Merge branch 'linus' into core/softlockup
[linux-2.6] / kernel / trace / trace.c
1 /*
2  * ring buffer based function tracer
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally taken from the RT patch by:
8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code from the latency_tracer, that is:
11  *  Copyright (C) 2004-2006 Ingo Molnar
12  *  Copyright (C) 2004 William Lee Irwin III
13  */
14 #include <linux/ring_buffer.h>
15 #include <linux/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/ctype.h>
34 #include <linux/init.h>
35 #include <linux/poll.h>
36 #include <linux/gfp.h>
37 #include <linux/fs.h>
38
39 #include "trace.h"
40 #include "trace_output.h"
41
42 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
43
44 unsigned long __read_mostly     tracing_max_latency;
45 unsigned long __read_mostly     tracing_thresh;
46
47 /*
48  * 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 | TRACE_ITER_SLEEP_TIME;
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         "sleep-time",
320         NULL
321 };
322
323 /*
324  * ftrace_max_lock is used to protect the swapping of buffers
325  * when taking a max snapshot. The buffers themselves are
326  * protected by per_cpu spinlocks. But the action of the swap
327  * needs its own lock.
328  *
329  * This is defined as a raw_spinlock_t in order to help
330  * with performance when lockdep debugging is enabled.
331  */
332 static raw_spinlock_t ftrace_max_lock =
333         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
334
335 /*
336  * Copy the new maximum trace into the separate maximum-trace
337  * structure. (this way the maximum trace is permanently saved,
338  * for later retrieval via /debugfs/tracing/latency_trace)
339  */
340 static void
341 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
342 {
343         struct trace_array_cpu *data = tr->data[cpu];
344
345         max_tr.cpu = cpu;
346         max_tr.time_start = data->preempt_timestamp;
347
348         data = max_tr.data[cpu];
349         data->saved_latency = tracing_max_latency;
350
351         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
352         data->pid = tsk->pid;
353         data->uid = task_uid(tsk);
354         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
355         data->policy = tsk->policy;
356         data->rt_priority = tsk->rt_priority;
357
358         /* record this tasks comm */
359         tracing_record_cmdline(tsk);
360 }
361
362 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
363 {
364         int len;
365         int ret;
366
367         if (!cnt)
368                 return 0;
369
370         if (s->len <= s->readpos)
371                 return -EBUSY;
372
373         len = s->len - s->readpos;
374         if (cnt > len)
375                 cnt = len;
376         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
377         if (ret == cnt)
378                 return -EFAULT;
379
380         cnt -= ret;
381
382         s->readpos += cnt;
383         return cnt;
384 }
385
386 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
387 {
388         int len;
389         void *ret;
390
391         if (s->len <= s->readpos)
392                 return -EBUSY;
393
394         len = s->len - s->readpos;
395         if (cnt > len)
396                 cnt = len;
397         ret = memcpy(buf, s->buffer + s->readpos, cnt);
398         if (!ret)
399                 return -EFAULT;
400
401         s->readpos += cnt;
402         return cnt;
403 }
404
405 static void
406 trace_print_seq(struct seq_file *m, struct trace_seq *s)
407 {
408         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
409
410         s->buffer[len] = 0;
411         seq_puts(m, s->buffer);
412
413         trace_seq_init(s);
414 }
415
416 /**
417  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
418  * @tr: tracer
419  * @tsk: the task with the latency
420  * @cpu: The cpu that initiated the trace.
421  *
422  * Flip the buffers between the @tr and the max_tr and record information
423  * about which task was the cause of this latency.
424  */
425 void
426 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
427 {
428         struct ring_buffer *buf = tr->buffer;
429
430         WARN_ON_ONCE(!irqs_disabled());
431         __raw_spin_lock(&ftrace_max_lock);
432
433         tr->buffer = max_tr.buffer;
434         max_tr.buffer = buf;
435
436         ftrace_disable_cpu();
437         ring_buffer_reset(tr->buffer);
438         ftrace_enable_cpu();
439
440         __update_max_tr(tr, tsk, cpu);
441         __raw_spin_unlock(&ftrace_max_lock);
442 }
443
444 /**
445  * update_max_tr_single - only copy one trace over, and reset the rest
446  * @tr - tracer
447  * @tsk - task with the latency
448  * @cpu - the cpu of the buffer to copy.
449  *
450  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
451  */
452 void
453 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
454 {
455         int ret;
456
457         WARN_ON_ONCE(!irqs_disabled());
458         __raw_spin_lock(&ftrace_max_lock);
459
460         ftrace_disable_cpu();
461
462         ring_buffer_reset(max_tr.buffer);
463         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
464
465         ftrace_enable_cpu();
466
467         WARN_ON_ONCE(ret && ret != -EAGAIN);
468
469         __update_max_tr(tr, tsk, cpu);
470         __raw_spin_unlock(&ftrace_max_lock);
471 }
472
473 /**
474  * register_tracer - register a tracer with the ftrace system.
475  * @type - the plugin for the tracer
476  *
477  * Register a new plugin tracer.
478  */
479 int register_tracer(struct tracer *type)
480 __releases(kernel_lock)
481 __acquires(kernel_lock)
482 {
483         struct tracer *t;
484         int len;
485         int ret = 0;
486
487         if (!type->name) {
488                 pr_info("Tracer must have a name\n");
489                 return -1;
490         }
491
492         /*
493          * When this gets called we hold the BKL which means that
494          * preemption is disabled. Various trace selftests however
495          * need to disable and enable preemption for successful tests.
496          * So we drop the BKL here and grab it after the tests again.
497          */
498         unlock_kernel();
499         mutex_lock(&trace_types_lock);
500
501         tracing_selftest_running = true;
502
503         for (t = trace_types; t; t = t->next) {
504                 if (strcmp(type->name, t->name) == 0) {
505                         /* already found */
506                         pr_info("Trace %s already registered\n",
507                                 type->name);
508                         ret = -1;
509                         goto out;
510                 }
511         }
512
513         if (!type->set_flag)
514                 type->set_flag = &dummy_set_flag;
515         if (!type->flags)
516                 type->flags = &dummy_tracer_flags;
517         else
518                 if (!type->flags->opts)
519                         type->flags->opts = dummy_tracer_opt;
520         if (!type->wait_pipe)
521                 type->wait_pipe = default_wait_pipe;
522
523
524 #ifdef CONFIG_FTRACE_STARTUP_TEST
525         if (type->selftest && !tracing_selftest_disabled) {
526                 struct tracer *saved_tracer = current_trace;
527                 struct trace_array *tr = &global_trace;
528                 int i;
529
530                 /*
531                  * Run a selftest on this tracer.
532                  * Here we reset the trace buffer, and set the current
533                  * tracer to be this tracer. The tracer can then run some
534                  * internal tracing to verify that everything is in order.
535                  * If we fail, we do not register this tracer.
536                  */
537                 for_each_tracing_cpu(i)
538                         tracing_reset(tr, i);
539
540                 current_trace = type;
541                 /* the test is responsible for initializing and enabling */
542                 pr_info("Testing tracer %s: ", type->name);
543                 ret = type->selftest(type, tr);
544                 /* the test is responsible for resetting too */
545                 current_trace = saved_tracer;
546                 if (ret) {
547                         printk(KERN_CONT "FAILED!\n");
548                         goto out;
549                 }
550                 /* Only reset on passing, to avoid touching corrupted buffers */
551                 for_each_tracing_cpu(i)
552                         tracing_reset(tr, i);
553
554                 printk(KERN_CONT "PASSED\n");
555         }
556 #endif
557
558         type->next = trace_types;
559         trace_types = type;
560         len = strlen(type->name);
561         if (len > max_tracer_type_len)
562                 max_tracer_type_len = len;
563
564  out:
565         tracing_selftest_running = false;
566         mutex_unlock(&trace_types_lock);
567
568         if (ret || !default_bootup_tracer)
569                 goto out_unlock;
570
571         if (strncmp(default_bootup_tracer, type->name, BOOTUP_TRACER_SIZE))
572                 goto out_unlock;
573
574         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
575         /* Do we want this tracer to start on bootup? */
576         tracing_set_tracer(type->name);
577         default_bootup_tracer = NULL;
578         /* disable other selftests, since this will break it. */
579         tracing_selftest_disabled = 1;
580 #ifdef CONFIG_FTRACE_STARTUP_TEST
581         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
582                type->name);
583 #endif
584
585  out_unlock:
586         lock_kernel();
587         return ret;
588 }
589
590 void unregister_tracer(struct tracer *type)
591 {
592         struct tracer **t;
593         int len;
594
595         mutex_lock(&trace_types_lock);
596         for (t = &trace_types; *t; t = &(*t)->next) {
597                 if (*t == type)
598                         goto found;
599         }
600         pr_info("Trace %s not registered\n", type->name);
601         goto out;
602
603  found:
604         *t = (*t)->next;
605
606         if (type == current_trace && tracer_enabled) {
607                 tracer_enabled = 0;
608                 tracing_stop();
609                 if (current_trace->stop)
610                         current_trace->stop(&global_trace);
611                 current_trace = &nop_trace;
612         }
613
614         if (strlen(type->name) != max_tracer_type_len)
615                 goto out;
616
617         max_tracer_type_len = 0;
618         for (t = &trace_types; *t; t = &(*t)->next) {
619                 len = strlen((*t)->name);
620                 if (len > max_tracer_type_len)
621                         max_tracer_type_len = len;
622         }
623  out:
624         mutex_unlock(&trace_types_lock);
625 }
626
627 void tracing_reset(struct trace_array *tr, int cpu)
628 {
629         ftrace_disable_cpu();
630         ring_buffer_reset_cpu(tr->buffer, cpu);
631         ftrace_enable_cpu();
632 }
633
634 void tracing_reset_online_cpus(struct trace_array *tr)
635 {
636         int cpu;
637
638         tr->time_start = ftrace_now(tr->cpu);
639
640         for_each_online_cpu(cpu)
641                 tracing_reset(tr, cpu);
642 }
643
644 #define SAVED_CMDLINES 128
645 #define NO_CMDLINE_MAP UINT_MAX
646 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
647 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
648 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
649 static int cmdline_idx;
650 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
651
652 /* temporary disable recording */
653 static atomic_t trace_record_cmdline_disabled __read_mostly;
654
655 static void trace_init_cmdlines(void)
656 {
657         memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
658         memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
659         cmdline_idx = 0;
660 }
661
662 static int trace_stop_count;
663 static DEFINE_SPINLOCK(tracing_start_lock);
664
665 /**
666  * ftrace_off_permanent - disable all ftrace code permanently
667  *
668  * This should only be called when a serious anomally has
669  * been detected.  This will turn off the function tracing,
670  * ring buffers, and other tracing utilites. It takes no
671  * locks and can be called from any context.
672  */
673 void ftrace_off_permanent(void)
674 {
675         tracing_disabled = 1;
676         ftrace_stop();
677         tracing_off_permanent();
678 }
679
680 /**
681  * tracing_start - quick start of the tracer
682  *
683  * If tracing is enabled but was stopped by tracing_stop,
684  * this will start the tracer back up.
685  */
686 void tracing_start(void)
687 {
688         struct ring_buffer *buffer;
689         unsigned long flags;
690
691         if (tracing_disabled)
692                 return;
693
694         spin_lock_irqsave(&tracing_start_lock, flags);
695         if (--trace_stop_count) {
696                 if (trace_stop_count < 0) {
697                         /* Someone screwed up their debugging */
698                         WARN_ON_ONCE(1);
699                         trace_stop_count = 0;
700                 }
701                 goto out;
702         }
703
704
705         buffer = global_trace.buffer;
706         if (buffer)
707                 ring_buffer_record_enable(buffer);
708
709         buffer = max_tr.buffer;
710         if (buffer)
711                 ring_buffer_record_enable(buffer);
712
713         ftrace_start();
714  out:
715         spin_unlock_irqrestore(&tracing_start_lock, flags);
716 }
717
718 /**
719  * tracing_stop - quick stop of the tracer
720  *
721  * Light weight way to stop tracing. Use in conjunction with
722  * tracing_start.
723  */
724 void tracing_stop(void)
725 {
726         struct ring_buffer *buffer;
727         unsigned long flags;
728
729         ftrace_stop();
730         spin_lock_irqsave(&tracing_start_lock, flags);
731         if (trace_stop_count++)
732                 goto out;
733
734         buffer = global_trace.buffer;
735         if (buffer)
736                 ring_buffer_record_disable(buffer);
737
738         buffer = max_tr.buffer;
739         if (buffer)
740                 ring_buffer_record_disable(buffer);
741
742  out:
743         spin_unlock_irqrestore(&tracing_start_lock, flags);
744 }
745
746 void trace_stop_cmdline_recording(void);
747
748 static void trace_save_cmdline(struct task_struct *tsk)
749 {
750         unsigned pid, idx;
751
752         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
753                 return;
754
755         /*
756          * It's not the end of the world if we don't get
757          * the lock, but we also don't want to spin
758          * nor do we want to disable interrupts,
759          * so if we miss here, then better luck next time.
760          */
761         if (!__raw_spin_trylock(&trace_cmdline_lock))
762                 return;
763
764         idx = map_pid_to_cmdline[tsk->pid];
765         if (idx == NO_CMDLINE_MAP) {
766                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
767
768                 /*
769                  * Check whether the cmdline buffer at idx has a pid
770                  * mapped. We are going to overwrite that entry so we
771                  * need to clear the map_pid_to_cmdline. Otherwise we
772                  * would read the new comm for the old pid.
773                  */
774                 pid = map_cmdline_to_pid[idx];
775                 if (pid != NO_CMDLINE_MAP)
776                         map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
777
778                 map_cmdline_to_pid[idx] = tsk->pid;
779                 map_pid_to_cmdline[tsk->pid] = idx;
780
781                 cmdline_idx = idx;
782         }
783
784         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
785
786         __raw_spin_unlock(&trace_cmdline_lock);
787 }
788
789 void trace_find_cmdline(int pid, char comm[])
790 {
791         unsigned map;
792
793         if (!pid) {
794                 strcpy(comm, "<idle>");
795                 return;
796         }
797
798         if (pid > PID_MAX_DEFAULT) {
799                 strcpy(comm, "<...>");
800                 return;
801         }
802
803         __raw_spin_lock(&trace_cmdline_lock);
804         map = map_pid_to_cmdline[pid];
805         if (map != NO_CMDLINE_MAP)
806                 strcpy(comm, saved_cmdlines[map]);
807         else
808                 strcpy(comm, "<...>");
809
810         __raw_spin_unlock(&trace_cmdline_lock);
811 }
812
813 void tracing_record_cmdline(struct task_struct *tsk)
814 {
815         if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
816             !tracing_is_on())
817                 return;
818
819         trace_save_cmdline(tsk);
820 }
821
822 void
823 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
824                              int pc)
825 {
826         struct task_struct *tsk = current;
827
828         entry->preempt_count            = pc & 0xff;
829         entry->pid                      = (tsk) ? tsk->pid : 0;
830         entry->tgid                     = (tsk) ? tsk->tgid : 0;
831         entry->flags =
832 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
833                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
834 #else
835                 TRACE_FLAG_IRQS_NOSUPPORT |
836 #endif
837                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
838                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
839                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
840 }
841
842 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
843                                                     unsigned char type,
844                                                     unsigned long len,
845                                                     unsigned long flags, int pc)
846 {
847         struct ring_buffer_event *event;
848
849         event = ring_buffer_lock_reserve(tr->buffer, len);
850         if (event != NULL) {
851                 struct trace_entry *ent = ring_buffer_event_data(event);
852
853                 tracing_generic_entry_update(ent, flags, pc);
854                 ent->type = type;
855         }
856
857         return event;
858 }
859 static void ftrace_trace_stack(struct trace_array *tr,
860                                unsigned long flags, int skip, int pc);
861 static void ftrace_trace_userstack(struct trace_array *tr,
862                                    unsigned long flags, int pc);
863
864 static inline void __trace_buffer_unlock_commit(struct trace_array *tr,
865                                         struct ring_buffer_event *event,
866                                         unsigned long flags, int pc,
867                                         int wake)
868 {
869         ring_buffer_unlock_commit(tr->buffer, event);
870
871         ftrace_trace_stack(tr, flags, 6, pc);
872         ftrace_trace_userstack(tr, flags, pc);
873
874         if (wake)
875                 trace_wake_up();
876 }
877
878 void trace_buffer_unlock_commit(struct trace_array *tr,
879                                         struct ring_buffer_event *event,
880                                         unsigned long flags, int pc)
881 {
882         __trace_buffer_unlock_commit(tr, event, flags, pc, 1);
883 }
884
885 struct ring_buffer_event *
886 trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
887                                   unsigned long flags, int pc)
888 {
889         return trace_buffer_lock_reserve(&global_trace,
890                                          type, len, flags, pc);
891 }
892
893 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
894                                         unsigned long flags, int pc)
895 {
896         return __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 1);
897 }
898
899 void trace_nowake_buffer_unlock_commit(struct ring_buffer_event *event,
900                                         unsigned long flags, int pc)
901 {
902         return __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 0);
903 }
904
905 void
906 trace_function(struct trace_array *tr,
907                unsigned long ip, unsigned long parent_ip, unsigned long flags,
908                int pc)
909 {
910         struct ring_buffer_event *event;
911         struct ftrace_entry *entry;
912
913         /* If we are reading the ring buffer, don't trace */
914         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
915                 return;
916
917         event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
918                                           flags, pc);
919         if (!event)
920                 return;
921         entry   = ring_buffer_event_data(event);
922         entry->ip                       = ip;
923         entry->parent_ip                = parent_ip;
924         ring_buffer_unlock_commit(tr->buffer, event);
925 }
926
927 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
928 static int __trace_graph_entry(struct trace_array *tr,
929                                 struct ftrace_graph_ent *trace,
930                                 unsigned long flags,
931                                 int pc)
932 {
933         struct ring_buffer_event *event;
934         struct ftrace_graph_ent_entry *entry;
935
936         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
937                 return 0;
938
939         event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
940                                           sizeof(*entry), flags, pc);
941         if (!event)
942                 return 0;
943         entry   = ring_buffer_event_data(event);
944         entry->graph_ent                        = *trace;
945         ring_buffer_unlock_commit(global_trace.buffer, event);
946
947         return 1;
948 }
949
950 static void __trace_graph_return(struct trace_array *tr,
951                                 struct ftrace_graph_ret *trace,
952                                 unsigned long flags,
953                                 int pc)
954 {
955         struct ring_buffer_event *event;
956         struct ftrace_graph_ret_entry *entry;
957
958         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
959                 return;
960
961         event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
962                                           sizeof(*entry), flags, pc);
963         if (!event)
964                 return;
965         entry   = ring_buffer_event_data(event);
966         entry->ret                              = *trace;
967         ring_buffer_unlock_commit(global_trace.buffer, event);
968 }
969 #endif
970
971 void
972 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
973        unsigned long ip, unsigned long parent_ip, unsigned long flags,
974        int pc)
975 {
976         if (likely(!atomic_read(&data->disabled)))
977                 trace_function(tr, ip, parent_ip, flags, pc);
978 }
979
980 static void __ftrace_trace_stack(struct trace_array *tr,
981                                  unsigned long flags,
982                                  int skip, int pc)
983 {
984 #ifdef CONFIG_STACKTRACE
985         struct ring_buffer_event *event;
986         struct stack_entry *entry;
987         struct stack_trace trace;
988
989         event = trace_buffer_lock_reserve(tr, TRACE_STACK,
990                                           sizeof(*entry), flags, pc);
991         if (!event)
992                 return;
993         entry   = ring_buffer_event_data(event);
994         memset(&entry->caller, 0, sizeof(entry->caller));
995
996         trace.nr_entries        = 0;
997         trace.max_entries       = FTRACE_STACK_ENTRIES;
998         trace.skip              = skip;
999         trace.entries           = entry->caller;
1000
1001         save_stack_trace(&trace);
1002         ring_buffer_unlock_commit(tr->buffer, event);
1003 #endif
1004 }
1005
1006 static void ftrace_trace_stack(struct trace_array *tr,
1007                                unsigned long flags,
1008                                int skip, int pc)
1009 {
1010         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1011                 return;
1012
1013         __ftrace_trace_stack(tr, flags, skip, pc);
1014 }
1015
1016 void __trace_stack(struct trace_array *tr,
1017                    unsigned long flags,
1018                    int skip, int pc)
1019 {
1020         __ftrace_trace_stack(tr, flags, skip, pc);
1021 }
1022
1023 static void ftrace_trace_userstack(struct trace_array *tr,
1024                                    unsigned long flags, int pc)
1025 {
1026 #ifdef CONFIG_STACKTRACE
1027         struct ring_buffer_event *event;
1028         struct userstack_entry *entry;
1029         struct stack_trace trace;
1030
1031         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1032                 return;
1033
1034         event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
1035                                           sizeof(*entry), flags, pc);
1036         if (!event)
1037                 return;
1038         entry   = ring_buffer_event_data(event);
1039
1040         memset(&entry->caller, 0, sizeof(entry->caller));
1041
1042         trace.nr_entries        = 0;
1043         trace.max_entries       = FTRACE_STACK_ENTRIES;
1044         trace.skip              = 0;
1045         trace.entries           = entry->caller;
1046
1047         save_stack_trace_user(&trace);
1048         ring_buffer_unlock_commit(tr->buffer, event);
1049 #endif
1050 }
1051
1052 #ifdef UNUSED
1053 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1054 {
1055         ftrace_trace_userstack(tr, flags, preempt_count());
1056 }
1057 #endif /* UNUSED */
1058
1059 static void
1060 ftrace_trace_special(void *__tr,
1061                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1062                      int pc)
1063 {
1064         struct ring_buffer_event *event;
1065         struct trace_array *tr = __tr;
1066         struct special_entry *entry;
1067
1068         event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1069                                           sizeof(*entry), 0, pc);
1070         if (!event)
1071                 return;
1072         entry   = ring_buffer_event_data(event);
1073         entry->arg1                     = arg1;
1074         entry->arg2                     = arg2;
1075         entry->arg3                     = arg3;
1076         trace_buffer_unlock_commit(tr, event, 0, pc);
1077 }
1078
1079 void
1080 __trace_special(void *__tr, void *__data,
1081                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1082 {
1083         ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1084 }
1085
1086 void
1087 tracing_sched_switch_trace(struct trace_array *tr,
1088                            struct task_struct *prev,
1089                            struct task_struct *next,
1090                            unsigned long flags, int pc)
1091 {
1092         struct ring_buffer_event *event;
1093         struct ctx_switch_entry *entry;
1094
1095         event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1096                                           sizeof(*entry), flags, pc);
1097         if (!event)
1098                 return;
1099         entry   = ring_buffer_event_data(event);
1100         entry->prev_pid                 = prev->pid;
1101         entry->prev_prio                = prev->prio;
1102         entry->prev_state               = prev->state;
1103         entry->next_pid                 = next->pid;
1104         entry->next_prio                = next->prio;
1105         entry->next_state               = next->state;
1106         entry->next_cpu = task_cpu(next);
1107         trace_buffer_unlock_commit(tr, event, flags, pc);
1108 }
1109
1110 void
1111 tracing_sched_wakeup_trace(struct trace_array *tr,
1112                            struct task_struct *wakee,
1113                            struct task_struct *curr,
1114                            unsigned long flags, int pc)
1115 {
1116         struct ring_buffer_event *event;
1117         struct ctx_switch_entry *entry;
1118
1119         event = trace_buffer_lock_reserve(tr, TRACE_WAKE,
1120                                           sizeof(*entry), flags, pc);
1121         if (!event)
1122                 return;
1123         entry   = ring_buffer_event_data(event);
1124         entry->prev_pid                 = curr->pid;
1125         entry->prev_prio                = curr->prio;
1126         entry->prev_state               = curr->state;
1127         entry->next_pid                 = wakee->pid;
1128         entry->next_prio                = wakee->prio;
1129         entry->next_state               = wakee->state;
1130         entry->next_cpu                 = task_cpu(wakee);
1131
1132         ring_buffer_unlock_commit(tr->buffer, event);
1133         ftrace_trace_stack(tr, flags, 6, pc);
1134         ftrace_trace_userstack(tr, flags, pc);
1135 }
1136
1137 void
1138 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1139 {
1140         struct trace_array *tr = &global_trace;
1141         struct trace_array_cpu *data;
1142         unsigned long flags;
1143         int cpu;
1144         int pc;
1145
1146         if (tracing_disabled)
1147                 return;
1148
1149         pc = preempt_count();
1150         local_irq_save(flags);
1151         cpu = raw_smp_processor_id();
1152         data = tr->data[cpu];
1153
1154         if (likely(atomic_inc_return(&data->disabled) == 1))
1155                 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1156
1157         atomic_dec(&data->disabled);
1158         local_irq_restore(flags);
1159 }
1160
1161 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1162 int trace_graph_entry(struct ftrace_graph_ent *trace)
1163 {
1164         struct trace_array *tr = &global_trace;
1165         struct trace_array_cpu *data;
1166         unsigned long flags;
1167         long disabled;
1168         int ret;
1169         int cpu;
1170         int pc;
1171
1172         if (!ftrace_trace_task(current))
1173                 return 0;
1174
1175         if (!ftrace_graph_addr(trace->func))
1176                 return 0;
1177
1178         local_irq_save(flags);
1179         cpu = raw_smp_processor_id();
1180         data = tr->data[cpu];
1181         disabled = atomic_inc_return(&data->disabled);
1182         if (likely(disabled == 1)) {
1183                 pc = preempt_count();
1184                 ret = __trace_graph_entry(tr, trace, flags, pc);
1185         } else {
1186                 ret = 0;
1187         }
1188         /* Only do the atomic if it is not already set */
1189         if (!test_tsk_trace_graph(current))
1190                 set_tsk_trace_graph(current);
1191
1192         atomic_dec(&data->disabled);
1193         local_irq_restore(flags);
1194
1195         return ret;
1196 }
1197
1198 void trace_graph_return(struct ftrace_graph_ret *trace)
1199 {
1200         struct trace_array *tr = &global_trace;
1201         struct trace_array_cpu *data;
1202         unsigned long flags;
1203         long disabled;
1204         int cpu;
1205         int pc;
1206
1207         local_irq_save(flags);
1208         cpu = raw_smp_processor_id();
1209         data = tr->data[cpu];
1210         disabled = atomic_inc_return(&data->disabled);
1211         if (likely(disabled == 1)) {
1212                 pc = preempt_count();
1213                 __trace_graph_return(tr, trace, flags, pc);
1214         }
1215         if (!trace->depth)
1216                 clear_tsk_trace_graph(current);
1217         atomic_dec(&data->disabled);
1218         local_irq_restore(flags);
1219 }
1220 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1221
1222
1223 /**
1224  * trace_vbprintk - write binary msg to tracing buffer
1225  *
1226  */
1227 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1228 {
1229         static raw_spinlock_t trace_buf_lock =
1230                 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1231         static u32 trace_buf[TRACE_BUF_SIZE];
1232
1233         struct ring_buffer_event *event;
1234         struct trace_array *tr = &global_trace;
1235         struct trace_array_cpu *data;
1236         struct bprint_entry *entry;
1237         unsigned long flags;
1238         int resched;
1239         int cpu, len = 0, size, pc;
1240
1241         if (unlikely(tracing_selftest_running || tracing_disabled))
1242                 return 0;
1243
1244         /* Don't pollute graph traces with trace_vprintk internals */
1245         pause_graph_tracing();
1246
1247         pc = preempt_count();
1248         resched = ftrace_preempt_disable();
1249         cpu = raw_smp_processor_id();
1250         data = tr->data[cpu];
1251
1252         if (unlikely(atomic_read(&data->disabled)))
1253                 goto out;
1254
1255         /* Lockdep uses trace_printk for lock tracing */
1256         local_irq_save(flags);
1257         __raw_spin_lock(&trace_buf_lock);
1258         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1259
1260         if (len > TRACE_BUF_SIZE || len < 0)
1261                 goto out_unlock;
1262
1263         size = sizeof(*entry) + sizeof(u32) * len;
1264         event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc);
1265         if (!event)
1266                 goto out_unlock;
1267         entry = ring_buffer_event_data(event);
1268         entry->ip                       = ip;
1269         entry->fmt                      = fmt;
1270
1271         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1272         ring_buffer_unlock_commit(tr->buffer, event);
1273
1274 out_unlock:
1275         __raw_spin_unlock(&trace_buf_lock);
1276         local_irq_restore(flags);
1277
1278 out:
1279         ftrace_preempt_enable(resched);
1280         unpause_graph_tracing();
1281
1282         return len;
1283 }
1284 EXPORT_SYMBOL_GPL(trace_vbprintk);
1285
1286 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1287 {
1288         static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1289         static char trace_buf[TRACE_BUF_SIZE];
1290
1291         struct ring_buffer_event *event;
1292         struct trace_array *tr = &global_trace;
1293         struct trace_array_cpu *data;
1294         int cpu, len = 0, size, pc;
1295         struct print_entry *entry;
1296         unsigned long irq_flags;
1297
1298         if (tracing_disabled || tracing_selftest_running)
1299                 return 0;
1300
1301         pc = preempt_count();
1302         preempt_disable_notrace();
1303         cpu = raw_smp_processor_id();
1304         data = tr->data[cpu];
1305
1306         if (unlikely(atomic_read(&data->disabled)))
1307                 goto out;
1308
1309         pause_graph_tracing();
1310         raw_local_irq_save(irq_flags);
1311         __raw_spin_lock(&trace_buf_lock);
1312         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1313
1314         len = min(len, TRACE_BUF_SIZE-1);
1315         trace_buf[len] = 0;
1316
1317         size = sizeof(*entry) + len + 1;
1318         event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
1319         if (!event)
1320                 goto out_unlock;
1321         entry = ring_buffer_event_data(event);
1322         entry->ip                       = ip;
1323
1324         memcpy(&entry->buf, trace_buf, len);
1325         entry->buf[len] = 0;
1326         ring_buffer_unlock_commit(tr->buffer, event);
1327
1328  out_unlock:
1329         __raw_spin_unlock(&trace_buf_lock);
1330         raw_local_irq_restore(irq_flags);
1331         unpause_graph_tracing();
1332  out:
1333         preempt_enable_notrace();
1334
1335         return len;
1336 }
1337 EXPORT_SYMBOL_GPL(trace_vprintk);
1338
1339 enum trace_file_type {
1340         TRACE_FILE_LAT_FMT      = 1,
1341         TRACE_FILE_ANNOTATE     = 2,
1342 };
1343
1344 static void trace_iterator_increment(struct trace_iterator *iter)
1345 {
1346         /* Don't allow ftrace to trace into the ring buffers */
1347         ftrace_disable_cpu();
1348
1349         iter->idx++;
1350         if (iter->buffer_iter[iter->cpu])
1351                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1352
1353         ftrace_enable_cpu();
1354 }
1355
1356 static struct trace_entry *
1357 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1358 {
1359         struct ring_buffer_event *event;
1360         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1361
1362         /* Don't allow ftrace to trace into the ring buffers */
1363         ftrace_disable_cpu();
1364
1365         if (buf_iter)
1366                 event = ring_buffer_iter_peek(buf_iter, ts);
1367         else
1368                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1369
1370         ftrace_enable_cpu();
1371
1372         return event ? ring_buffer_event_data(event) : NULL;
1373 }
1374
1375 static struct trace_entry *
1376 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1377 {
1378         struct ring_buffer *buffer = iter->tr->buffer;
1379         struct trace_entry *ent, *next = NULL;
1380         int cpu_file = iter->cpu_file;
1381         u64 next_ts = 0, ts;
1382         int next_cpu = -1;
1383         int cpu;
1384
1385         /*
1386          * If we are in a per_cpu trace file, don't bother by iterating over
1387          * all cpu and peek directly.
1388          */
1389         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1390                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1391                         return NULL;
1392                 ent = peek_next_entry(iter, cpu_file, ent_ts);
1393                 if (ent_cpu)
1394                         *ent_cpu = cpu_file;
1395
1396                 return ent;
1397         }
1398
1399         for_each_tracing_cpu(cpu) {
1400
1401                 if (ring_buffer_empty_cpu(buffer, cpu))
1402                         continue;
1403
1404                 ent = peek_next_entry(iter, cpu, &ts);
1405
1406                 /*
1407                  * Pick the entry with the smallest timestamp:
1408                  */
1409                 if (ent && (!next || ts < next_ts)) {
1410                         next = ent;
1411                         next_cpu = cpu;
1412                         next_ts = ts;
1413                 }
1414         }
1415
1416         if (ent_cpu)
1417                 *ent_cpu = next_cpu;
1418
1419         if (ent_ts)
1420                 *ent_ts = next_ts;
1421
1422         return next;
1423 }
1424
1425 /* Find the next real entry, without updating the iterator itself */
1426 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1427                                           int *ent_cpu, u64 *ent_ts)
1428 {
1429         return __find_next_entry(iter, ent_cpu, ent_ts);
1430 }
1431
1432 /* Find the next real entry, and increment the iterator to the next entry */
1433 static void *find_next_entry_inc(struct trace_iterator *iter)
1434 {
1435         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1436
1437         if (iter->ent)
1438                 trace_iterator_increment(iter);
1439
1440         return iter->ent ? iter : NULL;
1441 }
1442
1443 static void trace_consume(struct trace_iterator *iter)
1444 {
1445         /* Don't allow ftrace to trace into the ring buffers */
1446         ftrace_disable_cpu();
1447         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1448         ftrace_enable_cpu();
1449 }
1450
1451 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1452 {
1453         struct trace_iterator *iter = m->private;
1454         int i = (int)*pos;
1455         void *ent;
1456
1457         (*pos)++;
1458
1459         /* can't go backwards */
1460         if (iter->idx > i)
1461                 return NULL;
1462
1463         if (iter->idx < 0)
1464                 ent = find_next_entry_inc(iter);
1465         else
1466                 ent = iter;
1467
1468         while (ent && iter->idx < i)
1469                 ent = find_next_entry_inc(iter);
1470
1471         iter->pos = *pos;
1472
1473         return ent;
1474 }
1475
1476 /*
1477  * No necessary locking here. The worst thing which can
1478  * happen is loosing events consumed at the same time
1479  * by a trace_pipe reader.
1480  * Other than that, we don't risk to crash the ring buffer
1481  * because it serializes the readers.
1482  *
1483  * The current tracer is copied to avoid a global locking
1484  * all around.
1485  */
1486 static void *s_start(struct seq_file *m, loff_t *pos)
1487 {
1488         struct trace_iterator *iter = m->private;
1489         static struct tracer *old_tracer;
1490         int cpu_file = iter->cpu_file;
1491         void *p = NULL;
1492         loff_t l = 0;
1493         int cpu;
1494
1495         /* copy the tracer to avoid using a global lock all around */
1496         mutex_lock(&trace_types_lock);
1497         if (unlikely(old_tracer != current_trace && current_trace)) {
1498                 old_tracer = current_trace;
1499                 *iter->trace = *current_trace;
1500         }
1501         mutex_unlock(&trace_types_lock);
1502
1503         atomic_inc(&trace_record_cmdline_disabled);
1504
1505         if (*pos != iter->pos) {
1506                 iter->ent = NULL;
1507                 iter->cpu = 0;
1508                 iter->idx = -1;
1509
1510                 ftrace_disable_cpu();
1511
1512                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1513                         for_each_tracing_cpu(cpu)
1514                                 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1515                 } else
1516                         ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1517
1518
1519                 ftrace_enable_cpu();
1520
1521                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1522                         ;
1523
1524         } else {
1525                 l = *pos - 1;
1526                 p = s_next(m, p, &l);
1527         }
1528
1529         return p;
1530 }
1531
1532 static void s_stop(struct seq_file *m, void *p)
1533 {
1534         atomic_dec(&trace_record_cmdline_disabled);
1535 }
1536
1537 static void print_lat_help_header(struct seq_file *m)
1538 {
1539         seq_puts(m, "#                  _------=> CPU#            \n");
1540         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1541         seq_puts(m, "#                | / _----=> need-resched    \n");
1542         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1543         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1544         seq_puts(m, "#                |||| /                      \n");
1545         seq_puts(m, "#                |||||     delay             \n");
1546         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1547         seq_puts(m, "#     \\   /      |||||   \\   |   /           \n");
1548 }
1549
1550 static void print_func_help_header(struct seq_file *m)
1551 {
1552         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1553         seq_puts(m, "#              | |       |          |         |\n");
1554 }
1555
1556
1557 static void
1558 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1559 {
1560         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1561         struct trace_array *tr = iter->tr;
1562         struct trace_array_cpu *data = tr->data[tr->cpu];
1563         struct tracer *type = current_trace;
1564         unsigned long total;
1565         unsigned long entries;
1566         const char *name = "preemption";
1567
1568         if (type)
1569                 name = type->name;
1570
1571         entries = ring_buffer_entries(iter->tr->buffer);
1572         total = entries +
1573                 ring_buffer_overruns(iter->tr->buffer);
1574
1575         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1576                    name, UTS_RELEASE);
1577         seq_puts(m, "# -----------------------------------"
1578                  "---------------------------------\n");
1579         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1580                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1581                    nsecs_to_usecs(data->saved_latency),
1582                    entries,
1583                    total,
1584                    tr->cpu,
1585 #if defined(CONFIG_PREEMPT_NONE)
1586                    "server",
1587 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1588                    "desktop",
1589 #elif defined(CONFIG_PREEMPT)
1590                    "preempt",
1591 #else
1592                    "unknown",
1593 #endif
1594                    /* These are reserved for later use */
1595                    0, 0, 0, 0);
1596 #ifdef CONFIG_SMP
1597         seq_printf(m, " #P:%d)\n", num_online_cpus());
1598 #else
1599         seq_puts(m, ")\n");
1600 #endif
1601         seq_puts(m, "#    -----------------\n");
1602         seq_printf(m, "#    | task: %.16s-%d "
1603                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1604                    data->comm, data->pid, data->uid, data->nice,
1605                    data->policy, data->rt_priority);
1606         seq_puts(m, "#    -----------------\n");
1607
1608         if (data->critical_start) {
1609                 seq_puts(m, "#  => started at: ");
1610                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1611                 trace_print_seq(m, &iter->seq);
1612                 seq_puts(m, "\n#  => ended at:   ");
1613                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1614                 trace_print_seq(m, &iter->seq);
1615                 seq_puts(m, "#\n");
1616         }
1617
1618         seq_puts(m, "#\n");
1619 }
1620
1621 static void test_cpu_buff_start(struct trace_iterator *iter)
1622 {
1623         struct trace_seq *s = &iter->seq;
1624
1625         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1626                 return;
1627
1628         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1629                 return;
1630
1631         if (cpumask_test_cpu(iter->cpu, iter->started))
1632                 return;
1633
1634         cpumask_set_cpu(iter->cpu, iter->started);
1635         trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1636 }
1637
1638 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1639 {
1640         struct trace_seq *s = &iter->seq;
1641         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1642         struct trace_entry *entry;
1643         struct trace_event *event;
1644
1645         entry = iter->ent;
1646
1647         test_cpu_buff_start(iter);
1648
1649         event = ftrace_find_event(entry->type);
1650
1651         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1652                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1653                         if (!trace_print_lat_context(iter))
1654                                 goto partial;
1655                 } else {
1656                         if (!trace_print_context(iter))
1657                                 goto partial;
1658                 }
1659         }
1660
1661         if (event)
1662                 return event->trace(iter, sym_flags);
1663
1664         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1665                 goto partial;
1666
1667         return TRACE_TYPE_HANDLED;
1668 partial:
1669         return TRACE_TYPE_PARTIAL_LINE;
1670 }
1671
1672 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1673 {
1674         struct trace_seq *s = &iter->seq;
1675         struct trace_entry *entry;
1676         struct trace_event *event;
1677
1678         entry = iter->ent;
1679
1680         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1681                 if (!trace_seq_printf(s, "%d %d %llu ",
1682                                       entry->pid, iter->cpu, iter->ts))
1683                         goto partial;
1684         }
1685
1686         event = ftrace_find_event(entry->type);
1687         if (event)
1688                 return event->raw(iter, 0);
1689
1690         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1691                 goto partial;
1692
1693         return TRACE_TYPE_HANDLED;
1694 partial:
1695         return TRACE_TYPE_PARTIAL_LINE;
1696 }
1697
1698 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1699 {
1700         struct trace_seq *s = &iter->seq;
1701         unsigned char newline = '\n';
1702         struct trace_entry *entry;
1703         struct trace_event *event;
1704
1705         entry = iter->ent;
1706
1707         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1708                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1709                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1710                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1711         }
1712
1713         event = ftrace_find_event(entry->type);
1714         if (event) {
1715                 enum print_line_t ret = event->hex(iter, 0);
1716                 if (ret != TRACE_TYPE_HANDLED)
1717                         return ret;
1718         }
1719
1720         SEQ_PUT_FIELD_RET(s, newline);
1721
1722         return TRACE_TYPE_HANDLED;
1723 }
1724
1725 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1726 {
1727         struct trace_seq *s = &iter->seq;
1728         struct trace_entry *entry;
1729         struct trace_event *event;
1730
1731         entry = iter->ent;
1732
1733         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1734                 SEQ_PUT_FIELD_RET(s, entry->pid);
1735                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1736                 SEQ_PUT_FIELD_RET(s, iter->ts);
1737         }
1738
1739         event = ftrace_find_event(entry->type);
1740         return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1741 }
1742
1743 static int trace_empty(struct trace_iterator *iter)
1744 {
1745         int cpu;
1746
1747         /* If we are looking at one CPU buffer, only check that one */
1748         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1749                 cpu = iter->cpu_file;
1750                 if (iter->buffer_iter[cpu]) {
1751                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1752                                 return 0;
1753                 } else {
1754                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1755                                 return 0;
1756                 }
1757                 return 1;
1758         }
1759
1760         for_each_tracing_cpu(cpu) {
1761                 if (iter->buffer_iter[cpu]) {
1762                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1763                                 return 0;
1764                 } else {
1765                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1766                                 return 0;
1767                 }
1768         }
1769
1770         return 1;
1771 }
1772
1773 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1774 {
1775         enum print_line_t ret;
1776
1777         if (iter->trace && iter->trace->print_line) {
1778                 ret = iter->trace->print_line(iter);
1779                 if (ret != TRACE_TYPE_UNHANDLED)
1780                         return ret;
1781         }
1782
1783         if (iter->ent->type == TRACE_BPRINT &&
1784                         trace_flags & TRACE_ITER_PRINTK &&
1785                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1786                 return trace_print_bprintk_msg_only(iter);
1787
1788         if (iter->ent->type == TRACE_PRINT &&
1789                         trace_flags & TRACE_ITER_PRINTK &&
1790                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1791                 return trace_print_printk_msg_only(iter);
1792
1793         if (trace_flags & TRACE_ITER_BIN)
1794                 return print_bin_fmt(iter);
1795
1796         if (trace_flags & TRACE_ITER_HEX)
1797                 return print_hex_fmt(iter);
1798
1799         if (trace_flags & TRACE_ITER_RAW)
1800                 return print_raw_fmt(iter);
1801
1802         return print_trace_fmt(iter);
1803 }
1804
1805 static int s_show(struct seq_file *m, void *v)
1806 {
1807         struct trace_iterator *iter = v;
1808
1809         if (iter->ent == NULL) {
1810                 if (iter->tr) {
1811                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1812                         seq_puts(m, "#\n");
1813                 }
1814                 if (iter->trace && iter->trace->print_header)
1815                         iter->trace->print_header(m);
1816                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1817                         /* print nothing if the buffers are empty */
1818                         if (trace_empty(iter))
1819                                 return 0;
1820                         print_trace_header(m, iter);
1821                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1822                                 print_lat_help_header(m);
1823                 } else {
1824                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1825                                 print_func_help_header(m);
1826                 }
1827         } else {
1828                 print_trace_line(iter);
1829                 trace_print_seq(m, &iter->seq);
1830         }
1831
1832         return 0;
1833 }
1834
1835 static struct seq_operations tracer_seq_ops = {
1836         .start          = s_start,
1837         .next           = s_next,
1838         .stop           = s_stop,
1839         .show           = s_show,
1840 };
1841
1842 static struct trace_iterator *
1843 __tracing_open(struct inode *inode, struct file *file)
1844 {
1845         long cpu_file = (long) inode->i_private;
1846         void *fail_ret = ERR_PTR(-ENOMEM);
1847         struct trace_iterator *iter;
1848         struct seq_file *m;
1849         int cpu, ret;
1850
1851         if (tracing_disabled)
1852                 return ERR_PTR(-ENODEV);
1853
1854         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1855         if (!iter)
1856                 return ERR_PTR(-ENOMEM);
1857
1858         /*
1859          * We make a copy of the current tracer to avoid concurrent
1860          * changes on it while we are reading.
1861          */
1862         mutex_lock(&trace_types_lock);
1863         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1864         if (!iter->trace)
1865                 goto fail;
1866
1867         if (current_trace)
1868                 *iter->trace = *current_trace;
1869
1870         if (current_trace && current_trace->print_max)
1871                 iter->tr = &max_tr;
1872         else
1873                 iter->tr = &global_trace;
1874         iter->pos = -1;
1875         mutex_init(&iter->mutex);
1876         iter->cpu_file = cpu_file;
1877
1878         /* Notify the tracer early; before we stop tracing. */
1879         if (iter->trace && iter->trace->open)
1880                 iter->trace->open(iter);
1881
1882         /* Annotate start of buffers if we had overruns */
1883         if (ring_buffer_overruns(iter->tr->buffer))
1884                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1885
1886         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1887                 for_each_tracing_cpu(cpu) {
1888
1889                         iter->buffer_iter[cpu] =
1890                                 ring_buffer_read_start(iter->tr->buffer, cpu);
1891                 }
1892         } else {
1893                 cpu = iter->cpu_file;
1894                 iter->buffer_iter[cpu] =
1895                                 ring_buffer_read_start(iter->tr->buffer, cpu);
1896         }
1897
1898         /* TODO stop tracer */
1899         ret = seq_open(file, &tracer_seq_ops);
1900         if (ret < 0) {
1901                 fail_ret = ERR_PTR(ret);
1902                 goto fail_buffer;
1903         }
1904
1905         m = file->private_data;
1906         m->private = iter;
1907
1908         /* stop the trace while dumping */
1909         tracing_stop();
1910
1911         mutex_unlock(&trace_types_lock);
1912
1913         return iter;
1914
1915  fail_buffer:
1916         for_each_tracing_cpu(cpu) {
1917                 if (iter->buffer_iter[cpu])
1918                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
1919         }
1920  fail:
1921         mutex_unlock(&trace_types_lock);
1922         kfree(iter->trace);
1923         kfree(iter);
1924
1925         return fail_ret;
1926 }
1927
1928 int tracing_open_generic(struct inode *inode, struct file *filp)
1929 {
1930         if (tracing_disabled)
1931                 return -ENODEV;
1932
1933         filp->private_data = inode->i_private;
1934         return 0;
1935 }
1936
1937 static int tracing_release(struct inode *inode, struct file *file)
1938 {
1939         struct seq_file *m = (struct seq_file *)file->private_data;
1940         struct trace_iterator *iter;
1941         int cpu;
1942
1943         if (!(file->f_mode & FMODE_READ))
1944                 return 0;
1945
1946         iter = m->private;
1947
1948         mutex_lock(&trace_types_lock);
1949         for_each_tracing_cpu(cpu) {
1950                 if (iter->buffer_iter[cpu])
1951                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
1952         }
1953
1954         if (iter->trace && iter->trace->close)
1955                 iter->trace->close(iter);
1956
1957         /* reenable tracing if it was previously enabled */
1958         tracing_start();
1959         mutex_unlock(&trace_types_lock);
1960
1961         seq_release(inode, file);
1962         mutex_destroy(&iter->mutex);
1963         kfree(iter->trace);
1964         kfree(iter);
1965         return 0;
1966 }
1967
1968 static int tracing_open(struct inode *inode, struct file *file)
1969 {
1970         struct trace_iterator *iter;
1971         int ret = 0;
1972
1973         /* If this file was open for write, then erase contents */
1974         if ((file->f_mode & FMODE_WRITE) &&
1975             !(file->f_flags & O_APPEND)) {
1976                 long cpu = (long) inode->i_private;
1977
1978                 if (cpu == TRACE_PIPE_ALL_CPU)
1979                         tracing_reset_online_cpus(&global_trace);
1980                 else
1981                         tracing_reset(&global_trace, cpu);
1982         }
1983
1984         if (file->f_mode & FMODE_READ) {
1985                 iter = __tracing_open(inode, file);
1986                 if (IS_ERR(iter))
1987                         ret = PTR_ERR(iter);
1988                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
1989                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
1990         }
1991         return ret;
1992 }
1993
1994 static void *
1995 t_next(struct seq_file *m, void *v, loff_t *pos)
1996 {
1997         struct tracer *t = m->private;
1998
1999         (*pos)++;
2000
2001         if (t)
2002                 t = t->next;
2003
2004         m->private = t;
2005
2006         return t;
2007 }
2008
2009 static void *t_start(struct seq_file *m, loff_t *pos)
2010 {
2011         struct tracer *t = m->private;
2012         loff_t l = 0;
2013
2014         mutex_lock(&trace_types_lock);
2015         for (; t && l < *pos; t = t_next(m, t, &l))
2016                 ;
2017
2018         return t;
2019 }
2020
2021 static void t_stop(struct seq_file *m, void *p)
2022 {
2023         mutex_unlock(&trace_types_lock);
2024 }
2025
2026 static int t_show(struct seq_file *m, void *v)
2027 {
2028         struct tracer *t = v;
2029
2030         if (!t)
2031                 return 0;
2032
2033         seq_printf(m, "%s", t->name);
2034         if (t->next)
2035                 seq_putc(m, ' ');
2036         else
2037                 seq_putc(m, '\n');
2038
2039         return 0;
2040 }
2041
2042 static struct seq_operations show_traces_seq_ops = {
2043         .start          = t_start,
2044         .next           = t_next,
2045         .stop           = t_stop,
2046         .show           = t_show,
2047 };
2048
2049 static int show_traces_open(struct inode *inode, struct file *file)
2050 {
2051         int ret;
2052
2053         if (tracing_disabled)
2054                 return -ENODEV;
2055
2056         ret = seq_open(file, &show_traces_seq_ops);
2057         if (!ret) {
2058                 struct seq_file *m = file->private_data;
2059                 m->private = trace_types;
2060         }
2061
2062         return ret;
2063 }
2064
2065 static ssize_t
2066 tracing_write_stub(struct file *filp, const char __user *ubuf,
2067                    size_t count, loff_t *ppos)
2068 {
2069         return count;
2070 }
2071
2072 static const struct file_operations tracing_fops = {
2073         .open           = tracing_open,
2074         .read           = seq_read,
2075         .write          = tracing_write_stub,
2076         .llseek         = seq_lseek,
2077         .release        = tracing_release,
2078 };
2079
2080 static const struct file_operations show_traces_fops = {
2081         .open           = show_traces_open,
2082         .read           = seq_read,
2083         .release        = seq_release,
2084 };
2085
2086 /*
2087  * Only trace on a CPU if the bitmask is set:
2088  */
2089 static cpumask_var_t tracing_cpumask;
2090
2091 /*
2092  * The tracer itself will not take this lock, but still we want
2093  * to provide a consistent cpumask to user-space:
2094  */
2095 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2096
2097 /*
2098  * Temporary storage for the character representation of the
2099  * CPU bitmask (and one more byte for the newline):
2100  */
2101 static char mask_str[NR_CPUS + 1];
2102
2103 static ssize_t
2104 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2105                      size_t count, loff_t *ppos)
2106 {
2107         int len;
2108
2109         mutex_lock(&tracing_cpumask_update_lock);
2110
2111         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2112         if (count - len < 2) {
2113                 count = -EINVAL;
2114                 goto out_err;
2115         }
2116         len += sprintf(mask_str + len, "\n");
2117         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2118
2119 out_err:
2120         mutex_unlock(&tracing_cpumask_update_lock);
2121
2122         return count;
2123 }
2124
2125 static ssize_t
2126 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2127                       size_t count, loff_t *ppos)
2128 {
2129         int err, cpu;
2130         cpumask_var_t tracing_cpumask_new;
2131
2132         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2133                 return -ENOMEM;
2134
2135         mutex_lock(&tracing_cpumask_update_lock);
2136         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2137         if (err)
2138                 goto err_unlock;
2139
2140         local_irq_disable();
2141         __raw_spin_lock(&ftrace_max_lock);
2142         for_each_tracing_cpu(cpu) {
2143                 /*
2144                  * Increase/decrease the disabled counter if we are
2145                  * about to flip a bit in the cpumask:
2146                  */
2147                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2148                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2149                         atomic_inc(&global_trace.data[cpu]->disabled);
2150                 }
2151                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2152                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2153                         atomic_dec(&global_trace.data[cpu]->disabled);
2154                 }
2155         }
2156         __raw_spin_unlock(&ftrace_max_lock);
2157         local_irq_enable();
2158
2159         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2160
2161         mutex_unlock(&tracing_cpumask_update_lock);
2162         free_cpumask_var(tracing_cpumask_new);
2163
2164         return count;
2165
2166 err_unlock:
2167         mutex_unlock(&tracing_cpumask_update_lock);
2168         free_cpumask_var(tracing_cpumask);
2169
2170         return err;
2171 }
2172
2173 static const struct file_operations tracing_cpumask_fops = {
2174         .open           = tracing_open_generic,
2175         .read           = tracing_cpumask_read,
2176         .write          = tracing_cpumask_write,
2177 };
2178
2179 static ssize_t
2180 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2181                        size_t cnt, loff_t *ppos)
2182 {
2183         struct tracer_opt *trace_opts;
2184         u32 tracer_flags;
2185         int len = 0;
2186         char *buf;
2187         int r = 0;
2188         int i;
2189
2190
2191         /* calculate max size */
2192         for (i = 0; trace_options[i]; i++) {
2193                 len += strlen(trace_options[i]);
2194                 len += 3; /* "no" and newline */
2195         }
2196
2197         mutex_lock(&trace_types_lock);
2198         tracer_flags = current_trace->flags->val;
2199         trace_opts = current_trace->flags->opts;
2200
2201         /*
2202          * Increase the size with names of options specific
2203          * of the current tracer.
2204          */
2205         for (i = 0; trace_opts[i].name; i++) {
2206                 len += strlen(trace_opts[i].name);
2207                 len += 3; /* "no" and newline */
2208         }
2209
2210         /* +2 for \n and \0 */
2211         buf = kmalloc(len + 2, GFP_KERNEL);
2212         if (!buf) {
2213                 mutex_unlock(&trace_types_lock);
2214                 return -ENOMEM;
2215         }
2216
2217         for (i = 0; trace_options[i]; i++) {
2218                 if (trace_flags & (1 << i))
2219                         r += sprintf(buf + r, "%s\n", trace_options[i]);
2220                 else
2221                         r += sprintf(buf + r, "no%s\n", trace_options[i]);
2222         }
2223
2224         for (i = 0; trace_opts[i].name; i++) {
2225                 if (tracer_flags & trace_opts[i].bit)
2226                         r += sprintf(buf + r, "%s\n",
2227                                 trace_opts[i].name);
2228                 else
2229                         r += sprintf(buf + r, "no%s\n",
2230                                 trace_opts[i].name);
2231         }
2232         mutex_unlock(&trace_types_lock);
2233
2234         WARN_ON(r >= len + 2);
2235
2236         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2237
2238         kfree(buf);
2239         return r;
2240 }
2241
2242 /* Try to assign a tracer specific option */
2243 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2244 {
2245         struct tracer_flags *trace_flags = trace->flags;
2246         struct tracer_opt *opts = NULL;
2247         int ret = 0, i = 0;
2248         int len;
2249
2250         for (i = 0; trace_flags->opts[i].name; i++) {
2251                 opts = &trace_flags->opts[i];
2252                 len = strlen(opts->name);
2253
2254                 if (strncmp(cmp, opts->name, len) == 0) {
2255                         ret = trace->set_flag(trace_flags->val,
2256                                 opts->bit, !neg);
2257                         break;
2258                 }
2259         }
2260         /* Not found */
2261         if (!trace_flags->opts[i].name)
2262                 return -EINVAL;
2263
2264         /* Refused to handle */
2265         if (ret)
2266                 return ret;
2267
2268         if (neg)
2269                 trace_flags->val &= ~opts->bit;
2270         else
2271                 trace_flags->val |= opts->bit;
2272
2273         return 0;
2274 }
2275
2276 static void set_tracer_flags(unsigned int mask, int enabled)
2277 {
2278         /* do nothing if flag is already set */
2279         if (!!(trace_flags & mask) == !!enabled)
2280                 return;
2281
2282         if (enabled)
2283                 trace_flags |= mask;
2284         else
2285                 trace_flags &= ~mask;
2286
2287         if (mask == TRACE_ITER_GLOBAL_CLK) {
2288                 u64 (*func)(void);
2289
2290                 if (enabled)
2291                         func = trace_clock_global;
2292                 else
2293                         func = trace_clock_local;
2294
2295                 mutex_lock(&trace_types_lock);
2296                 ring_buffer_set_clock(global_trace.buffer, func);
2297
2298                 if (max_tr.buffer)
2299                         ring_buffer_set_clock(max_tr.buffer, func);
2300                 mutex_unlock(&trace_types_lock);
2301         }
2302 }
2303
2304 static ssize_t
2305 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2306                         size_t cnt, loff_t *ppos)
2307 {
2308         char buf[64];
2309         char *cmp = buf;
2310         int neg = 0;
2311         int ret;
2312         int i;
2313
2314         if (cnt >= sizeof(buf))
2315                 return -EINVAL;
2316
2317         if (copy_from_user(&buf, ubuf, cnt))
2318                 return -EFAULT;
2319
2320         buf[cnt] = 0;
2321
2322         if (strncmp(buf, "no", 2) == 0) {
2323                 neg = 1;
2324                 cmp += 2;
2325         }
2326
2327         for (i = 0; trace_options[i]; i++) {
2328                 int len = strlen(trace_options[i]);
2329
2330                 if (strncmp(cmp, trace_options[i], len) == 0) {
2331                         set_tracer_flags(1 << i, !neg);
2332                         break;
2333                 }
2334         }
2335
2336         /* If no option could be set, test the specific tracer options */
2337         if (!trace_options[i]) {
2338                 mutex_lock(&trace_types_lock);
2339                 ret = set_tracer_option(current_trace, cmp, neg);
2340                 mutex_unlock(&trace_types_lock);
2341                 if (ret)
2342                         return ret;
2343         }
2344
2345         filp->f_pos += cnt;
2346
2347         return cnt;
2348 }
2349
2350 static const struct file_operations tracing_iter_fops = {
2351         .open           = tracing_open_generic,
2352         .read           = tracing_trace_options_read,
2353         .write          = tracing_trace_options_write,
2354 };
2355
2356 static const char readme_msg[] =
2357         "tracing mini-HOWTO:\n\n"
2358         "# mkdir /debug\n"
2359         "# mount -t debugfs nodev /debug\n\n"
2360         "# cat /debug/tracing/available_tracers\n"
2361         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2362         "# cat /debug/tracing/current_tracer\n"
2363         "none\n"
2364         "# echo sched_switch > /debug/tracing/current_tracer\n"
2365         "# cat /debug/tracing/current_tracer\n"
2366         "sched_switch\n"
2367         "# cat /debug/tracing/trace_options\n"
2368         "noprint-parent nosym-offset nosym-addr noverbose\n"
2369         "# echo print-parent > /debug/tracing/trace_options\n"
2370         "# echo 1 > /debug/tracing/tracing_enabled\n"
2371         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2372         "echo 0 > /debug/tracing/tracing_enabled\n"
2373 ;
2374
2375 static ssize_t
2376 tracing_readme_read(struct file *filp, char __user *ubuf,
2377                        size_t cnt, loff_t *ppos)
2378 {
2379         return simple_read_from_buffer(ubuf, cnt, ppos,
2380                                         readme_msg, strlen(readme_msg));
2381 }
2382
2383 static const struct file_operations tracing_readme_fops = {
2384         .open           = tracing_open_generic,
2385         .read           = tracing_readme_read,
2386 };
2387
2388 static ssize_t
2389 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2390                   size_t cnt, loff_t *ppos)
2391 {
2392         char buf[64];
2393         int r;
2394
2395         r = sprintf(buf, "%u\n", tracer_enabled);
2396         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2397 }
2398
2399 static ssize_t
2400 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2401                    size_t cnt, loff_t *ppos)
2402 {
2403         struct trace_array *tr = filp->private_data;
2404         char buf[64];
2405         unsigned long val;
2406         int ret;
2407
2408         if (cnt >= sizeof(buf))
2409                 return -EINVAL;
2410
2411         if (copy_from_user(&buf, ubuf, cnt))
2412                 return -EFAULT;
2413
2414         buf[cnt] = 0;
2415
2416         ret = strict_strtoul(buf, 10, &val);
2417         if (ret < 0)
2418                 return ret;
2419
2420         val = !!val;
2421
2422         mutex_lock(&trace_types_lock);
2423         if (tracer_enabled ^ val) {
2424                 if (val) {
2425                         tracer_enabled = 1;
2426                         if (current_trace->start)
2427                                 current_trace->start(tr);
2428                         tracing_start();
2429                 } else {
2430                         tracer_enabled = 0;
2431                         tracing_stop();
2432                         if (current_trace->stop)
2433                                 current_trace->stop(tr);
2434                 }
2435         }
2436         mutex_unlock(&trace_types_lock);
2437
2438         filp->f_pos += cnt;
2439
2440         return cnt;
2441 }
2442
2443 static ssize_t
2444 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2445                        size_t cnt, loff_t *ppos)
2446 {
2447         char buf[max_tracer_type_len+2];
2448         int r;
2449
2450         mutex_lock(&trace_types_lock);
2451         if (current_trace)
2452                 r = sprintf(buf, "%s\n", current_trace->name);
2453         else
2454                 r = sprintf(buf, "\n");
2455         mutex_unlock(&trace_types_lock);
2456
2457         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2458 }
2459
2460 int tracer_init(struct tracer *t, struct trace_array *tr)
2461 {
2462         tracing_reset_online_cpus(tr);
2463         return t->init(tr);
2464 }
2465
2466 static int tracing_resize_ring_buffer(unsigned long size)
2467 {
2468         int ret;
2469
2470         /*
2471          * If kernel or user changes the size of the ring buffer
2472          * we use the size that was given, and we can forget about
2473          * expanding it later.
2474          */
2475         ring_buffer_expanded = 1;
2476
2477         ret = ring_buffer_resize(global_trace.buffer, size);
2478         if (ret < 0)
2479                 return ret;
2480
2481         ret = ring_buffer_resize(max_tr.buffer, size);
2482         if (ret < 0) {
2483                 int r;
2484
2485                 r = ring_buffer_resize(global_trace.buffer,
2486                                        global_trace.entries);
2487                 if (r < 0) {
2488                         /*
2489                          * AARGH! We are left with different
2490                          * size max buffer!!!!
2491                          * The max buffer is our "snapshot" buffer.
2492                          * When a tracer needs a snapshot (one of the
2493                          * latency tracers), it swaps the max buffer
2494                          * with the saved snap shot. We succeeded to
2495                          * update the size of the main buffer, but failed to
2496                          * update the size of the max buffer. But when we tried
2497                          * to reset the main buffer to the original size, we
2498                          * failed there too. This is very unlikely to
2499                          * happen, but if it does, warn and kill all
2500                          * tracing.
2501                          */
2502                         WARN_ON(1);
2503                         tracing_disabled = 1;
2504                 }
2505                 return ret;
2506         }
2507
2508         global_trace.entries = size;
2509
2510         return ret;
2511 }
2512
2513 /**
2514  * tracing_update_buffers - used by tracing facility to expand ring buffers
2515  *
2516  * To save on memory when the tracing is never used on a system with it
2517  * configured in. The ring buffers are set to a minimum size. But once
2518  * a user starts to use the tracing facility, then they need to grow
2519  * to their default size.
2520  *
2521  * This function is to be called when a tracer is about to be used.
2522  */
2523 int tracing_update_buffers(void)
2524 {
2525         int ret = 0;
2526
2527         mutex_lock(&trace_types_lock);
2528         if (!ring_buffer_expanded)
2529                 ret = tracing_resize_ring_buffer(trace_buf_size);
2530         mutex_unlock(&trace_types_lock);
2531
2532         return ret;
2533 }
2534
2535 struct trace_option_dentry;
2536
2537 static struct trace_option_dentry *
2538 create_trace_option_files(struct tracer *tracer);
2539
2540 static void
2541 destroy_trace_option_files(struct trace_option_dentry *topts);
2542
2543 static int tracing_set_tracer(const char *buf)
2544 {
2545         static struct trace_option_dentry *topts;
2546         struct trace_array *tr = &global_trace;
2547         struct tracer *t;
2548         int ret = 0;
2549
2550         mutex_lock(&trace_types_lock);
2551
2552         if (!ring_buffer_expanded) {
2553                 ret = tracing_resize_ring_buffer(trace_buf_size);
2554                 if (ret < 0)
2555                         goto out;
2556                 ret = 0;
2557         }
2558
2559         for (t = trace_types; t; t = t->next) {
2560                 if (strcmp(t->name, buf) == 0)
2561                         break;
2562         }
2563         if (!t) {
2564                 ret = -EINVAL;
2565                 goto out;
2566         }
2567         if (t == current_trace)
2568                 goto out;
2569
2570         trace_branch_disable();
2571         if (current_trace && current_trace->reset)
2572                 current_trace->reset(tr);
2573
2574         destroy_trace_option_files(topts);
2575
2576         current_trace = t;
2577
2578         topts = create_trace_option_files(current_trace);
2579
2580         if (t->init) {
2581                 ret = tracer_init(t, tr);
2582                 if (ret)
2583                         goto out;
2584         }
2585
2586         trace_branch_enable(tr);
2587  out:
2588         mutex_unlock(&trace_types_lock);
2589
2590         return ret;
2591 }
2592
2593 static ssize_t
2594 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2595                         size_t cnt, loff_t *ppos)
2596 {
2597         char buf[max_tracer_type_len+1];
2598         int i;
2599         size_t ret;
2600         int err;
2601
2602         ret = cnt;
2603
2604         if (cnt > max_tracer_type_len)
2605                 cnt = max_tracer_type_len;
2606
2607         if (copy_from_user(&buf, ubuf, cnt))
2608                 return -EFAULT;
2609
2610         buf[cnt] = 0;
2611
2612         /* strip ending whitespace. */
2613         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2614                 buf[i] = 0;
2615
2616         err = tracing_set_tracer(buf);
2617         if (err)
2618                 return err;
2619
2620         filp->f_pos += ret;
2621
2622         return ret;
2623 }
2624
2625 static ssize_t
2626 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2627                      size_t cnt, loff_t *ppos)
2628 {
2629         unsigned long *ptr = filp->private_data;
2630         char buf[64];
2631         int r;
2632
2633         r = snprintf(buf, sizeof(buf), "%ld\n",
2634                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2635         if (r > sizeof(buf))
2636                 r = sizeof(buf);
2637         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2638 }
2639
2640 static ssize_t
2641 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2642                       size_t cnt, loff_t *ppos)
2643 {
2644         unsigned long *ptr = filp->private_data;
2645         char buf[64];
2646         unsigned long val;
2647         int ret;
2648
2649         if (cnt >= sizeof(buf))
2650                 return -EINVAL;
2651
2652         if (copy_from_user(&buf, ubuf, cnt))
2653                 return -EFAULT;
2654
2655         buf[cnt] = 0;
2656
2657         ret = strict_strtoul(buf, 10, &val);
2658         if (ret < 0)
2659                 return ret;
2660
2661         *ptr = val * 1000;
2662
2663         return cnt;
2664 }
2665
2666 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2667 {
2668         long cpu_file = (long) inode->i_private;
2669         struct trace_iterator *iter;
2670         int ret = 0;
2671
2672         if (tracing_disabled)
2673                 return -ENODEV;
2674
2675         mutex_lock(&trace_types_lock);
2676
2677         /* We only allow one reader per cpu */
2678         if (cpu_file == TRACE_PIPE_ALL_CPU) {
2679                 if (!cpumask_empty(tracing_reader_cpumask)) {
2680                         ret = -EBUSY;
2681                         goto out;
2682                 }
2683                 cpumask_setall(tracing_reader_cpumask);
2684         } else {
2685                 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2686                         cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2687                 else {
2688                         ret = -EBUSY;
2689                         goto out;
2690                 }
2691         }
2692
2693         /* create a buffer to store the information to pass to userspace */
2694         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2695         if (!iter) {
2696                 ret = -ENOMEM;
2697                 goto out;
2698         }
2699
2700         /*
2701          * We make a copy of the current tracer to avoid concurrent
2702          * changes on it while we are reading.
2703          */
2704         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2705         if (!iter->trace) {
2706                 ret = -ENOMEM;
2707                 goto fail;
2708         }
2709         if (current_trace)
2710                 *iter->trace = *current_trace;
2711
2712         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2713                 ret = -ENOMEM;
2714                 goto fail;
2715         }
2716
2717         /* trace pipe does not show start of buffer */
2718         cpumask_setall(iter->started);
2719
2720         iter->cpu_file = cpu_file;
2721         iter->tr = &global_trace;
2722         mutex_init(&iter->mutex);
2723         filp->private_data = iter;
2724
2725         if (iter->trace->pipe_open)
2726                 iter->trace->pipe_open(iter);
2727
2728 out:
2729         mutex_unlock(&trace_types_lock);
2730         return ret;
2731
2732 fail:
2733         kfree(iter->trace);
2734         kfree(iter);
2735         mutex_unlock(&trace_types_lock);
2736         return ret;
2737 }
2738
2739 static int tracing_release_pipe(struct inode *inode, struct file *file)
2740 {
2741         struct trace_iterator *iter = file->private_data;
2742
2743         mutex_lock(&trace_types_lock);
2744
2745         if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2746                 cpumask_clear(tracing_reader_cpumask);
2747         else
2748                 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2749
2750         mutex_unlock(&trace_types_lock);
2751
2752         free_cpumask_var(iter->started);
2753         mutex_destroy(&iter->mutex);
2754         kfree(iter->trace);
2755         kfree(iter);
2756
2757         return 0;
2758 }
2759
2760 static unsigned int
2761 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2762 {
2763         struct trace_iterator *iter = filp->private_data;
2764
2765         if (trace_flags & TRACE_ITER_BLOCK) {
2766                 /*
2767                  * Always select as readable when in blocking mode
2768                  */
2769                 return POLLIN | POLLRDNORM;
2770         } else {
2771                 if (!trace_empty(iter))
2772                         return POLLIN | POLLRDNORM;
2773                 poll_wait(filp, &trace_wait, poll_table);
2774                 if (!trace_empty(iter))
2775                         return POLLIN | POLLRDNORM;
2776
2777                 return 0;
2778         }
2779 }
2780
2781
2782 void default_wait_pipe(struct trace_iterator *iter)
2783 {
2784         DEFINE_WAIT(wait);
2785
2786         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2787
2788         if (trace_empty(iter))
2789                 schedule();
2790
2791         finish_wait(&trace_wait, &wait);
2792 }
2793
2794 /*
2795  * This is a make-shift waitqueue.
2796  * A tracer might use this callback on some rare cases:
2797  *
2798  *  1) the current tracer might hold the runqueue lock when it wakes up
2799  *     a reader, hence a deadlock (sched, function, and function graph tracers)
2800  *  2) the function tracers, trace all functions, we don't want
2801  *     the overhead of calling wake_up and friends
2802  *     (and tracing them too)
2803  *
2804  *     Anyway, this is really very primitive wakeup.
2805  */
2806 void poll_wait_pipe(struct trace_iterator *iter)
2807 {
2808         set_current_state(TASK_INTERRUPTIBLE);
2809         /* sleep for 100 msecs, and try again. */
2810         schedule_timeout(HZ / 10);
2811 }
2812
2813 /* Must be called with trace_types_lock mutex held. */
2814 static int tracing_wait_pipe(struct file *filp)
2815 {
2816         struct trace_iterator *iter = filp->private_data;
2817
2818         while (trace_empty(iter)) {
2819
2820                 if ((filp->f_flags & O_NONBLOCK)) {
2821                         return -EAGAIN;
2822                 }
2823
2824                 mutex_unlock(&iter->mutex);
2825
2826                 iter->trace->wait_pipe(iter);
2827
2828                 mutex_lock(&iter->mutex);
2829
2830                 if (signal_pending(current))
2831                         return -EINTR;
2832
2833                 /*
2834                  * We block until we read something and tracing is disabled.
2835                  * We still block if tracing is disabled, but we have never
2836                  * read anything. This allows a user to cat this file, and
2837                  * then enable tracing. But after we have read something,
2838                  * we give an EOF when tracing is again disabled.
2839                  *
2840                  * iter->pos will be 0 if we haven't read anything.
2841                  */
2842                 if (!tracer_enabled && iter->pos)
2843                         break;
2844         }
2845
2846         return 1;
2847 }
2848
2849 /*
2850  * Consumer reader.
2851  */
2852 static ssize_t
2853 tracing_read_pipe(struct file *filp, char __user *ubuf,
2854                   size_t cnt, loff_t *ppos)
2855 {
2856         struct trace_iterator *iter = filp->private_data;
2857         static struct tracer *old_tracer;
2858         ssize_t sret;
2859
2860         /* return any leftover data */
2861         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2862         if (sret != -EBUSY)
2863                 return sret;
2864
2865         trace_seq_init(&iter->seq);
2866
2867         /* copy the tracer to avoid using a global lock all around */
2868         mutex_lock(&trace_types_lock);
2869         if (unlikely(old_tracer != current_trace && current_trace)) {
2870                 old_tracer = current_trace;
2871                 *iter->trace = *current_trace;
2872         }
2873         mutex_unlock(&trace_types_lock);
2874
2875         /*
2876          * Avoid more than one consumer on a single file descriptor
2877          * This is just a matter of traces coherency, the ring buffer itself
2878          * is protected.
2879          */
2880         mutex_lock(&iter->mutex);
2881         if (iter->trace->read) {
2882                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2883                 if (sret)
2884                         goto out;
2885         }
2886
2887 waitagain:
2888         sret = tracing_wait_pipe(filp);
2889         if (sret <= 0)
2890                 goto out;
2891
2892         /* stop when tracing is finished */
2893         if (trace_empty(iter)) {
2894                 sret = 0;
2895                 goto out;
2896         }
2897
2898         if (cnt >= PAGE_SIZE)
2899                 cnt = PAGE_SIZE - 1;
2900
2901         /* reset all but tr, trace, and overruns */
2902         memset(&iter->seq, 0,
2903                sizeof(struct trace_iterator) -
2904                offsetof(struct trace_iterator, seq));
2905         iter->pos = -1;
2906
2907         while (find_next_entry_inc(iter) != NULL) {
2908                 enum print_line_t ret;
2909                 int len = iter->seq.len;
2910
2911                 ret = print_trace_line(iter);
2912                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2913                         /* don't print partial lines */
2914                         iter->seq.len = len;
2915                         break;
2916                 }
2917                 if (ret != TRACE_TYPE_NO_CONSUME)
2918                         trace_consume(iter);
2919
2920                 if (iter->seq.len >= cnt)
2921                         break;
2922         }
2923
2924         /* Now copy what we have to the user */
2925         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2926         if (iter->seq.readpos >= iter->seq.len)
2927                 trace_seq_init(&iter->seq);
2928
2929         /*
2930          * If there was nothing to send to user, inspite of consuming trace
2931          * entries, go back to wait for more entries.
2932          */
2933         if (sret == -EBUSY)
2934                 goto waitagain;
2935
2936 out:
2937         mutex_unlock(&iter->mutex);
2938
2939         return sret;
2940 }
2941
2942 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
2943                                      struct pipe_buffer *buf)
2944 {
2945         __free_page(buf->page);
2946 }
2947
2948 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
2949                                      unsigned int idx)
2950 {
2951         __free_page(spd->pages[idx]);
2952 }
2953
2954 static struct pipe_buf_operations tracing_pipe_buf_ops = {
2955         .can_merge              = 0,
2956         .map                    = generic_pipe_buf_map,
2957         .unmap                  = generic_pipe_buf_unmap,
2958         .confirm                = generic_pipe_buf_confirm,
2959         .release                = tracing_pipe_buf_release,
2960         .steal                  = generic_pipe_buf_steal,
2961         .get                    = generic_pipe_buf_get,
2962 };
2963
2964 static size_t
2965 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
2966 {
2967         size_t count;
2968         int ret;
2969
2970         /* Seq buffer is page-sized, exactly what we need. */
2971         for (;;) {
2972                 count = iter->seq.len;
2973                 ret = print_trace_line(iter);
2974                 count = iter->seq.len - count;
2975                 if (rem < count) {
2976                         rem = 0;
2977                         iter->seq.len -= count;
2978                         break;
2979                 }
2980                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2981                         iter->seq.len -= count;
2982                         break;
2983                 }
2984
2985                 trace_consume(iter);
2986                 rem -= count;
2987                 if (!find_next_entry_inc(iter)) {
2988                         rem = 0;
2989                         iter->ent = NULL;
2990                         break;
2991                 }
2992         }
2993
2994         return rem;
2995 }
2996
2997 static ssize_t tracing_splice_read_pipe(struct file *filp,
2998                                         loff_t *ppos,
2999                                         struct pipe_inode_info *pipe,
3000                                         size_t len,
3001                                         unsigned int flags)
3002 {
3003         struct page *pages[PIPE_BUFFERS];
3004         struct partial_page partial[PIPE_BUFFERS];
3005         struct trace_iterator *iter = filp->private_data;
3006         struct splice_pipe_desc spd = {
3007                 .pages          = pages,
3008                 .partial        = partial,
3009                 .nr_pages       = 0, /* This gets updated below. */
3010                 .flags          = flags,
3011                 .ops            = &tracing_pipe_buf_ops,
3012                 .spd_release    = tracing_spd_release_pipe,
3013         };
3014         static struct tracer *old_tracer;
3015         ssize_t ret;
3016         size_t rem;
3017         unsigned int i;
3018
3019         /* copy the tracer to avoid using a global lock all around */
3020         mutex_lock(&trace_types_lock);
3021         if (unlikely(old_tracer != current_trace && current_trace)) {
3022                 old_tracer = current_trace;
3023                 *iter->trace = *current_trace;
3024         }
3025         mutex_unlock(&trace_types_lock);
3026
3027         mutex_lock(&iter->mutex);
3028
3029         if (iter->trace->splice_read) {
3030                 ret = iter->trace->splice_read(iter, filp,
3031                                                ppos, pipe, len, flags);
3032                 if (ret)
3033                         goto out_err;
3034         }
3035
3036         ret = tracing_wait_pipe(filp);
3037         if (ret <= 0)
3038                 goto out_err;
3039
3040         if (!iter->ent && !find_next_entry_inc(iter)) {
3041                 ret = -EFAULT;
3042                 goto out_err;
3043         }
3044
3045         /* Fill as many pages as possible. */
3046         for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3047                 pages[i] = alloc_page(GFP_KERNEL);
3048                 if (!pages[i])
3049                         break;
3050
3051                 rem = tracing_fill_pipe_page(rem, iter);
3052
3053                 /* Copy the data into the page, so we can start over. */
3054                 ret = trace_seq_to_buffer(&iter->seq,
3055                                           page_address(pages[i]),
3056                                           iter->seq.len);
3057                 if (ret < 0) {
3058                         __free_page(pages[i]);
3059                         break;
3060                 }
3061                 partial[i].offset = 0;
3062                 partial[i].len = iter->seq.len;
3063
3064                 trace_seq_init(&iter->seq);
3065         }
3066
3067         mutex_unlock(&iter->mutex);
3068
3069         spd.nr_pages = i;
3070
3071         return splice_to_pipe(pipe, &spd);
3072
3073 out_err:
3074         mutex_unlock(&iter->mutex);
3075
3076         return ret;
3077 }
3078
3079 static ssize_t
3080 tracing_entries_read(struct file *filp, char __user *ubuf,
3081                      size_t cnt, loff_t *ppos)
3082 {
3083         struct trace_array *tr = filp->private_data;
3084         char buf[96];
3085         int r;
3086
3087         mutex_lock(&trace_types_lock);
3088         if (!ring_buffer_expanded)
3089                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3090                             tr->entries >> 10,
3091                             trace_buf_size >> 10);
3092         else
3093                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3094         mutex_unlock(&trace_types_lock);
3095
3096         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3097 }
3098
3099 static ssize_t
3100 tracing_entries_write(struct file *filp, const char __user *ubuf,
3101                       size_t cnt, loff_t *ppos)
3102 {
3103         unsigned long val;
3104         char buf[64];
3105         int ret, cpu;
3106
3107         if (cnt >= sizeof(buf))
3108                 return -EINVAL;
3109
3110         if (copy_from_user(&buf, ubuf, cnt))
3111                 return -EFAULT;
3112
3113         buf[cnt] = 0;
3114
3115         ret = strict_strtoul(buf, 10, &val);
3116         if (ret < 0)
3117                 return ret;
3118
3119         /* must have at least 1 entry */
3120         if (!val)
3121                 return -EINVAL;
3122
3123         mutex_lock(&trace_types_lock);
3124
3125         tracing_stop();
3126
3127         /* disable all cpu buffers */
3128         for_each_tracing_cpu(cpu) {
3129                 if (global_trace.data[cpu])
3130                         atomic_inc(&global_trace.data[cpu]->disabled);
3131                 if (max_tr.data[cpu])
3132                         atomic_inc(&max_tr.data[cpu]->disabled);
3133         }
3134
3135         /* value is in KB */
3136         val <<= 10;
3137
3138         if (val != global_trace.entries) {
3139                 ret = tracing_resize_ring_buffer(val);
3140                 if (ret < 0) {
3141                         cnt = ret;
3142                         goto out;
3143                 }
3144         }
3145
3146         filp->f_pos += cnt;
3147
3148         /* If check pages failed, return ENOMEM */
3149         if (tracing_disabled)
3150                 cnt = -ENOMEM;
3151  out:
3152         for_each_tracing_cpu(cpu) {
3153                 if (global_trace.data[cpu])
3154                         atomic_dec(&global_trace.data[cpu]->disabled);
3155                 if (max_tr.data[cpu])
3156                         atomic_dec(&max_tr.data[cpu]->disabled);
3157         }
3158
3159         tracing_start();
3160         max_tr.entries = global_trace.entries;
3161         mutex_unlock(&trace_types_lock);
3162
3163         return cnt;
3164 }
3165
3166 static int mark_printk(const char *fmt, ...)
3167 {
3168         int ret;
3169         va_list args;
3170         va_start(args, fmt);
3171         ret = trace_vprintk(0, fmt, args);
3172         va_end(args);
3173         return ret;
3174 }
3175
3176 static ssize_t
3177 tracing_mark_write(struct file *filp, const char __user *ubuf,
3178                                         size_t cnt, loff_t *fpos)
3179 {
3180         char *buf;
3181         char *end;
3182
3183         if (tracing_disabled)
3184                 return -EINVAL;
3185
3186         if (cnt > TRACE_BUF_SIZE)
3187                 cnt = TRACE_BUF_SIZE;
3188
3189         buf = kmalloc(cnt + 1, GFP_KERNEL);
3190         if (buf == NULL)
3191                 return -ENOMEM;
3192
3193         if (copy_from_user(buf, ubuf, cnt)) {
3194                 kfree(buf);
3195                 return -EFAULT;
3196         }
3197
3198         /* Cut from the first nil or newline. */
3199         buf[cnt] = '\0';
3200         end = strchr(buf, '\n');
3201         if (end)
3202                 *end = '\0';
3203
3204         cnt = mark_printk("%s\n", buf);
3205         kfree(buf);
3206         *fpos += cnt;
3207
3208         return cnt;
3209 }
3210
3211 static const struct file_operations tracing_max_lat_fops = {
3212         .open           = tracing_open_generic,
3213         .read           = tracing_max_lat_read,
3214         .write          = tracing_max_lat_write,
3215 };
3216
3217 static const struct file_operations tracing_ctrl_fops = {
3218         .open           = tracing_open_generic,
3219         .read           = tracing_ctrl_read,
3220         .write          = tracing_ctrl_write,
3221 };
3222
3223 static const struct file_operations set_tracer_fops = {
3224         .open           = tracing_open_generic,
3225         .read           = tracing_set_trace_read,
3226         .write          = tracing_set_trace_write,
3227 };
3228
3229 static const struct file_operations tracing_pipe_fops = {
3230         .open           = tracing_open_pipe,
3231         .poll           = tracing_poll_pipe,
3232         .read           = tracing_read_pipe,
3233         .splice_read    = tracing_splice_read_pipe,
3234         .release        = tracing_release_pipe,
3235 };
3236
3237 static const struct file_operations tracing_entries_fops = {
3238         .open           = tracing_open_generic,
3239         .read           = tracing_entries_read,
3240         .write          = tracing_entries_write,
3241 };
3242
3243 static const struct file_operations tracing_mark_fops = {
3244         .open           = tracing_open_generic,
3245         .write          = tracing_mark_write,
3246 };
3247
3248 struct ftrace_buffer_info {
3249         struct trace_array      *tr;
3250         void                    *spare;
3251         int                     cpu;
3252         unsigned int            read;
3253 };
3254
3255 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3256 {
3257         int cpu = (int)(long)inode->i_private;
3258         struct ftrace_buffer_info *info;
3259
3260         if (tracing_disabled)
3261                 return -ENODEV;
3262
3263         info = kzalloc(sizeof(*info), GFP_KERNEL);
3264         if (!info)
3265                 return -ENOMEM;
3266
3267         info->tr        = &global_trace;
3268         info->cpu       = cpu;
3269         info->spare     = ring_buffer_alloc_read_page(info->tr->buffer);
3270         /* Force reading ring buffer for first read */
3271         info->read      = (unsigned int)-1;
3272         if (!info->spare)
3273                 goto out;
3274
3275         filp->private_data = info;
3276
3277         return 0;
3278
3279  out:
3280         kfree(info);
3281         return -ENOMEM;
3282 }
3283
3284 static ssize_t
3285 tracing_buffers_read(struct file *filp, char __user *ubuf,
3286                      size_t count, loff_t *ppos)
3287 {
3288         struct ftrace_buffer_info *info = filp->private_data;
3289         unsigned int pos;
3290         ssize_t ret;
3291         size_t size;
3292
3293         if (!count)
3294                 return 0;
3295
3296         /* Do we have previous read data to read? */
3297         if (info->read < PAGE_SIZE)
3298                 goto read;
3299
3300         info->read = 0;
3301
3302         ret = ring_buffer_read_page(info->tr->buffer,
3303                                     &info->spare,
3304                                     count,
3305                                     info->cpu, 0);
3306         if (ret < 0)
3307                 return 0;
3308
3309         pos = ring_buffer_page_len(info->spare);
3310
3311         if (pos < PAGE_SIZE)
3312                 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3313
3314 read:
3315         size = PAGE_SIZE - info->read;
3316         if (size > count)
3317                 size = count;
3318
3319         ret = copy_to_user(ubuf, info->spare + info->read, size);
3320         if (ret == size)
3321                 return -EFAULT;
3322         size -= ret;
3323
3324         *ppos += size;
3325         info->read += size;
3326
3327         return size;
3328 }
3329
3330 static int tracing_buffers_release(struct inode *inode, struct file *file)
3331 {
3332         struct ftrace_buffer_info *info = file->private_data;
3333
3334         ring_buffer_free_read_page(info->tr->buffer, info->spare);
3335         kfree(info);
3336
3337         return 0;
3338 }
3339
3340 struct buffer_ref {
3341         struct ring_buffer      *buffer;
3342         void                    *page;
3343         int                     ref;
3344 };
3345
3346 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3347                                     struct pipe_buffer *buf)
3348 {
3349         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3350
3351         if (--ref->ref)
3352                 return;
3353
3354         ring_buffer_free_read_page(ref->buffer, ref->page);
3355         kfree(ref);
3356         buf->private = 0;
3357 }
3358
3359 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3360                                  struct pipe_buffer *buf)
3361 {
3362         return 1;
3363 }
3364
3365 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3366                                 struct pipe_buffer *buf)
3367 {
3368         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3369
3370         ref->ref++;
3371 }
3372
3373 /* Pipe buffer operations for a buffer. */
3374 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3375         .can_merge              = 0,
3376         .map                    = generic_pipe_buf_map,
3377         .unmap                  = generic_pipe_buf_unmap,
3378         .confirm                = generic_pipe_buf_confirm,
3379         .release                = buffer_pipe_buf_release,
3380         .steal                  = buffer_pipe_buf_steal,
3381         .get                    = buffer_pipe_buf_get,
3382 };
3383
3384 /*
3385  * Callback from splice_to_pipe(), if we need to release some pages
3386  * at the end of the spd in case we error'ed out in filling the pipe.
3387  */
3388 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3389 {
3390         struct buffer_ref *ref =
3391                 (struct buffer_ref *)spd->partial[i].private;
3392
3393         if (--ref->ref)
3394                 return;
3395
3396         ring_buffer_free_read_page(ref->buffer, ref->page);
3397         kfree(ref);
3398         spd->partial[i].private = 0;
3399 }
3400
3401 static ssize_t
3402 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3403                             struct pipe_inode_info *pipe, size_t len,
3404                             unsigned int flags)
3405 {
3406         struct ftrace_buffer_info *info = file->private_data;
3407         struct partial_page partial[PIPE_BUFFERS];
3408         struct page *pages[PIPE_BUFFERS];
3409         struct splice_pipe_desc spd = {
3410                 .pages          = pages,
3411                 .partial        = partial,
3412                 .flags          = flags,
3413                 .ops            = &buffer_pipe_buf_ops,
3414                 .spd_release    = buffer_spd_release,
3415         };
3416         struct buffer_ref *ref;
3417         int size, i;
3418         size_t ret;
3419
3420         /*
3421          * We can't seek on a buffer input
3422          */
3423         if (unlikely(*ppos))
3424                 return -ESPIPE;
3425
3426
3427         for (i = 0; i < PIPE_BUFFERS && len; i++, len -= size) {
3428                 struct page *page;
3429                 int r;
3430
3431                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3432                 if (!ref)
3433                         break;
3434
3435                 ref->buffer = info->tr->buffer;
3436                 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3437                 if (!ref->page) {
3438                         kfree(ref);
3439                         break;
3440                 }
3441
3442                 r = ring_buffer_read_page(ref->buffer, &ref->page,
3443                                           len, info->cpu, 0);
3444                 if (r < 0) {
3445                         ring_buffer_free_read_page(ref->buffer,
3446                                                    ref->page);
3447                         kfree(ref);
3448                         break;
3449                 }
3450
3451                 /*
3452                  * zero out any left over data, this is going to
3453                  * user land.
3454                  */
3455                 size = ring_buffer_page_len(ref->page);
3456                 if (size < PAGE_SIZE)
3457                         memset(ref->page + size, 0, PAGE_SIZE - size);
3458
3459                 page = virt_to_page(ref->page);
3460
3461                 spd.pages[i] = page;
3462                 spd.partial[i].len = PAGE_SIZE;
3463                 spd.partial[i].offset = 0;
3464                 spd.partial[i].private = (unsigned long)ref;
3465                 spd.nr_pages++;
3466         }
3467
3468         spd.nr_pages = i;
3469
3470         /* did we read anything? */
3471         if (!spd.nr_pages) {
3472                 if (flags & SPLICE_F_NONBLOCK)
3473                         ret = -EAGAIN;
3474                 else
3475                         ret = 0;
3476                 /* TODO: block */
3477                 return ret;
3478         }
3479
3480         ret = splice_to_pipe(pipe, &spd);
3481
3482         return ret;
3483 }
3484
3485 static const struct file_operations tracing_buffers_fops = {
3486         .open           = tracing_buffers_open,
3487         .read           = tracing_buffers_read,
3488         .release        = tracing_buffers_release,
3489         .splice_read    = tracing_buffers_splice_read,
3490         .llseek         = no_llseek,
3491 };
3492
3493 #ifdef CONFIG_DYNAMIC_FTRACE
3494
3495 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3496 {
3497         return 0;
3498 }
3499
3500 static ssize_t
3501 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3502                   size_t cnt, loff_t *ppos)
3503 {
3504         static char ftrace_dyn_info_buffer[1024];
3505         static DEFINE_MUTEX(dyn_info_mutex);
3506         unsigned long *p = filp->private_data;
3507         char *buf = ftrace_dyn_info_buffer;
3508         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3509         int r;
3510
3511         mutex_lock(&dyn_info_mutex);
3512         r = sprintf(buf, "%ld ", *p);
3513
3514         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3515         buf[r++] = '\n';
3516
3517         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3518
3519         mutex_unlock(&dyn_info_mutex);
3520
3521         return r;
3522 }
3523
3524 static const struct file_operations tracing_dyn_info_fops = {
3525         .open           = tracing_open_generic,
3526         .read           = tracing_read_dyn_info,
3527 };
3528 #endif
3529
3530 static struct dentry *d_tracer;
3531
3532 struct dentry *tracing_init_dentry(void)
3533 {
3534         static int once;
3535
3536         if (d_tracer)
3537                 return d_tracer;
3538
3539         if (!debugfs_initialized())
3540                 return NULL;
3541
3542         d_tracer = debugfs_create_dir("tracing", NULL);
3543
3544         if (!d_tracer && !once) {
3545                 once = 1;
3546                 pr_warning("Could not create debugfs directory 'tracing'\n");
3547                 return NULL;
3548         }
3549
3550         return d_tracer;
3551 }
3552
3553 static struct dentry *d_percpu;
3554
3555 struct dentry *tracing_dentry_percpu(void)
3556 {
3557         static int once;
3558         struct dentry *d_tracer;
3559
3560         if (d_percpu)
3561                 return d_percpu;
3562
3563         d_tracer = tracing_init_dentry();
3564
3565         if (!d_tracer)
3566                 return NULL;
3567
3568         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3569
3570         if (!d_percpu && !once) {
3571                 once = 1;
3572                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3573                 return NULL;
3574         }
3575
3576         return d_percpu;
3577 }
3578
3579 static void tracing_init_debugfs_percpu(long cpu)
3580 {
3581         struct dentry *d_percpu = tracing_dentry_percpu();
3582         struct dentry *entry, *d_cpu;
3583         /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3584         char cpu_dir[7];
3585
3586         if (cpu > 999 || cpu < 0)
3587                 return;
3588
3589         sprintf(cpu_dir, "cpu%ld", cpu);
3590         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3591         if (!d_cpu) {
3592                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3593                 return;
3594         }
3595
3596         /* per cpu trace_pipe */
3597         entry = debugfs_create_file("trace_pipe", 0444, d_cpu,
3598                                 (void *) cpu, &tracing_pipe_fops);
3599         if (!entry)
3600                 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3601
3602         /* per cpu trace */
3603         entry = debugfs_create_file("trace", 0644, d_cpu,
3604                                 (void *) cpu, &tracing_fops);
3605         if (!entry)
3606                 pr_warning("Could not create debugfs 'trace' entry\n");
3607
3608         entry = debugfs_create_file("trace_pipe_raw", 0444, d_cpu,
3609                                     (void *) cpu, &tracing_buffers_fops);
3610         if (!entry)
3611                 pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n");
3612 }
3613
3614 #ifdef CONFIG_FTRACE_SELFTEST
3615 /* Let selftest have access to static functions in this file */
3616 #include "trace_selftest.c"
3617 #endif
3618
3619 struct trace_option_dentry {
3620         struct tracer_opt               *opt;
3621         struct tracer_flags             *flags;
3622         struct dentry                   *entry;
3623 };
3624
3625 static ssize_t
3626 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3627                         loff_t *ppos)
3628 {
3629         struct trace_option_dentry *topt = filp->private_data;
3630         char *buf;
3631
3632         if (topt->flags->val & topt->opt->bit)
3633                 buf = "1\n";
3634         else
3635                 buf = "0\n";
3636
3637         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3638 }
3639
3640 static ssize_t
3641 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3642                          loff_t *ppos)
3643 {
3644         struct trace_option_dentry *topt = filp->private_data;
3645         unsigned long val;
3646         char buf[64];
3647         int ret;
3648
3649         if (cnt >= sizeof(buf))
3650                 return -EINVAL;
3651
3652         if (copy_from_user(&buf, ubuf, cnt))
3653                 return -EFAULT;
3654
3655         buf[cnt] = 0;
3656
3657         ret = strict_strtoul(buf, 10, &val);
3658         if (ret < 0)
3659                 return ret;
3660
3661         ret = 0;
3662         switch (val) {
3663         case 0:
3664                 /* do nothing if already cleared */
3665                 if (!(topt->flags->val & topt->opt->bit))
3666                         break;
3667
3668                 mutex_lock(&trace_types_lock);
3669                 if (current_trace->set_flag)
3670                         ret = current_trace->set_flag(topt->flags->val,
3671                                                       topt->opt->bit, 0);
3672                 mutex_unlock(&trace_types_lock);
3673                 if (ret)
3674                         return ret;
3675                 topt->flags->val &= ~topt->opt->bit;
3676                 break;
3677         case 1:
3678                 /* do nothing if already set */
3679                 if (topt->flags->val & topt->opt->bit)
3680                         break;
3681
3682                 mutex_lock(&trace_types_lock);
3683                 if (current_trace->set_flag)
3684                         ret = current_trace->set_flag(topt->flags->val,
3685                                                       topt->opt->bit, 1);
3686                 mutex_unlock(&trace_types_lock);
3687                 if (ret)
3688                         return ret;
3689                 topt->flags->val |= topt->opt->bit;
3690                 break;
3691
3692         default:
3693                 return -EINVAL;
3694         }
3695
3696         *ppos += cnt;
3697
3698         return cnt;
3699 }
3700
3701
3702 static const struct file_operations trace_options_fops = {
3703         .open = tracing_open_generic,
3704         .read = trace_options_read,
3705         .write = trace_options_write,
3706 };
3707
3708 static ssize_t
3709 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3710                         loff_t *ppos)
3711 {
3712         long index = (long)filp->private_data;
3713         char *buf;
3714
3715         if (trace_flags & (1 << index))
3716                 buf = "1\n";
3717         else
3718                 buf = "0\n";
3719
3720         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3721 }
3722
3723 static ssize_t
3724 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3725                          loff_t *ppos)
3726 {
3727         long index = (long)filp->private_data;
3728         char buf[64];
3729         unsigned long val;
3730         int ret;
3731
3732         if (cnt >= sizeof(buf))
3733                 return -EINVAL;
3734
3735         if (copy_from_user(&buf, ubuf, cnt))
3736                 return -EFAULT;
3737
3738         buf[cnt] = 0;
3739
3740         ret = strict_strtoul(buf, 10, &val);
3741         if (ret < 0)
3742                 return ret;
3743
3744         switch (val) {
3745         case 0:
3746                 trace_flags &= ~(1 << index);
3747                 break;
3748         case 1:
3749                 trace_flags |= 1 << index;
3750                 break;
3751
3752         default:
3753                 return -EINVAL;
3754         }
3755
3756         *ppos += cnt;
3757
3758         return cnt;
3759 }
3760
3761 static const struct file_operations trace_options_core_fops = {
3762         .open = tracing_open_generic,
3763         .read = trace_options_core_read,
3764         .write = trace_options_core_write,
3765 };
3766
3767 static struct dentry *trace_options_init_dentry(void)
3768 {
3769         struct dentry *d_tracer;
3770         static struct dentry *t_options;
3771
3772         if (t_options)
3773                 return t_options;
3774
3775         d_tracer = tracing_init_dentry();
3776         if (!d_tracer)
3777                 return NULL;
3778
3779         t_options = debugfs_create_dir("options", d_tracer);
3780         if (!t_options) {
3781                 pr_warning("Could not create debugfs directory 'options'\n");
3782                 return NULL;
3783         }
3784
3785         return t_options;
3786 }
3787
3788 static void
3789 create_trace_option_file(struct trace_option_dentry *topt,
3790                          struct tracer_flags *flags,
3791                          struct tracer_opt *opt)
3792 {
3793         struct dentry *t_options;
3794         struct dentry *entry;
3795
3796         t_options = trace_options_init_dentry();
3797         if (!t_options)
3798                 return;
3799
3800         topt->flags = flags;
3801         topt->opt = opt;
3802
3803         entry = debugfs_create_file(opt->name, 0644, t_options, topt,
3804                                     &trace_options_fops);
3805
3806         topt->entry = entry;
3807
3808 }
3809
3810 static struct trace_option_dentry *
3811 create_trace_option_files(struct tracer *tracer)
3812 {
3813         struct trace_option_dentry *topts;
3814         struct tracer_flags *flags;
3815         struct tracer_opt *opts;
3816         int cnt;
3817
3818         if (!tracer)
3819                 return NULL;
3820
3821         flags = tracer->flags;
3822
3823         if (!flags || !flags->opts)
3824                 return NULL;
3825
3826         opts = flags->opts;
3827
3828         for (cnt = 0; opts[cnt].name; cnt++)
3829                 ;
3830
3831         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3832         if (!topts)
3833                 return NULL;
3834
3835         for (cnt = 0; opts[cnt].name; cnt++)
3836                 create_trace_option_file(&topts[cnt], flags,
3837                                          &opts[cnt]);
3838
3839         return topts;
3840 }
3841
3842 static void
3843 destroy_trace_option_files(struct trace_option_dentry *topts)
3844 {
3845         int cnt;
3846
3847         if (!topts)
3848                 return;
3849
3850         for (cnt = 0; topts[cnt].opt; cnt++) {
3851                 if (topts[cnt].entry)
3852                         debugfs_remove(topts[cnt].entry);
3853         }
3854
3855         kfree(topts);
3856 }
3857
3858 static struct dentry *
3859 create_trace_option_core_file(const char *option, long index)
3860 {
3861         struct dentry *t_options;
3862         struct dentry *entry;
3863
3864         t_options = trace_options_init_dentry();
3865         if (!t_options)
3866                 return NULL;
3867
3868         entry = debugfs_create_file(option, 0644, t_options, (void *)index,
3869                                     &trace_options_core_fops);
3870
3871         return entry;
3872 }
3873
3874 static __init void create_trace_options_dir(void)
3875 {
3876         struct dentry *t_options;
3877         struct dentry *entry;
3878         int i;
3879
3880         t_options = trace_options_init_dentry();
3881         if (!t_options)
3882                 return;
3883
3884         for (i = 0; trace_options[i]; i++) {
3885                 entry = create_trace_option_core_file(trace_options[i], i);
3886                 if (!entry)
3887                         pr_warning("Could not create debugfs %s entry\n",
3888                                    trace_options[i]);
3889         }
3890 }
3891
3892 static __init int tracer_init_debugfs(void)
3893 {
3894         struct dentry *d_tracer;
3895         struct dentry *entry;
3896         int cpu;
3897
3898         d_tracer = tracing_init_dentry();
3899
3900         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3901                                     &global_trace, &tracing_ctrl_fops);
3902         if (!entry)
3903                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3904
3905         entry = debugfs_create_file("trace_options", 0644, d_tracer,
3906                                     NULL, &tracing_iter_fops);
3907         if (!entry)
3908                 pr_warning("Could not create debugfs 'trace_options' entry\n");
3909
3910         create_trace_options_dir();
3911
3912         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3913                                     NULL, &tracing_cpumask_fops);
3914         if (!entry)
3915                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3916
3917         entry = debugfs_create_file("trace", 0644, d_tracer,
3918                                  (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
3919         if (!entry)
3920                 pr_warning("Could not create debugfs 'trace' entry\n");
3921
3922         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3923                                     &global_trace, &show_traces_fops);
3924         if (!entry)
3925                 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3926
3927         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3928                                     &global_trace, &set_tracer_fops);
3929         if (!entry)
3930                 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3931
3932         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3933                                     &tracing_max_latency,
3934                                     &tracing_max_lat_fops);
3935         if (!entry)
3936                 pr_warning("Could not create debugfs "
3937                            "'tracing_max_latency' entry\n");
3938
3939         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3940                                     &tracing_thresh, &tracing_max_lat_fops);
3941         if (!entry)
3942                 pr_warning("Could not create debugfs "
3943                            "'tracing_thresh' entry\n");
3944         entry = debugfs_create_file("README", 0644, d_tracer,
3945                                     NULL, &tracing_readme_fops);
3946         if (!entry)
3947                 pr_warning("Could not create debugfs 'README' entry\n");
3948
3949         entry = debugfs_create_file("trace_pipe", 0444, d_tracer,
3950                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
3951         if (!entry)
3952                 pr_warning("Could not create debugfs "
3953                            "'trace_pipe' entry\n");
3954
3955         entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3956                                     &global_trace, &tracing_entries_fops);
3957         if (!entry)
3958                 pr_warning("Could not create debugfs "
3959                            "'buffer_size_kb' entry\n");
3960
3961         entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3962                                     NULL, &tracing_mark_fops);
3963         if (!entry)
3964                 pr_warning("Could not create debugfs "
3965                            "'trace_marker' entry\n");
3966
3967 #ifdef CONFIG_DYNAMIC_FTRACE
3968         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3969                                     &ftrace_update_tot_cnt,
3970                                     &tracing_dyn_info_fops);
3971         if (!entry)
3972                 pr_warning("Could not create debugfs "
3973                            "'dyn_ftrace_total_info' entry\n");
3974 #endif
3975 #ifdef CONFIG_SYSPROF_TRACER
3976         init_tracer_sysprof_debugfs(d_tracer);
3977 #endif
3978
3979         for_each_tracing_cpu(cpu)
3980                 tracing_init_debugfs_percpu(cpu);
3981
3982         return 0;
3983 }
3984
3985 static int trace_panic_handler(struct notifier_block *this,
3986                                unsigned long event, void *unused)
3987 {
3988         if (ftrace_dump_on_oops)
3989                 ftrace_dump();
3990         return NOTIFY_OK;
3991 }
3992
3993 static struct notifier_block trace_panic_notifier = {
3994         .notifier_call  = trace_panic_handler,
3995         .next           = NULL,
3996         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
3997 };
3998
3999 static int trace_die_handler(struct notifier_block *self,
4000                              unsigned long val,
4001                              void *data)
4002 {
4003         switch (val) {
4004         case DIE_OOPS:
4005                 if (ftrace_dump_on_oops)
4006                         ftrace_dump();
4007                 break;
4008         default:
4009                 break;
4010         }
4011         return NOTIFY_OK;
4012 }
4013
4014 static struct notifier_block trace_die_notifier = {
4015         .notifier_call = trace_die_handler,
4016         .priority = 200
4017 };
4018
4019 /*
4020  * printk is set to max of 1024, we really don't need it that big.
4021  * Nothing should be printing 1000 characters anyway.
4022  */
4023 #define TRACE_MAX_PRINT         1000
4024
4025 /*
4026  * Define here KERN_TRACE so that we have one place to modify
4027  * it if we decide to change what log level the ftrace dump
4028  * should be at.
4029  */
4030 #define KERN_TRACE              KERN_EMERG
4031
4032 static void
4033 trace_printk_seq(struct trace_seq *s)
4034 {
4035         /* Probably should print a warning here. */
4036         if (s->len >= 1000)
4037                 s->len = 1000;
4038
4039         /* should be zero ended, but we are paranoid. */
4040         s->buffer[s->len] = 0;
4041
4042         printk(KERN_TRACE "%s", s->buffer);
4043
4044         trace_seq_init(s);
4045 }
4046
4047 static void __ftrace_dump(bool disable_tracing)
4048 {
4049         static DEFINE_SPINLOCK(ftrace_dump_lock);
4050         /* use static because iter can be a bit big for the stack */
4051         static struct trace_iterator iter;
4052         unsigned int old_userobj;
4053         static int dump_ran;
4054         unsigned long flags;
4055         int cnt = 0, cpu;
4056
4057         /* only one dump */
4058         spin_lock_irqsave(&ftrace_dump_lock, flags);
4059         if (dump_ran)
4060                 goto out;
4061
4062         dump_ran = 1;
4063
4064         tracing_off();
4065
4066         if (disable_tracing)
4067                 ftrace_kill();
4068
4069         for_each_tracing_cpu(cpu) {
4070                 atomic_inc(&global_trace.data[cpu]->disabled);
4071         }
4072
4073         old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4074
4075         /* don't look at user memory in panic mode */
4076         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4077
4078         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4079
4080         /* Simulate the iterator */
4081         iter.tr = &global_trace;
4082         iter.trace = current_trace;
4083         iter.cpu_file = TRACE_PIPE_ALL_CPU;
4084
4085         /*
4086          * We need to stop all tracing on all CPUS to read the
4087          * the next buffer. This is a bit expensive, but is
4088          * not done often. We fill all what we can read,
4089          * and then release the locks again.
4090          */
4091
4092         while (!trace_empty(&iter)) {
4093
4094                 if (!cnt)
4095                         printk(KERN_TRACE "---------------------------------\n");
4096
4097                 cnt++;
4098
4099                 /* reset all but tr, trace, and overruns */
4100                 memset(&iter.seq, 0,
4101                        sizeof(struct trace_iterator) -
4102                        offsetof(struct trace_iterator, seq));
4103                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4104                 iter.pos = -1;
4105
4106                 if (find_next_entry_inc(&iter) != NULL) {
4107                         print_trace_line(&iter);
4108                         trace_consume(&iter);
4109                 }
4110
4111                 trace_printk_seq(&iter.seq);
4112         }
4113
4114         if (!cnt)
4115                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4116         else
4117                 printk(KERN_TRACE "---------------------------------\n");
4118
4119         /* Re-enable tracing if requested */
4120         if (!disable_tracing) {
4121                 trace_flags |= old_userobj;
4122
4123                 for_each_tracing_cpu(cpu) {
4124                         atomic_dec(&global_trace.data[cpu]->disabled);
4125                 }
4126                 tracing_on();
4127         }
4128
4129  out:
4130         spin_unlock_irqrestore(&ftrace_dump_lock, flags);
4131 }
4132
4133 /* By default: disable tracing after the dump */
4134 void ftrace_dump(void)
4135 {
4136         __ftrace_dump(true);
4137 }
4138
4139 __init static int tracer_alloc_buffers(void)
4140 {
4141         struct trace_array_cpu *data;
4142         int ring_buf_size;
4143         int i;
4144         int ret = -ENOMEM;
4145
4146         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4147                 goto out;
4148
4149         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4150                 goto out_free_buffer_mask;
4151
4152         if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4153                 goto out_free_tracing_cpumask;
4154
4155         /* To save memory, keep the ring buffer size to its minimum */
4156         if (ring_buffer_expanded)
4157                 ring_buf_size = trace_buf_size;
4158         else
4159                 ring_buf_size = 1;
4160
4161         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4162         cpumask_copy(tracing_cpumask, cpu_all_mask);
4163         cpumask_clear(tracing_reader_cpumask);
4164
4165         /* TODO: make the number of buffers hot pluggable with CPUS */
4166         global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4167                                                    TRACE_BUFFER_FLAGS);
4168         if (!global_trace.buffer) {
4169                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4170                 WARN_ON(1);
4171                 goto out_free_cpumask;
4172         }
4173         global_trace.entries = ring_buffer_size(global_trace.buffer);
4174
4175
4176 #ifdef CONFIG_TRACER_MAX_TRACE
4177         max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4178                                              TRACE_BUFFER_FLAGS);
4179         if (!max_tr.buffer) {
4180                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4181                 WARN_ON(1);
4182                 ring_buffer_free(global_trace.buffer);
4183                 goto out_free_cpumask;
4184         }
4185         max_tr.entries = ring_buffer_size(max_tr.buffer);
4186         WARN_ON(max_tr.entries != global_trace.entries);
4187 #endif
4188
4189         /* Allocate the first page for all buffers */
4190         for_each_tracing_cpu(i) {
4191                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4192                 max_tr.data[i] = &per_cpu(max_data, i);
4193         }
4194
4195         trace_init_cmdlines();
4196
4197         register_tracer(&nop_trace);
4198         current_trace = &nop_trace;
4199 #ifdef CONFIG_BOOT_TRACER
4200         register_tracer(&boot_tracer);
4201 #endif
4202         /* All seems OK, enable tracing */
4203         tracing_disabled = 0;
4204
4205         atomic_notifier_chain_register(&panic_notifier_list,
4206                                        &trace_panic_notifier);
4207
4208         register_die_notifier(&trace_die_notifier);
4209
4210         return 0;
4211
4212 out_free_cpumask:
4213         free_cpumask_var(tracing_reader_cpumask);
4214 out_free_tracing_cpumask:
4215         free_cpumask_var(tracing_cpumask);
4216 out_free_buffer_mask:
4217         free_cpumask_var(tracing_buffer_mask);
4218 out:
4219         return ret;
4220 }
4221
4222 __init static int clear_boot_tracer(void)
4223 {
4224         /*
4225          * The default tracer at boot buffer is an init section.
4226          * This function is called in lateinit. If we did not
4227          * find the boot tracer, then clear it out, to prevent
4228          * later registration from accessing the buffer that is
4229          * about to be freed.
4230          */
4231         if (!default_bootup_tracer)
4232                 return 0;
4233
4234         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4235                default_bootup_tracer);
4236         default_bootup_tracer = NULL;
4237
4238         return 0;
4239 }
4240
4241 early_initcall(tracer_alloc_buffers);
4242 fs_initcall(tracer_init_debugfs);
4243 late_initcall(clear_boot_tracer);