Merge branch 'linus' into tracing/mmiotrace-mergefixups
[linux-2.6] / kernel / trace / trace.c
1 /*
2  * ring buffer based function tracer
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally taken from the RT patch by:
8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code from the latency_tracer, that is:
11  *  Copyright (C) 2004-2006 Ingo Molnar
12  *  Copyright (C) 2004 William Lee Irwin III
13  */
14 #include <linux/utsrelease.h>
15 #include <linux/kallsyms.h>
16 #include <linux/seq_file.h>
17 #include <linux/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 #ifdef CONFIG_MMIOTRACE
835 void __trace_mmiotrace_rw(struct trace_array *tr, struct trace_array_cpu *data,
836                                                 struct mmiotrace_rw *rw)
837 {
838         struct trace_entry *entry;
839         unsigned long irq_flags;
840
841         raw_local_irq_save(irq_flags);
842         __raw_spin_lock(&data->lock);
843
844         entry                   = tracing_get_trace_entry(tr, data);
845         tracing_generic_entry_update(entry, 0);
846         entry->type             = TRACE_MMIO_RW;
847         entry->mmiorw           = *rw;
848
849         __raw_spin_unlock(&data->lock);
850         raw_local_irq_restore(irq_flags);
851
852         trace_wake_up();
853 }
854
855 void __trace_mmiotrace_map(struct trace_array *tr, struct trace_array_cpu *data,
856                                                 struct mmiotrace_map *map)
857 {
858         struct trace_entry *entry;
859         unsigned long irq_flags;
860
861         raw_local_irq_save(irq_flags);
862         __raw_spin_lock(&data->lock);
863
864         entry                   = tracing_get_trace_entry(tr, data);
865         tracing_generic_entry_update(entry, 0);
866         entry->type             = TRACE_MMIO_MAP;
867         entry->mmiomap          = *map;
868
869         __raw_spin_unlock(&data->lock);
870         raw_local_irq_restore(irq_flags);
871
872         trace_wake_up();
873 }
874 #endif
875
876 void __trace_stack(struct trace_array *tr,
877                    struct trace_array_cpu *data,
878                    unsigned long flags,
879                    int skip)
880 {
881         struct trace_entry *entry;
882         struct stack_trace trace;
883
884         if (!(trace_flags & TRACE_ITER_STACKTRACE))
885                 return;
886
887         entry                   = tracing_get_trace_entry(tr, data);
888         tracing_generic_entry_update(entry, flags);
889         entry->type             = TRACE_STACK;
890
891         memset(&entry->stack, 0, sizeof(entry->stack));
892
893         trace.nr_entries        = 0;
894         trace.max_entries       = FTRACE_STACK_ENTRIES;
895         trace.skip              = skip;
896         trace.entries           = entry->stack.caller;
897
898         save_stack_trace(&trace);
899 }
900
901 void
902 __trace_special(void *__tr, void *__data,
903                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
904 {
905         struct trace_array_cpu *data = __data;
906         struct trace_array *tr = __tr;
907         struct trace_entry *entry;
908         unsigned long irq_flags;
909
910         raw_local_irq_save(irq_flags);
911         __raw_spin_lock(&data->lock);
912         entry                   = tracing_get_trace_entry(tr, data);
913         tracing_generic_entry_update(entry, 0);
914         entry->type             = TRACE_SPECIAL;
915         entry->special.arg1     = arg1;
916         entry->special.arg2     = arg2;
917         entry->special.arg3     = arg3;
918         __trace_stack(tr, data, irq_flags, 4);
919         __raw_spin_unlock(&data->lock);
920         raw_local_irq_restore(irq_flags);
921
922         trace_wake_up();
923 }
924
925 void
926 tracing_sched_switch_trace(struct trace_array *tr,
927                            struct trace_array_cpu *data,
928                            struct task_struct *prev,
929                            struct task_struct *next,
930                            unsigned long flags)
931 {
932         struct trace_entry *entry;
933         unsigned long irq_flags;
934
935         raw_local_irq_save(irq_flags);
936         __raw_spin_lock(&data->lock);
937         entry                   = tracing_get_trace_entry(tr, data);
938         tracing_generic_entry_update(entry, flags);
939         entry->type             = TRACE_CTX;
940         entry->ctx.prev_pid     = prev->pid;
941         entry->ctx.prev_prio    = prev->prio;
942         entry->ctx.prev_state   = prev->state;
943         entry->ctx.next_pid     = next->pid;
944         entry->ctx.next_prio    = next->prio;
945         entry->ctx.next_state   = next->state;
946         __trace_stack(tr, data, flags, 5);
947         __raw_spin_unlock(&data->lock);
948         raw_local_irq_restore(irq_flags);
949 }
950
951 void
952 tracing_sched_wakeup_trace(struct trace_array *tr,
953                            struct trace_array_cpu *data,
954                            struct task_struct *wakee,
955                            struct task_struct *curr,
956                            unsigned long flags)
957 {
958         struct trace_entry *entry;
959         unsigned long irq_flags;
960
961         raw_local_irq_save(irq_flags);
962         __raw_spin_lock(&data->lock);
963         entry                   = tracing_get_trace_entry(tr, data);
964         tracing_generic_entry_update(entry, flags);
965         entry->type             = TRACE_WAKE;
966         entry->ctx.prev_pid     = curr->pid;
967         entry->ctx.prev_prio    = curr->prio;
968         entry->ctx.prev_state   = curr->state;
969         entry->ctx.next_pid     = wakee->pid;
970         entry->ctx.next_prio    = wakee->prio;
971         entry->ctx.next_state   = wakee->state;
972         __trace_stack(tr, data, flags, 6);
973         __raw_spin_unlock(&data->lock);
974         raw_local_irq_restore(irq_flags);
975
976         trace_wake_up();
977 }
978
979 #ifdef CONFIG_FTRACE
980 static void
981 function_trace_call(unsigned long ip, unsigned long parent_ip)
982 {
983         struct trace_array *tr = &global_trace;
984         struct trace_array_cpu *data;
985         unsigned long flags;
986         long disabled;
987         int cpu;
988
989         if (unlikely(!tracer_enabled))
990                 return;
991
992         local_irq_save(flags);
993         cpu = raw_smp_processor_id();
994         data = tr->data[cpu];
995         disabled = atomic_inc_return(&data->disabled);
996
997         if (likely(disabled == 1))
998                 trace_function(tr, data, ip, parent_ip, flags);
999
1000         atomic_dec(&data->disabled);
1001         local_irq_restore(flags);
1002 }
1003
1004 static struct ftrace_ops trace_ops __read_mostly =
1005 {
1006         .func = function_trace_call,
1007 };
1008
1009 void tracing_start_function_trace(void)
1010 {
1011         register_ftrace_function(&trace_ops);
1012 }
1013
1014 void tracing_stop_function_trace(void)
1015 {
1016         unregister_ftrace_function(&trace_ops);
1017 }
1018 #endif
1019
1020 enum trace_file_type {
1021         TRACE_FILE_LAT_FMT      = 1,
1022 };
1023
1024 static struct trace_entry *
1025 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
1026                 struct trace_iterator *iter, int cpu)
1027 {
1028         struct page *page;
1029         struct trace_entry *array;
1030
1031         if (iter->next_idx[cpu] >= tr->entries ||
1032             iter->next_idx[cpu] >= data->trace_idx ||
1033             (data->trace_head == data->trace_tail &&
1034              data->trace_head_idx == data->trace_tail_idx))
1035                 return NULL;
1036
1037         if (!iter->next_page[cpu]) {
1038                 /* Initialize the iterator for this cpu trace buffer */
1039                 WARN_ON(!data->trace_tail);
1040                 page = virt_to_page(data->trace_tail);
1041                 iter->next_page[cpu] = &page->lru;
1042                 iter->next_page_idx[cpu] = data->trace_tail_idx;
1043         }
1044
1045         page = list_entry(iter->next_page[cpu], struct page, lru);
1046         BUG_ON(&data->trace_pages == &page->lru);
1047
1048         array = page_address(page);
1049
1050         WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
1051         return &array[iter->next_page_idx[cpu]];
1052 }
1053
1054 static struct trace_entry *
1055 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
1056 {
1057         struct trace_array *tr = iter->tr;
1058         struct trace_entry *ent, *next = NULL;
1059         int next_cpu = -1;
1060         int cpu;
1061
1062         for_each_tracing_cpu(cpu) {
1063                 if (!head_page(tr->data[cpu]))
1064                         continue;
1065                 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
1066                 /*
1067                  * Pick the entry with the smallest timestamp:
1068                  */
1069                 if (ent && (!next || ent->t < next->t)) {
1070                         next = ent;
1071                         next_cpu = cpu;
1072                 }
1073         }
1074
1075         if (ent_cpu)
1076                 *ent_cpu = next_cpu;
1077
1078         return next;
1079 }
1080
1081 static void trace_iterator_increment(struct trace_iterator *iter)
1082 {
1083         iter->idx++;
1084         iter->next_idx[iter->cpu]++;
1085         iter->next_page_idx[iter->cpu]++;
1086
1087         if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
1088                 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1089
1090                 iter->next_page_idx[iter->cpu] = 0;
1091                 iter->next_page[iter->cpu] =
1092                         trace_next_list(data, iter->next_page[iter->cpu]);
1093         }
1094 }
1095
1096 static void trace_consume(struct trace_iterator *iter)
1097 {
1098         struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1099
1100         data->trace_tail_idx++;
1101         if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
1102                 data->trace_tail = trace_next_page(data, data->trace_tail);
1103                 data->trace_tail_idx = 0;
1104         }
1105
1106         /* Check if we empty it, then reset the index */
1107         if (data->trace_head == data->trace_tail &&
1108             data->trace_head_idx == data->trace_tail_idx)
1109                 data->trace_idx = 0;
1110 }
1111
1112 static void *find_next_entry_inc(struct trace_iterator *iter)
1113 {
1114         struct trace_entry *next;
1115         int next_cpu = -1;
1116
1117         next = find_next_entry(iter, &next_cpu);
1118
1119         iter->prev_ent = iter->ent;
1120         iter->prev_cpu = iter->cpu;
1121
1122         iter->ent = next;
1123         iter->cpu = next_cpu;
1124
1125         if (next)
1126                 trace_iterator_increment(iter);
1127
1128         return next ? iter : NULL;
1129 }
1130
1131 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1132 {
1133         struct trace_iterator *iter = m->private;
1134         void *last_ent = iter->ent;
1135         int i = (int)*pos;
1136         void *ent;
1137
1138         (*pos)++;
1139
1140         /* can't go backwards */
1141         if (iter->idx > i)
1142                 return NULL;
1143
1144         if (iter->idx < 0)
1145                 ent = find_next_entry_inc(iter);
1146         else
1147                 ent = iter;
1148
1149         while (ent && iter->idx < i)
1150                 ent = find_next_entry_inc(iter);
1151
1152         iter->pos = *pos;
1153
1154         if (last_ent && !ent)
1155                 seq_puts(m, "\n\nvim:ft=help\n");
1156
1157         return ent;
1158 }
1159
1160 static void *s_start(struct seq_file *m, loff_t *pos)
1161 {
1162         struct trace_iterator *iter = m->private;
1163         void *p = NULL;
1164         loff_t l = 0;
1165         int i;
1166
1167         mutex_lock(&trace_types_lock);
1168
1169         if (!current_trace || current_trace != iter->trace) {
1170                 mutex_unlock(&trace_types_lock);
1171                 return NULL;
1172         }
1173
1174         atomic_inc(&trace_record_cmdline_disabled);
1175
1176         /* let the tracer grab locks here if needed */
1177         if (current_trace->start)
1178                 current_trace->start(iter);
1179
1180         if (*pos != iter->pos) {
1181                 iter->ent = NULL;
1182                 iter->cpu = 0;
1183                 iter->idx = -1;
1184                 iter->prev_ent = NULL;
1185                 iter->prev_cpu = -1;
1186
1187                 for_each_tracing_cpu(i) {
1188                         iter->next_idx[i] = 0;
1189                         iter->next_page[i] = NULL;
1190                 }
1191
1192                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1193                         ;
1194
1195         } else {
1196                 l = *pos - 1;
1197                 p = s_next(m, p, &l);
1198         }
1199
1200         return p;
1201 }
1202
1203 static void s_stop(struct seq_file *m, void *p)
1204 {
1205         struct trace_iterator *iter = m->private;
1206
1207         atomic_dec(&trace_record_cmdline_disabled);
1208
1209         /* let the tracer release locks here if needed */
1210         if (current_trace && current_trace == iter->trace && iter->trace->stop)
1211                 iter->trace->stop(iter);
1212
1213         mutex_unlock(&trace_types_lock);
1214 }
1215
1216 static int
1217 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1218 {
1219 #ifdef CONFIG_KALLSYMS
1220         char str[KSYM_SYMBOL_LEN];
1221
1222         kallsyms_lookup(address, NULL, NULL, NULL, str);
1223
1224         return trace_seq_printf(s, fmt, str);
1225 #endif
1226         return 1;
1227 }
1228
1229 static int
1230 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1231                      unsigned long address)
1232 {
1233 #ifdef CONFIG_KALLSYMS
1234         char str[KSYM_SYMBOL_LEN];
1235
1236         sprint_symbol(str, address);
1237         return trace_seq_printf(s, fmt, str);
1238 #endif
1239         return 1;
1240 }
1241
1242 #ifndef CONFIG_64BIT
1243 # define IP_FMT "%08lx"
1244 #else
1245 # define IP_FMT "%016lx"
1246 #endif
1247
1248 static int
1249 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1250 {
1251         int ret;
1252
1253         if (!ip)
1254                 return trace_seq_printf(s, "0");
1255
1256         if (sym_flags & TRACE_ITER_SYM_OFFSET)
1257                 ret = seq_print_sym_offset(s, "%s", ip);
1258         else
1259                 ret = seq_print_sym_short(s, "%s", ip);
1260
1261         if (!ret)
1262                 return 0;
1263
1264         if (sym_flags & TRACE_ITER_SYM_ADDR)
1265                 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1266         return ret;
1267 }
1268
1269 static void print_lat_help_header(struct seq_file *m)
1270 {
1271         seq_puts(m, "#                _------=> CPU#            \n");
1272         seq_puts(m, "#               / _-----=> irqs-off        \n");
1273         seq_puts(m, "#              | / _----=> need-resched    \n");
1274         seq_puts(m, "#              || / _---=> hardirq/softirq \n");
1275         seq_puts(m, "#              ||| / _--=> preempt-depth   \n");
1276         seq_puts(m, "#              |||| /                      \n");
1277         seq_puts(m, "#              |||||     delay             \n");
1278         seq_puts(m, "#  cmd     pid ||||| time  |   caller      \n");
1279         seq_puts(m, "#     \\   /    |||||   \\   |   /           \n");
1280 }
1281
1282 static void print_func_help_header(struct seq_file *m)
1283 {
1284         seq_puts(m, "#           TASK-PID   CPU#    TIMESTAMP  FUNCTION\n");
1285         seq_puts(m, "#              | |      |          |         |\n");
1286 }
1287
1288
1289 static void
1290 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1291 {
1292         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1293         struct trace_array *tr = iter->tr;
1294         struct trace_array_cpu *data = tr->data[tr->cpu];
1295         struct tracer *type = current_trace;
1296         unsigned long total   = 0;
1297         unsigned long entries = 0;
1298         int cpu;
1299         const char *name = "preemption";
1300
1301         if (type)
1302                 name = type->name;
1303
1304         for_each_tracing_cpu(cpu) {
1305                 if (head_page(tr->data[cpu])) {
1306                         total += tr->data[cpu]->trace_idx;
1307                         if (tr->data[cpu]->trace_idx > tr->entries)
1308                                 entries += tr->entries;
1309                         else
1310                                 entries += tr->data[cpu]->trace_idx;
1311                 }
1312         }
1313
1314         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1315                    name, UTS_RELEASE);
1316         seq_puts(m, "-----------------------------------"
1317                  "---------------------------------\n");
1318         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1319                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1320                    nsecs_to_usecs(data->saved_latency),
1321                    entries,
1322                    total,
1323                    tr->cpu,
1324 #if defined(CONFIG_PREEMPT_NONE)
1325                    "server",
1326 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1327                    "desktop",
1328 #elif defined(CONFIG_PREEMPT_DESKTOP)
1329                    "preempt",
1330 #else
1331                    "unknown",
1332 #endif
1333                    /* These are reserved for later use */
1334                    0, 0, 0, 0);
1335 #ifdef CONFIG_SMP
1336         seq_printf(m, " #P:%d)\n", num_online_cpus());
1337 #else
1338         seq_puts(m, ")\n");
1339 #endif
1340         seq_puts(m, "    -----------------\n");
1341         seq_printf(m, "    | task: %.16s-%d "
1342                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1343                    data->comm, data->pid, data->uid, data->nice,
1344                    data->policy, data->rt_priority);
1345         seq_puts(m, "    -----------------\n");
1346
1347         if (data->critical_start) {
1348                 seq_puts(m, " => started at: ");
1349                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1350                 trace_print_seq(m, &iter->seq);
1351                 seq_puts(m, "\n => ended at:   ");
1352                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1353                 trace_print_seq(m, &iter->seq);
1354                 seq_puts(m, "\n");
1355         }
1356
1357         seq_puts(m, "\n");
1358 }
1359
1360 static void
1361 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1362 {
1363         int hardirq, softirq;
1364         char *comm;
1365
1366         comm = trace_find_cmdline(entry->pid);
1367
1368         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1369         trace_seq_printf(s, "%d", cpu);
1370         trace_seq_printf(s, "%c%c",
1371                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1372                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1373
1374         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1375         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1376         if (hardirq && softirq) {
1377                 trace_seq_putc(s, 'H');
1378         } else {
1379                 if (hardirq) {
1380                         trace_seq_putc(s, 'h');
1381                 } else {
1382                         if (softirq)
1383                                 trace_seq_putc(s, 's');
1384                         else
1385                                 trace_seq_putc(s, '.');
1386                 }
1387         }
1388
1389         if (entry->preempt_count)
1390                 trace_seq_printf(s, "%x", entry->preempt_count);
1391         else
1392                 trace_seq_puts(s, ".");
1393 }
1394
1395 unsigned long preempt_mark_thresh = 100;
1396
1397 static void
1398 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1399                     unsigned long rel_usecs)
1400 {
1401         trace_seq_printf(s, " %4lldus", abs_usecs);
1402         if (rel_usecs > preempt_mark_thresh)
1403                 trace_seq_puts(s, "!: ");
1404         else if (rel_usecs > 1)
1405                 trace_seq_puts(s, "+: ");
1406         else
1407                 trace_seq_puts(s, " : ");
1408 }
1409
1410 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1411
1412 static int
1413 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1414 {
1415         struct trace_seq *s = &iter->seq;
1416         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1417         struct trace_entry *next_entry = find_next_entry(iter, NULL);
1418         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1419         struct trace_entry *entry = iter->ent;
1420         unsigned long abs_usecs;
1421         unsigned long rel_usecs;
1422         char *comm;
1423         int S, T;
1424         int i;
1425         unsigned state;
1426
1427         if (!next_entry)
1428                 next_entry = entry;
1429         rel_usecs = ns2usecs(next_entry->t - entry->t);
1430         abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1431
1432         if (verbose) {
1433                 comm = trace_find_cmdline(entry->pid);
1434                 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1435                                  " %ld.%03ldms (+%ld.%03ldms): ",
1436                                  comm,
1437                                  entry->pid, cpu, entry->flags,
1438                                  entry->preempt_count, trace_idx,
1439                                  ns2usecs(entry->t),
1440                                  abs_usecs/1000,
1441                                  abs_usecs % 1000, rel_usecs/1000,
1442                                  rel_usecs % 1000);
1443         } else {
1444                 lat_print_generic(s, entry, cpu);
1445                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1446         }
1447         switch (entry->type) {
1448         case TRACE_FN:
1449                 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1450                 trace_seq_puts(s, " (");
1451                 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1452                 trace_seq_puts(s, ")\n");
1453                 break;
1454         case TRACE_CTX:
1455         case TRACE_WAKE:
1456                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1457                         state_to_char[entry->ctx.next_state] : 'X';
1458
1459                 state = entry->ctx.prev_state ? __ffs(entry->ctx.prev_state) + 1 : 0;
1460                 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1461                 comm = trace_find_cmdline(entry->ctx.next_pid);
1462                 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1463                                  entry->ctx.prev_pid,
1464                                  entry->ctx.prev_prio,
1465                                  S, entry->type == TRACE_CTX ? "==>" : "  +",
1466                                  entry->ctx.next_pid,
1467                                  entry->ctx.next_prio,
1468                                  T, comm);
1469                 break;
1470         case TRACE_SPECIAL:
1471                 trace_seq_printf(s, "# %ld %ld %ld\n",
1472                                  entry->special.arg1,
1473                                  entry->special.arg2,
1474                                  entry->special.arg3);
1475                 break;
1476         case TRACE_STACK:
1477                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1478                         if (i)
1479                                 trace_seq_puts(s, " <= ");
1480                         seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1481                 }
1482                 trace_seq_puts(s, "\n");
1483                 break;
1484         default:
1485                 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1486         }
1487         return 1;
1488 }
1489
1490 static int print_trace_fmt(struct trace_iterator *iter)
1491 {
1492         struct trace_seq *s = &iter->seq;
1493         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1494         struct trace_entry *entry;
1495         unsigned long usec_rem;
1496         unsigned long long t;
1497         unsigned long secs;
1498         char *comm;
1499         int ret;
1500         int S, T;
1501         int i;
1502
1503         entry = iter->ent;
1504
1505         comm = trace_find_cmdline(iter->ent->pid);
1506
1507         t = ns2usecs(entry->t);
1508         usec_rem = do_div(t, 1000000ULL);
1509         secs = (unsigned long)t;
1510
1511         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1512         if (!ret)
1513                 return 0;
1514         ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1515         if (!ret)
1516                 return 0;
1517         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1518         if (!ret)
1519                 return 0;
1520
1521         switch (entry->type) {
1522         case TRACE_FN:
1523                 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1524                 if (!ret)
1525                         return 0;
1526                 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1527                                                 entry->fn.parent_ip) {
1528                         ret = trace_seq_printf(s, " <-");
1529                         if (!ret)
1530                                 return 0;
1531                         ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1532                                                sym_flags);
1533                         if (!ret)
1534                                 return 0;
1535                 }
1536                 ret = trace_seq_printf(s, "\n");
1537                 if (!ret)
1538                         return 0;
1539                 break;
1540         case TRACE_CTX:
1541         case TRACE_WAKE:
1542                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1543                         state_to_char[entry->ctx.prev_state] : 'X';
1544                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1545                         state_to_char[entry->ctx.next_state] : 'X';
1546                 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1547                                        entry->ctx.prev_pid,
1548                                        entry->ctx.prev_prio,
1549                                        S,
1550                                        entry->type == TRACE_CTX ? "==>" : "  +",
1551                                        entry->ctx.next_pid,
1552                                        entry->ctx.next_prio,
1553                                        T);
1554                 if (!ret)
1555                         return 0;
1556                 break;
1557         case TRACE_SPECIAL:
1558                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1559                                  entry->special.arg1,
1560                                  entry->special.arg2,
1561                                  entry->special.arg3);
1562                 if (!ret)
1563                         return 0;
1564                 break;
1565         case TRACE_STACK:
1566                 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1567                         if (i) {
1568                                 ret = trace_seq_puts(s, " <= ");
1569                                 if (!ret)
1570                                         return 0;
1571                         }
1572                         ret = seq_print_ip_sym(s, entry->stack.caller[i],
1573                                                sym_flags);
1574                         if (!ret)
1575                                 return 0;
1576                 }
1577                 ret = trace_seq_puts(s, "\n");
1578                 if (!ret)
1579                         return 0;
1580                 break;
1581         }
1582         return 1;
1583 }
1584
1585 static int print_raw_fmt(struct trace_iterator *iter)
1586 {
1587         struct trace_seq *s = &iter->seq;
1588         struct trace_entry *entry;
1589         int ret;
1590         int S, T;
1591
1592         entry = iter->ent;
1593
1594         ret = trace_seq_printf(s, "%d %d %llu ",
1595                 entry->pid, iter->cpu, entry->t);
1596         if (!ret)
1597                 return 0;
1598
1599         switch (entry->type) {
1600         case TRACE_FN:
1601                 ret = trace_seq_printf(s, "%x %x\n",
1602                                         entry->fn.ip, entry->fn.parent_ip);
1603                 if (!ret)
1604                         return 0;
1605                 break;
1606         case TRACE_CTX:
1607         case TRACE_WAKE:
1608                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1609                         state_to_char[entry->ctx.prev_state] : 'X';
1610                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1611                         state_to_char[entry->ctx.next_state] : 'X';
1612                 if (entry->type == TRACE_WAKE)
1613                         S = '+';
1614                 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1615                                        entry->ctx.prev_pid,
1616                                        entry->ctx.prev_prio,
1617                                        S,
1618                                        entry->ctx.next_pid,
1619                                        entry->ctx.next_prio,
1620                                        T);
1621                 if (!ret)
1622                         return 0;
1623                 break;
1624         case TRACE_SPECIAL:
1625         case TRACE_STACK:
1626                 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1627                                  entry->special.arg1,
1628                                  entry->special.arg2,
1629                                  entry->special.arg3);
1630                 if (!ret)
1631                         return 0;
1632                 break;
1633         }
1634         return 1;
1635 }
1636
1637 #define SEQ_PUT_FIELD_RET(s, x)                         \
1638 do {                                                    \
1639         if (!trace_seq_putmem(s, &(x), sizeof(x)))      \
1640                 return 0;                               \
1641 } while (0)
1642
1643 #define SEQ_PUT_HEX_FIELD_RET(s, x)                     \
1644 do {                                                    \
1645         if (!trace_seq_putmem_hex(s, &(x), sizeof(x)))  \
1646                 return 0;                               \
1647 } while (0)
1648
1649 static int print_hex_fmt(struct trace_iterator *iter)
1650 {
1651         struct trace_seq *s = &iter->seq;
1652         unsigned char newline = '\n';
1653         struct trace_entry *entry;
1654         int S, T;
1655
1656         entry = iter->ent;
1657
1658         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1659         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1660         SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1661
1662         switch (entry->type) {
1663         case TRACE_FN:
1664                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1665                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1666                 break;
1667         case TRACE_CTX:
1668         case TRACE_WAKE:
1669                 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1670                         state_to_char[entry->ctx.prev_state] : 'X';
1671                 T = entry->ctx.next_state < sizeof(state_to_char) ?
1672                         state_to_char[entry->ctx.next_state] : 'X';
1673                 if (entry->type == TRACE_WAKE)
1674                         S = '+';
1675                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1676                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1677                 SEQ_PUT_HEX_FIELD_RET(s, S);
1678                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1679                 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1680                 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1681                 SEQ_PUT_HEX_FIELD_RET(s, T);
1682                 break;
1683         case TRACE_SPECIAL:
1684         case TRACE_STACK:
1685                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1686                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1687                 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1688                 break;
1689         }
1690         SEQ_PUT_FIELD_RET(s, newline);
1691
1692         return 1;
1693 }
1694
1695 static int print_bin_fmt(struct trace_iterator *iter)
1696 {
1697         struct trace_seq *s = &iter->seq;
1698         struct trace_entry *entry;
1699
1700         entry = iter->ent;
1701
1702         SEQ_PUT_FIELD_RET(s, entry->pid);
1703         SEQ_PUT_FIELD_RET(s, entry->cpu);
1704         SEQ_PUT_FIELD_RET(s, entry->t);
1705
1706         switch (entry->type) {
1707         case TRACE_FN:
1708                 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1709                 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1710                 break;
1711         case TRACE_CTX:
1712                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1713                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1714                 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1715                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1716                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1717                 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1718                 break;
1719         case TRACE_SPECIAL:
1720         case TRACE_STACK:
1721                 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1722                 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1723                 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1724                 break;
1725         }
1726         return 1;
1727 }
1728
1729 static int trace_empty(struct trace_iterator *iter)
1730 {
1731         struct trace_array_cpu *data;
1732         int cpu;
1733
1734         for_each_tracing_cpu(cpu) {
1735                 data = iter->tr->data[cpu];
1736
1737                 if (head_page(data) && data->trace_idx &&
1738                     (data->trace_tail != data->trace_head ||
1739                      data->trace_tail_idx != data->trace_head_idx))
1740                         return 0;
1741         }
1742         return 1;
1743 }
1744
1745 static int print_trace_line(struct trace_iterator *iter)
1746 {
1747         if (iter->trace && iter->trace->print_line)
1748                 return iter->trace->print_line(iter);
1749
1750         if (trace_flags & TRACE_ITER_BIN)
1751                 return print_bin_fmt(iter);
1752
1753         if (trace_flags & TRACE_ITER_HEX)
1754                 return print_hex_fmt(iter);
1755
1756         if (trace_flags & TRACE_ITER_RAW)
1757                 return print_raw_fmt(iter);
1758
1759         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1760                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1761
1762         return print_trace_fmt(iter);
1763 }
1764
1765 static int s_show(struct seq_file *m, void *v)
1766 {
1767         struct trace_iterator *iter = v;
1768
1769         if (iter->ent == NULL) {
1770                 if (iter->tr) {
1771                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1772                         seq_puts(m, "#\n");
1773                 }
1774                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1775                         /* print nothing if the buffers are empty */
1776                         if (trace_empty(iter))
1777                                 return 0;
1778                         print_trace_header(m, iter);
1779                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1780                                 print_lat_help_header(m);
1781                 } else {
1782                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1783                                 print_func_help_header(m);
1784                 }
1785         } else {
1786                 print_trace_line(iter);
1787                 trace_print_seq(m, &iter->seq);
1788         }
1789
1790         return 0;
1791 }
1792
1793 static struct seq_operations tracer_seq_ops = {
1794         .start          = s_start,
1795         .next           = s_next,
1796         .stop           = s_stop,
1797         .show           = s_show,
1798 };
1799
1800 static struct trace_iterator *
1801 __tracing_open(struct inode *inode, struct file *file, int *ret)
1802 {
1803         struct trace_iterator *iter;
1804
1805         if (tracing_disabled) {
1806                 *ret = -ENODEV;
1807                 return NULL;
1808         }
1809
1810         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1811         if (!iter) {
1812                 *ret = -ENOMEM;
1813                 goto out;
1814         }
1815
1816         mutex_lock(&trace_types_lock);
1817         if (current_trace && current_trace->print_max)
1818                 iter->tr = &max_tr;
1819         else
1820                 iter->tr = inode->i_private;
1821         iter->trace = current_trace;
1822         iter->pos = -1;
1823
1824         /* TODO stop tracer */
1825         *ret = seq_open(file, &tracer_seq_ops);
1826         if (!*ret) {
1827                 struct seq_file *m = file->private_data;
1828                 m->private = iter;
1829
1830                 /* stop the trace while dumping */
1831                 if (iter->tr->ctrl)
1832                         tracer_enabled = 0;
1833
1834                 if (iter->trace && iter->trace->open)
1835                         iter->trace->open(iter);
1836         } else {
1837                 kfree(iter);
1838                 iter = NULL;
1839         }
1840         mutex_unlock(&trace_types_lock);
1841
1842  out:
1843         return iter;
1844 }
1845
1846 int tracing_open_generic(struct inode *inode, struct file *filp)
1847 {
1848         if (tracing_disabled)
1849                 return -ENODEV;
1850
1851         filp->private_data = inode->i_private;
1852         return 0;
1853 }
1854
1855 int tracing_release(struct inode *inode, struct file *file)
1856 {
1857         struct seq_file *m = (struct seq_file *)file->private_data;
1858         struct trace_iterator *iter = m->private;
1859
1860         mutex_lock(&trace_types_lock);
1861         if (iter->trace && iter->trace->close)
1862                 iter->trace->close(iter);
1863
1864         /* reenable tracing if it was previously enabled */
1865         if (iter->tr->ctrl)
1866                 tracer_enabled = 1;
1867         mutex_unlock(&trace_types_lock);
1868
1869         seq_release(inode, file);
1870         kfree(iter);
1871         return 0;
1872 }
1873
1874 static int tracing_open(struct inode *inode, struct file *file)
1875 {
1876         int ret;
1877
1878         __tracing_open(inode, file, &ret);
1879
1880         return ret;
1881 }
1882
1883 static int tracing_lt_open(struct inode *inode, struct file *file)
1884 {
1885         struct trace_iterator *iter;
1886         int ret;
1887
1888         iter = __tracing_open(inode, file, &ret);
1889
1890         if (!ret)
1891                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1892
1893         return ret;
1894 }
1895
1896
1897 static void *
1898 t_next(struct seq_file *m, void *v, loff_t *pos)
1899 {
1900         struct tracer *t = m->private;
1901
1902         (*pos)++;
1903
1904         if (t)
1905                 t = t->next;
1906
1907         m->private = t;
1908
1909         return t;
1910 }
1911
1912 static void *t_start(struct seq_file *m, loff_t *pos)
1913 {
1914         struct tracer *t = m->private;
1915         loff_t l = 0;
1916
1917         mutex_lock(&trace_types_lock);
1918         for (; t && l < *pos; t = t_next(m, t, &l))
1919                 ;
1920
1921         return t;
1922 }
1923
1924 static void t_stop(struct seq_file *m, void *p)
1925 {
1926         mutex_unlock(&trace_types_lock);
1927 }
1928
1929 static int t_show(struct seq_file *m, void *v)
1930 {
1931         struct tracer *t = v;
1932
1933         if (!t)
1934                 return 0;
1935
1936         seq_printf(m, "%s", t->name);
1937         if (t->next)
1938                 seq_putc(m, ' ');
1939         else
1940                 seq_putc(m, '\n');
1941
1942         return 0;
1943 }
1944
1945 static struct seq_operations show_traces_seq_ops = {
1946         .start          = t_start,
1947         .next           = t_next,
1948         .stop           = t_stop,
1949         .show           = t_show,
1950 };
1951
1952 static int show_traces_open(struct inode *inode, struct file *file)
1953 {
1954         int ret;
1955
1956         if (tracing_disabled)
1957                 return -ENODEV;
1958
1959         ret = seq_open(file, &show_traces_seq_ops);
1960         if (!ret) {
1961                 struct seq_file *m = file->private_data;
1962                 m->private = trace_types;
1963         }
1964
1965         return ret;
1966 }
1967
1968 static struct file_operations tracing_fops = {
1969         .open           = tracing_open,
1970         .read           = seq_read,
1971         .llseek         = seq_lseek,
1972         .release        = tracing_release,
1973 };
1974
1975 static struct file_operations tracing_lt_fops = {
1976         .open           = tracing_lt_open,
1977         .read           = seq_read,
1978         .llseek         = seq_lseek,
1979         .release        = tracing_release,
1980 };
1981
1982 static struct file_operations show_traces_fops = {
1983         .open           = show_traces_open,
1984         .read           = seq_read,
1985         .release        = seq_release,
1986 };
1987
1988 /*
1989  * Only trace on a CPU if the bitmask is set:
1990  */
1991 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1992
1993 /*
1994  * When tracing/tracing_cpu_mask is modified then this holds
1995  * the new bitmask we are about to install:
1996  */
1997 static cpumask_t tracing_cpumask_new;
1998
1999 /*
2000  * The tracer itself will not take this lock, but still we want
2001  * to provide a consistent cpumask to user-space:
2002  */
2003 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2004
2005 /*
2006  * Temporary storage for the character representation of the
2007  * CPU bitmask (and one more byte for the newline):
2008  */
2009 static char mask_str[NR_CPUS + 1];
2010
2011 static ssize_t
2012 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2013                      size_t count, loff_t *ppos)
2014 {
2015         int len;
2016
2017         mutex_lock(&tracing_cpumask_update_lock);
2018
2019         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2020         if (count - len < 2) {
2021                 count = -EINVAL;
2022                 goto out_err;
2023         }
2024         len += sprintf(mask_str + len, "\n");
2025         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2026
2027 out_err:
2028         mutex_unlock(&tracing_cpumask_update_lock);
2029
2030         return count;
2031 }
2032
2033 static ssize_t
2034 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2035                       size_t count, loff_t *ppos)
2036 {
2037         int err, cpu;
2038
2039         mutex_lock(&tracing_cpumask_update_lock);
2040         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2041         if (err)
2042                 goto err_unlock;
2043
2044         raw_local_irq_disable();
2045         __raw_spin_lock(&ftrace_max_lock);
2046         for_each_tracing_cpu(cpu) {
2047                 /*
2048                  * Increase/decrease the disabled counter if we are
2049                  * about to flip a bit in the cpumask:
2050                  */
2051                 if (cpu_isset(cpu, tracing_cpumask) &&
2052                                 !cpu_isset(cpu, tracing_cpumask_new)) {
2053                         atomic_inc(&global_trace.data[cpu]->disabled);
2054                 }
2055                 if (!cpu_isset(cpu, tracing_cpumask) &&
2056                                 cpu_isset(cpu, tracing_cpumask_new)) {
2057                         atomic_dec(&global_trace.data[cpu]->disabled);
2058                 }
2059         }
2060         __raw_spin_unlock(&ftrace_max_lock);
2061         raw_local_irq_enable();
2062
2063         tracing_cpumask = tracing_cpumask_new;
2064
2065         mutex_unlock(&tracing_cpumask_update_lock);
2066
2067         return count;
2068
2069 err_unlock:
2070         mutex_unlock(&tracing_cpumask_update_lock);
2071
2072         return err;
2073 }
2074
2075 static struct file_operations tracing_cpumask_fops = {
2076         .open           = tracing_open_generic,
2077         .read           = tracing_cpumask_read,
2078         .write          = tracing_cpumask_write,
2079 };
2080
2081 static ssize_t
2082 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2083                        size_t cnt, loff_t *ppos)
2084 {
2085         char *buf;
2086         int r = 0;
2087         int len = 0;
2088         int i;
2089
2090         /* calulate max size */
2091         for (i = 0; trace_options[i]; i++) {
2092                 len += strlen(trace_options[i]);
2093                 len += 3; /* "no" and space */
2094         }
2095
2096         /* +2 for \n and \0 */
2097         buf = kmalloc(len + 2, GFP_KERNEL);
2098         if (!buf)
2099                 return -ENOMEM;
2100
2101         for (i = 0; trace_options[i]; i++) {
2102                 if (trace_flags & (1 << i))
2103                         r += sprintf(buf + r, "%s ", trace_options[i]);
2104                 else
2105                         r += sprintf(buf + r, "no%s ", trace_options[i]);
2106         }
2107
2108         r += sprintf(buf + r, "\n");
2109         WARN_ON(r >= len + 2);
2110
2111         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2112
2113         kfree(buf);
2114
2115         return r;
2116 }
2117
2118 static ssize_t
2119 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2120                         size_t cnt, loff_t *ppos)
2121 {
2122         char buf[64];
2123         char *cmp = buf;
2124         int neg = 0;
2125         int i;
2126
2127         if (cnt >= sizeof(buf))
2128                 return -EINVAL;
2129
2130         if (copy_from_user(&buf, ubuf, cnt))
2131                 return -EFAULT;
2132
2133         buf[cnt] = 0;
2134
2135         if (strncmp(buf, "no", 2) == 0) {
2136                 neg = 1;
2137                 cmp += 2;
2138         }
2139
2140         for (i = 0; trace_options[i]; i++) {
2141                 int len = strlen(trace_options[i]);
2142
2143                 if (strncmp(cmp, trace_options[i], len) == 0) {
2144                         if (neg)
2145                                 trace_flags &= ~(1 << i);
2146                         else
2147                                 trace_flags |= (1 << i);
2148                         break;
2149                 }
2150         }
2151         /*
2152          * If no option could be set, return an error:
2153          */
2154         if (!trace_options[i])
2155                 return -EINVAL;
2156
2157         filp->f_pos += cnt;
2158
2159         return cnt;
2160 }
2161
2162 static struct file_operations tracing_iter_fops = {
2163         .open           = tracing_open_generic,
2164         .read           = tracing_iter_ctrl_read,
2165         .write          = tracing_iter_ctrl_write,
2166 };
2167
2168 static const char readme_msg[] =
2169         "tracing mini-HOWTO:\n\n"
2170         "# mkdir /debug\n"
2171         "# mount -t debugfs nodev /debug\n\n"
2172         "# cat /debug/tracing/available_tracers\n"
2173         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2174         "# cat /debug/tracing/current_tracer\n"
2175         "none\n"
2176         "# echo sched_switch > /debug/tracing/current_tracer\n"
2177         "# cat /debug/tracing/current_tracer\n"
2178         "sched_switch\n"
2179         "# cat /debug/tracing/iter_ctrl\n"
2180         "noprint-parent nosym-offset nosym-addr noverbose\n"
2181         "# echo print-parent > /debug/tracing/iter_ctrl\n"
2182         "# echo 1 > /debug/tracing/tracing_enabled\n"
2183         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2184         "echo 0 > /debug/tracing/tracing_enabled\n"
2185 ;
2186
2187 static ssize_t
2188 tracing_readme_read(struct file *filp, char __user *ubuf,
2189                        size_t cnt, loff_t *ppos)
2190 {
2191         return simple_read_from_buffer(ubuf, cnt, ppos,
2192                                         readme_msg, strlen(readme_msg));
2193 }
2194
2195 static struct file_operations tracing_readme_fops = {
2196         .open           = tracing_open_generic,
2197         .read           = tracing_readme_read,
2198 };
2199
2200 static ssize_t
2201 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2202                   size_t cnt, loff_t *ppos)
2203 {
2204         struct trace_array *tr = filp->private_data;
2205         char buf[64];
2206         int r;
2207
2208         r = sprintf(buf, "%ld\n", tr->ctrl);
2209         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2210 }
2211
2212 static ssize_t
2213 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2214                    size_t cnt, loff_t *ppos)
2215 {
2216         struct trace_array *tr = filp->private_data;
2217         char buf[64];
2218         long val;
2219         int ret;
2220
2221         if (cnt >= sizeof(buf))
2222                 return -EINVAL;
2223
2224         if (copy_from_user(&buf, ubuf, cnt))
2225                 return -EFAULT;
2226
2227         buf[cnt] = 0;
2228
2229         ret = strict_strtoul(buf, 10, &val);
2230         if (ret < 0)
2231                 return ret;
2232
2233         val = !!val;
2234
2235         mutex_lock(&trace_types_lock);
2236         if (tr->ctrl ^ val) {
2237                 if (val)
2238                         tracer_enabled = 1;
2239                 else
2240                         tracer_enabled = 0;
2241
2242                 tr->ctrl = val;
2243
2244                 if (current_trace && current_trace->ctrl_update)
2245                         current_trace->ctrl_update(tr);
2246         }
2247         mutex_unlock(&trace_types_lock);
2248
2249         filp->f_pos += cnt;
2250
2251         return cnt;
2252 }
2253
2254 static ssize_t
2255 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2256                        size_t cnt, loff_t *ppos)
2257 {
2258         char buf[max_tracer_type_len+2];
2259         int r;
2260
2261         mutex_lock(&trace_types_lock);
2262         if (current_trace)
2263                 r = sprintf(buf, "%s\n", current_trace->name);
2264         else
2265                 r = sprintf(buf, "\n");
2266         mutex_unlock(&trace_types_lock);
2267
2268         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2269 }
2270
2271 static ssize_t
2272 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2273                         size_t cnt, loff_t *ppos)
2274 {
2275         struct trace_array *tr = &global_trace;
2276         struct tracer *t;
2277         char buf[max_tracer_type_len+1];
2278         int i;
2279
2280         if (cnt > max_tracer_type_len)
2281                 cnt = max_tracer_type_len;
2282
2283         if (copy_from_user(&buf, ubuf, cnt))
2284                 return -EFAULT;
2285
2286         buf[cnt] = 0;
2287
2288         /* strip ending whitespace. */
2289         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2290                 buf[i] = 0;
2291
2292         mutex_lock(&trace_types_lock);
2293         for (t = trace_types; t; t = t->next) {
2294                 if (strcmp(t->name, buf) == 0)
2295                         break;
2296         }
2297         if (!t || t == current_trace)
2298                 goto out;
2299
2300         if (current_trace && current_trace->reset)
2301                 current_trace->reset(tr);
2302
2303         current_trace = t;
2304         if (t->init)
2305                 t->init(tr);
2306
2307  out:
2308         mutex_unlock(&trace_types_lock);
2309
2310         filp->f_pos += cnt;
2311
2312         return cnt;
2313 }
2314
2315 static ssize_t
2316 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2317                      size_t cnt, loff_t *ppos)
2318 {
2319         unsigned long *ptr = filp->private_data;
2320         char buf[64];
2321         int r;
2322
2323         r = snprintf(buf, sizeof(buf), "%ld\n",
2324                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2325         if (r > sizeof(buf))
2326                 r = sizeof(buf);
2327         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2328 }
2329
2330 static ssize_t
2331 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2332                       size_t cnt, loff_t *ppos)
2333 {
2334         long *ptr = filp->private_data;
2335         char buf[64];
2336         long val;
2337         int ret;
2338
2339         if (cnt >= sizeof(buf))
2340                 return -EINVAL;
2341
2342         if (copy_from_user(&buf, ubuf, cnt))
2343                 return -EFAULT;
2344
2345         buf[cnt] = 0;
2346
2347         ret = strict_strtoul(buf, 10, &val);
2348         if (ret < 0)
2349                 return ret;
2350
2351         *ptr = val * 1000;
2352
2353         return cnt;
2354 }
2355
2356 static atomic_t tracing_reader;
2357
2358 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2359 {
2360         struct trace_iterator *iter;
2361
2362         if (tracing_disabled)
2363                 return -ENODEV;
2364
2365         /* We only allow for reader of the pipe */
2366         if (atomic_inc_return(&tracing_reader) != 1) {
2367                 atomic_dec(&tracing_reader);
2368                 return -EBUSY;
2369         }
2370
2371         /* create a buffer to store the information to pass to userspace */
2372         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2373         if (!iter)
2374                 return -ENOMEM;
2375
2376         mutex_lock(&trace_types_lock);
2377         iter->tr = &global_trace;
2378         iter->trace = current_trace;
2379         filp->private_data = iter;
2380
2381         if (iter->trace->pipe_open)
2382                 iter->trace->pipe_open(iter);
2383         mutex_unlock(&trace_types_lock);
2384
2385         return 0;
2386 }
2387
2388 static int tracing_release_pipe(struct inode *inode, struct file *file)
2389 {
2390         struct trace_iterator *iter = file->private_data;
2391
2392         kfree(iter);
2393         atomic_dec(&tracing_reader);
2394
2395         return 0;
2396 }
2397
2398 static unsigned int
2399 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2400 {
2401         struct trace_iterator *iter = filp->private_data;
2402
2403         if (trace_flags & TRACE_ITER_BLOCK) {
2404                 /*
2405                  * Always select as readable when in blocking mode
2406                  */
2407                 return POLLIN | POLLRDNORM;
2408         } else {
2409                 if (!trace_empty(iter))
2410                         return POLLIN | POLLRDNORM;
2411                 poll_wait(filp, &trace_wait, poll_table);
2412                 if (!trace_empty(iter))
2413                         return POLLIN | POLLRDNORM;
2414
2415                 return 0;
2416         }
2417 }
2418
2419 /*
2420  * Consumer reader.
2421  */
2422 static ssize_t
2423 tracing_read_pipe(struct file *filp, char __user *ubuf,
2424                   size_t cnt, loff_t *ppos)
2425 {
2426         struct trace_iterator *iter = filp->private_data;
2427         struct trace_array_cpu *data;
2428         static cpumask_t mask;
2429         unsigned long flags;
2430 #ifdef CONFIG_FTRACE
2431         int ftrace_save;
2432 #endif
2433         int cpu;
2434         ssize_t sret;
2435
2436         /* return any leftover data */
2437         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2438         if (sret != -EBUSY)
2439                 return sret;
2440         sret = 0;
2441
2442         trace_seq_reset(&iter->seq);
2443
2444         mutex_lock(&trace_types_lock);
2445         if (iter->trace->read) {
2446                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2447                 if (sret)
2448                         goto out;
2449         }
2450
2451         while (trace_empty(iter)) {
2452
2453                 if ((filp->f_flags & O_NONBLOCK)) {
2454                         sret = -EAGAIN;
2455                         goto out;
2456                 }
2457
2458                 /*
2459                  * This is a make-shift waitqueue. The reason we don't use
2460                  * an actual wait queue is because:
2461                  *  1) we only ever have one waiter
2462                  *  2) the tracing, traces all functions, we don't want
2463                  *     the overhead of calling wake_up and friends
2464                  *     (and tracing them too)
2465                  *     Anyway, this is really very primitive wakeup.
2466                  */
2467                 set_current_state(TASK_INTERRUPTIBLE);
2468                 iter->tr->waiter = current;
2469
2470                 mutex_unlock(&trace_types_lock);
2471
2472                 /* sleep for 100 msecs, and try again. */
2473                 schedule_timeout(HZ/10);
2474
2475                 mutex_lock(&trace_types_lock);
2476
2477                 iter->tr->waiter = NULL;
2478
2479                 if (signal_pending(current)) {
2480                         sret = -EINTR;
2481                         goto out;
2482                 }
2483
2484                 if (iter->trace != current_trace)
2485                         goto out;
2486
2487                 /*
2488                  * We block until we read something and tracing is disabled.
2489                  * We still block if tracing is disabled, but we have never
2490                  * read anything. This allows a user to cat this file, and
2491                  * then enable tracing. But after we have read something,
2492                  * we give an EOF when tracing is again disabled.
2493                  *
2494                  * iter->pos will be 0 if we haven't read anything.
2495                  */
2496                 if (!tracer_enabled && iter->pos)
2497                         break;
2498
2499                 continue;
2500         }
2501
2502         /* stop when tracing is finished */
2503         if (trace_empty(iter))
2504                 goto out;
2505
2506         if (cnt >= PAGE_SIZE)
2507                 cnt = PAGE_SIZE - 1;
2508
2509         /* reset all but tr, trace, and overruns */
2510         memset(&iter->seq, 0,
2511                sizeof(struct trace_iterator) -
2512                offsetof(struct trace_iterator, seq));
2513         iter->pos = -1;
2514
2515         /*
2516          * We need to stop all tracing on all CPUS to read the
2517          * the next buffer. This is a bit expensive, but is
2518          * not done often. We fill all what we can read,
2519          * and then release the locks again.
2520          */
2521
2522         cpus_clear(mask);
2523         local_irq_save(flags);
2524 #ifdef CONFIG_FTRACE
2525         ftrace_save = ftrace_enabled;
2526         ftrace_enabled = 0;
2527 #endif
2528         smp_wmb();
2529         for_each_tracing_cpu(cpu) {
2530                 data = iter->tr->data[cpu];
2531
2532                 if (!head_page(data) || !data->trace_idx)
2533                         continue;
2534
2535                 atomic_inc(&data->disabled);
2536                 cpu_set(cpu, mask);
2537         }
2538
2539         for_each_cpu_mask(cpu, mask) {
2540                 data = iter->tr->data[cpu];
2541                 __raw_spin_lock(&data->lock);
2542
2543                 if (data->overrun > iter->last_overrun[cpu])
2544                         iter->overrun[cpu] +=
2545                                 data->overrun - iter->last_overrun[cpu];
2546                 iter->last_overrun[cpu] = data->overrun;
2547         }
2548
2549         while (find_next_entry_inc(iter) != NULL) {
2550                 int ret;
2551                 int len = iter->seq.len;
2552
2553                 ret = print_trace_line(iter);
2554                 if (!ret) {
2555                         /* don't print partial lines */
2556                         iter->seq.len = len;
2557                         break;
2558                 }
2559
2560                 trace_consume(iter);
2561
2562                 if (iter->seq.len >= cnt)
2563                         break;
2564         }
2565
2566         for_each_cpu_mask(cpu, mask) {
2567                 data = iter->tr->data[cpu];
2568                 __raw_spin_unlock(&data->lock);
2569         }
2570
2571         for_each_cpu_mask(cpu, mask) {
2572                 data = iter->tr->data[cpu];
2573                 atomic_dec(&data->disabled);
2574         }
2575 #ifdef CONFIG_FTRACE
2576         ftrace_enabled = ftrace_save;
2577 #endif
2578         local_irq_restore(flags);
2579
2580         /* Now copy what we have to the user */
2581         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2582         if (iter->seq.readpos >= iter->seq.len)
2583                 trace_seq_reset(&iter->seq);
2584         if (sret == -EBUSY)
2585                 sret = 0;
2586
2587 out:
2588         mutex_unlock(&trace_types_lock);
2589
2590         return sret;
2591 }
2592
2593 static ssize_t
2594 tracing_entries_read(struct file *filp, char __user *ubuf,
2595                      size_t cnt, loff_t *ppos)
2596 {
2597         struct trace_array *tr = filp->private_data;
2598         char buf[64];
2599         int r;
2600
2601         r = sprintf(buf, "%lu\n", tr->entries);
2602         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2603 }
2604
2605 static ssize_t
2606 tracing_entries_write(struct file *filp, const char __user *ubuf,
2607                       size_t cnt, loff_t *ppos)
2608 {
2609         unsigned long val;
2610         char buf[64];
2611         int ret;
2612
2613         if (cnt >= sizeof(buf))
2614                 return -EINVAL;
2615
2616         if (copy_from_user(&buf, ubuf, cnt))
2617                 return -EFAULT;
2618
2619         buf[cnt] = 0;
2620
2621         ret = strict_strtoul(buf, 10, &val);
2622         if (ret < 0)
2623                 return ret;
2624
2625         /* must have at least 1 entry */
2626         if (!val)
2627                 return -EINVAL;
2628
2629         mutex_lock(&trace_types_lock);
2630
2631         if (current_trace != &no_tracer) {
2632                 cnt = -EBUSY;
2633                 pr_info("ftrace: set current_tracer to none"
2634                         " before modifying buffer size\n");
2635                 goto out;
2636         }
2637
2638         if (val > global_trace.entries) {
2639                 long pages_requested;
2640                 unsigned long freeable_pages;
2641
2642                 /* make sure we have enough memory before mapping */
2643                 pages_requested =
2644                         (val + (ENTRIES_PER_PAGE-1)) / ENTRIES_PER_PAGE;
2645
2646                 /* account for each buffer (and max_tr) */
2647                 pages_requested *= tracing_nr_buffers * 2;
2648
2649                 /* Check for overflow */
2650                 if (pages_requested < 0) {
2651                         cnt = -ENOMEM;
2652                         goto out;
2653                 }
2654
2655                 freeable_pages = determine_dirtyable_memory();
2656
2657                 /* we only allow to request 1/4 of useable memory */
2658                 if (pages_requested >
2659                     ((freeable_pages + tracing_pages_allocated) / 4)) {
2660                         cnt = -ENOMEM;
2661                         goto out;
2662                 }
2663
2664                 while (global_trace.entries < val) {
2665                         if (trace_alloc_page()) {
2666                                 cnt = -ENOMEM;
2667                                 goto out;
2668                         }
2669                         /* double check that we don't go over the known pages */
2670                         if (tracing_pages_allocated > pages_requested)
2671                                 break;
2672                 }
2673
2674         } else {
2675                 /* include the number of entries in val (inc of page entries) */
2676                 while (global_trace.entries > val + (ENTRIES_PER_PAGE - 1))
2677                         trace_free_page();
2678         }
2679
2680         filp->f_pos += cnt;
2681
2682  out:
2683         max_tr.entries = global_trace.entries;
2684         mutex_unlock(&trace_types_lock);
2685
2686         return cnt;
2687 }
2688
2689 static struct file_operations tracing_max_lat_fops = {
2690         .open           = tracing_open_generic,
2691         .read           = tracing_max_lat_read,
2692         .write          = tracing_max_lat_write,
2693 };
2694
2695 static struct file_operations tracing_ctrl_fops = {
2696         .open           = tracing_open_generic,
2697         .read           = tracing_ctrl_read,
2698         .write          = tracing_ctrl_write,
2699 };
2700
2701 static struct file_operations set_tracer_fops = {
2702         .open           = tracing_open_generic,
2703         .read           = tracing_set_trace_read,
2704         .write          = tracing_set_trace_write,
2705 };
2706
2707 static struct file_operations tracing_pipe_fops = {
2708         .open           = tracing_open_pipe,
2709         .poll           = tracing_poll_pipe,
2710         .read           = tracing_read_pipe,
2711         .release        = tracing_release_pipe,
2712 };
2713
2714 static struct file_operations tracing_entries_fops = {
2715         .open           = tracing_open_generic,
2716         .read           = tracing_entries_read,
2717         .write          = tracing_entries_write,
2718 };
2719
2720 #ifdef CONFIG_DYNAMIC_FTRACE
2721
2722 static ssize_t
2723 tracing_read_long(struct file *filp, char __user *ubuf,
2724                   size_t cnt, loff_t *ppos)
2725 {
2726         unsigned long *p = filp->private_data;
2727         char buf[64];
2728         int r;
2729
2730         r = sprintf(buf, "%ld\n", *p);
2731
2732         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2733 }
2734
2735 static struct file_operations tracing_read_long_fops = {
2736         .open           = tracing_open_generic,
2737         .read           = tracing_read_long,
2738 };
2739 #endif
2740
2741 static struct dentry *d_tracer;
2742
2743 struct dentry *tracing_init_dentry(void)
2744 {
2745         static int once;
2746
2747         if (d_tracer)
2748                 return d_tracer;
2749
2750         d_tracer = debugfs_create_dir("tracing", NULL);
2751
2752         if (!d_tracer && !once) {
2753                 once = 1;
2754                 pr_warning("Could not create debugfs directory 'tracing'\n");
2755                 return NULL;
2756         }
2757
2758         return d_tracer;
2759 }
2760
2761 #ifdef CONFIG_FTRACE_SELFTEST
2762 /* Let selftest have access to static functions in this file */
2763 #include "trace_selftest.c"
2764 #endif
2765
2766 static __init void tracer_init_debugfs(void)
2767 {
2768         struct dentry *d_tracer;
2769         struct dentry *entry;
2770
2771         d_tracer = tracing_init_dentry();
2772
2773         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2774                                     &global_trace, &tracing_ctrl_fops);
2775         if (!entry)
2776                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2777
2778         entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2779                                     NULL, &tracing_iter_fops);
2780         if (!entry)
2781                 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2782
2783         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2784                                     NULL, &tracing_cpumask_fops);
2785         if (!entry)
2786                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2787
2788         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2789                                     &global_trace, &tracing_lt_fops);
2790         if (!entry)
2791                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2792
2793         entry = debugfs_create_file("trace", 0444, d_tracer,
2794                                     &global_trace, &tracing_fops);
2795         if (!entry)
2796                 pr_warning("Could not create debugfs 'trace' entry\n");
2797
2798         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2799                                     &global_trace, &show_traces_fops);
2800         if (!entry)
2801                 pr_warning("Could not create debugfs 'trace' entry\n");
2802
2803         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2804                                     &global_trace, &set_tracer_fops);
2805         if (!entry)
2806                 pr_warning("Could not create debugfs 'trace' entry\n");
2807
2808         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2809                                     &tracing_max_latency,
2810                                     &tracing_max_lat_fops);
2811         if (!entry)
2812                 pr_warning("Could not create debugfs "
2813                            "'tracing_max_latency' entry\n");
2814
2815         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2816                                     &tracing_thresh, &tracing_max_lat_fops);
2817         if (!entry)
2818                 pr_warning("Could not create debugfs "
2819                            "'tracing_threash' entry\n");
2820         entry = debugfs_create_file("README", 0644, d_tracer,
2821                                     NULL, &tracing_readme_fops);
2822         if (!entry)
2823                 pr_warning("Could not create debugfs 'README' entry\n");
2824
2825         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2826                                     NULL, &tracing_pipe_fops);
2827         if (!entry)
2828                 pr_warning("Could not create debugfs "
2829                            "'tracing_threash' entry\n");
2830
2831         entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2832                                     &global_trace, &tracing_entries_fops);
2833         if (!entry)
2834                 pr_warning("Could not create debugfs "
2835                            "'tracing_threash' entry\n");
2836
2837 #ifdef CONFIG_DYNAMIC_FTRACE
2838         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2839                                     &ftrace_update_tot_cnt,
2840                                     &tracing_read_long_fops);
2841         if (!entry)
2842                 pr_warning("Could not create debugfs "
2843                            "'dyn_ftrace_total_info' entry\n");
2844 #endif
2845 }
2846
2847 static int trace_alloc_page(void)
2848 {
2849         struct trace_array_cpu *data;
2850         struct page *page, *tmp;
2851         LIST_HEAD(pages);
2852         void *array;
2853         unsigned pages_allocated = 0;
2854         int i;
2855
2856         /* first allocate a page for each CPU */
2857         for_each_tracing_cpu(i) {
2858                 array = (void *)__get_free_page(GFP_KERNEL);
2859                 if (array == NULL) {
2860                         printk(KERN_ERR "tracer: failed to allocate page"
2861                                "for trace buffer!\n");
2862                         goto free_pages;
2863                 }
2864
2865                 pages_allocated++;
2866                 page = virt_to_page(array);
2867                 list_add(&page->lru, &pages);
2868
2869 /* Only allocate if we are actually using the max trace */
2870 #ifdef CONFIG_TRACER_MAX_TRACE
2871                 array = (void *)__get_free_page(GFP_KERNEL);
2872                 if (array == NULL) {
2873                         printk(KERN_ERR "tracer: failed to allocate page"
2874                                "for trace buffer!\n");
2875                         goto free_pages;
2876                 }
2877                 pages_allocated++;
2878                 page = virt_to_page(array);
2879                 list_add(&page->lru, &pages);
2880 #endif
2881         }
2882
2883         /* Now that we successfully allocate a page per CPU, add them */
2884         for_each_tracing_cpu(i) {
2885                 data = global_trace.data[i];
2886                 page = list_entry(pages.next, struct page, lru);
2887                 list_del_init(&page->lru);
2888                 list_add_tail(&page->lru, &data->trace_pages);
2889                 ClearPageLRU(page);
2890
2891 #ifdef CONFIG_TRACER_MAX_TRACE
2892                 data = max_tr.data[i];
2893                 page = list_entry(pages.next, struct page, lru);
2894                 list_del_init(&page->lru);
2895                 list_add_tail(&page->lru, &data->trace_pages);
2896                 SetPageLRU(page);
2897 #endif
2898         }
2899         tracing_pages_allocated += pages_allocated;
2900         global_trace.entries += ENTRIES_PER_PAGE;
2901
2902         return 0;
2903
2904  free_pages:
2905         list_for_each_entry_safe(page, tmp, &pages, lru) {
2906                 list_del_init(&page->lru);
2907                 __free_page(page);
2908         }
2909         return -ENOMEM;
2910 }
2911
2912 static int trace_free_page(void)
2913 {
2914         struct trace_array_cpu *data;
2915         struct page *page;
2916         struct list_head *p;
2917         int i;
2918         int ret = 0;
2919
2920         /* free one page from each buffer */
2921         for_each_tracing_cpu(i) {
2922                 data = global_trace.data[i];
2923                 p = data->trace_pages.next;
2924                 if (p == &data->trace_pages) {
2925                         /* should never happen */
2926                         WARN_ON(1);
2927                         tracing_disabled = 1;
2928                         ret = -1;
2929                         break;
2930                 }
2931                 page = list_entry(p, struct page, lru);
2932                 ClearPageLRU(page);
2933                 list_del(&page->lru);
2934                 tracing_pages_allocated--;
2935                 tracing_pages_allocated--;
2936                 __free_page(page);
2937
2938                 tracing_reset(data);
2939
2940 #ifdef CONFIG_TRACER_MAX_TRACE
2941                 data = max_tr.data[i];
2942                 p = data->trace_pages.next;
2943                 if (p == &data->trace_pages) {
2944                         /* should never happen */
2945                         WARN_ON(1);
2946                         tracing_disabled = 1;
2947                         ret = -1;
2948                         break;
2949                 }
2950                 page = list_entry(p, struct page, lru);
2951                 ClearPageLRU(page);
2952                 list_del(&page->lru);
2953                 __free_page(page);
2954
2955                 tracing_reset(data);
2956 #endif
2957         }
2958         global_trace.entries -= ENTRIES_PER_PAGE;
2959
2960         return ret;
2961 }
2962
2963 __init static int tracer_alloc_buffers(void)
2964 {
2965         struct trace_array_cpu *data;
2966         void *array;
2967         struct page *page;
2968         int pages = 0;
2969         int ret = -ENOMEM;
2970         int i;
2971
2972         global_trace.ctrl = tracer_enabled;
2973
2974         /* TODO: make the number of buffers hot pluggable with CPUS */
2975         tracing_nr_buffers = num_possible_cpus();
2976         tracing_buffer_mask = cpu_possible_map;
2977
2978         /* Allocate the first page for all buffers */
2979         for_each_tracing_cpu(i) {
2980                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2981                 max_tr.data[i] = &per_cpu(max_data, i);
2982
2983                 array = (void *)__get_free_page(GFP_KERNEL);
2984                 if (array == NULL) {
2985                         printk(KERN_ERR "tracer: failed to allocate page"
2986                                "for trace buffer!\n");
2987                         goto free_buffers;
2988                 }
2989
2990                 /* set the array to the list */
2991                 INIT_LIST_HEAD(&data->trace_pages);
2992                 page = virt_to_page(array);
2993                 list_add(&page->lru, &data->trace_pages);
2994                 /* use the LRU flag to differentiate the two buffers */
2995                 ClearPageLRU(page);
2996
2997                 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2998                 max_tr.data[i]->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2999
3000 /* Only allocate if we are actually using the max trace */
3001 #ifdef CONFIG_TRACER_MAX_TRACE
3002                 array = (void *)__get_free_page(GFP_KERNEL);
3003                 if (array == NULL) {
3004                         printk(KERN_ERR "tracer: failed to allocate page"
3005                                "for trace buffer!\n");
3006                         goto free_buffers;
3007                 }
3008
3009                 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
3010                 page = virt_to_page(array);
3011                 list_add(&page->lru, &max_tr.data[i]->trace_pages);
3012                 SetPageLRU(page);
3013 #endif
3014         }
3015
3016         /*
3017          * Since we allocate by orders of pages, we may be able to
3018          * round up a bit.
3019          */
3020         global_trace.entries = ENTRIES_PER_PAGE;
3021         pages++;
3022
3023         while (global_trace.entries < trace_nr_entries) {
3024                 if (trace_alloc_page())
3025                         break;
3026                 pages++;
3027         }
3028         max_tr.entries = global_trace.entries;
3029
3030         pr_info("tracer: %d pages allocated for %ld",
3031                 pages, trace_nr_entries);
3032         pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
3033         pr_info("   actual entries %ld\n", global_trace.entries);
3034
3035         tracer_init_debugfs();
3036
3037         trace_init_cmdlines();
3038
3039         register_tracer(&no_tracer);
3040         current_trace = &no_tracer;
3041
3042         /* All seems OK, enable tracing */
3043         tracing_disabled = 0;
3044
3045         return 0;
3046
3047  free_buffers:
3048         for (i-- ; i >= 0; i--) {
3049                 struct page *page, *tmp;
3050                 struct trace_array_cpu *data = global_trace.data[i];
3051
3052                 if (data) {
3053                         list_for_each_entry_safe(page, tmp,
3054                                                  &data->trace_pages, lru) {
3055                                 list_del_init(&page->lru);
3056                                 __free_page(page);
3057                         }
3058                 }
3059
3060 #ifdef CONFIG_TRACER_MAX_TRACE
3061                 data = max_tr.data[i];
3062                 if (data) {
3063                         list_for_each_entry_safe(page, tmp,
3064                                                  &data->trace_pages, lru) {
3065                                 list_del_init(&page->lru);
3066                                 __free_page(page);
3067                         }
3068                 }
3069 #endif
3070         }
3071         return ret;
3072 }
3073 fs_initcall(tracer_alloc_buffers);