ftrace: stacktrace fix
[linux-2.6.git] / kernel / trace / trace.c
1 /*
2  * ring buffer based function tracer
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally taken from the RT patch by:
8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code from the latency_tracer, that is:
11  *  Copyright (C) 2004-2006 Ingo Molnar
12  *  Copyright (C) 2004 William Lee Irwin III
13  */
14 #include <linux/utsrelease.h>
15 #include <linux/kallsyms.h>
16 #include <linux/seq_file.h>
17 #include <linux/debugfs.h>
18 #include <linux/pagemap.h>
19 #include <linux/hardirq.h>
20 #include <linux/linkage.h>
21 #include <linux/uaccess.h>
22 #include <linux/ftrace.h>
23 #include <linux/module.h>
24 #include <linux/percpu.h>
25 #include <linux/ctype.h>
26 #include <linux/init.h>
27 #include <linux/poll.h>
28 #include <linux/gfp.h>
29 #include <linux/fs.h>
30 #include <linux/writeback.h>
31
32 #include <linux/stacktrace.h>
33
34 #include "trace.h"
35
36 unsigned long __read_mostly     tracing_max_latency = (cycle_t)ULONG_MAX;
37 unsigned long __read_mostly     tracing_thresh;
38
39 static unsigned long __read_mostly      tracing_nr_buffers;
40 static cpumask_t __read_mostly          tracing_buffer_mask;
41
42 #define for_each_tracing_cpu(cpu)       \
43         for_each_cpu_mask(cpu, tracing_buffer_mask)
44
45 /* dummy trace to disable tracing */
46 static struct tracer no_tracer __read_mostly = {
47         .name           = "none",
48 };
49
50 static int trace_alloc_page(void);
51 static int trace_free_page(void);
52
53 static int tracing_disabled = 1;
54
55 static unsigned long tracing_pages_allocated;
56
57 long
58 ns2usecs(cycle_t nsec)
59 {
60         nsec += 500;
61         do_div(nsec, 1000);
62         return nsec;
63 }
64
65 cycle_t ftrace_now(int cpu)
66 {
67         return cpu_clock(cpu);
68 }
69
70 /*
71  * The global_trace is the descriptor that holds the tracing
72  * buffers for the live tracing. For each CPU, it contains
73  * a link list of pages that will store trace entries. The
74  * page descriptor of the pages in the memory is used to hold
75  * the link list by linking the lru item in the page descriptor
76  * to each of the pages in the buffer per CPU.
77  *
78  * For each active CPU there is a data field that holds the
79  * pages for the buffer for that CPU. Each CPU has the same number
80  * of pages allocated for its buffer.
81  */
82 static struct trace_array       global_trace;
83
84 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
85
86 /*
87  * The max_tr is used to snapshot the global_trace when a maximum
88  * latency is reached. Some tracers will use this to store a maximum
89  * trace while it continues examining live traces.
90  *
91  * The buffers for the max_tr are set up the same as the global_trace.
92  * When a snapshot is taken, the link list of the max_tr is swapped
93  * with the link list of the global_trace and the buffers are reset for
94  * the global_trace so the tracing can continue.
95  */
96 static struct trace_array       max_tr;
97
98 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
99
100 /* tracer_enabled is used to toggle activation of a tracer */
101 static int                      tracer_enabled = 1;
102
103 /*
104  * trace_nr_entries is the number of entries that is allocated
105  * for a buffer. Note, the number of entries is always rounded
106  * to ENTRIES_PER_PAGE.
107  */
108 static unsigned long            trace_nr_entries = 65536UL;
109
110 /* trace_types holds a link list of available tracers. */
111 static struct tracer            *trace_types __read_mostly;
112
113 /* current_trace points to the tracer that is currently active */
114 static struct tracer            *current_trace __read_mostly;
115
116 /*
117  * max_tracer_type_len is used to simplify the allocating of
118  * buffers to read userspace tracer names. We keep track of
119  * the longest tracer name registered.
120  */
121 static int                      max_tracer_type_len;
122
123 /*
124  * trace_types_lock is used to protect the trace_types list.
125  * This lock is also used to keep user access serialized.
126  * Accesses from userspace will grab this lock while userspace
127  * activities happen inside the kernel.
128  */
129 static DEFINE_MUTEX(trace_types_lock);
130
131 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
132 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
133
134 /* trace_flags holds iter_ctrl options */
135 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
136
137 /**
138  * trace_wake_up - wake up tasks waiting for trace input
139  *
140  * Simply wakes up any task that is blocked on the trace_wait
141  * queue. These is used with trace_poll for tasks polling the trace.
142  */
143 void trace_wake_up(void)
144 {
145         /*
146          * The runqueue_is_locked() can fail, but this is the best we
147          * have for now:
148          */
149         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
150                 wake_up(&trace_wait);
151 }
152
153 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
154
155 static int __init set_nr_entries(char *str)
156 {
157         unsigned long nr_entries;
158         int ret;
159
160         if (!str)
161                 return 0;
162         ret = strict_strtoul(str, 0, &nr_entries);
163         /* nr_entries can not be zero */
164         if (ret < 0 || nr_entries == 0)
165                 return 0;
166         trace_nr_entries = nr_entries;
167         return 1;
168 }
169 __setup("trace_entries=", set_nr_entries);
170
171 unsigned long nsecs_to_usecs(unsigned long nsecs)
172 {
173         return nsecs / 1000;
174 }
175
176 /*
177  * trace_flag_type is an enumeration that holds different
178  * states when a trace occurs. These are:
179  *  IRQS_OFF    - interrupts were disabled
180  *  NEED_RESCED - reschedule is requested
181  *  HARDIRQ     - inside an interrupt handler
182  *  SOFTIRQ     - inside a softirq handler
183  */
184 enum trace_flag_type {
185         TRACE_FLAG_IRQS_OFF             = 0x01,
186         TRACE_FLAG_NEED_RESCHED         = 0x02,
187         TRACE_FLAG_HARDIRQ              = 0x04,
188         TRACE_FLAG_SOFTIRQ              = 0x08,
189 };
190
191 /*
192  * TRACE_ITER_SYM_MASK masks the options in trace_flags that
193  * control the output of kernel symbols.
194  */
195 #define TRACE_ITER_SYM_MASK \
196         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
197
198 /* These must match the bit postions in trace_iterator_flags */
199 static const char *trace_options[] = {
200         "print-parent",
201         "sym-offset",
202         "sym-addr",
203         "verbose",
204         "raw",
205         "hex",
206         "bin",
207         "block",
208         "stacktrace",
209         "sched-tree",
210         NULL
211 };
212
213 /*
214  * ftrace_max_lock is used to protect the swapping of buffers
215  * when taking a max snapshot. The buffers themselves are
216  * protected by per_cpu spinlocks. But the action of the swap
217  * needs its own lock.
218  *
219  * This is defined as a raw_spinlock_t in order to help
220  * with performance when lockdep debugging is enabled.
221  */
222 static raw_spinlock_t ftrace_max_lock =
223         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
224
225 /*
226  * Copy the new maximum trace into the separate maximum-trace
227  * structure. (this way the maximum trace is permanently saved,
228  * for later retrieval via /debugfs/tracing/latency_trace)
229  */
230 static void
231 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
232 {
233         struct trace_array_cpu *data = tr->data[cpu];
234
235         max_tr.cpu = cpu;
236         max_tr.time_start = data->preempt_timestamp;
237
238         data = max_tr.data[cpu];
239         data->saved_latency = tracing_max_latency;
240
241         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
242         data->pid = tsk->pid;
243         data->uid = tsk->uid;
244         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
245         data->policy = tsk->policy;
246         data->rt_priority = tsk->rt_priority;
247
248         /* record this tasks comm */
249         tracing_record_cmdline(current);
250 }
251
252 /**
253  * check_pages - integrity check of trace buffers
254  *
255  * As a safty measure we check to make sure the data pages have not
256  * been corrupted. TODO: configure to disable this because it adds
257  * a bit of overhead.
258  */
259 void check_pages(struct trace_array_cpu *data)
260 {
261         struct page *page, *tmp;
262
263         BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
264         BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
265
266         list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
267                 BUG_ON(page->lru.next->prev != &page->lru);
268                 BUG_ON(page->lru.prev->next != &page->lru);
269         }
270 }
271
272 /**
273  * head_page - page address of the first page in per_cpu buffer.
274  *
275  * head_page returns the page address of the first page in
276  * a per_cpu buffer. This also preforms various consistency
277  * checks to make sure the buffer has not been corrupted.
278  */
279 void *head_page(struct trace_array_cpu *data)
280 {
281         struct page *page;
282
283         check_pages(data);
284         if (list_empty(&data->trace_pages))
285                 return NULL;
286
287         page = list_entry(data->trace_pages.next, struct page, lru);
288         BUG_ON(&page->lru == &data->trace_pages);
289
290         return page_address(page);
291 }
292
293 /**
294  * trace_seq_printf - sequence printing of trace information
295  * @s: trace sequence descriptor
296  * @fmt: printf format string
297  *
298  * The tracer may use either sequence operations or its own
299  * copy to user routines. To simplify formating of a trace
300  * trace_seq_printf is used to store strings into a special
301  * buffer (@s). Then the output may be either used by
302  * the sequencer or pulled into another buffer.
303  */
304 int
305 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
306 {
307         int len = (PAGE_SIZE - 1) - s->len;
308         va_list ap;
309         int ret;
310
311         if (!len)
312                 return 0;
313
314         va_start(ap, fmt);
315         ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
316         va_end(ap);
317
318         /* If we can't write it all, don't bother writing anything */
319         if (ret >= len)
320                 return 0;
321
322         s->len += ret;
323
324         return len;
325 }
326
327 /**
328  * trace_seq_puts - trace sequence printing of simple string
329  * @s: trace sequence descriptor
330  * @str: simple string to record
331  *
332  * The tracer may use either the sequence operations or its own
333  * copy to user routines. This function records a simple string
334  * into a special buffer (@s) for later retrieval by a sequencer
335  * or other mechanism.
336  */
337 static int
338 trace_seq_puts(struct trace_seq *s, const char *str)
339 {
340         int len = strlen(str);
341
342         if (len > ((PAGE_SIZE - 1) - s->len))
343                 return 0;
344
345         memcpy(s->buffer + s->len, str, len);
346         s->len += len;
347
348         return len;
349 }
350
351 static int
352 trace_seq_putc(struct trace_seq *s, unsigned char c)
353 {
354         if (s->len >= (PAGE_SIZE - 1))
355                 return 0;
356
357         s->buffer[s->len++] = c;
358
359         return 1;
360 }
361
362 static int
363 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
364 {
365         if (len > ((PAGE_SIZE - 1) - s->len))
366                 return 0;
367
368         memcpy(s->buffer + s->len, mem, len);
369         s->len += len;
370
371         return len;
372 }
373
374 #define HEX_CHARS 17
375 static const char hex2asc[] = "0123456789abcdef";
376
377 static int
378 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
379 {
380         unsigned char hex[HEX_CHARS];
381         unsigned char *data = mem;
382         unsigned char byte;
383         int i, j;
384
385         BUG_ON(len >= HEX_CHARS);
386
387 #ifdef __BIG_ENDIAN
388         for (i = 0, j = 0; i < len; i++) {
389 #else
390         for (i = len-1, j = 0; i >= 0; i--) {
391 #endif
392                 byte = data[i];
393
394                 hex[j++] = hex2asc[byte & 0x0f];
395                 hex[j++] = hex2asc[byte >> 4];
396         }
397         hex[j++] = ' ';
398
399         return trace_seq_putmem(s, hex, j);
400 }
401
402 static void
403 trace_seq_reset(struct trace_seq *s)
404 {
405         s->len = 0;
406         s->readpos = 0;
407 }
408
409 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
410 {
411         int len;
412         int ret;
413
414         if (s->len <= s->readpos)
415                 return -EBUSY;
416
417         len = s->len - s->readpos;
418         if (cnt > len)
419                 cnt = len;
420         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
421         if (ret)
422                 return -EFAULT;
423
424         s->readpos += len;
425         return cnt;
426 }
427
428 static void
429 trace_print_seq(struct seq_file *m, struct trace_seq *s)
430 {
431         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
432
433         s->buffer[len] = 0;
434         seq_puts(m, s->buffer);
435
436         trace_seq_reset(s);
437 }
438
439 /*
440  * flip the trace buffers between two trace descriptors.
441  * This usually is the buffers between the global_trace and
442  * the max_tr to record a snapshot of a current trace.
443  *
444  * The ftrace_max_lock must be held.
445  */
446 static void
447 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
448 {
449         struct list_head flip_pages;
450
451         INIT_LIST_HEAD(&flip_pages);
452
453         memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
454                 sizeof(struct trace_array_cpu) -
455                 offsetof(struct trace_array_cpu, trace_head_idx));
456
457         check_pages(tr1);
458         check_pages(tr2);
459         list_splice_init(&tr1->trace_pages, &flip_pages);
460         list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
461         list_splice_init(&flip_pages, &tr2->trace_pages);
462         BUG_ON(!list_empty(&flip_pages));
463         check_pages(tr1);
464         check_pages(tr2);
465 }
466
467 /**
468  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
469  * @tr: tracer
470  * @tsk: the task with the latency
471  * @cpu: The cpu that initiated the trace.
472  *
473  * Flip the buffers between the @tr and the max_tr and record information
474  * about which task was the cause of this latency.
475  */
476 void
477 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
478 {
479         struct trace_array_cpu *data;
480         int i;
481
482         WARN_ON_ONCE(!irqs_disabled());
483         __raw_spin_lock(&ftrace_max_lock);
484         /* clear out all the previous traces */
485         for_each_tracing_cpu(i) {
486                 data = tr->data[i];
487                 flip_trace(max_tr.data[i], data);
488                 tracing_reset(data);
489         }
490
491         __update_max_tr(tr, tsk, cpu);
492         __raw_spin_unlock(&ftrace_max_lock);
493 }
494
495 /**
496  * update_max_tr_single - only copy one trace over, and reset the rest
497  * @tr - tracer
498  * @tsk - task with the latency
499  * @cpu - the cpu of the buffer to copy.
500  *
501  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
502  */
503 void
504 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
505 {
506         struct trace_array_cpu *data = tr->data[cpu];
507         int i;
508
509         WARN_ON_ONCE(!irqs_disabled());
510         __raw_spin_lock(&ftrace_max_lock);
511         for_each_tracing_cpu(i)
512                 tracing_reset(max_tr.data[i]);
513
514         flip_trace(max_tr.data[cpu], data);
515         tracing_reset(data);
516
517         __update_max_tr(tr, tsk, cpu);
518         __raw_spin_unlock(&ftrace_max_lock);
519 }
520
521 /**
522  * register_tracer - register a tracer with the ftrace system.
523  * @type - the plugin for the tracer
524  *
525  * Register a new plugin tracer.
526  */
527 int register_tracer(struct tracer *type)
528 {
529         struct tracer *t;
530         int len;
531         int ret = 0;
532
533         if (!type->name) {
534                 pr_info("Tracer must have a name\n");
535                 return -1;
536         }
537
538         mutex_lock(&trace_types_lock);
539         for (t = trace_types; t; t = t->next) {
540                 if (strcmp(type->name, t->name) == 0) {
541                         /* already found */
542                         pr_info("Trace %s already registered\n",
543                                 type->name);
544                         ret = -1;
545                         goto out;
546                 }
547         }
548
549 #ifdef CONFIG_FTRACE_STARTUP_TEST
550         if (type->selftest) {
551                 struct tracer *saved_tracer = current_trace;
552                 struct trace_array_cpu *data;
553                 struct trace_array *tr = &global_trace;
554                 int saved_ctrl = tr->ctrl;
555                 int i;
556                 /*
557                  * Run a selftest on this tracer.
558                  * Here we reset the trace buffer, and set the current
559                  * tracer to be this tracer. The tracer can then run some
560                  * internal tracing to verify that everything is in order.
561                  * If we fail, we do not register this tracer.
562                  */
563                 for_each_tracing_cpu(i) {
564                         data = tr->data[i];
565                         if (!head_page(data))
566                                 continue;
567                         tracing_reset(data);
568                 }
569                 current_trace = type;
570                 tr->ctrl = 0;
571                 /* the test is responsible for initializing and enabling */
572                 pr_info("Testing tracer %s: ", type->name);
573                 ret = type->selftest(type, tr);
574                 /* the test is responsible for resetting too */
575                 current_trace = saved_tracer;
576                 tr->ctrl = saved_ctrl;
577                 if (ret) {
578                         printk(KERN_CONT "FAILED!\n");
579                         goto out;
580                 }
581                 /* Only reset on passing, to avoid touching corrupted buffers */
582                 for_each_tracing_cpu(i) {
583                         data = tr->data[i];
584                         if (!head_page(data))
585                                 continue;
586                         tracing_reset(data);
587                 }
588                 printk(KERN_CONT "PASSED\n");
589         }
590 #endif
591
592         type->next = trace_types;
593         trace_types = type;
594         len = strlen(type->name);
595         if (len > max_tracer_type_len)
596                 max_tracer_type_len = len;
597
598  out:
599         mutex_unlock(&trace_types_lock);
600
601         return ret;
602 }
603
604 void unregister_tracer(struct tracer *type)
605 {
606         struct tracer **t;
607         int len;
608
609         mutex_lock(&trace_types_lock);
610         for (t = &trace_types; *t; t = &(*t)->next) {
611                 if (*t == type)
612                         goto found;
613         }
614         pr_info("Trace %s not registered\n", type->name);
615         goto out;
616
617  found:
618         *t = (*t)->next;
619         if (strlen(type->name) != max_tracer_type_len)
620                 goto out;
621
622         max_tracer_type_len = 0;
623         for (t = &trace_types; *t; t = &(*t)->next) {
624                 len = strlen((*t)->name);
625                 if (len > max_tracer_type_len)
626                         max_tracer_type_len = len;
627         }
628  out:
629         mutex_unlock(&trace_types_lock);
630 }
631
632 void tracing_reset(struct trace_array_cpu *data)
633 {
634         data->trace_idx = 0;
635         data->overrun = 0;
636         data->trace_head = data->trace_tail = head_page(data);
637         data->trace_head_idx = 0;
638         data->trace_tail_idx = 0;
639 }
640
641 #define SAVED_CMDLINES 128
642 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
643 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
644 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
645 static int cmdline_idx;
646 static DEFINE_SPINLOCK(trace_cmdline_lock);
647
648 /* trace in all context switches */
649 atomic_t trace_record_cmdline_enabled __read_mostly;
650
651 /* temporary disable recording */
652 atomic_t trace_record_cmdline_disabled __read_mostly;
653
654 static void trace_init_cmdlines(void)
655 {
656         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
657         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
658         cmdline_idx = 0;
659 }
660
661 void trace_stop_cmdline_recording(void);
662
663 static void trace_save_cmdline(struct task_struct *tsk)
664 {
665         unsigned map;
666         unsigned idx;
667
668         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
669                 return;
670
671         /*
672          * It's not the end of the world if we don't get
673          * the lock, but we also don't want to spin
674          * nor do we want to disable interrupts,
675          * so if we miss here, then better luck next time.
676          */
677         if (!spin_trylock(&trace_cmdline_lock))
678                 return;
679
680         idx = map_pid_to_cmdline[tsk->pid];
681         if (idx >= SAVED_CMDLINES) {
682                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
683
684                 map = map_cmdline_to_pid[idx];
685                 if (map <= PID_MAX_DEFAULT)
686                         map_pid_to_cmdline[map] = (unsigned)-1;
687
688                 map_pid_to_cmdline[tsk->pid] = idx;
689
690                 cmdline_idx = idx;
691         }
692
693         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
694
695         spin_unlock(&trace_cmdline_lock);
696 }
697
698 static char *trace_find_cmdline(int pid)
699 {
700         char *cmdline = "<...>";
701         unsigned map;
702
703         if (!pid)
704                 return "<idle>";
705
706         if (pid > PID_MAX_DEFAULT)
707                 goto out;
708
709         map = map_pid_to_cmdline[pid];
710         if (map >= SAVED_CMDLINES)
711                 goto out;
712
713         cmdline = saved_cmdlines[map];
714
715  out:
716         return cmdline;
717 }
718
719 void tracing_record_cmdline(struct task_struct *tsk)
720 {
721         if (atomic_read(&trace_record_cmdline_disabled))
722                 return;
723
724         trace_save_cmdline(tsk);
725 }
726
727 static inline struct list_head *
728 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
729 {
730         /*
731          * Roundrobin - but skip the head (which is not a real page):
732          */
733         next = next->next;
734         if (unlikely(next == &data->trace_pages))
735                 next = next->next;
736         BUG_ON(next == &data->trace_pages);
737
738         return next;
739 }
740
741 static inline void *
742 trace_next_page(struct trace_array_cpu *data, void *addr)
743 {
744         struct list_head *next;
745         struct page *page;
746
747         page = virt_to_page(addr);
748
749         next = trace_next_list(data, &page->lru);
750         page = list_entry(next, struct page, lru);
751
752         return page_address(page);
753 }
754
755 static inline struct trace_entry *
756 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
757 {
758         unsigned long idx, idx_next;
759         struct trace_entry *entry;
760
761         data->trace_idx++;
762         idx = data->trace_head_idx;
763         idx_next = idx + 1;
764
765         BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
766
767         entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
768
769         if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
770                 data->trace_head = trace_next_page(data, data->trace_head);
771                 idx_next = 0;
772         }
773
774         if (data->trace_head == data->trace_tail &&
775             idx_next == data->trace_tail_idx) {
776                 /* overrun */
777                 data->overrun++;
778                 data->trace_tail_idx++;
779                 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
780                         data->trace_tail =
781                                 trace_next_page(data, data->trace_tail);
782                         data->trace_tail_idx = 0;
783                 }
784         }
785
786         data->trace_head_idx = idx_next;
787
788         return entry;
789 }
790
791 static inline void
792 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
793 {
794         struct task_struct *tsk = current;
795         unsigned long pc;
796
797         pc = preempt_count();
798
799         entry->preempt_count    = pc & 0xff;
800         entry->pid              = (tsk) ? tsk->pid : 0;
801         entry->t                = ftrace_now(raw_smp_processor_id());
802         entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
803                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
804                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
805                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
806 }
807
808 void
809 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
810                unsigned long ip, unsigned long parent_ip, unsigned long flags)
811 {
812         struct trace_entry *entry;
813         unsigned long irq_flags;
814
815         raw_local_irq_save(irq_flags);
816         __raw_spin_lock(&data->lock);
817         entry                   = tracing_get_trace_entry(tr, data);
818         tracing_generic_entry_update(entry, flags);
819         entry->type             = TRACE_FN;
820         entry->fn.ip            = ip;
821         entry->fn.parent_ip     = parent_ip;
822         __raw_spin_unlock(&data->lock);
823         raw_local_irq_restore(irq_flags);
824 }
825
826 void
827 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
828        unsigned long ip, unsigned long parent_ip, unsigned long flags)
829 {
830         if (likely(!atomic_read(&data->disabled)))
831                 trace_function(tr, data, ip, parent_ip, flags);
832 }
833
834 void __trace_stack(struct trace_array *tr,
835                    struct trace_array_cpu *data,
836                    unsigned long flags,
837                    int skip)
838 {
839         struct trace_entry *entry;
840         struct stack_trace trace;
841
842         if (!(trace_flags & TRACE_ITER_STACKTRACE))
843                 return;
844
845         entry                   = tracing_get_trace_entry(tr, data);
846         tracing_generic_entry_update(entry, flags);
847         entry->type             = TRACE_STACK;
848
849         memset(&entry->stack, 0, sizeof(entry->stack));
850
851         trace.nr_entries        = 0;
852         trace.max_entries       = FTRACE_STACK_ENTRIES;
853         trace.skip              = skip;
854         trace.entries           = entry->stack.caller;
855
856         save_stack_trace(&trace);
857 }
858
859 void
860 __trace_special(void *__tr, void *__data,
861                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
862 {
863         struct trace_array_cpu *data = __data;
864         struct trace_array *tr = __tr;
865         struct trace_entry *entry;
866         unsigned long irq_flags;
867
868         raw_local_irq_save(irq_flags);
869         __raw_spin_lock(&data->lock);
870         entry                   = tracing_get_trace_entry(tr, data);
871         tracing_generic_entry_update(entry, 0);
872         entry->type             = TRACE_SPECIAL;
873         entry->special.arg1     = arg1;
874         entry->special.arg2     = arg2;
875         entry->special.arg3     = arg3;
876         __trace_stack(tr, data, irq_flags, 4);
877         __raw_spin_unlock(&data->lock);
878         raw_local_irq_restore(irq_flags);
879
880         trace_wake_up();
881 }
882
883 void
884 tracing_sched_switch_trace(struct trace_array *tr,
885                            struct trace_array_cpu *data,
886                            struct task_struct *prev,
887                            struct task_struct *next,
888                            unsigned long flags)
889 {
890         struct trace_entry *entry;
891         unsigned long irq_flags;
892
893         raw_local_irq_save(irq_flags);
894         __raw_spin_lock(&data->lock);
895         entry                   = tracing_get_trace_entry(tr, data);
896         tracing_generic_entry_update(entry, flags);
897         entry->type             = TRACE_CTX;
898         entry->ctx.prev_pid     = prev->pid;
899         entry->ctx.prev_prio    = prev->prio;
900         entry->ctx.prev_state   = prev->state;
901         entry->ctx.next_pid     = next->pid;
902         entry->ctx.next_prio    = next->prio;
903         entry->ctx.next_state   = next->state;
904         __trace_stack(tr, data, flags, 5);
905         __raw_spin_unlock(&data->lock);
906         raw_local_irq_restore(irq_flags);
907 }
908
909 void
910 tracing_sched_wakeup_trace(struct trace_array *tr,
911                            struct trace_array_cpu *data,
912                            struct task_struct *wakee,
913                            struct task_struct *curr,
914                            unsigned long flags)
915 {
916         struct trace_entry *entry;
917         unsigned long irq_flags;
918
919         raw_local_irq_save(irq_flags);
920         __raw_spin_lock(&data->lock);
921         entry                   = tracing_get_trace_entry(tr, data);
922         tracing_generic_entry_update(entry, flags);
923         entry->type             = TRACE_WAKE;
924         entry->ctx.prev_pid     = curr->pid;
925         entry->ctx.prev_prio    = curr->prio;
926         entry->ctx.prev_state   = curr->state;
927         entry->ctx.next_pid     = wakee->pid;
928         entry->ctx.next_prio    = wakee->prio;
929         entry->ctx.next_state   = wakee->state;
930         __trace_stack(tr, data, flags, 6);
931         __raw_spin_unlock(&data->lock);
932         raw_local_irq_restore(irq_flags);
933
934         trace_wake_up();
935 }
936
937 #ifdef CONFIG_FTRACE
938 static void
939 function_trace_call(unsigned long ip, unsigned long parent_ip)
940 {
941         struct trace_array *tr = &global_trace;
942         struct trace_array_cpu *data;
943         unsigned long flags;
944         long disabled;
945         int cpu;
946
947         if (unlikely(!tracer_enabled))
948                 return;
949
950         local_irq_save(flags);
951         cpu = raw_smp_processor_id();
952         data = tr->data[cpu];
953         disabled = atomic_inc_return(&data->disabled);
954
955         if (likely(disabled == 1))
956                 trace_function(tr, data, ip, parent_ip, flags);
957
958         atomic_dec(&data->disabled);
959         local_irq_restore(flags);
960 }
961
962 static struct ftrace_ops trace_ops __read_mostly =
963 {
964         .func = function_trace_call,
965 };
966
967 void tracing_start_function_trace(void)
968 {
969         register_ftrace_function(&trace_ops);
970 }
971
972 void tracing_stop_function_trace(void)
973 {
974         unregister_ftrace_function(&trace_ops);
975 }
976 #endif
977
978 enum trace_file_type {
979         TRACE_FILE_LAT_FMT      = 1,
980 };
981
982 static struct trace_entry *
983 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
984                 struct trace_iterator *iter, int cpu)
985 {
986         struct page *page;
987         struct trace_entry *array;
988
989         if (iter->next_idx[cpu] >= tr->entries ||
990             iter->next_idx[cpu] >= data->trace_idx ||
991             (data->trace_head == data->trace_tail &&
992              data->trace_head_idx == data->trace_tail_idx))
993                 return NULL;
994
995         if (!iter->next_page[cpu]) {
996                 /* Initialize the iterator for this cpu trace buffer */
997                 WARN_ON(!data->trace_tail);
998                 page = virt_to_page(data->trace_tail);
999                 iter->next_page[cpu] = &page->lru;
1000                 iter->next_page_idx[cpu] = data->trace_tail_idx;
1001         }
1002
1003         page = list_entry(iter->next_page[cpu], struct page, lru);
1004         BUG_ON(&data->trace_pages == &page->lru);
1005
1006         array = page_address(page);
1007
1008         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
1009         return &array[iter->next_page_idx[cpu]];
1010 }
1011
1012 static struct trace_entry *
1013 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
1014 {
1015         struct trace_array *tr = iter->tr;
1016         struct trace_entry *ent, *next = NULL;
1017         int next_cpu = -1;
1018         int cpu;
1019
1020         for_each_tracing_cpu(cpu) {
1021                 if (!head_page(tr->data[cpu]))
1022                         continue;
1023                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
1024                 /*
1025                  * Pick the entry with the smallest timestamp:
1026                  */
1027                 if (ent && (!next || ent->t < next->t)) {
1028                         next = ent;
1029                         next_cpu = cpu;
1030                 }
1031         }
1032
1033         if (ent_cpu)
1034                 *ent_cpu = next_cpu;
1035
1036         return next;
1037 }
1038
1039 static void trace_iterator_increment(struct trace_iterator *iter)
1040 {
1041         iter->idx++;
1042         iter->next_idx[iter->cpu]++;
1043         iter->next_page_idx[iter->cpu]++;
1044
1045         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
1046                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1047
1048                 iter->next_page_idx[iter->cpu] = 0;
1049                 iter->next_page[iter->cpu] =
1050                         trace_next_list(data, iter->next_page[iter->cpu]);
1051         }
1052 }
1053
1054 static void trace_consume(struct trace_iterator *iter)
1055 {
1056         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1057
1058         data->trace_tail_idx++;
1059         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
1060                 data->trace_tail = trace_next_page(data, data->trace_tail);
1061                 data->trace_tail_idx = 0;
1062         }
1063
1064         /* Check if we empty it, then reset the index */
1065         if (data->trace_head == data->trace_tail &&
1066             data->trace_head_idx == data->trace_tail_idx)
1067                 data->trace_idx = 0;
1068 }
1069
1070 static void *find_next_entry_inc(struct trace_iterator *iter)
1071 {
1072         struct trace_entry *next;
1073         int next_cpu = -1;
1074
1075         next = find_next_entry(iter, &next_cpu);
1076
1077         iter->prev_ent = iter->ent;
1078         iter->prev_cpu = iter->cpu;
1079
1080         iter->ent = next;
1081         iter->cpu = next_cpu;
1082
1083         if (next)
1084                 trace_iterator_increment(iter);
1085
1086         return next ? iter : NULL;
1087 }
1088
1089 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1090 {
1091         struct trace_iterator *iter = m->private;
1092         void *last_ent = iter->ent;
1093         int i = (int)*pos;
1094         void *ent;
1095
1096         (*pos)++;
1097
1098         /* can't go backwards */
1099         if (iter->idx > i)
1100                 return NULL;
1101
1102         if (iter->idx < 0)
1103                 ent = find_next_entry_inc(iter);
1104         else
1105                 ent = iter;
1106
1107         while (ent && iter->idx < i)
1108                 ent = find_next_entry_inc(iter);
1109
1110         iter->pos = *pos;
1111
1112         if (last_ent && !ent)
1113                 seq_puts(m, "\n\nvim:ft=help\n");
1114
1115         return ent;
1116 }
1117
1118 static void *s_start(struct seq_file *m, loff_t *pos)
1119 {
1120         struct trace_iterator *iter = m->private;
1121         void *p = NULL;
1122         loff_t l = 0;
1123         int i;
1124
1125         mutex_lock(&trace_types_lock);
1126
1127         if (!current_trace || current_trace != iter->trace) {
1128                 mutex_unlock(&trace_types_lock);
1129                 return NULL;
1130         }
1131
1132         atomic_inc(&trace_record_cmdline_disabled);
1133
1134         /* let the tracer grab locks here if needed */
1135         if (current_trace->start)
1136                 current_trace->start(iter);
1137
1138         if (*pos != iter->pos) {
1139                 iter->ent = NULL;
1140                 iter->cpu = 0;
1141                 iter->idx = -1;
1142                 iter->prev_ent = NULL;
1143                 iter->prev_cpu = -1;
1144
1145                 for_each_tracing_cpu(i) {
1146                         iter->next_idx[i] = 0;
1147                         iter->next_page[i] = NULL;
1148                 }
1149
1150                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1151                         ;
1152
1153         } else {
1154                 l = *pos - 1;
1155                 p = s_next(m, p, &l);
1156         }
1157
1158         return p;
1159 }
1160
1161 static void s_stop(struct seq_file *m, void *p)
1162 {
1163         struct trace_iterator *iter = m->private;
1164
1165         atomic_dec(&trace_record_cmdline_disabled);
1166
1167         /* let the tracer release locks here if needed */
1168         if (current_trace && current_trace == iter->trace && iter->trace->stop)
1169                 iter->trace->stop(iter);
1170
1171         mutex_unlock(&trace_types_lock);
1172 }
1173
1174 static int
1175 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1176 {
1177 #ifdef CONFIG_KALLSYMS
1178         char str[KSYM_SYMBOL_LEN];
1179
1180         kallsyms_lookup(address, NULL, NULL, NULL, str);
1181
1182         return trace_seq_printf(s, fmt, str);
1183 #endif
1184         return 1;
1185 }
1186
1187 static int
1188 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1189                      unsigned long address)
1190 {
1191 #ifdef CONFIG_KALLSYMS
1192         char str[KSYM_SYMBOL_LEN];
1193
1194         sprint_symbol(str, address);
1195         return trace_seq_printf(s, fmt, str);
1196 #endif
1197         return 1;
1198 }
1199
1200 #ifndef CONFIG_64BIT
1201 # define IP_FMT "%08lx"
1202 #else
1203 # define IP_FMT "%016lx"
1204 #endif
1205
1206 static int
1207 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1208 {
1209         int ret;
1210
1211         if (!ip)
1212                 return trace_seq_printf(s, "0");
1213
1214         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1215                 ret = seq_print_sym_offset(s, "%s", ip);
1216         else
1217                 ret = seq_print_sym_short(s, "%s", ip);
1218
1219         if (!ret)
1220                 return 0;
1221
1222         if (sym_flags & TRACE_ITER_SYM_ADDR)
1223                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1224         return ret;
1225 }
1226
1227 static void print_lat_help_header(struct seq_file *m)
1228 {
1229         seq_puts(m, "#                _------=> CPU#            \n");
1230         seq_puts(m, "#               / _-----=> irqs-off        \n");
1231         seq_puts(m, "#              | / _----=> need-resched    \n");
1232         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1233         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1234         seq_puts(m, "#              |||| /                      \n");
1235         seq_puts(m, "#              |||||     delay             \n");
1236         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1237         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1238 }
1239
1240 static void print_func_help_header(struct seq_file *m)
1241 {
1242         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1243         seq_puts(m, "#              | |      |          |         |\n");
1244 }
1245
1246
1247 static void
1248 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1249 {
1250         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1251         struct trace_array *tr = iter->tr;
1252         struct trace_array_cpu *data = tr->data[tr->cpu];
1253         struct tracer *type = current_trace;
1254         unsigned long total   = 0;
1255         unsigned long entries = 0;
1256         int cpu;
1257         const char *name = "preemption";
1258
1259         if (type)
1260                 name = type->name;
1261
1262         for_each_tracing_cpu(cpu) {
1263                 if (head_page(tr->data[cpu])) {
1264                         total += tr->data[cpu]->trace_idx;
1265                         if (tr->data[cpu]->trace_idx > tr->entries)
1266                                 entries += tr->entries;
1267                         else
1268                                 entries += tr->data[cpu]->trace_idx;
1269                 }
1270         }
1271
1272         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1273                    name, UTS_RELEASE);
1274         seq_puts(m, "-----------------------------------"
1275                  "---------------------------------\n");
1276         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1277                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1278                    nsecs_to_usecs(data->saved_latency),
1279                    entries,
1280                    total,
1281                    tr->cpu,
1282 #if defined(CONFIG_PREEMPT_NONE)
1283                    "server",
1284 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1285                    "desktop",
1286 #elif defined(CONFIG_PREEMPT_DESKTOP)
1287                    "preempt",
1288 #else
1289                    "unknown",
1290 #endif
1291                    /* These are reserved for later use */
1292                    0, 0, 0, 0);
1293 #ifdef CONFIG_SMP
1294         seq_printf(m, " #P:%d)\n", num_online_cpus());
1295 #else
1296         seq_puts(m, ")\n");
1297 #endif
1298         seq_puts(m, "    -----------------\n");
1299         seq_printf(m, "    | task: %.16s-%d "
1300                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1301                    data->comm, data->pid, data->uid, data->nice,
1302                    data->policy, data->rt_priority);
1303         seq_puts(m, "    -----------------\n");
1304
1305         if (data->critical_start) {
1306                 seq_puts(m, " => started at: ");
1307                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1308                 trace_print_seq(m, &iter->seq);
1309                 seq_puts(m, "\n => ended at:   ");
1310                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1311                 trace_print_seq(m, &iter->seq);
1312                 seq_puts(m, "\n");
1313         }
1314
1315         seq_puts(m, "\n");
1316 }
1317
1318 static void
1319 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1320 {
1321         int hardirq, softirq;
1322         char *comm;
1323
1324         comm = trace_find_cmdline(entry->pid);
1325
1326         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1327         trace_seq_printf(s, "%d", cpu);
1328         trace_seq_printf(s, "%c%c",
1329                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1330                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1331
1332         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1333         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1334         if (hardirq && softirq) {
1335                 trace_seq_putc(s, 'H');
1336         } else {
1337                 if (hardirq) {
1338                         trace_seq_putc(s, 'h');
1339                 } else {
1340                         if (softirq)
1341                                 trace_seq_putc(s, 's');
1342                         else
1343                                 trace_seq_putc(s, '.');
1344                 }
1345         }
1346
1347         if (entry->preempt_count)
1348                 trace_seq_printf(s, "%x", entry->preempt_count);
1349         else
1350                 trace_seq_puts(s, ".");
1351 }
1352
1353 unsigned long preempt_mark_thresh = 100;
1354
1355 static void
1356 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1357                     unsigned long rel_usecs)
1358 {
1359         trace_seq_printf(s, " %4lldus", abs_usecs);
1360         if (rel_usecs > preempt_mark_thresh)
1361                 trace_seq_puts(s, "!: ");
1362         else if (rel_usecs > 1)
1363                 trace_seq_puts(s, "+: ");
1364         else
1365                 trace_seq_puts(s, " : ");
1366 }
1367
1368 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1369
1370 static int
1371 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1372 {
1373         struct trace_seq *s = &iter->seq;
1374         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1375         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1376         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1377         struct trace_entry *entry = iter->ent;
1378         unsigned long abs_usecs;
1379         unsigned long rel_usecs;
1380         char *comm;
1381         int S, T;
1382         int i;
1383         unsigned state;
1384
1385         if (!next_entry)
1386                 next_entry = entry;
1387         rel_usecs = ns2usecs(next_entry->t - entry->t);
1388         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1389
1390         if (verbose) {
1391                 comm = trace_find_cmdline(entry->pid);
1392                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1393                                  " %ld.%03ldms (+%ld.%03ldms): ",
1394                                  comm,
1395                                  entry->pid, cpu, entry->flags,
1396                                  entry->preempt_count, trace_idx,
1397                                  ns2usecs(entry->t),
1398                                  abs_usecs/1000,
1399                                  abs_usecs % 1000, rel_usecs/1000,
1400                                  rel_usecs % 1000);
1401         } else {
1402                 lat_print_generic(s, entry, cpu);
1403                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1404         }
1405         switch (entry->type) {
1406         case TRACE_FN:
1407                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1408                 trace_seq_puts(s, " (");
1409                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1410                 trace_seq_puts(s, ")\n");
1411                 break;
1412         case TRACE_CTX:
1413         case TRACE_WAKE:
1414                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1415                         state_to_char[entry->ctx.next_state] : 'X';
1416
1417                 state = entry->ctx.prev_state ? __ffs(entry->ctx.prev_state) + 1 : 0;
1418                 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1419                 comm = trace_find_cmdline(entry->ctx.next_pid);
1420                 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1421                                  entry->ctx.prev_pid,
1422                                  entry->ctx.prev_prio,
1423                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1424                                  entry->ctx.next_pid,
1425                                  entry->ctx.next_prio,
1426                                  T, comm);
1427                 break;
1428         case TRACE_SPECIAL:
1429                 trace_seq_printf(s, "# %ld %ld %ld\n",
1430                                  entry->special.arg1,
1431                                  entry->special.arg2,
1432                                  entry->special.arg3);
1433                 break;
1434         case TRACE_STACK:
1435                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1436                         if (i)
1437                                 trace_seq_puts(s, " <= ");
1438                         seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1439                 }
1440                 trace_seq_puts(s, "\n");
1441                 break;
1442         default:
1443                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1444         }
1445         return 1;
1446 }
1447
1448 static int print_trace_fmt(struct trace_iterator *iter)
1449 {
1450         struct trace_seq *s = &iter->seq;
1451         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1452         struct trace_entry *entry;
1453         unsigned long usec_rem;
1454         unsigned long long t;
1455         unsigned long secs;
1456         char *comm;
1457         int ret;
1458         int S, T;
1459         int i;
1460
1461         entry = iter->ent;
1462
1463         comm = trace_find_cmdline(iter->ent->pid);
1464
1465         t = ns2usecs(entry->t);
1466         usec_rem = do_div(t, 1000000ULL);
1467         secs = (unsigned long)t;
1468
1469         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1470         if (!ret)
1471                 return 0;
1472         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1473         if (!ret)
1474                 return 0;
1475         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1476         if (!ret)
1477                 return 0;
1478
1479         switch (entry->type) {
1480         case TRACE_FN:
1481                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1482                 if (!ret)
1483                         return 0;
1484                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1485                                                 entry->fn.parent_ip) {
1486                         ret = trace_seq_printf(s, " <-");
1487                         if (!ret)
1488                                 return 0;
1489                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1490                                                sym_flags);
1491                         if (!ret)
1492                                 return 0;
1493                 }
1494                 ret = trace_seq_printf(s, "\n");
1495                 if (!ret)
1496                         return 0;
1497                 break;
1498         case TRACE_CTX:
1499         case TRACE_WAKE:
1500                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1501                         state_to_char[entry->ctx.prev_state] : 'X';
1502                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1503                         state_to_char[entry->ctx.next_state] : 'X';
1504                 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1505                                        entry->ctx.prev_pid,
1506                                        entry->ctx.prev_prio,
1507                                        S,
1508                                        entry->type == TRACE_CTX ? "==>" : "  +",
1509                                        entry->ctx.next_pid,
1510                                        entry->ctx.next_prio,
1511                                        T);
1512                 if (!ret)
1513                         return 0;
1514                 break;
1515         case TRACE_SPECIAL:
1516                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1517                                  entry->special.arg1,
1518                                  entry->special.arg2,
1519                                  entry->special.arg3);
1520                 if (!ret)
1521                         return 0;
1522                 break;
1523         case TRACE_STACK:
1524                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1525                         if (i) {
1526                                 ret = trace_seq_puts(s, " <= ");
1527                                 if (!ret)
1528                                         return 0;
1529                         }
1530                         ret = seq_print_ip_sym(s, entry->stack.caller[i],
1531                                                sym_flags);
1532                         if (!ret)
1533                                 return 0;
1534                 }
1535                 ret = trace_seq_puts(s, "\n");
1536                 if (!ret)
1537                         return 0;
1538                 break;
1539         }
1540         return 1;
1541 }
1542
1543 static int print_raw_fmt(struct trace_iterator *iter)
1544 {
1545         struct trace_seq *s = &iter->seq;
1546         struct trace_entry *entry;
1547         int ret;
1548         int S, T;
1549
1550         entry = iter->ent;
1551
1552         ret = trace_seq_printf(s, "%d %d %llu ",
1553                 entry->pid, iter->cpu, entry->t);
1554         if (!ret)
1555                 return 0;
1556
1557         switch (entry->type) {
1558         case TRACE_FN:
1559                 ret = trace_seq_printf(s, "%x %x\n",
1560                                         entry->fn.ip, entry->fn.parent_ip);
1561                 if (!ret)
1562                         return 0;
1563                 break;
1564         case TRACE_CTX:
1565         case TRACE_WAKE:
1566                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1567                         state_to_char[entry->ctx.prev_state] : 'X';
1568                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1569                         state_to_char[entry->ctx.next_state] : 'X';
1570                 if (entry->type == TRACE_WAKE)
1571                         S = '+';
1572                 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1573                                        entry->ctx.prev_pid,
1574                                        entry->ctx.prev_prio,
1575                                        S,
1576                                        entry->ctx.next_pid,
1577                                        entry->ctx.next_prio,
1578                                        T);
1579                 if (!ret)
1580                         return 0;
1581                 break;
1582         case TRACE_SPECIAL:
1583         case TRACE_STACK:
1584                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1585                                  entry->special.arg1,
1586                                  entry->special.arg2,
1587                                  entry->special.arg3);
1588                 if (!ret)
1589                         return 0;
1590                 break;
1591         }
1592         return 1;
1593 }
1594
1595 #define SEQ_PUT_FIELD_RET(s, x)                         \
1596 do {                                                    \
1597         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1598                 return 0;                               \
1599 } while (0)
1600
1601 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1602 do {                                                    \
1603         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1604                 return 0;                               \
1605 } while (0)
1606
1607 static int print_hex_fmt(struct trace_iterator *iter)
1608 {
1609         struct trace_seq *s = &iter->seq;
1610         unsigned char newline = '\n';
1611         struct trace_entry *entry;
1612         int S, T;
1613
1614         entry = iter->ent;
1615
1616         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1617         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1618         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1619
1620         switch (entry->type) {
1621         case TRACE_FN:
1622                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1623                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1624                 break;
1625         case TRACE_CTX:
1626         case TRACE_WAKE:
1627                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1628                         state_to_char[entry->ctx.prev_state] : 'X';
1629                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1630                         state_to_char[entry->ctx.next_state] : 'X';
1631                 if (entry->type == TRACE_WAKE)
1632                         S = '+';
1633                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1634                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1635                 SEQ_PUT_HEX_FIELD_RET(s, S);
1636                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1637                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1638                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1639                 SEQ_PUT_HEX_FIELD_RET(s, T);
1640                 break;
1641         case TRACE_SPECIAL:
1642         case TRACE_STACK:
1643                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1644                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1645                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1646                 break;
1647         }
1648         SEQ_PUT_FIELD_RET(s, newline);
1649
1650         return 1;
1651 }
1652
1653 static int print_bin_fmt(struct trace_iterator *iter)
1654 {
1655         struct trace_seq *s = &iter->seq;
1656         struct trace_entry *entry;
1657
1658         entry = iter->ent;
1659
1660         SEQ_PUT_FIELD_RET(s, entry->pid);
1661         SEQ_PUT_FIELD_RET(s, entry->cpu);
1662         SEQ_PUT_FIELD_RET(s, entry->t);
1663
1664         switch (entry->type) {
1665         case TRACE_FN:
1666                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1667                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1668                 break;
1669         case TRACE_CTX:
1670                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1671                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1672                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1673                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1674                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1675                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1676                 break;
1677         case TRACE_SPECIAL:
1678         case TRACE_STACK:
1679                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1680                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1681                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1682                 break;
1683         }
1684         return 1;
1685 }
1686
1687 static int trace_empty(struct trace_iterator *iter)
1688 {
1689         struct trace_array_cpu *data;
1690         int cpu;
1691
1692         for_each_tracing_cpu(cpu) {
1693                 data = iter->tr->data[cpu];
1694
1695                 if (head_page(data) && data->trace_idx &&
1696                     (data->trace_tail != data->trace_head ||
1697                      data->trace_tail_idx != data->trace_head_idx))
1698                         return 0;
1699         }
1700         return 1;
1701 }
1702
1703 static int print_trace_line(struct trace_iterator *iter)
1704 {
1705         if (iter->trace && iter->trace->print_line)
1706                 return iter->trace->print_line(iter);
1707
1708         if (trace_flags & TRACE_ITER_BIN)
1709                 return print_bin_fmt(iter);
1710
1711         if (trace_flags & TRACE_ITER_HEX)
1712                 return print_hex_fmt(iter);
1713
1714         if (trace_flags & TRACE_ITER_RAW)
1715                 return print_raw_fmt(iter);
1716
1717         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1718                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1719
1720         return print_trace_fmt(iter);
1721 }
1722
1723 static int s_show(struct seq_file *m, void *v)
1724 {
1725         struct trace_iterator *iter = v;
1726
1727         if (iter->ent == NULL) {
1728                 if (iter->tr) {
1729                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1730                         seq_puts(m, "#\n");
1731                 }
1732                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1733                         /* print nothing if the buffers are empty */
1734                         if (trace_empty(iter))
1735                                 return 0;
1736                         print_trace_header(m, iter);
1737                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1738                                 print_lat_help_header(m);
1739                 } else {
1740                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1741                                 print_func_help_header(m);
1742                 }
1743         } else {
1744                 print_trace_line(iter);
1745                 trace_print_seq(m, &iter->seq);
1746         }
1747
1748         return 0;
1749 }
1750
1751 static struct seq_operations tracer_seq_ops = {
1752         .start          = s_start,
1753         .next           = s_next,
1754         .stop           = s_stop,
1755         .show           = s_show,
1756 };
1757
1758 static struct trace_iterator *
1759 __tracing_open(struct inode *inode, struct file *file, int *ret)
1760 {
1761         struct trace_iterator *iter;
1762
1763         if (tracing_disabled) {
1764                 *ret = -ENODEV;
1765                 return NULL;
1766         }
1767
1768         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1769         if (!iter) {
1770                 *ret = -ENOMEM;
1771                 goto out;
1772         }
1773
1774         mutex_lock(&trace_types_lock);
1775         if (current_trace && current_trace->print_max)
1776                 iter->tr = &max_tr;
1777         else
1778                 iter->tr = inode->i_private;
1779         iter->trace = current_trace;
1780         iter->pos = -1;
1781
1782         /* TODO stop tracer */
1783         *ret = seq_open(file, &tracer_seq_ops);
1784         if (!*ret) {
1785                 struct seq_file *m = file->private_data;
1786                 m->private = iter;
1787
1788                 /* stop the trace while dumping */
1789                 if (iter->tr->ctrl)
1790                         tracer_enabled = 0;
1791
1792                 if (iter->trace && iter->trace->open)
1793                         iter->trace->open(iter);
1794         } else {
1795                 kfree(iter);
1796                 iter = NULL;
1797         }
1798         mutex_unlock(&trace_types_lock);
1799
1800  out:
1801         return iter;
1802 }
1803
1804 int tracing_open_generic(struct inode *inode, struct file *filp)
1805 {
1806         if (tracing_disabled)
1807                 return -ENODEV;
1808
1809         filp->private_data = inode->i_private;
1810         return 0;
1811 }
1812
1813 int tracing_release(struct inode *inode, struct file *file)
1814 {
1815         struct seq_file *m = (struct seq_file *)file->private_data;
1816         struct trace_iterator *iter = m->private;
1817
1818         mutex_lock(&trace_types_lock);
1819         if (iter->trace && iter->trace->close)
1820                 iter->trace->close(iter);
1821
1822         /* reenable tracing if it was previously enabled */
1823         if (iter->tr->ctrl)
1824                 tracer_enabled = 1;
1825         mutex_unlock(&trace_types_lock);
1826
1827         seq_release(inode, file);
1828         kfree(iter);
1829         return 0;
1830 }
1831
1832 static int tracing_open(struct inode *inode, struct file *file)
1833 {
1834         int ret;
1835
1836         __tracing_open(inode, file, &ret);
1837
1838         return ret;
1839 }
1840
1841 static int tracing_lt_open(struct inode *inode, struct file *file)
1842 {
1843         struct trace_iterator *iter;
1844         int ret;
1845
1846         iter = __tracing_open(inode, file, &ret);
1847
1848         if (!ret)
1849                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1850
1851         return ret;
1852 }
1853
1854
1855 static void *
1856 t_next(struct seq_file *m, void *v, loff_t *pos)
1857 {
1858         struct tracer *t = m->private;
1859
1860         (*pos)++;
1861
1862         if (t)
1863                 t = t->next;
1864
1865         m->private = t;
1866
1867         return t;
1868 }
1869
1870 static void *t_start(struct seq_file *m, loff_t *pos)
1871 {
1872         struct tracer *t = m->private;
1873         loff_t l = 0;
1874
1875         mutex_lock(&trace_types_lock);
1876         for (; t && l < *pos; t = t_next(m, t, &l))
1877                 ;
1878
1879         return t;
1880 }
1881
1882 static void t_stop(struct seq_file *m, void *p)
1883 {
1884         mutex_unlock(&trace_types_lock);
1885 }
1886
1887 static int t_show(struct seq_file *m, void *v)
1888 {
1889         struct tracer *t = v;
1890
1891         if (!t)
1892                 return 0;
1893
1894         seq_printf(m, "%s", t->name);
1895         if (t->next)
1896                 seq_putc(m, ' ');
1897         else
1898                 seq_putc(m, '\n');
1899
1900         return 0;
1901 }
1902
1903 static struct seq_operations show_traces_seq_ops = {
1904         .start          = t_start,
1905         .next           = t_next,
1906         .stop           = t_stop,
1907         .show           = t_show,
1908 };
1909
1910 static int show_traces_open(struct inode *inode, struct file *file)
1911 {
1912         int ret;
1913
1914         if (tracing_disabled)
1915                 return -ENODEV;
1916
1917         ret = seq_open(file, &show_traces_seq_ops);
1918         if (!ret) {
1919                 struct seq_file *m = file->private_data;
1920                 m->private = trace_types;
1921         }
1922
1923         return ret;
1924 }
1925
1926 static struct file_operations tracing_fops = {
1927         .open           = tracing_open,
1928         .read           = seq_read,
1929         .llseek         = seq_lseek,
1930         .release        = tracing_release,
1931 };
1932
1933 static struct file_operations tracing_lt_fops = {
1934         .open           = tracing_lt_open,
1935         .read           = seq_read,
1936         .llseek         = seq_lseek,
1937         .release        = tracing_release,
1938 };
1939
1940 static struct file_operations show_traces_fops = {
1941         .open           = show_traces_open,
1942         .read           = seq_read,
1943         .release        = seq_release,
1944 };
1945
1946 /*
1947  * Only trace on a CPU if the bitmask is set:
1948  */
1949 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1950
1951 /*
1952  * When tracing/tracing_cpu_mask is modified then this holds
1953  * the new bitmask we are about to install:
1954  */
1955 static cpumask_t tracing_cpumask_new;
1956
1957 /*
1958  * The tracer itself will not take this lock, but still we want
1959  * to provide a consistent cpumask to user-space:
1960  */
1961 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1962
1963 /*
1964  * Temporary storage for the character representation of the
1965  * CPU bitmask (and one more byte for the newline):
1966  */
1967 static char mask_str[NR_CPUS + 1];
1968
1969 static ssize_t
1970 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1971                      size_t count, loff_t *ppos)
1972 {
1973         int len;
1974
1975         mutex_lock(&tracing_cpumask_update_lock);
1976
1977         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1978         if (count - len < 2) {
1979                 count = -EINVAL;
1980                 goto out_err;
1981         }
1982         len += sprintf(mask_str + len, "\n");
1983         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1984
1985 out_err:
1986         mutex_unlock(&tracing_cpumask_update_lock);
1987
1988         return count;
1989 }
1990
1991 static ssize_t
1992 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1993                       size_t count, loff_t *ppos)
1994 {
1995         int err, cpu;
1996
1997         mutex_lock(&tracing_cpumask_update_lock);
1998         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
1999         if (err)
2000                 goto err_unlock;
2001
2002         raw_local_irq_disable();
2003         __raw_spin_lock(&ftrace_max_lock);
2004         for_each_tracing_cpu(cpu) {
2005                 /*
2006                  * Increase/decrease the disabled counter if we are
2007                  * about to flip a bit in the cpumask:
2008                  */
2009                 if (cpu_isset(cpu, tracing_cpumask) &&
2010                                 !cpu_isset(cpu, tracing_cpumask_new)) {
2011                         atomic_inc(&global_trace.data[cpu]->disabled);
2012                 }
2013                 if (!cpu_isset(cpu, tracing_cpumask) &&
2014                                 cpu_isset(cpu, tracing_cpumask_new)) {
2015                         atomic_dec(&global_trace.data[cpu]->disabled);
2016                 }
2017         }
2018         __raw_spin_unlock(&ftrace_max_lock);
2019         raw_local_irq_enable();
2020
2021         tracing_cpumask = tracing_cpumask_new;
2022
2023         mutex_unlock(&tracing_cpumask_update_lock);
2024
2025         return count;
2026
2027 err_unlock:
2028         mutex_unlock(&tracing_cpumask_update_lock);
2029
2030         return err;
2031 }
2032
2033 static struct file_operations tracing_cpumask_fops = {
2034         .open           = tracing_open_generic,
2035         .read           = tracing_cpumask_read,
2036         .write          = tracing_cpumask_write,
2037 };
2038
2039 static ssize_t
2040 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2041                        size_t cnt, loff_t *ppos)
2042 {
2043         char *buf;
2044         int r = 0;
2045         int len = 0;
2046         int i;
2047
2048         /* calulate max size */
2049         for (i = 0; trace_options[i]; i++) {
2050                 len += strlen(trace_options[i]);
2051                 len += 3; /* "no" and space */
2052         }
2053
2054         /* +2 for \n and \0 */
2055         buf = kmalloc(len + 2, GFP_KERNEL);
2056         if (!buf)
2057                 return -ENOMEM;
2058
2059         for (i = 0; trace_options[i]; i++) {
2060                 if (trace_flags & (1 << i))
2061                         r += sprintf(buf + r, "%s ", trace_options[i]);
2062                 else
2063                         r += sprintf(buf + r, "no%s ", trace_options[i]);
2064         }
2065
2066         r += sprintf(buf + r, "\n");
2067         WARN_ON(r >= len + 2);
2068
2069         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2070
2071         kfree(buf);
2072
2073         return r;
2074 }
2075
2076 static ssize_t
2077 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2078                         size_t cnt, loff_t *ppos)
2079 {
2080         char buf[64];
2081         char *cmp = buf;
2082         int neg = 0;
2083         int i;
2084
2085         if (cnt >= sizeof(buf))
2086                 return -EINVAL;
2087
2088         if (copy_from_user(&buf, ubuf, cnt))
2089                 return -EFAULT;
2090
2091         buf[cnt] = 0;
2092
2093         if (strncmp(buf, "no", 2) == 0) {
2094                 neg = 1;
2095                 cmp += 2;
2096         }
2097
2098         for (i = 0; trace_options[i]; i++) {
2099                 int len = strlen(trace_options[i]);
2100
2101                 if (strncmp(cmp, trace_options[i], len) == 0) {
2102                         if (neg)
2103                                 trace_flags &= ~(1 << i);
2104                         else
2105                                 trace_flags |= (1 << i);
2106                         break;
2107                 }
2108         }
2109         /*
2110          * If no option could be set, return an error:
2111          */
2112         if (!trace_options[i])
2113                 return -EINVAL;
2114
2115         filp->f_pos += cnt;
2116
2117         return cnt;
2118 }
2119
2120 static struct file_operations tracing_iter_fops = {
2121         .open           = tracing_open_generic,
2122         .read           = tracing_iter_ctrl_read,
2123         .write          = tracing_iter_ctrl_write,
2124 };
2125
2126 static const char readme_msg[] =
2127         "tracing mini-HOWTO:\n\n"
2128         "# mkdir /debug\n"
2129         "# mount -t debugfs nodev /debug\n\n"
2130         "# cat /debug/tracing/available_tracers\n"
2131         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2132         "# cat /debug/tracing/current_tracer\n"
2133         "none\n"
2134         "# echo sched_switch > /debug/tracing/current_tracer\n"
2135         "# cat /debug/tracing/current_tracer\n"
2136         "sched_switch\n"
2137         "# cat /debug/tracing/iter_ctrl\n"
2138         "noprint-parent nosym-offset nosym-addr noverbose\n"
2139         "# echo print-parent > /debug/tracing/iter_ctrl\n"
2140         "# echo 1 > /debug/tracing/tracing_enabled\n"
2141         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2142         "echo 0 > /debug/tracing/tracing_enabled\n"
2143 ;
2144
2145 static ssize_t
2146 tracing_readme_read(struct file *filp, char __user *ubuf,
2147                        size_t cnt, loff_t *ppos)
2148 {
2149         return simple_read_from_buffer(ubuf, cnt, ppos,
2150                                         readme_msg, strlen(readme_msg));
2151 }
2152
2153 static struct file_operations tracing_readme_fops = {
2154         .open           = tracing_open_generic,
2155         .read           = tracing_readme_read,
2156 };
2157
2158 static ssize_t
2159 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2160                   size_t cnt, loff_t *ppos)
2161 {
2162         struct trace_array *tr = filp->private_data;
2163         char buf[64];
2164         int r;
2165
2166         r = sprintf(buf, "%ld\n", tr->ctrl);
2167         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2168 }
2169
2170 static ssize_t
2171 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2172                    size_t cnt, loff_t *ppos)
2173 {
2174         struct trace_array *tr = filp->private_data;
2175         char buf[64];
2176         long val;
2177         int ret;
2178
2179         if (cnt >= sizeof(buf))
2180                 return -EINVAL;
2181
2182         if (copy_from_user(&buf, ubuf, cnt))
2183                 return -EFAULT;
2184
2185         buf[cnt] = 0;
2186
2187         ret = strict_strtoul(buf, 10, &val);
2188         if (ret < 0)
2189                 return ret;
2190
2191         val = !!val;
2192
2193         mutex_lock(&trace_types_lock);
2194         if (tr->ctrl ^ val) {
2195                 if (val)
2196                         tracer_enabled = 1;
2197                 else
2198                         tracer_enabled = 0;
2199
2200                 tr->ctrl = val;
2201
2202                 if (current_trace && current_trace->ctrl_update)
2203                         current_trace->ctrl_update(tr);
2204         }
2205         mutex_unlock(&trace_types_lock);
2206
2207         filp->f_pos += cnt;
2208
2209         return cnt;
2210 }
2211
2212 static ssize_t
2213 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2214                        size_t cnt, loff_t *ppos)
2215 {
2216         char buf[max_tracer_type_len+2];
2217         int r;
2218
2219         mutex_lock(&trace_types_lock);
2220         if (current_trace)
2221                 r = sprintf(buf, "%s\n", current_trace->name);
2222         else
2223                 r = sprintf(buf, "\n");
2224         mutex_unlock(&trace_types_lock);
2225
2226         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2227 }
2228
2229 static ssize_t
2230 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2231                         size_t cnt, loff_t *ppos)
2232 {
2233         struct trace_array *tr = &global_trace;
2234         struct tracer *t;
2235         char buf[max_tracer_type_len+1];
2236         int i;
2237
2238         if (cnt > max_tracer_type_len)
2239                 cnt = max_tracer_type_len;
2240
2241         if (copy_from_user(&buf, ubuf, cnt))
2242                 return -EFAULT;
2243
2244         buf[cnt] = 0;
2245
2246         /* strip ending whitespace. */
2247         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2248                 buf[i] = 0;
2249
2250         mutex_lock(&trace_types_lock);
2251         for (t = trace_types; t; t = t->next) {
2252                 if (strcmp(t->name, buf) == 0)
2253                         break;
2254         }
2255         if (!t || t == current_trace)
2256                 goto out;
2257
2258         if (current_trace && current_trace->reset)
2259                 current_trace->reset(tr);
2260
2261         current_trace = t;
2262         if (t->init)
2263                 t->init(tr);
2264
2265  out:
2266         mutex_unlock(&trace_types_lock);
2267
2268         filp->f_pos += cnt;
2269
2270         return cnt;
2271 }
2272
2273 static ssize_t
2274 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2275                      size_t cnt, loff_t *ppos)
2276 {
2277         unsigned long *ptr = filp->private_data;
2278         char buf[64];
2279         int r;
2280
2281         r = snprintf(buf, sizeof(buf), "%ld\n",
2282                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2283         if (r > sizeof(buf))
2284                 r = sizeof(buf);
2285         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2286 }
2287
2288 static ssize_t
2289 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2290                       size_t cnt, loff_t *ppos)
2291 {
2292         long *ptr = filp->private_data;
2293         char buf[64];
2294         long val;
2295         int ret;
2296
2297         if (cnt >= sizeof(buf))
2298                 return -EINVAL;
2299
2300         if (copy_from_user(&buf, ubuf, cnt))
2301                 return -EFAULT;
2302
2303         buf[cnt] = 0;
2304
2305         ret = strict_strtoul(buf, 10, &val);
2306         if (ret < 0)
2307                 return ret;
2308
2309         *ptr = val * 1000;
2310
2311         return cnt;
2312 }
2313
2314 static atomic_t tracing_reader;
2315
2316 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2317 {
2318         struct trace_iterator *iter;
2319
2320         if (tracing_disabled)
2321                 return -ENODEV;
2322
2323         /* We only allow for reader of the pipe */
2324         if (atomic_inc_return(&tracing_reader) != 1) {
2325                 atomic_dec(&tracing_reader);
2326                 return -EBUSY;
2327         }
2328
2329         /* create a buffer to store the information to pass to userspace */
2330         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2331         if (!iter)
2332                 return -ENOMEM;
2333
2334         mutex_lock(&trace_types_lock);
2335         iter->tr = &global_trace;
2336         iter->trace = current_trace;
2337         filp->private_data = iter;
2338
2339         if (iter->trace->pipe_open)
2340                 iter->trace->pipe_open(iter);
2341         mutex_unlock(&trace_types_lock);
2342
2343         return 0;
2344 }
2345
2346 static int tracing_release_pipe(struct inode *inode, struct file *file)
2347 {
2348         struct trace_iterator *iter = file->private_data;
2349
2350         kfree(iter);
2351         atomic_dec(&tracing_reader);
2352
2353         return 0;
2354 }
2355
2356 static unsigned int
2357 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2358 {
2359         struct trace_iterator *iter = filp->private_data;
2360
2361         if (trace_flags & TRACE_ITER_BLOCK) {
2362                 /*
2363                  * Always select as readable when in blocking mode
2364                  */
2365                 return POLLIN | POLLRDNORM;
2366         } else {
2367                 if (!trace_empty(iter))
2368                         return POLLIN | POLLRDNORM;
2369                 poll_wait(filp, &trace_wait, poll_table);
2370                 if (!trace_empty(iter))
2371                         return POLLIN | POLLRDNORM;
2372
2373                 return 0;
2374         }
2375 }
2376
2377 /*
2378  * Consumer reader.
2379  */
2380 static ssize_t
2381 tracing_read_pipe(struct file *filp, char __user *ubuf,
2382                   size_t cnt, loff_t *ppos)
2383 {
2384         struct trace_iterator *iter = filp->private_data;
2385         struct trace_array_cpu *data;
2386         static cpumask_t mask;
2387         unsigned long flags;
2388 #ifdef CONFIG_FTRACE
2389         int ftrace_save;
2390 #endif
2391         int cpu;
2392         ssize_t sret;
2393
2394         /* return any leftover data */
2395         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2396         if (sret != -EBUSY)
2397                 return sret;
2398         sret = 0;
2399
2400         trace_seq_reset(&iter->seq);
2401
2402         mutex_lock(&trace_types_lock);
2403         if (iter->trace->read) {
2404                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2405                 if (sret)
2406                         goto out;
2407         }
2408
2409         while (trace_empty(iter)) {
2410
2411                 if ((filp->f_flags & O_NONBLOCK)) {
2412                         sret = -EAGAIN;
2413                         goto out;
2414                 }
2415
2416                 /*
2417                  * This is a make-shift waitqueue. The reason we don't use
2418                  * an actual wait queue is because:
2419                  *  1) we only ever have one waiter
2420                  *  2) the tracing, traces all functions, we don't want
2421                  *     the overhead of calling wake_up and friends
2422                  *     (and tracing them too)
2423                  *     Anyway, this is really very primitive wakeup.
2424                  */
2425                 set_current_state(TASK_INTERRUPTIBLE);
2426                 iter->tr->waiter = current;
2427
2428                 mutex_unlock(&trace_types_lock);
2429
2430                 /* sleep for 100 msecs, and try again. */
2431                 schedule_timeout(HZ/10);
2432
2433                 mutex_lock(&trace_types_lock);
2434
2435                 iter->tr->waiter = NULL;
2436
2437                 if (signal_pending(current)) {
2438                         sret = -EINTR;
2439                         goto out;
2440                 }
2441
2442                 if (iter->trace != current_trace)
2443                         goto out;
2444
2445                 /*
2446                  * We block until we read something and tracing is disabled.
2447                  * We still block if tracing is disabled, but we have never
2448                  * read anything. This allows a user to cat this file, and
2449                  * then enable tracing. But after we have read something,
2450                  * we give an EOF when tracing is again disabled.
2451                  *
2452                  * iter->pos will be 0 if we haven't read anything.
2453                  */
2454                 if (!tracer_enabled && iter->pos)
2455                         break;
2456
2457                 continue;
2458         }
2459
2460         /* stop when tracing is finished */
2461         if (trace_empty(iter))
2462                 goto out;
2463
2464         if (cnt >= PAGE_SIZE)
2465                 cnt = PAGE_SIZE - 1;
2466
2467         /* reset all but tr, trace, and overruns */
2468         memset(&iter->seq, 0,
2469                sizeof(struct trace_iterator) -
2470                offsetof(struct trace_iterator, seq));
2471         iter->pos = -1;
2472
2473         /*
2474          * We need to stop all tracing on all CPUS to read the
2475          * the next buffer. This is a bit expensive, but is
2476          * not done often. We fill all what we can read,
2477          * and then release the locks again.
2478          */
2479
2480         cpus_clear(mask);
2481         local_irq_save(flags);
2482 #ifdef CONFIG_FTRACE
2483         ftrace_save = ftrace_enabled;
2484         ftrace_enabled = 0;
2485 #endif
2486         smp_wmb();
2487         for_each_tracing_cpu(cpu) {
2488                 data = iter->tr->data[cpu];
2489
2490                 if (!head_page(data) || !data->trace_idx)
2491                         continue;
2492
2493                 atomic_inc(&data->disabled);
2494                 cpu_set(cpu, mask);
2495         }
2496
2497         for_each_cpu_mask(cpu, mask) {
2498                 data = iter->tr->data[cpu];
2499                 __raw_spin_lock(&data->lock);
2500
2501                 if (data->overrun > iter->last_overrun[cpu])
2502                         iter->overrun[cpu] +=
2503                                 data->overrun - iter->last_overrun[cpu];
2504                 iter->last_overrun[cpu] = data->overrun;
2505         }
2506
2507         while (find_next_entry_inc(iter) != NULL) {
2508                 int ret;
2509                 int len = iter->seq.len;
2510
2511                 ret = print_trace_line(iter);
2512                 if (!ret) {
2513                         /* don't print partial lines */
2514                         iter->seq.len = len;
2515                         break;
2516                 }
2517
2518                 trace_consume(iter);
2519
2520                 if (iter->seq.len >= cnt)
2521                         break;
2522         }
2523
2524         for_each_cpu_mask(cpu, mask) {
2525                 data = iter->tr->data[cpu];
2526                 __raw_spin_unlock(&data->lock);
2527         }
2528
2529         for_each_cpu_mask(cpu, mask) {
2530                 data = iter->tr->data[cpu];
2531                 atomic_dec(&data->disabled);
2532         }
2533 #ifdef CONFIG_FTRACE
2534         ftrace_enabled = ftrace_save;
2535 #endif
2536         local_irq_restore(flags);
2537
2538         /* Now copy what we have to the user */
2539         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2540         if (iter->seq.readpos >= iter->seq.len)
2541                 trace_seq_reset(&iter->seq);
2542         if (sret == -EBUSY)
2543                 sret = 0;
2544
2545 out:
2546         mutex_unlock(&trace_types_lock);
2547
2548         return sret;
2549 }
2550
2551 static ssize_t
2552 tracing_entries_read(struct file *filp, char __user *ubuf,
2553                      size_t cnt, loff_t *ppos)
2554 {
2555         struct trace_array *tr = filp->private_data;
2556         char buf[64];
2557         int r;
2558
2559         r = sprintf(buf, "%lu\n", tr->entries);
2560         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2561 }
2562
2563 static ssize_t
2564 tracing_entries_write(struct file *filp, const char __user *ubuf,
2565                       size_t cnt, loff_t *ppos)
2566 {
2567         unsigned long val;
2568         char buf[64];
2569         int ret;
2570
2571         if (cnt >= sizeof(buf))
2572                 return -EINVAL;
2573
2574         if (copy_from_user(&buf, ubuf, cnt))
2575                 return -EFAULT;
2576
2577         buf[cnt] = 0;
2578
2579         ret = strict_strtoul(buf, 10, &val);
2580         if (ret < 0)
2581                 return ret;
2582
2583         /* must have at least 1 entry */
2584         if (!val)
2585                 return -EINVAL;
2586
2587         mutex_lock(&trace_types_lock);
2588
2589         if (current_trace != &no_tracer) {
2590                 cnt = -EBUSY;
2591                 pr_info("ftrace: set current_tracer to none"
2592                         " before modifying buffer size\n");
2593                 goto out;
2594         }
2595
2596         if (val > global_trace.entries) {
2597                 long pages_requested;
2598                 unsigned long freeable_pages;
2599
2600                 /* make sure we have enough memory before mapping */
2601                 pages_requested =
2602                         (val + (ENTRIES_PER_PAGE-1)) / ENTRIES_PER_PAGE;
2603
2604                 /* account for each buffer (and max_tr) */
2605                 pages_requested *= tracing_nr_buffers * 2;
2606
2607                 /* Check for overflow */
2608                 if (pages_requested < 0) {
2609                         cnt = -ENOMEM;
2610                         goto out;
2611                 }
2612
2613                 freeable_pages = determine_dirtyable_memory();
2614
2615                 /* we only allow to request 1/4 of useable memory */
2616                 if (pages_requested >
2617                     ((freeable_pages + tracing_pages_allocated) / 4)) {
2618                         cnt = -ENOMEM;
2619                         goto out;
2620                 }
2621
2622                 while (global_trace.entries < val) {
2623                         if (trace_alloc_page()) {
2624                                 cnt = -ENOMEM;
2625                                 goto out;
2626                         }
2627                         /* double check that we don't go over the known pages */
2628                         if (tracing_pages_allocated > pages_requested)
2629                                 break;
2630                 }
2631
2632         } else {
2633                 /* include the number of entries in val (inc of page entries) */
2634                 while (global_trace.entries > val + (ENTRIES_PER_PAGE - 1))
2635                         trace_free_page();
2636         }
2637
2638         filp->f_pos += cnt;
2639
2640  out:
2641         max_tr.entries = global_trace.entries;
2642         mutex_unlock(&trace_types_lock);
2643
2644         return cnt;
2645 }
2646
2647 static struct file_operations tracing_max_lat_fops = {
2648         .open           = tracing_open_generic,
2649         .read           = tracing_max_lat_read,
2650         .write          = tracing_max_lat_write,
2651 };
2652
2653 static struct file_operations tracing_ctrl_fops = {
2654         .open           = tracing_open_generic,
2655         .read           = tracing_ctrl_read,
2656         .write          = tracing_ctrl_write,
2657 };
2658
2659 static struct file_operations set_tracer_fops = {
2660         .open           = tracing_open_generic,
2661         .read           = tracing_set_trace_read,
2662         .write          = tracing_set_trace_write,
2663 };
2664
2665 static struct file_operations tracing_pipe_fops = {
2666         .open           = tracing_open_pipe,
2667         .poll           = tracing_poll_pipe,
2668         .read           = tracing_read_pipe,
2669         .release        = tracing_release_pipe,
2670 };
2671
2672 static struct file_operations tracing_entries_fops = {
2673         .open           = tracing_open_generic,
2674         .read           = tracing_entries_read,
2675         .write          = tracing_entries_write,
2676 };
2677
2678 #ifdef CONFIG_DYNAMIC_FTRACE
2679
2680 static ssize_t
2681 tracing_read_long(struct file *filp, char __user *ubuf,
2682                   size_t cnt, loff_t *ppos)
2683 {
2684         unsigned long *p = filp->private_data;
2685         char buf[64];
2686         int r;
2687
2688         r = sprintf(buf, "%ld\n", *p);
2689
2690         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2691 }
2692
2693 static struct file_operations tracing_read_long_fops = {
2694         .open           = tracing_open_generic,
2695         .read           = tracing_read_long,
2696 };
2697 #endif
2698
2699 static struct dentry *d_tracer;
2700
2701 struct dentry *tracing_init_dentry(void)
2702 {
2703         static int once;
2704
2705         if (d_tracer)
2706                 return d_tracer;
2707
2708         d_tracer = debugfs_create_dir("tracing", NULL);
2709
2710         if (!d_tracer && !once) {
2711                 once = 1;
2712                 pr_warning("Could not create debugfs directory 'tracing'\n");
2713                 return NULL;
2714         }
2715
2716         return d_tracer;
2717 }
2718
2719 #ifdef CONFIG_FTRACE_SELFTEST
2720 /* Let selftest have access to static functions in this file */
2721 #include "trace_selftest.c"
2722 #endif
2723
2724 static __init void tracer_init_debugfs(void)
2725 {
2726         struct dentry *d_tracer;
2727         struct dentry *entry;
2728
2729         d_tracer = tracing_init_dentry();
2730
2731         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2732                                     &global_trace, &tracing_ctrl_fops);
2733         if (!entry)
2734                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2735
2736         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2737                                     NULL, &tracing_iter_fops);
2738         if (!entry)
2739                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2740
2741         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2742                                     NULL, &tracing_cpumask_fops);
2743         if (!entry)
2744                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2745
2746         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2747                                     &global_trace, &tracing_lt_fops);
2748         if (!entry)
2749                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2750
2751         entry = debugfs_create_file("trace", 0444, d_tracer,
2752                                     &global_trace, &tracing_fops);
2753         if (!entry)
2754                 pr_warning("Could not create debugfs 'trace' entry\n");
2755
2756         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2757                                     &global_trace, &show_traces_fops);
2758         if (!entry)
2759                 pr_warning("Could not create debugfs 'trace' entry\n");
2760
2761         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2762                                     &global_trace, &set_tracer_fops);
2763         if (!entry)
2764                 pr_warning("Could not create debugfs 'trace' entry\n");
2765
2766         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2767                                     &tracing_max_latency,
2768                                     &tracing_max_lat_fops);
2769         if (!entry)
2770                 pr_warning("Could not create debugfs "
2771                            "'tracing_max_latency' entry\n");
2772
2773         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2774                                     &tracing_thresh, &tracing_max_lat_fops);
2775         if (!entry)
2776                 pr_warning("Could not create debugfs "
2777                            "'tracing_threash' entry\n");
2778         entry = debugfs_create_file("README", 0644, d_tracer,
2779                                     NULL, &tracing_readme_fops);
2780         if (!entry)
2781                 pr_warning("Could not create debugfs 'README' entry\n");
2782
2783         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2784                                     NULL, &tracing_pipe_fops);
2785         if (!entry)
2786                 pr_warning("Could not create debugfs "
2787                            "'tracing_threash' entry\n");
2788
2789         entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2790                                     &global_trace, &tracing_entries_fops);
2791         if (!entry)
2792                 pr_warning("Could not create debugfs "
2793                            "'tracing_threash' entry\n");
2794
2795 #ifdef CONFIG_DYNAMIC_FTRACE
2796         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2797                                     &ftrace_update_tot_cnt,
2798                                     &tracing_read_long_fops);
2799         if (!entry)
2800                 pr_warning("Could not create debugfs "
2801                            "'dyn_ftrace_total_info' entry\n");
2802 #endif
2803 }
2804
2805 static int trace_alloc_page(void)
2806 {
2807         struct trace_array_cpu *data;
2808         struct page *page, *tmp;
2809         LIST_HEAD(pages);
2810         void *array;
2811         unsigned pages_allocated = 0;
2812         int i;
2813
2814         /* first allocate a page for each CPU */
2815         for_each_tracing_cpu(i) {
2816                 array = (void *)__get_free_page(GFP_KERNEL);
2817                 if (array == NULL) {
2818                         printk(KERN_ERR "tracer: failed to allocate page"
2819                                "for trace buffer!\n");
2820                         goto free_pages;
2821                 }
2822
2823                 pages_allocated++;
2824                 page = virt_to_page(array);
2825                 list_add(&page->lru, &pages);
2826
2827 /* Only allocate if we are actually using the max trace */
2828 #ifdef CONFIG_TRACER_MAX_TRACE
2829                 array = (void *)__get_free_page(GFP_KERNEL);
2830                 if (array == NULL) {
2831                         printk(KERN_ERR "tracer: failed to allocate page"
2832                                "for trace buffer!\n");
2833                         goto free_pages;
2834                 }
2835                 pages_allocated++;
2836                 page = virt_to_page(array);
2837                 list_add(&page->lru, &pages);
2838 #endif
2839         }
2840
2841         /* Now that we successfully allocate a page per CPU, add them */
2842         for_each_tracing_cpu(i) {
2843                 data = global_trace.data[i];
2844                 page = list_entry(pages.next, struct page, lru);
2845                 list_del_init(&page->lru);
2846                 list_add_tail(&page->lru, &data->trace_pages);
2847                 ClearPageLRU(page);
2848
2849 #ifdef CONFIG_TRACER_MAX_TRACE
2850                 data = max_tr.data[i];
2851                 page = list_entry(pages.next, struct page, lru);
2852                 list_del_init(&page->lru);
2853                 list_add_tail(&page->lru, &data->trace_pages);
2854                 SetPageLRU(page);
2855 #endif
2856         }
2857         tracing_pages_allocated += pages_allocated;
2858         global_trace.entries += ENTRIES_PER_PAGE;
2859
2860         return 0;
2861
2862  free_pages:
2863         list_for_each_entry_safe(page, tmp, &pages, lru) {
2864                 list_del_init(&page->lru);
2865                 __free_page(page);
2866         }
2867         return -ENOMEM;
2868 }
2869
2870 static int trace_free_page(void)
2871 {
2872         struct trace_array_cpu *data;
2873         struct page *page;
2874         struct list_head *p;
2875         int i;
2876         int ret = 0;
2877
2878         /* free one page from each buffer */
2879         for_each_tracing_cpu(i) {
2880                 data = global_trace.data[i];
2881                 p = data->trace_pages.next;
2882                 if (p == &data->trace_pages) {
2883                         /* should never happen */
2884                         WARN_ON(1);
2885                         tracing_disabled = 1;
2886                         ret = -1;
2887                         break;
2888                 }
2889                 page = list_entry(p, struct page, lru);
2890                 ClearPageLRU(page);
2891                 list_del(&page->lru);
2892                 tracing_pages_allocated--;
2893                 tracing_pages_allocated--;
2894                 __free_page(page);
2895
2896                 tracing_reset(data);
2897
2898 #ifdef CONFIG_TRACER_MAX_TRACE
2899                 data = max_tr.data[i];
2900                 p = data->trace_pages.next;
2901                 if (p == &data->trace_pages) {
2902                         /* should never happen */
2903                         WARN_ON(1);
2904                         tracing_disabled = 1;
2905                         ret = -1;
2906                         break;
2907                 }
2908                 page = list_entry(p, struct page, lru);
2909                 ClearPageLRU(page);
2910                 list_del(&page->lru);
2911                 __free_page(page);
2912
2913                 tracing_reset(data);
2914 #endif
2915         }
2916         global_trace.entries -= ENTRIES_PER_PAGE;
2917
2918         return ret;
2919 }
2920
2921 __init static int tracer_alloc_buffers(void)
2922 {
2923         struct trace_array_cpu *data;
2924         void *array;
2925         struct page *page;
2926         int pages = 0;
2927         int ret = -ENOMEM;
2928         int i;
2929
2930         global_trace.ctrl = tracer_enabled;
2931
2932         /* TODO: make the number of buffers hot pluggable with CPUS */
2933         tracing_nr_buffers = num_possible_cpus();
2934         tracing_buffer_mask = cpu_possible_map;
2935
2936         /* Allocate the first page for all buffers */
2937         for_each_tracing_cpu(i) {
2938                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2939                 max_tr.data[i] = &per_cpu(max_data, i);
2940
2941                 array = (void *)__get_free_page(GFP_KERNEL);
2942                 if (array == NULL) {
2943                         printk(KERN_ERR "tracer: failed to allocate page"
2944                                "for trace buffer!\n");
2945                         goto free_buffers;
2946                 }
2947
2948                 /* set the array to the list */
2949                 INIT_LIST_HEAD(&data->trace_pages);
2950                 page = virt_to_page(array);
2951                 list_add(&page->lru, &data->trace_pages);
2952                 /* use the LRU flag to differentiate the two buffers */
2953                 ClearPageLRU(page);
2954
2955                 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2956                 max_tr.data[i]->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2957
2958 /* Only allocate if we are actually using the max trace */
2959 #ifdef CONFIG_TRACER_MAX_TRACE
2960                 array = (void *)__get_free_page(GFP_KERNEL);
2961                 if (array == NULL) {
2962                         printk(KERN_ERR "tracer: failed to allocate page"
2963                                "for trace buffer!\n");
2964                         goto free_buffers;
2965                 }
2966
2967                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2968                 page = virt_to_page(array);
2969                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2970                 SetPageLRU(page);
2971 #endif
2972         }
2973
2974         /*
2975          * Since we allocate by orders of pages, we may be able to
2976          * round up a bit.
2977          */
2978         global_trace.entries = ENTRIES_PER_PAGE;
2979         pages++;
2980
2981         while (global_trace.entries < trace_nr_entries) {
2982                 if (trace_alloc_page())
2983                         break;
2984                 pages++;
2985         }
2986         max_tr.entries = global_trace.entries;
2987
2988         pr_info("tracer: %d pages allocated for %ld",
2989                 pages, trace_nr_entries);
2990         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2991         pr_info("   actual entries %ld\n", global_trace.entries);
2992
2993         tracer_init_debugfs();
2994
2995         trace_init_cmdlines();
2996
2997         register_tracer(&no_tracer);
2998         current_trace = &no_tracer;
2999
3000         /* All seems OK, enable tracing */
3001         tracing_disabled = 0;
3002
3003         return 0;
3004
3005  free_buffers:
3006         for (i-- ; i >= 0; i--) {
3007                 struct page *page, *tmp;
3008                 struct trace_array_cpu *data = global_trace.data[i];
3009
3010                 if (data) {
3011                         list_for_each_entry_safe(page, tmp,
3012                                                  &data->trace_pages, lru) {
3013                                 list_del_init(&page->lru);
3014                                 __free_page(page);
3015                         }
3016                 }
3017
3018 #ifdef CONFIG_TRACER_MAX_TRACE
3019                 data = max_tr.data[i];
3020                 if (data) {
3021                         list_for_each_entry_safe(page, tmp,
3022                                                  &data->trace_pages, lru) {
3023                                 list_del_init(&page->lru);
3024                                 __free_page(page);
3025                         }
3026                 }
3027 #endif
3028         }
3029         return ret;
3030 }
3031 fs_initcall(tracer_alloc_buffers);