTIAL_LINE;
1725
1726         return TRACE_TYPE_HANDLED;
1727 }
1728
1729 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1730 {
1731         struct trace_seq *s = &iter->seq;
1732         struct trace_entry *entry;
1733         struct trace_event *event;
1734
1735         entry = iter->ent;
1736
1737         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1738                 SEQ_PUT_FIELD_RET(s, entry->pid);
1739                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1740                 SEQ_PUT_FIELD_RET(s, iter->ts);
1741         }
1742
1743         event = ftrace_find_event(entry->type);
1744         return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1745 }
1746
1747 static int trace_empty(struct trace_iterator *iter)
1748 {
1749         int cpu;
1750
1751         /* If we are looking at one CPU buffer, only check that one */
1752         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1753                 cpu = iter->cpu_file;
1754                 if (iter->buffer_iter[cpu]) {
1755                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1756                                 return 0;
1757                 } else {
1758                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1759                                 return 0;
1760                 }
1761                 return 1;
1762         }
1763
1764         for_each_tracing_cpu(cpu) {
1765                 if (iter->buffer_iter[cpu]) {
1766                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1767                                 return 0;
1768                 } else {
1769                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1770                                 return 0;
1771                 }
1772         }
1773
1774         return 1;
1775 }
1776
1777 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1778 {
1779         enum print_line_t ret;
1780
1781         if (iter->trace && iter->trace->print_line) {
1782                 ret = iter->trace->print_line(iter);
1783                 if (ret != TRACE_TYPE_UNHANDLED)
1784                         return ret;
1785         }
1786
1787         if (iter->ent->type == TRACE_BPRINT &&
1788                         trace_flags & TRACE_ITER_PRINTK &&
1789                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1790                 return print_bprintk_msg_only(iter);
1791
1792         if (iter->ent->type == TRACE_PRINT &&
1793                         trace_flags & TRACE_ITER_PRINTK &&
1794                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1795                 return print_printk_msg_only(iter);
1796
1797         if (trace_flags & TRACE_ITER_BIN)
1798                 return print_bin_fmt(iter);
1799
1800         if (trace_flags & TRACE_ITER_HEX)
1801                 return print_hex_fmt(iter);
1802
1803         if (trace_flags & TRACE_ITER_RAW)
1804                 return print_raw_fmt(iter);
1805
1806         return print_trace_fmt(iter);
1807 }
1808
1809 static int s_show(struct seq_file *m, void *v)
1810 {
1811         struct trace_iterator *iter = v;
1812
1813         if (iter->ent == NULL) {
1814                 if (iter->tr) {
1815                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1816                         seq_puts(m, "#\n");
1817                 }
1818                 if (iter->trace && iter->trace->print_header)
1819                         iter->trace->print_header(m);
1820                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1821                         /* print nothing if the buffers are empty */
1822                         if (trace_empty(iter))
1823                                 return 0;
1824                         print_trace_header(m, iter);
1825                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1826                                 print_lat_help_header(m);
1827                 } else {
1828                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1829                                 print_func_help_header(m);
1830                 }
1831         } else {
1832                 print_trace_line(iter);
1833                 trace_print_seq(m, &iter->seq);
1834         }
1835
1836         return 0;
1837 }
1838
1839 static struct seq_operations tracer_seq_ops = {
1840         .start          = s_start,
1841         .next           = s_next,
1842         .stop           = s_stop,
1843         .show           = s_show,
1844 };
1845
1846 static struct trace_iterator *
1847 __tracing_open(struct inode *inode, struct file *file)
1848 {
1849         long cpu_file = (long) inode->i_private;
1850         void *fail_ret = ERR_PTR(-ENOMEM);
1851         struct trace_iterator *iter;
1852         struct seq_file *m;
1853         int cpu, ret;
1854
1855         if (tracing_disabled)
1856                 return ERR_PTR(-ENODEV);
1857
1858         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1859         if (!iter)
1860                 return ERR_PTR(-ENOMEM);
1861
1862         /*
1863          * We make a copy of the current tracer to avoid concurrent
1864          * changes on it while we are reading.
1865          */
1866         mutex_lock(&trace_types_lock);
1867         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1868         if (!iter->trace)
1869                 goto fail;
1870
1871         if (current_trace)
1872                 *iter->trace = *current_trace;
1873
1874         if (current_trace && current_trace->print_max)
1875                 iter->tr = &max_tr;
1876         else
1877                 iter->tr = &global_trace;
1878         iter->pos = -1;
1879         mutex_init(&iter->mutex);
1880         iter->cpu_file = cpu_file;
1881
1882         /* Notify the tracer early; before we stop tracing. */
1883         if (iter->trace && iter->trace->open)
1884                 iter->trace->open(iter);
1885
1886         /* Annotate start of buffers if we had overruns */
1887         if (ring_buffer_overruns(iter->tr->buffer))
1888                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1889
1890         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1891                 for_each_tracing_cpu(cpu) {
1892
1893                         iter->buffer_iter[cpu] =
1894                                 ring_buffer_read_start(iter->tr->buffer, cpu);
1895                 }
1896         } else {
1897                 cpu = iter->cpu_file;
1898                 iter->buffer_iter[cpu] =
1899                                 ring_buffer_read_start(iter->tr->buffer, cpu);
1900         }
1901
1902         /* TODO stop tracer */
1903         ret = seq_open(file, &tracer_seq_ops);
1904         if (ret < 0) {
1905                 fail_ret = ERR_PTR(ret);
1906                 goto fail_buffer;
1907         }
1908
1909         m = file->private_data;
1910         m->private = iter;
1911
1912         /* stop the trace while dumping */
1913         tracing_stop();
1914
1915         mutex_unlock(&trace_types_lock);
1916
1917         return iter;
1918
1919  fail_buffer:
1920         for_each_tracing_cpu(cpu) {
1921                 if (iter->buffer_iter[cpu])
1922                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
1923         }
1924  fail:
1925         mutex_unlock(&trace_types_lock);
1926         kfree(iter->trace);
1927         kfree(iter);
1928
1929         return fail_ret;
1930 }
1931
1932 int tracing_open_generic(struct inode *inode, struct file *filp)
1933 {
1934         if (tracing_disabled)
1935                 return -ENODEV;
1936
1937         filp->private_data = inode->i_private;
1938         return 0;
1939 }
1940
1941 static int tracing_release(struct inode *inode, struct file *file)
1942 {
1943         struct seq_file *m = (struct seq_file *)file->private_data;
1944         struct trace_iterator *iter = m->private;
1945         int cpu;
1946
1947         mutex_lock(&trace_types_lock);
1948         for_each_tracing_cpu(cpu) {
1949                 if (iter->buffer_iter[cpu])
1950                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
1951         }
1952
1953         if (iter->trace && iter->trace->close)
1954                 iter->trace->close(iter);
1955
1956         /* reenable tracing if it was previously enabled */
1957         tracing_start();
1958         mutex_unlock(&trace_types_lock);
1959
1960         seq_release(inode, file);
1961         mutex_destroy(&iter->mutex);
1962         kfree(iter->trace);
1963         kfree(iter);
1964         return 0;
1965 }
1966
1967 static int tracing_open(struct inode *inode, struct file *file)
1968 {
1969         struct trace_iterator *iter;
1970         int ret = 0;
1971
1972         iter = __tracing_open(inode, file);
1973         if (IS_ERR(iter))
1974                 ret = PTR_ERR(iter);
1975         else if (trace_flags & TRACE_ITER_LATENCY_FMT)
1976                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1977
1978         return ret;
1979 }
1980
1981 static void *
1982 t_next(struct seq_file *m, void *v, loff_t *pos)
1983 {
1984         struct tracer *t = m->private;
1985
1986         (*pos)++;
1987
1988         if (t)
1989                 t = t->next;
1990
1991         m->private = t;
1992
1993         return t;
1994 }
1995
1996 static void *t_start(struct seq_file *m, loff_t *pos)
1997 {
1998         struct tracer *t = m->private;
1999         loff_t l = 0;
2000
2001         mutex_lock(&trace_types_lock);
2002         for (; t && l < *pos; t = t_next(m, t, &l))
2003                 ;
2004
2005         return t;
2006 }
2007
2008 static void t_stop(struct seq_file *m, void *p)
2009 {
2010         mutex_unlock(&trace_types_lock);
2011 }
2012
2013 static int t_show(struct seq_file *m, void *v)
2014 {
2015         struct tracer *t = v;
2016
2017         if (!t)
2018                 return 0;
2019
2020         seq_printf(m, "%s", t->name);
2021         if (t->next)
2022                 seq_putc(m, ' ');
2023         else
2024                 seq_putc(m, '\n');
2025
2026         return 0;
2027 }
2028
2029 static struct seq_operations show_traces_seq_ops = {
2030         .start          = t_start,
2031         .next           = t_next,
2032         .stop           = t_stop,
2033         .show           = t_show,
2034 };
2035
2036 static int show_traces_open(struct inode *inode, struct file *file)
2037 {
2038         int ret;
2039
2040         if (tracing_disabled)
2041                 return -ENODEV;
2042
2043         ret = seq_open(file, &show_traces_seq_ops);
2044         if (!ret) {
2045                 struct seq_file *m = file->private_data;
2046                 m->private = trace_types;
2047         }
2048
2049         return ret;
2050 }
2051
2052 static const struct file_operations tracing_fops = {
2053         .open           = tracing_open,
2054         .read           = seq_read,
2055         .llseek         = seq_lseek,
2056         .release        = tracing_release,
2057 };
2058
2059 static const struct file_operations show_traces_fops = {
2060         .open           = show_traces_open,
2061         .read           = seq_read,
2062         .release        = seq_release,
2063 };
2064
2065 /*
2066  * Only trace on a CPU if the bitmask is set:
2067  */
2068 static cpumask_var_t tracing_cpumask;
2069
2070 /*
2071  * The tracer itself will not take this lock, but still we want
2072  * to provide a consistent cpumask to user-space:
2073  */
2074 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2075
2076 /*
2077  * Temporary storage for the character representation of the
2078  * CPU bitmask (and one more byte for the newline):
2079  */
2080 static char mask_str[NR_CPUS + 1];
2081
2082 static ssize_t
2083 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2084                      size_t count, loff_t *ppos)
2085 {
2086         int len;
2087
2088         mutex_lock(&tracing_cpumask_update_lock);
2089
2090         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2091         if (count - len < 2) {
2092                 count = -EINVAL;
2093                 goto out_err;
2094         }
2095         len += sprintf(mask_str + len, "\n");
2096         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2097
2098 out_err:
2099         mutex_unlock(&tracing_cpumask_update_lock);
2100
2101         return count;
2102 }
2103
2104 static ssize_t
2105 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2106                       size_t count, loff_t *ppos)
2107 {
2108         int err, cpu;
2109         cpumask_var_t tracing_cpumask_new;
2110
2111         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2112                 return -ENOMEM;
2113
2114         mutex_lock(&tracing_cpumask_update_lock);
2115         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2116         if (err)
2117                 goto err_unlock;
2118
2119         local_irq_disable();
2120         __raw_spin_lock(&ftrace_max_lock);
2121         for_each_tracing_cpu(cpu) {
2122                 /*
2123                  * Increase/decrease the disabled counter if we are
2124                  * about to flip a bit in the cpumask:
2125                  */
2126                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2127                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2128                         atomic_inc(&global_trace.data[cpu]->disabled);
2129                 }
2130                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2131                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2132                         atomic_dec(&global_trace.data[cpu]->disabled);
2133                 }
2134         }
2135         __raw_spin_unlock(&ftrace_max_lock);
2136         local_irq_enable();
2137
2138         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2139
2140         mutex_unlock(&tracing_cpumask_update_lock);
2141         free_cpumask_var(tracing_cpumask_new);
2142
2143         return count;
2144
2145 err_unlock:
2146         mutex_unlock(&tracing_cpumask_update_lock);
2147         free_cpumask_var(tracing_cpumask);
2148
2149         return err;
2150 }
2151
2152 static const struct file_operations tracing_cpumask_fops = {
2153         .open           = tracing_open_generic,
2154         .read           = tracing_cpumask_read,
2155         .write          = tracing_cpumask_write,
2156 };
2157
2158 static ssize_t
2159 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2160                        size_t cnt, loff_t *ppos)
2161 {
2162         struct tracer_opt *trace_opts;
2163         u32 tracer_flags;
2164         int len = 0;
2165         char *buf;
2166         int r = 0;
2167         int i;
2168
2169
2170         /* calculate max size */
2171         for (i = 0; trace_options[i]; i++) {
2172                 len += strlen(trace_options[i]);
2173                 len += 3; /* "no" and newline */
2174         }
2175
2176         mutex_lock(&trace_types_lock);
2177         tracer_flags = current_trace->flags->val;
2178         trace_opts = current_trace->flags->opts;
2179
2180         /*
2181          * Increase the size with names of options specific
2182          * of the current tracer.
2183          */
2184         for (i = 0; trace_opts[i].name; i++) {
2185                 len += strlen(trace_opts[i].name);
2186                 len += 3; /* "no" and newline */
2187         }
2188
2189         /* +2 for \n and \0 */
2190         buf = kmalloc(len + 2, GFP_KERNEL);
2191         if (!buf) {
2192                 mutex_unlock(&trace_types_lock);
2193                 return -ENOMEM;
2194         }
2195
2196         for (i = 0; trace_options[i]; i++) {
2197                 if (trace_flags & (1 << i))
2198                         r += sprintf(buf + r, "%s\n", trace_options[i]);
2199                 else
2200                         r += sprintf(buf + r, "no%s\n", trace_options[i]);
2201         }
2202
2203         for (i = 0; trace_opts[i].name; i++) {
2204                 if (tracer_flags & trace_opts[i].bit)
2205                         r += sprintf(buf + r, "%s\n",
2206                                 trace_opts[i].name);
2207                 else
2208                         r += sprintf(buf + r, "no%s\n",
2209                                 trace_opts[i].name);
2210         }
2211         mutex_unlock(&trace_types_lock);
2212
2213         WARN_ON(r >= len + 2);
2214
2215         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2216
2217         kfree(buf);
2218         return r;
2219 }
2220
2221 /* Try to assign a tracer specific option */
2222 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2223 {
2224         struct tracer_flags *trace_flags = trace->flags;
2225         struct tracer_opt *opts = NULL;
2226         int ret = 0, i = 0;
2227         int len;
2228
2229         for (i = 0; trace_flags->opts[i].name; i++) {
2230                 opts = &trace_flags->opts[i];
2231                 len = strlen(opts->name);
2232
2233                 if (strncmp(cmp, opts->name, len) == 0) {
2234                         ret = trace->set_flag(trace_flags->val,
2235                                 opts->bit, !neg);
2236                         break;
2237                 }
2238         }
2239         /* Not found */
2240         if (!trace_flags->opts[i].name)
2241                 return -EINVAL;
2242
2243         /* Refused to handle */
2244         if (ret)
2245                 return ret;
2246
2247         if (neg)
2248                 trace_flags->val &= ~opts->bit;
2249         else
2250                 trace_flags->val |= opts->bit;
2251
2252         return 0;
2253 }
2254
2255 static void set_tracer_flags(unsigned int mask, int enabled)
2256 {
2257         /* do nothing if flag is already set */
2258         if (!!(trace_flags & mask) == !!enabled)
2259                 return;
2260
2261         if (enabled)
2262                 trace_flags |= mask;
2263         else
2264                 trace_flags &= ~mask;
2265
2266         if (mask == TRACE_ITER_GLOBAL_CLK) {
2267                 u64 (*func)(void);
2268
2269                 if (enabled)
2270                         func = trace_clock_global;
2271                 else
2272                         func = trace_clock_local;
2273
2274                 mutex_lock(&trace_types_lock);
2275                 ring_buffer_set_clock(global_trace.buffer, func);
2276
2277                 if (max_tr.buffer)
2278                         ring_buffer_set_clock(max_tr.buffer, func);
2279                 mutex_unlock(&trace_types_lock);
2280         }
2281 }
2282
2283 static ssize_t
2284 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2285                         size_t cnt, loff_t *ppos)
2286 {
2287         char buf[64];
2288         char *cmp = buf;
2289         int neg = 0;
2290         int ret;
2291         int i;
2292
2293         if (cnt >= sizeof(buf))
2294                 return -EINVAL;
2295
2296         if (copy_from_user(&buf, ubuf, cnt))
2297                 return -EFAULT;
2298
2299         buf[cnt] = 0;
2300
2301         if (strncmp(buf, "no", 2) == 0) {
2302                 neg = 1;
2303                 cmp += 2;
2304         }
2305
2306         for (i = 0; trace_options[i]; i++) {
2307                 int len = strlen(trace_options[i]);
2308
2309                 if (strncmp(cmp, trace_options[i], len) == 0) {
2310                         set_tracer_flags(1 << i, !neg);
2311                         break;
2312                 }
2313         }
2314
2315         /* If no option could be set, test the specific tracer options */
2316         if (!trace_options[i]) {
2317                 mutex_lock(&trace_types_lock);
2318                 ret = set_tracer_option(current_trace, cmp, neg);
2319                 mutex_unlock(&trace_types_lock);
2320                 if (ret)
2321                         return ret;
2322         }
2323
2324         filp->f_pos += cnt;
2325
2326         return cnt;
2327 }
2328
2329 static const struct file_operations tracing_iter_fops = {
2330         .open           = tracing_open_generic,
2331         .read           = tracing_trace_options_read,
2332         .write          = tracing_trace_options_write,
2333 };
2334
2335 static const char readme_msg[] =
2336         "tracing mini-HOWTO:\n\n"
2337         "# mkdir /debug\n"
2338         "# mount -t debugfs nodev /debug\n\n"
2339         "# cat /debug/tracing/available_tracers\n"
2340         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2341         "# cat /debug/tracing/current_tracer\n"
2342         "none\n"
2343         "# echo sched_switch > /debug/tracing/current_tracer\n"
2344         "# cat /debug/tracing/current_tracer\n"
2345         "sched_switch\n"
2346         "# cat /debug/tracing/trace_options\n"
2347         "noprint-parent nosym-offset nosym-addr noverbose\n"
2348         "# echo print-parent > /debug/tracing/trace_options\n"
2349         "# echo 1 > /debug/tracing/tracing_enabled\n"
2350         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2351         "echo 0 > /debug/tracing/tracing_enabled\n"
2352 ;
2353
2354 static ssize_t
2355 tracing_readme_read(struct file *filp, char __user *ubuf,
2356                        size_t cnt, loff_t *ppos)
2357 {
2358         return simple_read_from_buffer(ubuf, cnt, ppos,
2359                                         readme_msg, strlen(readme_msg));
2360 }
2361
2362 static const struct file_operations tracing_readme_fops = {
2363         .open           = tracing_open_generic,
2364         .read           = tracing_readme_read,
2365 };
2366
2367 static ssize_t
2368 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2369                   size_t cnt, loff_t *ppos)
2370 {
2371         char buf[64];
2372         int r;
2373
2374         r = sprintf(buf, "%u\n", tracer_enabled);
2375         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2376 }
2377
2378 static ssize_t
2379 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2380                    size_t cnt, loff_t *ppos)
2381 {
2382         struct trace_array *tr = filp->private_data;
2383         char buf[64];
2384         unsigned long val;
2385         int ret;
2386
2387         if (cnt >= sizeof(buf))
2388                 return -EINVAL;
2389
2390         if (copy_from_user(&buf, ubuf, cnt))
2391                 return -EFAULT;
2392
2393         buf[cnt] = 0;
2394
2395         ret = strict_strtoul(buf, 10, &val);
2396         if (ret < 0)
2397                 return ret;
2398
2399         val = !!val;
2400
2401         mutex_lock(&trace_types_lock);
2402         if (tracer_enabled ^ val) {
2403                 if (val) {
2404                         tracer_enabled = 1;
2405                         if (current_trace->start)
2406                                 current_trace->start(tr);
2407                         tracing_start();
2408                 } else {
2409                         tracer_enabled = 0;
2410                         tracing_stop();
2411                         if (current_trace->stop)
2412                                 current_trace->stop(tr);
2413                 }
2414         }
2415         mutex_unlock(&trace_types_lock);
2416
2417         filp->f_pos += cnt;
2418
2419         return cnt;
2420 }
2421
2422 static ssize_t
2423 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2424                        size_t cnt, loff_t *ppos)
2425 {
2426         char buf[max_tracer_type_len+2];
2427         int r;
2428
2429         mutex_lock(&trace_types_lock);
2430         if (current_trace)
2431                 r = sprintf(buf, "%s\n", current_trace->name);
2432         else
2433                 r = sprintf(buf, "\n");
2434         mutex_unlock(&trace_types_lock);
2435
2436         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2437 }
2438
2439 int tracer_init(struct tracer *t, struct trace_array *tr)
2440 {
2441         tracing_reset_online_cpus(tr);
2442         return t->init(tr);
2443 }
2444
2445 static int tracing_resize_ring_buffer(unsigned long size)
2446 {
2447         int ret;
2448
2449         /*
2450          * If kernel or user changes the size of the ring buffer
2451          * we use the size that was given, and we can forget about
2452          * expanding it later.
2453          */
2454         ring_buffer_expanded = 1;
2455
2456         ret = ring_buffer_resize(global_trace.buffer, size);
2457         if (ret < 0)
2458                 return ret;
2459
2460         ret = ring_buffer_resize(max_tr.buffer, size);
2461         if (ret < 0) {
2462                 int r;
2463
2464                 r = ring_buffer_resize(global_trace.buffer,
2465                                        global_trace.entries);
2466                 if (r < 0) {
2467                         /*
2468                          * AARGH! We are left with different
2469                          * size max buffer!!!!
2470                          * The max buffer is our "snapshot" buffer.
2471                          * When a tracer needs a snapshot (one of the
2472                          * latency tracers), it swaps the max buffer
2473                          * with the saved snap shot. We succeeded to
2474                          * update the size of the main buffer, but failed to
2475                          * update the size of the max buffer. But when we tried
2476                          * to reset the main buffer to the original size, we
2477                          * failed there too. This is very unlikely to
2478                          * happen, but if it does, warn and kill all
2479                          * tracing.
2480                          */
2481                         WARN_ON(1);
2482                         tracing_disabled = 1;
2483                 }
2484                 return ret;
2485         }
2486
2487         global_trace.entries = size;
2488
2489         return ret;
2490 }
2491
2492 /**
2493  * tracing_update_buffers - used by tracing facility to expand ring buffers
2494  *
2495  * To save on memory when the tracing is never used on a system with it
2496  * configured in. The ring buffers are set to a minimum size. But once
2497  * a user starts to use the tracing facility, then they need to grow
2498  * to their default size.
2499  *
2500  * This function is to be called when a tracer is about to be used.
2501  */
2502 int tracing_update_buffers(void)
2503 {
2504         int ret = 0;
2505
2506         mutex_lock(&trace_types_lock);
2507         if (!ring_buffer_expanded)
2508                 ret = tracing_resize_ring_buffer(trace_buf_size);
2509         mutex_unlock(&trace_types_lock);
2510
2511         return ret;
2512 }
2513
2514 struct trace_option_dentry;
2515
2516 static struct trace_option_dentry *
2517 create_trace_option_files(struct tracer *tracer);
2518
2519 static void
2520 destroy_trace_option_files(struct trace_option_dentry *topts);
2521
2522 static int tracing_set_tracer(const char *buf)
2523 {
2524         static struct trace_option_dentry *topts;
2525         struct trace_array *tr = &global_trace;
2526         struct tracer *t;
2527         int ret = 0;
2528
2529         mutex_lock(&trace_types_lock);
2530
2531         if (!ring_buffer_expanded) {
2532                 ret = tracing_resize_ring_buffer(trace_buf_size);
2533                 if (ret < 0)
2534                         return ret;
2535                 ret = 0;
2536         }
2537
2538         for (t = trace_types; t; t = t->next) {
2539                 if (strcmp(t->name, buf) == 0)
2540                         break;
2541         }
2542         if (!t) {
2543                 ret = -EINVAL;
2544                 goto out;
2545         }
2546         if (t == current_trace)
2547                 goto out;
2548
2549         trace_branch_disable();
2550         if (current_trace && current_trace->reset)
2551                 current_trace->reset(tr);
2552
2553         destroy_trace_option_files(topts);
2554
2555         current_trace = t;
2556
2557         topts = create_trace_option_files(current_trace);
2558
2559         if (t->init) {
2560                 ret = tracer_init(t, tr);
2561                 if (ret)
2562                         goto out;
2563         }
2564
2565         trace_branch_enable(tr);
2566  out:
2567         mutex_unlock(&trace_types_lock);
2568
2569         return ret;
2570 }
2571
2572 static ssize_t
2573 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2574                         size_t cnt, loff_t *ppos)
2575 {
2576         char buf[max_tracer_type_len+1];
2577         int i;
2578         size_t ret;
2579         int err;
2580
2581         ret = cnt;
2582
2583         if (cnt > max_tracer_type_len)
2584                 cnt = max_tracer_type_len;
2585
2586         if (copy_from_user(&buf, ubuf, cnt))
2587                 return -EFAULT;
2588
2589         buf[cnt] = 0;
2590
2591         /* strip ending whitespace. */
2592         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2593                 buf[i] = 0;
2594
2595         err = tracing_set_tracer(buf);
2596         if (err)
2597                 return err;
2598
2599         filp->f_pos += ret;
2600
2601         return ret;
2602 }
2603
2604 static ssize_t
2605 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2606                      size_t cnt, loff_t *ppos)
2607 {
2608         unsigned long *ptr = filp->private_data;
2609         char buf[64];
2610         int r;
2611
2612         r = snprintf(buf, sizeof(buf), "%ld\n",
2613                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2614         if (r > sizeof(buf))
2615                 r = sizeof(buf);
2616         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2617 }
2618
2619 static ssize_t
2620 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2621                       size_t cnt, loff_t *ppos)
2622 {
2623         unsigned long *ptr = filp->private_data;
2624         char buf[64];
2625         unsigned long val;
2626         int ret;
2627
2628         if (cnt >= sizeof(buf))
2629                 return -EINVAL;
2630
2631         if (copy_from_user(&buf, ubuf, cnt))
2632                 return -EFAULT;
2633
2634         buf[cnt] = 0;
2635
2636         ret = strict_strtoul(buf, 10, &val);
2637         if (ret < 0)
2638                 return ret;
2639
2640         *ptr = val * 1000;
2641
2642         return cnt;
2643 }
2644
2645 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2646 {
2647         long cpu_file = (long) inode->i_private;
2648         struct trace_iterator *iter;
2649         int ret = 0;
2650
2651         if (tracing_disabled)
2652                 return -ENODEV;
2653
2654         mutex_lock(&trace_types_lock);
2655
2656         /* We only allow one reader per cpu */
2657         if (cpu_file == TRACE_PIPE_ALL_CPU) {
2658                 if (!cpumask_empty(tracing_reader_cpumask)) {
2659                         ret = -EBUSY;
2660                         goto out;
2661                 }
2662                 cpumask_setall(tracing_reader_cpumask);
2663         } else {
2664                 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2665                         cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2666                 else {
2667                         ret = -EBUSY;
2668                         goto out;
2669                 }
2670         }
2671
2672         /* create a buffer to store the information to pass to userspace */
2673         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2674         if (!iter) {
2675                 ret = -ENOMEM;
2676                 goto out;
2677         }
2678
2679         /*
2680          * We make a copy of the current tracer to avoid concurrent
2681          * changes on it while we are reading.
2682          */
2683         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2684         if (!iter->trace) {
2685                 ret = -ENOMEM;
2686                 goto fail;
2687         }
2688         if (current_trace)
2689                 *iter->trace = *current_trace;
2690
2691         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2692                 ret = -ENOMEM;
2693                 goto fail;
2694         }
2695
2696         /* trace pipe does not show start of buffer */
2697         cpumask_setall(iter->started);
2698
2699         iter->cpu_file = cpu_file;
2700         iter->tr = &global_trace;
2701         mutex_init(&iter->mutex);
2702         filp->private_data = iter;
2703
2704         if (iter->trace->pipe_open)
2705                 iter->trace->pipe_open(iter);
2706
2707 out:
2708         mutex_unlock(&trace_types_lock);
2709         return ret;
2710
2711 fail:
2712         kfree(iter->trace);
2713         kfree(iter);
2714         mutex_unlock(&trace_types_lock);
2715         return ret;
2716 }
2717
2718 static int tracing_release_pipe(struct inode *inode, struct file *file)
2719 {
2720         struct trace_iterator *iter = file->private_data;
2721
2722         mutex_lock(&trace_types_lock);
2723
2724         if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2725                 cpumask_clear(tracing_reader_cpumask);
2726         else
2727                 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2728
2729         mutex_unlock(&trace_types_lock);
2730
2731         free_cpumask_var(iter->started);
2732         mutex_destroy(&iter->mutex);
2733         kfree(iter->trace);
2734         kfree(iter);
2735
2736         return 0;
2737 }
2738
2739 static unsigned int
2740 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2741 {
2742         struct trace_iterator *iter = filp->private_data;
2743
2744         if (trace_flags & TRACE_ITER_BLOCK) {
2745                 /*
2746                  * Always select as readable when in blocking mode
2747                  */
2748                 return POLLIN | POLLRDNORM;
2749         } else {
2750                 if (!trace_empty(iter))
2751                         return POLLIN | POLLRDNORM;
2752                 poll_wait(filp, &trace_wait, poll_table);
2753                 if (!trace_empty(iter))
2754                         return POLLIN | POLLRDNORM;
2755
2756                 return 0;
2757         }
2758 }
2759
2760
2761 void default_wait_pipe(struct trace_iterator *iter)
2762 {
2763         DEFINE_WAIT(wait);
2764
2765         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2766
2767         if (trace_empty(iter))
2768                 schedule();
2769
2770         finish_wait(&trace_wait, &wait);
2771 }
2772
2773 /*
2774  * This is a make-shift waitqueue.
2775  * A tracer might use this callback on some rare cases:
2776  *
2777  *  1) the current tracer might hold the runqueue lock when it wakes up
2778  *     a reader, hence a deadlock (sched, function, and function graph tracers)
2779  *  2) the function tracers, trace all functions, we don't want
2780  *     the overhead of calling wake_up and friends
2781  *     (and tracing them too)
2782  *
2783  *     Anyway, this is really very primitive wakeup.
2784  */
2785 void poll_wait_pipe(struct trace_iterator *iter)
2786 {
2787         set_current_state(TASK_INTERRUPTIBLE);
2788         /* sleep for 100 msecs, and try again. */
2789         schedule_timeout(HZ / 10);
2790 }
2791
2792 /* Must be called with trace_types_lock mutex held. */
2793 static int tracing_wait_pipe(struct file *filp)
2794 {
2795         struct trace_iterator *iter = filp->private_data;
2796
2797         while (trace_empty(iter)) {
2798
2799                 if ((filp->f_flags & O_NONBLOCK)) {
2800                         return -EAGAIN;
2801                 }
2802
2803                 mutex_unlock(&iter->mutex);
2804
2805                 iter->trace->wait_pipe(iter);
2806
2807                 mutex_lock(&iter->mutex);
2808
2809                 if (signal_pending(current))
2810                         return -EINTR;
2811
2812                 /*
2813                  * We block until we read something and tracing is disabled.
2814                  * We still block if tracing is disabled, but we have never
2815                  * read anything. This allows a user to cat this file, and
2816                  * then enable tracing. But after we have read something,
2817                  * we give an EOF when tracing is again disabled.
2818                  *
2819                  * iter->pos will be 0 if we haven't read anything.
2820                  */
2821                 if (!tracer_enabled && iter->pos)
2822                         break;
2823         }
2824
2825         return 1;
2826 }
2827
2828 /*
2829  * Consumer reader.
2830  */
2831 static ssize_t
2832 tracing_read_pipe(struct file *filp, char __user *ubuf,
2833                   size_t cnt, loff_t *ppos)
2834 {
2835         struct trace_iterator *iter = filp->private_data;
2836         static struct tracer *old_tracer;
2837         ssize_t sret;
2838
2839         /* return any leftover data */
2840         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2841         if (sret != -EBUSY)
2842                 return sret;
2843
2844         trace_seq_init(&iter->seq);
2845
2846         /* copy the tracer to avoid using a global lock all around */
2847         mutex_lock(&trace_types_lock);
2848         if (unlikely(old_tracer != current_trace && current_trace)) {
2849                 old_tracer = current_trace;
2850                 *iter->trace = *current_trace;
2851         }
2852         mutex_unlock(&trace_types_lock);
2853
2854         /*
2855          * Avoid more than one consumer on a single file descriptor
2856          * This is just a matter of traces coherency, the ring buffer itself
2857          * is protected.
2858          */
2859         mutex_lock(&iter->mutex);
2860         if (iter->trace->read) {
2861                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2862                 if (sret)
2863                         goto out;
2864         }
2865
2866 waitagain:
2867         sret = tracing_wait_pipe(filp);
2868         if (sret <= 0)
2869                 goto out;
2870
2871         /* stop when tracing is finished */
2872         if (trace_empty(iter)) {
2873                 sret = 0;
2874                 goto out;
2875         }
2876
2877         if (cnt >= PAGE_SIZE)
2878                 cnt = PAGE_SIZE - 1;
2879
2880         /* reset all but tr, trace, and overruns */
2881         memset(&iter->seq, 0,
2882                sizeof(struct trace_iterator) -
2883                offsetof(struct trace_iterator, seq));
2884         iter->pos = -1;
2885
2886         while (find_next_entry_inc(iter) != NULL) {
2887                 enum print_line_t ret;
2888                 int len = iter->seq.len;
2889
2890                 ret = print_trace_line(iter);
2891                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2892                         /* don't print partial lines */
2893                         iter->seq.len = len;
2894                         break;
2895                 }
2896                 if (ret != TRACE_TYPE_NO_CONSUME)
2897                         trace_consume(iter);
2898
2899                 if (iter->seq.len >= cnt)
2900                         break;
2901         }
2902
2903         /* Now copy what we have to the user */
2904         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2905         if (iter->seq.readpos >= iter->seq.len)
2906                 trace_seq_init(&iter->seq);
2907
2908         /*
2909          * If there was nothing to send to user, inspite of consuming trace
2910          * entries, go back to wait for more entries.
2911          */
2912         if (sret == -EBUSY)
2913                 goto waitagain;
2914
2915 out:
2916         mutex_unlock(&iter->mutex);
2917
2918         return sret;
2919 }
2920
2921 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
2922                                      struct pipe_buffer *buf)
2923 {
2924         __free_page(buf->page);
2925 }
2926
2927 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
2928                                      unsigned int idx)
2929 {
2930         __free_page(spd->pages[idx]);
2931 }
2932
2933 static struct pipe_buf_operations tracing_pipe_buf_ops = {
2934         .can_merge              = 0,
2935         .map                    = generic_pipe_buf_map,
2936         .unmap                  = generic_pipe_buf_unmap,
2937         .confirm                = generic_pipe_buf_confirm,
2938         .release                = tracing_pipe_buf_release,
2939         .steal                  = generic_pipe_buf_steal,
2940         .get                    = generic_pipe_buf_get,
2941 };
2942
2943 static size_t
2944 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
2945 {
2946         size_t count;
2947         int ret;
2948
2949         /* Seq buffer is page-sized, exactly what we need. */
2950         for (;;) {
2951                 count = iter->seq.len;
2952                 ret = print_trace_line(iter);
2953                 count = iter->seq.len - count;
2954                 if (rem < count) {
2955                         rem = 0;
2956                         iter->seq.len -= count;
2957                         break;
2958                 }
2959                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2960                         iter->seq.len -= count;
2961                         break;
2962                 }
2963
2964                 trace_consume(iter);
2965                 rem -= count;
2966                 if (!find_next_entry_inc(iter)) {
2967                         rem = 0;
2968                         iter->ent = NULL;
2969                         break;
2970                 }
2971         }
2972
2973         return rem;
2974 }
2975
2976 static ssize_t tracing_splice_read_pipe(struct file *filp,
2977                                         loff_t *ppos,
2978                                         struct pipe_inode_info *pipe,
2979                                         size_t len,
2980                                         unsigned int flags)
2981 {
2982         struct page *pages[PIPE_BUFFERS];
2983         struct partial_page partial[PIPE_BUFFERS];
2984         struct trace_iterator *iter = filp->private_data;
2985         struct splice_pipe_desc spd = {
2986                 .pages          = pages,
2987                 .partial        = partial,
2988                 .nr_pages       = 0, /* This gets updated below. */
2989                 .flags          = flags,
2990                 .ops            = &tracing_pipe_buf_ops,
2991                 .spd_release    = tracing_spd_release_pipe,
2992         };
2993         static struct tracer *old_tracer;
2994         ssize_t ret;
2995         size_t rem;
2996         unsigned int i;
2997
2998         /* copy the tracer to avoid using a global lock all around */
2999         mutex_lock(&trace_types_lock);
3000         if (unlikely(old_tracer != current_trace && current_trace)) {
3001                 old_tracer = current_trace;
3002                 *iter->trace = *current_trace;
3003         }
3004         mutex_unlock(&trace_types_lock);
3005
3006         mutex_lock(&iter->mutex);
3007
3008         if (iter->trace->splice_read) {
3009                 ret = iter->trace->splice_read(iter, filp,
3010                                                ppos, pipe, len, flags);
3011                 if (ret)
3012                         goto out_err;
3013         }
3014
3015         ret = tracing_wait_pipe(filp);
3016         if (ret <= 0)
3017                 goto out_err;
3018
3019         if (!iter->ent && !find_next_entry_inc(iter)) {
3020                 ret = -EFAULT;
3021                 goto out_err;
3022         }
3023
3024         /* Fill as many pages as possible. */
3025         for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3026                 pages[i] = alloc_page(GFP_KERNEL);
3027                 if (!pages[i])
3028                         break;
3029
3030                 rem = tracing_fill_pipe_page(rem, iter);
3031
3032                 /* Copy the data into the page, so we can start over. */
3033                 ret = trace_seq_to_buffer(&iter->seq,
3034                                           page_address(pages[i]),
3035                                           iter->seq.len);
3036                 if (ret < 0) {
3037                         __free_page(pages[i]);
3038                         break;
3039                 }
3040                 partial[i].offset = 0;
3041                 partial[i].len = iter->seq.len;
3042
3043                 trace_seq_init(&iter->seq);
3044         }
3045
3046         mutex_unlock(&iter->mutex);
3047
3048         spd.nr_pages = i;
3049
3050         return splice_to_pipe(pipe, &spd);
3051
3052 out_err:
3053         mutex_unlock(&iter->mutex);
3054
3055         return ret;
3056 }
3057
3058 static ssize_t
3059 tracing_entries_read(struct file *filp, char __user *ubuf,
3060                      size_t cnt, loff_t *ppos)
3061 {
3062         struct trace_array *tr = filp->private_data;
3063         char buf[96];
3064         int r;
3065
3066         mutex_lock(&trace_types_lock);
3067         if (!ring_buffer_expanded)
3068                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3069                             tr->entries >> 10,
3070                             trace_buf_size >> 10);
3071         else
3072                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3073         mutex_unlock(&trace_types_lock);
3074
3075         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3076 }
3077
3078 static ssize_t
3079 tracing_entries_write(struct file *filp, const char __user *ubuf,
3080                       size_t cnt, loff_t *ppos)
3081 {
3082         unsigned long val;
3083         char buf[64];
3084         int ret, cpu;
3085
3086         if (cnt >= sizeof(buf))
3087                 return -EINVAL;
3088
3089         if (copy_from_user(&buf, ubuf, cnt))
3090                 return -EFAULT;
3091
3092         buf[cnt] = 0;
3093
3094         ret = strict_strtoul(buf, 10, &val);
3095         if (ret < 0)
3096                 return ret;
3097
3098         /* must have at least 1 entry */
3099         if (!val)
3100                 return -EINVAL;
3101
3102         mutex_lock(&trace_types_lock);
3103
3104         tracing_stop();
3105
3106         /* disable all cpu buffers */
3107         for_each_tracing_cpu(cpu) {
3108                 if (global_trace.data[cpu])
3109                         atomic_inc(&global_trace.data[cpu]->disabled);
3110                 if (max_tr.data[cpu])
3111                         atomic_inc(&max_tr.data[cpu]->disabled);
3112         }
3113
3114         /* value is in KB */
3115         val <<= 10;
3116
3117         if (val != global_trace.entries) {
3118                 ret = tracing_resize_ring_buffer(val);
3119                 if (ret < 0) {
3120                         cnt = ret;
3121                         goto out;
3122                 }
3123         }
3124
3125         filp->f_pos += cnt;
3126
3127         /* If check pages failed, return ENOMEM */
3128         if (tracing_disabled)
3129                 cnt = -ENOMEM;
3130  out:
3131         for_each_tracing_cpu(cpu) {
3132                 if (global_trace.data[cpu])
3133                         atomic_dec(&global_trace.data[cpu]->disabled);
3134                 if (max_tr.data[cpu])
3135                         atomic_dec(&max_tr.data[cpu]->disabled);
3136         }
3137
3138         tracing_start();
3139         max_tr.entries = global_trace.entries;
3140         mutex_unlock(&trace_types_lock);
3141
3142         return cnt;
3143 }
3144
3145 static int mark_printk(const char *fmt, ...)
3146 {
3147         int ret;
3148         va_list args;
3149         va_start(args, fmt);
3150         ret = trace_vprintk(0, -1, fmt, args);
3151         va_end(args);
3152         return ret;
3153 }
3154
3155 static ssize_t
3156 tracing_mark_write(struct file *filp, const char __user *ubuf,
3157                                         size_t cnt, loff_t *fpos)
3158 {
3159         char *buf;
3160         char *end;
3161
3162         if (tracing_disabled)
3163                 return -EINVAL;
3164
3165         if (cnt > TRACE_BUF_SIZE)
3166                 cnt = TRACE_BUF_SIZE;
3167
3168         buf = kmalloc(cnt + 1, GFP_KERNEL);
3169         if (buf == NULL)
3170                 return -ENOMEM;
3171
3172         if (copy_from_user(buf, ubuf, cnt)) {
3173                 kfree(buf);
3174                 return -EFAULT;
3175         }
3176
3177         /* Cut from the first nil or newline. */
3178         buf[cnt] = '\0';
3179         end = strchr(buf, '\n');
3180         if (end)
3181                 *end = '\0';
3182
3183         cnt = mark_printk("%s\n", buf);
3184         kfree(buf);
3185         *fpos += cnt;
3186
3187         return cnt;
3188 }
3189
3190 static const struct file_operations tracing_max_lat_fops = {
3191         .open           = tracing_open_generic,
3192         .read           = tracing_max_lat_read,
3193         .write          = tracing_max_lat_write,
3194 };
3195
3196 static const struct file_operations tracing_ctrl_fops = {
3197         .open           = tracing_open_generic,
3198         .read           = tracing_ctrl_read,
3199         .write          = tracing_ctrl_write,
3200 };
3201
3202 static const struct file_operations set_tracer_fops = {
3203         .open           = tracing_open_generic,
3204         .read           = tracing_set_trace_read,
3205         .write          = tracing_set_trace_write,
3206 };
3207
3208 static const struct file_operations tracing_pipe_fops = {
3209         .open           = tracing_open_pipe,
3210         .poll           = tracing_poll_pipe,
3211         .read           = tracing_read_pipe,
3212         .splice_read    = tracing_splice_read_pipe,
3213         .release        = tracing_release_pipe,
3214 };
3215
3216 static const struct file_operations tracing_entries_fops = {
3217         .open           = tracing_open_generic,
3218         .read           = tracing_entries_read,
3219         .write          = tracing_entries_write,
3220 };
3221
3222 static const struct file_operations tracing_mark_fops = {
3223         .open           = tracing_open_generic,
3224         .write          = tracing_mark_write,
3225 };
3226
3227 struct ftrace_buffer_info {
3228         struct trace_array      *tr;
3229         void                    *spare;
3230         int                     cpu;
3231         unsigned int            read;
3232 };
3233
3234 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3235 {
3236         int cpu = (int)(long)inode->i_private;
3237         struct ftrace_buffer_info *info;
3238
3239         if (tracing_disabled)
3240                 return -ENODEV;
3241
3242         info = kzalloc(sizeof(*info), GFP_KERNEL);
3243         if (!info)
3244                 return -ENOMEM;
3245
3246         info->tr        = &global_trace;
3247         info->cpu       = cpu;
3248         info->spare     = ring_buffer_alloc_read_page(info->tr->buffer);
3249         /* Force reading ring buffer for first read */
3250         info->read      = (unsigned int)-1;
3251         if (!info->spare)
3252                 goto out;
3253
3254         filp->private_data = info;
3255
3256         return 0;
3257
3258  out:
3259         kfree(info);
3260         return -ENOMEM;
3261 }
3262
3263 static ssize_t
3264 tracing_buffers_read(struct file *filp, char __user *ubuf,
3265                      size_t count, loff_t *ppos)
3266 {
3267         struct ftrace_buffer_info *info = filp->private_data;
3268         unsigned int pos;
3269         ssize_t ret;
3270         size_t size;
3271
3272         if (!count)
3273                 return 0;
3274
3275         /* Do we have previous read data to read? */
3276         if (info->read < PAGE_SIZE)
3277                 goto read;
3278
3279         info->read = 0;
3280
3281         ret = ring_buffer_read_page(info->tr->buffer,
3282                                     &info->spare,
3283                                     count,
3284                                     info->cpu, 0);
3285         if (ret < 0)
3286                 return 0;
3287
3288         pos = ring_buffer_page_len(info->spare);
3289
3290         if (pos < PAGE_SIZE)
3291                 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3292
3293 read:
3294         size = PAGE_SIZE - info->read;
3295         if (size > count)
3296                 size = count;
3297
3298         ret = copy_to_user(ubuf, info->spare + info->read, size);
3299         if (ret == size)
3300                 return -EFAULT;
3301         size -= ret;
3302
3303         *ppos += size;
3304         info->read += size;
3305
3306         return size;
3307 }
3308
3309 static int tracing_buffers_release(struct inode *inode, struct file *file)
3310 {
3311         struct ftrace_buffer_info *info = file->private_data;
3312
3313         ring_buffer_free_read_page(info->tr->buffer, info->spare);
3314         kfree(info);
3315
3316         return 0;
3317 }
3318
3319 struct buffer_ref {
3320         struct ring_buffer      *buffer;
3321         void                    *page;
3322         int                     ref;
3323 };
3324
3325 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3326                                     struct pipe_buffer *buf)
3327 {
3328         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3329
3330         if (--ref->ref)
3331                 return;
3332
3333         ring_buffer_free_read_page(ref->buffer, ref->page);
3334         kfree(ref);
3335         buf->private = 0;
3336 }
3337
3338 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3339                                  struct pipe_buffer *buf)
3340 {
3341         return 1;
3342 }
3343
3344 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3345                                 struct pipe_buffer *buf)
3346 {
3347         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3348
3349         ref->ref++;
3350 }
3351
3352 /* Pipe buffer operations for a buffer. */
3353 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3354         .can_merge              = 0,
3355         .map                    = generic_pipe_buf_map,
3356         .unmap                  = generic_pipe_buf_unmap,
3357         .confirm                = generic_pipe_buf_confirm,
3358         .release                = buffer_pipe_buf_release,
3359         .steal                  = buffer_pipe_buf_steal,
3360         .get                    = buffer_pipe_buf_get,
3361 };
3362
3363 /*
3364  * Callback from splice_to_pipe(), if we need to release some pages
3365  * at the end of the spd in case we error'ed out in filling the pipe.
3366  */
3367 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3368 {
3369         struct buffer_ref *ref =
3370                 (struct buffer_ref *)spd->partial[i].private;
3371
3372         if (--ref->ref)
3373                 return;
3374
3375         ring_buffer_free_read_page(ref->buffer, ref->page);
3376         kfree(ref);
3377         spd->partial[i].private = 0;
3378 }
3379
3380 static ssize_t
3381 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3382                             struct pipe_inode_info *pipe, size_t len,
3383                             unsigned int flags)
3384 {
3385         struct ftrace_buffer_info *info = file->private_data;
3386         struct partial_page partial[PIPE_BUFFERS];
3387         struct page *pages[PIPE_BUFFERS];
3388         struct splice_pipe_desc spd = {
3389                 .pages          = pages,
3390                 .partial        = partial,
3391                 .flags          = flags,
3392                 .ops            = &buffer_pipe_buf_ops,
3393                 .spd_release    = buffer_spd_release,
3394         };
3395         struct buffer_ref *ref;
3396         int size, i;
3397         size_t ret;
3398
3399         /*
3400          * We can't seek on a buffer input
3401          */
3402         if (unlikely(*ppos))
3403                 return -ESPIPE;
3404
3405
3406         for (i = 0; i < PIPE_BUFFERS && len; i++, len -= size) {
3407                 struct page *page;
3408                 int r;
3409
3410                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3411                 if (!ref)
3412                         break;
3413
3414                 ref->buffer = info->tr->buffer;
3415                 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3416                 if (!ref->page) {
3417                         kfree(ref);
3418                         break;
3419                 }
3420
3421                 r = ring_buffer_read_page(ref->buffer, &ref->page,
3422                                           len, info->cpu, 0);
3423                 if (r < 0) {
3424                         ring_buffer_free_read_page(ref->buffer,
3425                                                    ref->page);
3426                         kfree(ref);
3427                         break;
3428                 }
3429
3430                 /*
3431                  * zero out any left over data, this is going to
3432                  * user land.
3433                  */
3434                 size = ring_buffer_page_len(ref->page);
3435                 if (size < PAGE_SIZE)
3436                         memset(ref->page + size, 0, PAGE_SIZE - size);
3437
3438                 page = virt_to_page(ref->page);
3439
3440                 spd.pages[i] = page;
3441                 spd.partial[i].len = PAGE_SIZE;
3442                 spd.partial[i].offset = 0;
3443                 spd.partial[i].private = (unsigned long)ref;
3444                 spd.nr_pages++;
3445         }
3446
3447         spd.nr_pages = i;
3448
3449         /* did we read anything? */
3450         if (!spd.nr_pages) {
3451                 if (flags & SPLICE_F_NONBLOCK)
3452                         ret = -EAGAIN;
3453                 else
3454                         ret = 0;
3455                 /* TODO: block */
3456                 return ret;
3457         }
3458
3459         ret = splice_to_pipe(pipe, &spd);
3460
3461         return ret;
3462 }
3463
3464 static const struct file_operations tracing_buffers_fops = {
3465         .open           = tracing_buffers_open,
3466         .read           = tracing_buffers_read,
3467         .release        = tracing_buffers_release,
3468         .splice_read    = tracing_buffers_splice_read,
3469         .llseek         = no_llseek,
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 const 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 static struct dentry *d_percpu;
3530
3531 struct dentry *tracing_dentry_percpu(void)
3532 {
3533         static int once;
3534         struct dentry *d_tracer;
3535
3536         if (d_percpu)
3537                 return d_percpu;
3538
3539         d_tracer = tracing_init_dentry();
3540
3541         if (!d_tracer)
3542                 return NULL;
3543
3544         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3545
3546         if (!d_percpu && !once) {
3547                 once = 1;
3548                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3549                 return NULL;
3550         }
3551
3552         return d_percpu;
3553 }
3554
3555 static void tracing_init_debugfs_percpu(long cpu)
3556 {
3557         struct dentry *d_percpu = tracing_dentry_percpu();
3558         struct dentry *entry, *d_cpu;
3559         /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3560         char cpu_dir[7];
3561
3562         if (cpu > 999 || cpu < 0)
3563                 return;
3564
3565         sprintf(cpu_dir, "cpu%ld", cpu);
3566         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3567         if (!d_cpu) {
3568                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3569                 return;
3570         }
3571
3572         /* per cpu trace_pipe */
3573         entry = debugfs_create_file("trace_pipe", 0444, d_cpu,
3574                                 (void *) cpu, &tracing_pipe_fops);
3575         if (!entry)
3576                 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3577
3578         /* per cpu trace */
3579         entry = debugfs_create_file("trace", 0444, d_cpu,
3580                                 (void *) cpu, &tracing_fops);
3581         if (!entry)
3582                 pr_warning("Could not create debugfs 'trace' entry\n");
3583
3584         entry = debugfs_create_file("trace_pipe_raw", 0444, d_cpu,
3585                                     (void *) cpu, &tracing_buffers_fops);
3586         if (!entry)
3587                 pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n");
3588 }
3589
3590 #ifdef CONFIG_FTRACE_SELFTEST
3591 /* Let selftest have access to static functions in this file */
3592 #include "trace_selftest.c"
3593 #endif
3594
3595 struct trace_option_dentry {
3596         struct tracer_opt               *opt;
3597         struct tracer_flags             *flags;
3598         struct dentry                   *entry;
3599 };
3600
3601 static ssize_t
3602 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3603                         loff_t *ppos)
3604 {
3605         struct trace_option_dentry *topt = filp->private_data;
3606         char *buf;
3607
3608         if (topt->flags->val & topt->opt->bit)
3609                 buf = "1\n";
3610         else
3611                 buf = "0\n";
3612
3613         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3614 }
3615
3616 static ssize_t
3617 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3618                          loff_t *ppos)
3619 {
3620         struct trace_option_dentry *topt = filp->private_data;
3621         unsigned long val;
3622         char buf[64];
3623         int ret;
3624
3625         if (cnt >= sizeof(buf))
3626                 return -EINVAL;
3627
3628         if (copy_from_user(&buf, ubuf, cnt))
3629                 return -EFAULT;
3630
3631         buf[cnt] = 0;
3632
3633         ret = strict_strtoul(buf, 10, &val);
3634         if (ret < 0)
3635                 return ret;
3636
3637         ret = 0;
3638         switch (val) {
3639         case 0:
3640                 /* do nothing if already cleared */
3641                 if (!(topt->flags->val & topt->opt->bit))
3642                         break;
3643
3644                 mutex_lock(&trace_types_lock);
3645                 if (current_trace->set_flag)
3646                         ret = current_trace->set_flag(topt->flags->val,
3647                                                       topt->opt->bit, 0);
3648                 mutex_unlock(&trace_types_lock);
3649                 if (ret)
3650                         return ret;
3651                 topt->flags->val &= ~topt->opt->bit;
3652                 break;
3653         case 1:
3654                 /* do nothing if already set */
3655                 if (topt->flags->val & topt->opt->bit)
3656                         break;
3657
3658                 mutex_lock(&trace_types_lock);
3659                 if (current_trace->set_flag)
3660                         ret = current_trace->set_flag(topt->flags->val,
3661                                                       topt->opt->bit, 1);
3662                 mutex_unlock(&trace_types_lock);
3663                 if (ret)
3664                         return ret;
3665                 topt->flags->val |= topt->opt->bit;
3666                 break;
3667
3668         default:
3669                 return -EINVAL;
3670         }
3671
3672         *ppos += cnt;
3673
3674         return cnt;
3675 }
3676
3677
3678 static const struct file_operations trace_options_fops = {
3679         .open = tracing_open_generic,
3680         .read = trace_options_read,
3681         .write = trace_options_write,
3682 };
3683
3684 static ssize_t
3685 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3686                         loff_t *ppos)
3687 {
3688         long index = (long)filp->private_data;
3689         char *buf;
3690
3691         if (trace_flags & (1 << index))
3692                 buf = "1\n";
3693         else
3694                 buf = "0\n";
3695
3696         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3697 }
3698
3699 static ssize_t
3700 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3701                          loff_t *ppos)
3702 {
3703         long index = (long)filp->private_data;
3704         char buf[64];
3705         unsigned long val;
3706         int ret;
3707
3708         if (cnt >= sizeof(buf))
3709                 return -EINVAL;
3710
3711         if (copy_from_user(&buf, ubuf, cnt))
3712                 return -EFAULT;
3713
3714         buf[cnt] = 0;
3715
3716         ret = strict_strtoul(buf, 10, &val);
3717         if (ret < 0)
3718                 return ret;
3719
3720         switch (val) {
3721         case 0:
3722                 trace_flags &= ~(1 << index);
3723                 break;
3724         case 1:
3725                 trace_flags |= 1 << index;
3726                 break;
3727
3728         default:
3729                 return -EINVAL;
3730         }
3731
3732         *ppos += cnt;
3733
3734         return cnt;
3735 }
3736
3737 static const struct file_operations trace_options_core_fops = {
3738         .open = tracing_open_generic,
3739         .read = trace_options_core_read,
3740         .write = trace_options_core_write,
3741 };
3742
3743 static struct dentry *trace_options_init_dentry(void)
3744 {
3745         struct dentry *d_tracer;
3746         static struct dentry *t_options;
3747
3748         if (t_options)
3749                 return t_options;
3750
3751         d_tracer = tracing_init_dentry();
3752         if (!d_tracer)
3753                 return NULL;
3754
3755         t_options = debugfs_create_dir("options", d_tracer);
3756         if (!t_options) {
3757                 pr_warning("Could not create debugfs directory 'options'\n");
3758                 return NULL;
3759         }
3760
3761         return t_options;
3762 }
3763
3764 static void
3765 create_trace_option_file(struct trace_option_dentry *topt,
3766                          struct tracer_flags *flags,
3767                          struct tracer_opt *opt)
3768 {
3769         struct dentry *t_options;
3770         struct dentry *entry;
3771
3772         t_options = trace_options_init_dentry();
3773         if (!t_options)
3774                 return;
3775
3776         topt->flags = flags;
3777         topt->opt = opt;
3778
3779         entry = debugfs_create_file(opt->name, 0644, t_options, topt,
3780                                     &trace_options_fops);
3781
3782         topt->entry = entry;
3783
3784 }
3785
3786 static struct trace_option_dentry *
3787 create_trace_option_files(struct tracer *tracer)
3788 {
3789         struct trace_option_dentry *topts;
3790         struct tracer_flags *flags;
3791         struct tracer_opt *opts;
3792         int cnt;
3793
3794         if (!tracer)
3795                 return NULL;
3796
3797         flags = tracer->flags;
3798
3799         if (!flags || !flags->opts)
3800                 return NULL;
3801
3802         opts = flags->opts;
3803
3804         for (cnt = 0; opts[cnt].name; cnt++)
3805                 ;
3806
3807         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3808         if (!topts)
3809                 return NULL;
3810
3811         for (cnt = 0; opts[cnt].name; cnt++)
3812                 create_trace_option_file(&topts[cnt], flags,
3813                                          &opts[cnt]);
3814
3815         return topts;
3816 }
3817
3818 static void
3819 destroy_trace_option_files(struct trace_option_dentry *topts)
3820 {
3821         int cnt;
3822
3823         if (!topts)
3824                 return;
3825
3826         for (cnt = 0; topts[cnt].opt; cnt++) {
3827                 if (topts[cnt].entry)
3828                         debugfs_remove(topts[cnt].entry);
3829         }
3830
3831         kfree(topts);
3832 }
3833
3834 static struct dentry *
3835 create_trace_option_core_file(const char *option, long index)
3836 {
3837         struct dentry *t_options;
3838         struct dentry *entry;
3839
3840         t_options = trace_options_init_dentry();
3841         if (!t_options)
3842                 return NULL;
3843
3844         entry = debugfs_create_file(option, 0644, t_options, (void *)index,
3845                                     &trace_options_core_fops);
3846
3847         return entry;
3848 }
3849
3850 static __init void create_trace_options_dir(void)
3851 {
3852         struct dentry *t_options;
3853         struct dentry *entry;
3854         int i;
3855
3856         t_options = trace_options_init_dentry();
3857         if (!t_options)
3858                 return;
3859
3860         for (i = 0; trace_options[i]; i++) {
3861                 entry = create_trace_option_core_file(trace_options[i], i);
3862                 if (!entry)
3863                         pr_warning("Could not create debugfs %s entry\n",
3864                                    trace_options[i]);
3865         }
3866 }
3867
3868 static __init int tracer_init_debugfs(void)
3869 {
3870         struct dentry *d_tracer;
3871         struct dentry *entry;
3872         int cpu;
3873
3874         d_tracer = tracing_init_dentry();
3875
3876         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3877                                     &global_trace, &tracing_ctrl_fops);
3878         if (!entry)
3879                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3880
3881         entry = debugfs_create_file("trace_options", 0644, d_tracer,
3882                                     NULL, &tracing_iter_fops);
3883         if (!entry)
3884                 pr_warning("Could not create debugfs 'trace_options' entry\n");
3885
3886         create_trace_options_dir();
3887
3888         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3889                                     NULL, &tracing_cpumask_fops);
3890         if (!entry)
3891                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3892
3893         entry = debugfs_create_file("trace", 0444, d_tracer,
3894                                  (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
3895         if (!entry)
3896                 pr_warning("Could not create debugfs 'trace' entry\n");
3897
3898         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3899                                     &global_trace, &show_traces_fops);
3900         if (!entry)
3901                 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3902
3903         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3904                                     &global_trace, &set_tracer_fops);
3905         if (!entry)
3906                 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3907
3908         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3909                                     &tracing_max_latency,
3910                                     &tracing_max_lat_fops);
3911         if (!entry)
3912                 pr_warning("Could not create debugfs "
3913                            "'tracing_max_latency' entry\n");
3914
3915         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3916                                     &tracing_thresh, &tracing_max_lat_fops);
3917         if (!entry)
3918                 pr_warning("Could not create debugfs "
3919                            "'tracing_thresh' entry\n");
3920         entry = debugfs_create_file("README", 0644, d_tracer,
3921                                     NULL, &tracing_readme_fops);
3922         if (!entry)
3923                 pr_warning("Could not create debugfs 'README' entry\n");
3924
3925         entry = debugfs_create_file("trace_pipe", 0444, d_tracer,
3926                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
3927         if (!entry)
3928                 pr_warning("Could not create debugfs "
3929                            "'trace_pipe' entry\n");
3930
3931         entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3932                                     &global_trace, &tracing_entries_fops);
3933         if (!entry)
3934                 pr_warning("Could not create debugfs "
3935                            "'buffer_size_kb' entry\n");
3936
3937         entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3938                                     NULL, &tracing_mark_fops);
3939         if (!entry)
3940                 pr_warning("Could not create debugfs "
3941                            "'trace_marker' entry\n");
3942
3943 #ifdef CONFIG_DYNAMIC_FTRACE
3944         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3945                                     &ftrace_update_tot_cnt,
3946                                     &tracing_dyn_info_fops);
3947         if (!entry)
3948                 pr_warning("Could not create debugfs "
3949                            "'dyn_ftrace_total_info' entry\n");
3950 #endif
3951 #ifdef CONFIG_SYSPROF_TRACER
3952         init_tracer_sysprof_debugfs(d_tracer);
3953 #endif
3954
3955         for_each_tracing_cpu(cpu)
3956                 tracing_init_debugfs_percpu(cpu);
3957
3958         return 0;
3959 }
3960
3961 static int trace_panic_handler(struct notifier_block *this,
3962                                unsigned long event, void *unused)
3963 {
3964         if (ftrace_dump_on_oops)
3965                 ftrace_dump();
3966         return NOTIFY_OK;
3967 }
3968
3969 static struct notifier_block trace_panic_notifier = {
3970         .notifier_call  = trace_panic_handler,
3971         .next           = NULL,
3972         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
3973 };
3974
3975 static int trace_die_handler(struct notifier_block *self,
3976                              unsigned long val,
3977                              void *data)
3978 {
3979         switch (val) {
3980         case DIE_OOPS:
3981                 if (ftrace_dump_on_oops)
3982                         ftrace_dump();
3983                 break;
3984         default:
3985                 break;
3986         }
3987         return NOTIFY_OK;
3988 }
3989
3990 static struct notifier_block trace_die_notifier = {
3991         .notifier_call = trace_die_handler,
3992         .priority = 200
3993 };
3994
3995 /*
3996  * printk is set to max of 1024, we really don't need it that big.
3997  * Nothing should be printing 1000 characters anyway.
3998  */
3999 #define TRACE_MAX_PRINT         1000
4000
4001 /*
4002  * Define here KERN_TRACE so that we have one place to modify
4003  * it if we decide to change what log level the ftrace dump
4004  * should be at.
4005  */
4006 #define KERN_TRACE              KERN_EMERG
4007
4008 static void
4009 trace_printk_seq(struct trace_seq *s)
4010 {
4011         /* Probably should print a warning here. */
4012         if (s->len >= 1000)
4013                 s->len = 1000;
4014
4015         /* should be zero ended, but we are paranoid. */
4016         s->buffer[s->len] = 0;
4017
4018         printk(KERN_TRACE "%s", s->buffer);
4019
4020         trace_seq_init(s);
4021 }
4022
4023 void ftrace_dump(void)
4024 {
4025         static DEFINE_SPINLOCK(ftrace_dump_lock);
4026         /* use static because iter can be a bit big for the stack */
4027         static struct trace_iterator iter;
4028         static int dump_ran;
4029         unsigned long flags;
4030         int cnt = 0, cpu;
4031
4032         /* only one dump */
4033         spin_lock_irqsave(&ftrace_dump_lock, flags);
4034         if (dump_ran)
4035                 goto out;
4036
4037         dump_ran = 1;
4038
4039         /* No turning back! */
4040         tracing_off();
4041         ftrace_kill();
4042
4043         for_each_tracing_cpu(cpu) {
4044                 atomic_inc(&global_trace.data[cpu]->disabled);
4045         }
4046
4047         /* don't look at user memory in panic mode */
4048         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4049
4050         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4051
4052         /* Simulate the iterator */
4053         iter.tr = &global_trace;
4054         iter.trace = current_trace;
4055         iter.cpu_file = TRACE_PIPE_ALL_CPU;
4056
4057         /*
4058          * We need to stop all tracing on all CPUS to read the
4059          * the next buffer. This is a bit expensive, but is
4060          * not done often. We fill all what we can read,
4061          * and then release the locks again.
4062          */
4063
4064         while (!trace_empty(&iter)) {
4065
4066                 if (!cnt)
4067                         printk(KERN_TRACE "---------------------------------\n");
4068
4069                 cnt++;
4070
4071                 /* reset all but tr, trace, and overruns */
4072                 memset(&iter.seq, 0,
4073                        sizeof(struct trace_iterator) -
4074                        offsetof(struct trace_iterator, seq));
4075                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4076                 iter.pos = -1;
4077
4078                 if (find_next_entry_inc(&iter) != NULL) {
4079                         print_trace_line(&iter);
4080                         trace_consume(&iter);
4081                 }
4082
4083                 trace_printk_seq(&iter.seq);
4084         }
4085
4086         if (!cnt)
4087                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4088         else
4089                 printk(KERN_TRACE "---------------------------------\n");
4090
4091  out:
4092         spin_unlock_irqrestore(&ftrace_dump_lock, flags);
4093 }
4094
4095 __init static int tracer_alloc_buffers(void)
4096 {
4097         struct trace_array_cpu *data;
4098         int ring_buf_size;
4099         int i;
4100         int ret = -ENOMEM;
4101
4102         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4103                 goto out;
4104
4105         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4106                 goto out_free_buffer_mask;
4107
4108         if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4109                 goto out_free_tracing_cpumask;
4110
4111         /* To save memory, keep the ring buffer size to its minimum */
4112         if (ring_buffer_expanded)
4113                 ring_buf_size = trace_buf_size;
4114         else
4115                 ring_buf_size = 1;
4116
4117         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4118         cpumask_copy(tracing_cpumask, cpu_all_mask);
4119         cpumask_clear(tracing_reader_cpumask);
4120
4121         /* TODO: make the number of buffers hot pluggable with CPUS */
4122         global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4123                                                    TRACE_BUFFER_FLAGS);
4124         if (!global_trace.buffer) {
4125                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4126                 WARN_ON(1);
4127                 goto out_free_cpumask;
4128         }
4129         global_trace.entries = ring_buffer_size(global_trace.buffer);
4130
4131
4132 #ifdef CONFIG_TRACER_MAX_TRACE
4133         max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4134                                              TRACE_BUFFER_FLAGS);
4135         if (!max_tr.buffer) {
4136                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4137                 WARN_ON(1);
4138                 ring_buffer_free(global_trace.buffer);
4139                 goto out_free_cpumask;
4140         }
4141         max_tr.entries = ring_buffer_size(max_tr.buffer);
4142         WARN_ON(max_tr.entries != global_trace.entries);
4143 #endif
4144
4145         /* Allocate the first page for all buffers */
4146         for_each_tracing_cpu(i) {
4147                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4148                 max_tr.data[i] = &per_cpu(max_data, i);
4149         }
4150
4151         trace_init_cmdlines();
4152
4153         register_tracer(&nop_trace);
4154         current_trace = &nop_trace;
4155 #ifdef CONFIG_BOOT_TRACER
4156         register_tracer(&boot_tracer);
4157 #endif
4158         /* All seems OK, enable tracing */
4159         tracing_disabled = 0;
4160
4161         atomic_notifier_chain_register(&panic_notifier_list,
4162                                        &trace_panic_notifier);
4163
4164         register_die_notifier(&trace_die_notifier);
4165         ret = 0;
4166
4167 out_free_cpumask:
4168         free_cpumask_var(tracing_reader_cpumask);
4169 out_free_tracing_cpumask:
4170         free_cpumask_var(tracing_cpumask);
4171 out_free_buffer_mask:
4172         free_cpumask_var(tracing_buffer_mask);
4173 out:
4174         return ret;
4175 }
4176
4177 __init static int clear_boot_tracer(void)
4178 {
4179         /*
4180          * The default tracer at boot buffer is an init section.
4181          * This function is called in lateinit. If we did not
4182          * find the boot tracer, then clear it out, to prevent
4183          * later registration from accessing the buffer that is
4184          * about to be freed.
4185          */
4186         if (!default_bootup_tracer)
4187                 return 0;
4188
4189         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4190                default_bootup_tracer);
4191         default_bootup_tracer = NULL;
4192
4193         return 0;
4194 }
4195
4196 early_initcall(tracer_alloc_buffers);
4197 fs_initcall(tracer_init_debugfs);
4198 late_initcall(clear_boot_tracer);