2 * ring buffer based function tracer
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Originally taken from the RT patch by:
8 * Arnaldo Carvalho de Melo <acme@redhat.com>
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
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>
30 #include <linux/writeback.h>
32 #include <linux/stacktrace.h>
36 unsigned long __read_mostly tracing_max_latency = (cycle_t)ULONG_MAX;
37 unsigned long __read_mostly tracing_thresh;
39 static unsigned long __read_mostly tracing_nr_buffers;
40 static cpumask_t __read_mostly tracing_buffer_mask;
42 #define for_each_tracing_cpu(cpu) \
43 for_each_cpu_mask(cpu, tracing_buffer_mask)
45 /* dummy trace to disable tracing */
46 static struct tracer no_tracer __read_mostly = {
50 static int trace_alloc_page(void);
51 static int trace_free_page(void);
53 static int tracing_disabled = 1;
55 static unsigned long tracing_pages_allocated;
58 ns2usecs(cycle_t nsec)
65 cycle_t ftrace_now(int cpu)
67 return cpu_clock(cpu);
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.
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.
82 static struct trace_array global_trace;
84 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
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.
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.
96 static struct trace_array max_tr;
98 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
100 /* tracer_enabled is used to toggle activation of a tracer */
101 static int tracer_enabled = 1;
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.
108 static unsigned long trace_nr_entries = 65536UL;
110 /* trace_types holds a link list of available tracers. */
111 static struct tracer *trace_types __read_mostly;
113 /* current_trace points to the tracer that is currently active */
114 static struct tracer *current_trace __read_mostly;
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.
121 static int max_tracer_type_len;
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.
129 static DEFINE_MUTEX(trace_types_lock);
131 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
132 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
134 /* trace_flags holds iter_ctrl options */
135 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
138 * trace_wake_up - wake up tasks waiting for trace input
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.
143 void trace_wake_up(void)
146 * The runqueue_is_locked() can fail, but this is the best we
149 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
150 wake_up(&trace_wait);
153 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
155 static int __init set_nr_entries(char *str)
157 unsigned long nr_entries;
162 ret = strict_strtoul(str, 0, &nr_entries);
163 /* nr_entries can not be zero */
164 if (ret < 0 || nr_entries == 0)
166 trace_nr_entries = nr_entries;
169 __setup("trace_entries=", set_nr_entries);
171 unsigned long nsecs_to_usecs(unsigned long nsecs)
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
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,
192 * TRACE_ITER_SYM_MASK masks the options in trace_flags that
193 * control the output of kernel symbols.
195 #define TRACE_ITER_SYM_MASK \
196 (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
198 /* These must match the bit postions in trace_iterator_flags */
199 static const char *trace_options[] = {
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.
219 * This is defined as a raw_spinlock_t in order to help
220 * with performance when lockdep debugging is enabled.
222 static raw_spinlock_t ftrace_max_lock =
223 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
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)
231 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
233 struct trace_array_cpu *data = tr->data[cpu];
236 max_tr.time_start = data->preempt_timestamp;
238 data = max_tr.data[cpu];
239 data->saved_latency = tracing_max_latency;
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;
248 /* record this tasks comm */
249 tracing_record_cmdline(current);
253 * check_pages - integrity check of trace buffers
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
259 void check_pages(struct trace_array_cpu *data)
261 struct page *page, *tmp;
263 BUG_ON(data->trace_pages.next->prev != &data->trace_pages);
264 BUG_ON(data->trace_pages.prev->next != &data->trace_pages);
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);
273 * head_page - page address of the first page in per_cpu buffer.
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.
279 void *head_page(struct trace_array_cpu *data)
284 if (list_empty(&data->trace_pages))
287 page = list_entry(data->trace_pages.next, struct page, lru);
288 BUG_ON(&page->lru == &data->trace_pages);
290 return page_address(page);
294 * trace_seq_printf - sequence printing of trace information
295 * @s: trace sequence descriptor
296 * @fmt: printf format string
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.
305 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
307 int len = (PAGE_SIZE - 1) - s->len;
315 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
318 /* If we can't write it all, don't bother writing anything */
328 * trace_seq_puts - trace sequence printing of simple string
329 * @s: trace sequence descriptor
330 * @str: simple string to record
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.
338 trace_seq_puts(struct trace_seq *s, const char *str)
340 int len = strlen(str);
342 if (len > ((PAGE_SIZE - 1) - s->len))
345 memcpy(s->buffer + s->len, str, len);
352 trace_seq_putc(struct trace_seq *s, unsigned char c)
354 if (s->len >= (PAGE_SIZE - 1))
357 s->buffer[s->len++] = c;
363 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
365 if (len > ((PAGE_SIZE - 1) - s->len))
368 memcpy(s->buffer + s->len, mem, len);
375 static const char hex2asc[] = "0123456789abcdef";
378 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
380 unsigned char hex[HEX_CHARS];
381 unsigned char *data = mem;
385 BUG_ON(len >= HEX_CHARS);
388 for (i = 0, j = 0; i < len; i++) {
390 for (i = len-1, j = 0; i >= 0; i--) {
394 hex[j++] = hex2asc[byte & 0x0f];
395 hex[j++] = hex2asc[byte >> 4];
399 return trace_seq_putmem(s, hex, j);
403 trace_seq_reset(struct trace_seq *s)
409 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
414 if (s->len <= s->readpos)
417 len = s->len - s->readpos;
420 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
429 trace_print_seq(struct seq_file *m, struct trace_seq *s)
431 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
434 seq_puts(m, s->buffer);
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.
444 * The ftrace_max_lock must be held.
447 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
449 struct list_head flip_pages;
451 INIT_LIST_HEAD(&flip_pages);
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));
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));
468 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
470 * @tsk: the task with the latency
471 * @cpu: The cpu that initiated the trace.
473 * Flip the buffers between the @tr and the max_tr and record information
474 * about which task was the cause of this latency.
477 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
479 struct trace_array_cpu *data;
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) {
487 flip_trace(max_tr.data[i], data);
491 __update_max_tr(tr, tsk, cpu);
492 __raw_spin_unlock(&ftrace_max_lock);
496 * update_max_tr_single - only copy one trace over, and reset the rest
498 * @tsk - task with the latency
499 * @cpu - the cpu of the buffer to copy.
501 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
504 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
506 struct trace_array_cpu *data = tr->data[cpu];
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]);
514 flip_trace(max_tr.data[cpu], data);
517 __update_max_tr(tr, tsk, cpu);
518 __raw_spin_unlock(&ftrace_max_lock);
522 * register_tracer - register a tracer with the ftrace system.
523 * @type - the plugin for the tracer
525 * Register a new plugin tracer.
527 int register_tracer(struct tracer *type)
534 pr_info("Tracer must have a name\n");
538 mutex_lock(&trace_types_lock);
539 for (t = trace_types; t; t = t->next) {
540 if (strcmp(type->name, t->name) == 0) {
542 pr_info("Trace %s already registered\n",
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;
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.
563 for_each_tracing_cpu(i) {
565 if (!head_page(data))
569 current_trace = type;
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;
578 printk(KERN_CONT "FAILED!\n");
581 /* Only reset on passing, to avoid touching corrupted buffers */
582 for_each_tracing_cpu(i) {
584 if (!head_page(data))
588 printk(KERN_CONT "PASSED\n");
592 type->next = trace_types;
594 len = strlen(type->name);
595 if (len > max_tracer_type_len)
596 max_tracer_type_len = len;
599 mutex_unlock(&trace_types_lock);
604 void unregister_tracer(struct tracer *type)
609 mutex_lock(&trace_types_lock);
610 for (t = &trace_types; *t; t = &(*t)->next) {
614 pr_info("Trace %s not registered\n", type->name);
619 if (strlen(type->name) != max_tracer_type_len)
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;
629 mutex_unlock(&trace_types_lock);
632 void tracing_reset(struct trace_array_cpu *data)
636 data->trace_head = data->trace_tail = head_page(data);
637 data->trace_head_idx = 0;
638 data->trace_tail_idx = 0;
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);
648 /* trace in all context switches */
649 atomic_t trace_record_cmdline_enabled __read_mostly;
651 /* temporary disable recording */
652 atomic_t trace_record_cmdline_disabled __read_mostly;
654 static void trace_init_cmdlines(void)
656 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
657 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
661 void trace_stop_cmdline_recording(void);
663 static void trace_save_cmdline(struct task_struct *tsk)
668 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
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.
677 if (!spin_trylock(&trace_cmdline_lock))
680 idx = map_pid_to_cmdline[tsk->pid];
681 if (idx >= SAVED_CMDLINES) {
682 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
684 map = map_cmdline_to_pid[idx];
685 if (map <= PID_MAX_DEFAULT)
686 map_pid_to_cmdline[map] = (unsigned)-1;
688 map_pid_to_cmdline[tsk->pid] = idx;
693 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
695 spin_unlock(&trace_cmdline_lock);
698 static char *trace_find_cmdline(int pid)
700 char *cmdline = "<...>";
706 if (pid > PID_MAX_DEFAULT)
709 map = map_pid_to_cmdline[pid];
710 if (map >= SAVED_CMDLINES)
713 cmdline = saved_cmdlines[map];
719 void tracing_record_cmdline(struct task_struct *tsk)
721 if (atomic_read(&trace_record_cmdline_disabled))
724 trace_save_cmdline(tsk);
727 static inline struct list_head *
728 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
731 * Roundrobin - but skip the head (which is not a real page):
734 if (unlikely(next == &data->trace_pages))
736 BUG_ON(next == &data->trace_pages);
742 trace_next_page(struct trace_array_cpu *data, void *addr)
744 struct list_head *next;
747 page = virt_to_page(addr);
749 next = trace_next_list(data, &page->lru);
750 page = list_entry(next, struct page, lru);
752 return page_address(page);
755 static inline struct trace_entry *
756 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
758 unsigned long idx, idx_next;
759 struct trace_entry *entry;
762 idx = data->trace_head_idx;
765 BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
767 entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
769 if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
770 data->trace_head = trace_next_page(data, data->trace_head);
774 if (data->trace_head == data->trace_tail &&
775 idx_next == data->trace_tail_idx) {
778 data->trace_tail_idx++;
779 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
781 trace_next_page(data, data->trace_tail);
782 data->trace_tail_idx = 0;
786 data->trace_head_idx = idx_next;
792 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
794 struct task_struct *tsk = current;
797 pc = preempt_count();
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);
809 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
810 unsigned long ip, unsigned long parent_ip, unsigned long flags)
812 struct trace_entry *entry;
813 unsigned long irq_flags;
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;
821 entry->fn.parent_ip = parent_ip;
822 __raw_spin_unlock(&data->lock);
823 raw_local_irq_restore(irq_flags);
827 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
828 unsigned long ip, unsigned long parent_ip, unsigned long flags)
830 if (likely(!atomic_read(&data->disabled)))
831 trace_function(tr, data, ip, parent_ip, flags);
834 void __trace_stack(struct trace_array *tr,
835 struct trace_array_cpu *data,
839 struct trace_entry *entry;
840 struct stack_trace trace;
842 if (!(trace_flags & TRACE_ITER_STACKTRACE))
845 entry = tracing_get_trace_entry(tr, data);
846 tracing_generic_entry_update(entry, flags);
847 entry->type = TRACE_STACK;
849 memset(&entry->stack, 0, sizeof(entry->stack));
851 trace.nr_entries = 0;
852 trace.max_entries = FTRACE_STACK_ENTRIES;
854 trace.entries = entry->stack.caller;
856 save_stack_trace(&trace);
860 __trace_special(void *__tr, void *__data,
861 unsigned long arg1, unsigned long arg2, unsigned long arg3)
863 struct trace_array_cpu *data = __data;
864 struct trace_array *tr = __tr;
865 struct trace_entry *entry;
866 unsigned long irq_flags;
868 raw_local_irq_save(irq_flags);
869 __raw_spin_lock(&data->lock);
870 entry = tracing_get_trace_entry(tr, data);
871 tracing_generic_entry_update(entry, 0);
872 entry->type = TRACE_SPECIAL;
873 entry->special.arg1 = arg1;
874 entry->special.arg2 = arg2;
875 entry->special.arg3 = arg3;
876 __trace_stack(tr, data, irq_flags, 4);
877 __raw_spin_unlock(&data->lock);
878 raw_local_irq_restore(irq_flags);
884 tracing_sched_switch_trace(struct trace_array *tr,
885 struct trace_array_cpu *data,
886 struct task_struct *prev,
887 struct task_struct *next,
890 struct trace_entry *entry;
891 unsigned long irq_flags;
893 raw_local_irq_save(irq_flags);
894 __raw_spin_lock(&data->lock);
895 entry = tracing_get_trace_entry(tr, data);
896 tracing_generic_entry_update(entry, flags);
897 entry->type = TRACE_CTX;
898 entry->ctx.prev_pid = prev->pid;
899 entry->ctx.prev_prio = prev->prio;
900 entry->ctx.prev_state = prev->state;
901 entry->ctx.next_pid = next->pid;
902 entry->ctx.next_prio = next->prio;
903 entry->ctx.next_state = next->state;
904 __trace_stack(tr, data, flags, 4);
905 __raw_spin_unlock(&data->lock);
906 raw_local_irq_restore(irq_flags);
910 tracing_sched_wakeup_trace(struct trace_array *tr,
911 struct trace_array_cpu *data,
912 struct task_struct *wakee,
913 struct task_struct *curr,
916 struct trace_entry *entry;
917 unsigned long irq_flags;
919 raw_local_irq_save(irq_flags);
920 __raw_spin_lock(&data->lock);
921 entry = tracing_get_trace_entry(tr, data);
922 tracing_generic_entry_update(entry, flags);
923 entry->type = TRACE_WAKE;
924 entry->ctx.prev_pid = curr->pid;
925 entry->ctx.prev_prio = curr->prio;
926 entry->ctx.prev_state = curr->state;
927 entry->ctx.next_pid = wakee->pid;
928 entry->ctx.next_prio = wakee->prio;
929 entry->ctx.next_state = wakee->state;
930 __trace_stack(tr, data, flags, 5);
931 __raw_spin_unlock(&data->lock);
932 raw_local_irq_restore(irq_flags);
939 function_trace_call(unsigned long ip, unsigned long parent_ip)
941 struct trace_array *tr = &global_trace;
942 struct trace_array_cpu *data;
947 if (unlikely(!tracer_enabled))
950 local_irq_save(flags);
951 cpu = raw_smp_processor_id();
952 data = tr->data[cpu];
953 disabled = atomic_inc_return(&data->disabled);
955 if (likely(disabled == 1))
956 trace_function(tr, data, ip, parent_ip, flags);
958 atomic_dec(&data->disabled);
959 local_irq_restore(flags);
962 static struct ftrace_ops trace_ops __read_mostly =
964 .func = function_trace_call,
967 void tracing_start_function_trace(void)
969 register_ftrace_function(&trace_ops);
972 void tracing_stop_function_trace(void)
974 unregister_ftrace_function(&trace_ops);
978 enum trace_file_type {
979 TRACE_FILE_LAT_FMT = 1,
982 static struct trace_entry *
983 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
984 struct trace_iterator *iter, int cpu)
987 struct trace_entry *array;
989 if (iter->next_idx[cpu] >= tr->entries ||
990 iter->next_idx[cpu] >= data->trace_idx ||
991 (data->trace_head == data->trace_tail &&
992 data->trace_head_idx == data->trace_tail_idx))
995 if (!iter->next_page[cpu]) {
996 /* Initialize the iterator for this cpu trace buffer */
997 WARN_ON(!data->trace_tail);
998 page = virt_to_page(data->trace_tail);
999 iter->next_page[cpu] = &page->lru;
1000 iter->next_page_idx[cpu] = data->trace_tail_idx;
1003 page = list_entry(iter->next_page[cpu], struct page, lru);
1004 BUG_ON(&data->trace_pages == &page->lru);
1006 array = page_address(page);
1008 WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
1009 return &array[iter->next_page_idx[cpu]];
1012 static struct trace_entry *
1013 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
1015 struct trace_array *tr = iter->tr;
1016 struct trace_entry *ent, *next = NULL;
1020 for_each_tracing_cpu(cpu) {
1021 if (!head_page(tr->data[cpu]))
1023 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
1025 * Pick the entry with the smallest timestamp:
1027 if (ent && (!next || ent->t < next->t)) {
1034 *ent_cpu = next_cpu;
1039 static void trace_iterator_increment(struct trace_iterator *iter)
1042 iter->next_idx[iter->cpu]++;
1043 iter->next_page_idx[iter->cpu]++;
1045 if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
1046 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1048 iter->next_page_idx[iter->cpu] = 0;
1049 iter->next_page[iter->cpu] =
1050 trace_next_list(data, iter->next_page[iter->cpu]);
1054 static void trace_consume(struct trace_iterator *iter)
1056 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1058 data->trace_tail_idx++;
1059 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
1060 data->trace_tail = trace_next_page(data, data->trace_tail);
1061 data->trace_tail_idx = 0;
1064 /* Check if we empty it, then reset the index */
1065 if (data->trace_head == data->trace_tail &&
1066 data->trace_head_idx == data->trace_tail_idx)
1067 data->trace_idx = 0;
1070 static void *find_next_entry_inc(struct trace_iterator *iter)
1072 struct trace_entry *next;
1075 next = find_next_entry(iter, &next_cpu);
1077 iter->prev_ent = iter->ent;
1078 iter->prev_cpu = iter->cpu;
1081 iter->cpu = next_cpu;
1084 trace_iterator_increment(iter);
1086 return next ? iter : NULL;
1089 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1091 struct trace_iterator *iter = m->private;
1092 void *last_ent = iter->ent;
1098 /* can't go backwards */
1103 ent = find_next_entry_inc(iter);
1107 while (ent && iter->idx < i)
1108 ent = find_next_entry_inc(iter);
1112 if (last_ent && !ent)
1113 seq_puts(m, "\n\nvim:ft=help\n");
1118 static void *s_start(struct seq_file *m, loff_t *pos)
1120 struct trace_iterator *iter = m->private;
1125 mutex_lock(&trace_types_lock);
1127 if (!current_trace || current_trace != iter->trace) {
1128 mutex_unlock(&trace_types_lock);
1132 atomic_inc(&trace_record_cmdline_disabled);
1134 /* let the tracer grab locks here if needed */
1135 if (current_trace->start)
1136 current_trace->start(iter);
1138 if (*pos != iter->pos) {
1142 iter->prev_ent = NULL;
1143 iter->prev_cpu = -1;
1145 for_each_tracing_cpu(i) {
1146 iter->next_idx[i] = 0;
1147 iter->next_page[i] = NULL;
1150 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1155 p = s_next(m, p, &l);
1161 static void s_stop(struct seq_file *m, void *p)
1163 struct trace_iterator *iter = m->private;
1165 atomic_dec(&trace_record_cmdline_disabled);
1167 /* let the tracer release locks here if needed */
1168 if (current_trace && current_trace == iter->trace && iter->trace->stop)
1169 iter->trace->stop(iter);
1171 mutex_unlock(&trace_types_lock);
1175 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1177 #ifdef CONFIG_KALLSYMS
1178 char str[KSYM_SYMBOL_LEN];
1180 kallsyms_lookup(address, NULL, NULL, NULL, str);
1182 return trace_seq_printf(s, fmt, str);
1188 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1189 unsigned long address)
1191 #ifdef CONFIG_KALLSYMS
1192 char str[KSYM_SYMBOL_LEN];
1194 sprint_symbol(str, address);
1195 return trace_seq_printf(s, fmt, str);
1200 #ifndef CONFIG_64BIT
1201 # define IP_FMT "%08lx"
1203 # define IP_FMT "%016lx"
1207 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1212 return trace_seq_printf(s, "0");
1214 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1215 ret = seq_print_sym_offset(s, "%s", ip);
1217 ret = seq_print_sym_short(s, "%s", ip);
1222 if (sym_flags & TRACE_ITER_SYM_ADDR)
1223 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1227 static void print_lat_help_header(struct seq_file *m)
1229 seq_puts(m, "# _------=> CPU# \n");
1230 seq_puts(m, "# / _-----=> irqs-off \n");
1231 seq_puts(m, "# | / _----=> need-resched \n");
1232 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1233 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1234 seq_puts(m, "# |||| / \n");
1235 seq_puts(m, "# ||||| delay \n");
1236 seq_puts(m, "# cmd pid ||||| time | caller \n");
1237 seq_puts(m, "# \\ / ||||| \\ | / \n");
1240 static void print_func_help_header(struct seq_file *m)
1242 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1243 seq_puts(m, "# | | | | |\n");
1248 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1250 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1251 struct trace_array *tr = iter->tr;
1252 struct trace_array_cpu *data = tr->data[tr->cpu];
1253 struct tracer *type = current_trace;
1254 unsigned long total = 0;
1255 unsigned long entries = 0;
1257 const char *name = "preemption";
1262 for_each_tracing_cpu(cpu) {
1263 if (head_page(tr->data[cpu])) {
1264 total += tr->data[cpu]->trace_idx;
1265 if (tr->data[cpu]->trace_idx > tr->entries)
1266 entries += tr->entries;
1268 entries += tr->data[cpu]->trace_idx;
1272 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1274 seq_puts(m, "-----------------------------------"
1275 "---------------------------------\n");
1276 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1277 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1278 nsecs_to_usecs(data->saved_latency),
1282 #if defined(CONFIG_PREEMPT_NONE)
1284 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1286 #elif defined(CONFIG_PREEMPT_DESKTOP)
1291 /* These are reserved for later use */
1294 seq_printf(m, " #P:%d)\n", num_online_cpus());
1298 seq_puts(m, " -----------------\n");
1299 seq_printf(m, " | task: %.16s-%d "
1300 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1301 data->comm, data->pid, data->uid, data->nice,
1302 data->policy, data->rt_priority);
1303 seq_puts(m, " -----------------\n");
1305 if (data->critical_start) {
1306 seq_puts(m, " => started at: ");
1307 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1308 trace_print_seq(m, &iter->seq);
1309 seq_puts(m, "\n => ended at: ");
1310 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1311 trace_print_seq(m, &iter->seq);
1319 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1321 int hardirq, softirq;
1324 comm = trace_find_cmdline(entry->pid);
1326 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1327 trace_seq_printf(s, "%d", cpu);
1328 trace_seq_printf(s, "%c%c",
1329 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1330 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1332 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1333 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1334 if (hardirq && softirq) {
1335 trace_seq_putc(s, 'H');
1338 trace_seq_putc(s, 'h');
1341 trace_seq_putc(s, 's');
1343 trace_seq_putc(s, '.');
1347 if (entry->preempt_count)
1348 trace_seq_printf(s, "%x", entry->preempt_count);
1350 trace_seq_puts(s, ".");
1353 unsigned long preempt_mark_thresh = 100;
1356 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1357 unsigned long rel_usecs)
1359 trace_seq_printf(s, " %4lldus", abs_usecs);
1360 if (rel_usecs > preempt_mark_thresh)
1361 trace_seq_puts(s, "!: ");
1362 else if (rel_usecs > 1)
1363 trace_seq_puts(s, "+: ");
1365 trace_seq_puts(s, " : ");
1368 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1371 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1373 struct trace_seq *s = &iter->seq;
1374 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1375 struct trace_entry *next_entry = find_next_entry(iter, NULL);
1376 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1377 struct trace_entry *entry = iter->ent;
1378 unsigned long abs_usecs;
1379 unsigned long rel_usecs;
1387 rel_usecs = ns2usecs(next_entry->t - entry->t);
1388 abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1391 comm = trace_find_cmdline(entry->pid);
1392 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1393 " %ld.%03ldms (+%ld.%03ldms): ",
1395 entry->pid, cpu, entry->flags,
1396 entry->preempt_count, trace_idx,
1399 abs_usecs % 1000, rel_usecs/1000,
1402 lat_print_generic(s, entry, cpu);
1403 lat_print_timestamp(s, abs_usecs, rel_usecs);
1405 switch (entry->type) {
1407 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1408 trace_seq_puts(s, " (");
1409 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1410 trace_seq_puts(s, ")\n");
1414 T = entry->ctx.next_state < sizeof(state_to_char) ?
1415 state_to_char[entry->ctx.next_state] : 'X';
1417 state = entry->ctx.prev_state ? __ffs(entry->ctx.prev_state) + 1 : 0;
1418 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1419 comm = trace_find_cmdline(entry->ctx.next_pid);
1420 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1421 entry->ctx.prev_pid,
1422 entry->ctx.prev_prio,
1423 S, entry->type == TRACE_CTX ? "==>" : " +",
1424 entry->ctx.next_pid,
1425 entry->ctx.next_prio,
1429 trace_seq_printf(s, "# %ld %ld %ld\n",
1430 entry->special.arg1,
1431 entry->special.arg2,
1432 entry->special.arg3);
1435 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1437 trace_seq_puts(s, " <= ");
1438 seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1440 trace_seq_puts(s, "\n");
1443 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1448 static int print_trace_fmt(struct trace_iterator *iter)
1450 struct trace_seq *s = &iter->seq;
1451 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1452 struct trace_entry *entry;
1453 unsigned long usec_rem;
1454 unsigned long long t;
1463 comm = trace_find_cmdline(iter->ent->pid);
1465 t = ns2usecs(entry->t);
1466 usec_rem = do_div(t, 1000000ULL);
1467 secs = (unsigned long)t;
1469 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1472 ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1475 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1479 switch (entry->type) {
1481 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1484 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1485 entry->fn.parent_ip) {
1486 ret = trace_seq_printf(s, " <-");
1489 ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1494 ret = trace_seq_printf(s, "\n");
1500 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1501 state_to_char[entry->ctx.prev_state] : 'X';
1502 T = entry->ctx.next_state < sizeof(state_to_char) ?
1503 state_to_char[entry->ctx.next_state] : 'X';
1504 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1505 entry->ctx.prev_pid,
1506 entry->ctx.prev_prio,
1508 entry->type == TRACE_CTX ? "==>" : " +",
1509 entry->ctx.next_pid,
1510 entry->ctx.next_prio,
1516 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1517 entry->special.arg1,
1518 entry->special.arg2,
1519 entry->special.arg3);
1524 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1526 ret = trace_seq_puts(s, " <= ");
1530 ret = seq_print_ip_sym(s, entry->stack.caller[i],
1535 ret = trace_seq_puts(s, "\n");
1543 static int print_raw_fmt(struct trace_iterator *iter)
1545 struct trace_seq *s = &iter->seq;
1546 struct trace_entry *entry;
1552 ret = trace_seq_printf(s, "%d %d %llu ",
1553 entry->pid, iter->cpu, entry->t);
1557 switch (entry->type) {
1559 ret = trace_seq_printf(s, "%x %x\n",
1560 entry->fn.ip, entry->fn.parent_ip);
1566 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1567 state_to_char[entry->ctx.prev_state] : 'X';
1568 T = entry->ctx.next_state < sizeof(state_to_char) ?
1569 state_to_char[entry->ctx.next_state] : 'X';
1570 if (entry->type == TRACE_WAKE)
1572 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1573 entry->ctx.prev_pid,
1574 entry->ctx.prev_prio,
1576 entry->ctx.next_pid,
1577 entry->ctx.next_prio,
1584 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1585 entry->special.arg1,
1586 entry->special.arg2,
1587 entry->special.arg3);
1595 #define SEQ_PUT_FIELD_RET(s, x) \
1597 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1601 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1603 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1607 static int print_hex_fmt(struct trace_iterator *iter)
1609 struct trace_seq *s = &iter->seq;
1610 unsigned char newline = '\n';
1611 struct trace_entry *entry;
1616 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1617 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1618 SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1620 switch (entry->type) {
1622 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1623 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1627 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1628 state_to_char[entry->ctx.prev_state] : 'X';
1629 T = entry->ctx.next_state < sizeof(state_to_char) ?
1630 state_to_char[entry->ctx.next_state] : 'X';
1631 if (entry->type == TRACE_WAKE)
1633 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1634 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1635 SEQ_PUT_HEX_FIELD_RET(s, S);
1636 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1637 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1638 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1639 SEQ_PUT_HEX_FIELD_RET(s, T);
1643 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1644 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1645 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1648 SEQ_PUT_FIELD_RET(s, newline);
1653 static int print_bin_fmt(struct trace_iterator *iter)
1655 struct trace_seq *s = &iter->seq;
1656 struct trace_entry *entry;
1660 SEQ_PUT_FIELD_RET(s, entry->pid);
1661 SEQ_PUT_FIELD_RET(s, entry->cpu);
1662 SEQ_PUT_FIELD_RET(s, entry->t);
1664 switch (entry->type) {
1666 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1667 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1670 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1671 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1672 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1673 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1674 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1675 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1679 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1680 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1681 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1687 static int trace_empty(struct trace_iterator *iter)
1689 struct trace_array_cpu *data;
1692 for_each_tracing_cpu(cpu) {
1693 data = iter->tr->data[cpu];
1695 if (head_page(data) && data->trace_idx &&
1696 (data->trace_tail != data->trace_head ||
1697 data->trace_tail_idx != data->trace_head_idx))
1703 static int print_trace_line(struct trace_iterator *iter)
1705 if (iter->trace && iter->trace->print_line)
1706 return iter->trace->print_line(iter);
1708 if (trace_flags & TRACE_ITER_BIN)
1709 return print_bin_fmt(iter);
1711 if (trace_flags & TRACE_ITER_HEX)
1712 return print_hex_fmt(iter);
1714 if (trace_flags & TRACE_ITER_RAW)
1715 return print_raw_fmt(iter);
1717 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1718 return print_lat_fmt(iter, iter->idx, iter->cpu);
1720 return print_trace_fmt(iter);
1723 static int s_show(struct seq_file *m, void *v)
1725 struct trace_iterator *iter = v;
1727 if (iter->ent == NULL) {
1729 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1732 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1733 /* print nothing if the buffers are empty */
1734 if (trace_empty(iter))
1736 print_trace_header(m, iter);
1737 if (!(trace_flags & TRACE_ITER_VERBOSE))
1738 print_lat_help_header(m);
1740 if (!(trace_flags & TRACE_ITER_VERBOSE))
1741 print_func_help_header(m);
1744 print_trace_line(iter);
1745 trace_print_seq(m, &iter->seq);
1751 static struct seq_operations tracer_seq_ops = {
1758 static struct trace_iterator *
1759 __tracing_open(struct inode *inode, struct file *file, int *ret)
1761 struct trace_iterator *iter;
1763 if (tracing_disabled) {
1768 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1774 mutex_lock(&trace_types_lock);
1775 if (current_trace && current_trace->print_max)
1778 iter->tr = inode->i_private;
1779 iter->trace = current_trace;
1782 /* TODO stop tracer */
1783 *ret = seq_open(file, &tracer_seq_ops);
1785 struct seq_file *m = file->private_data;
1788 /* stop the trace while dumping */
1792 if (iter->trace && iter->trace->open)
1793 iter->trace->open(iter);
1798 mutex_unlock(&trace_types_lock);
1804 int tracing_open_generic(struct inode *inode, struct file *filp)
1806 if (tracing_disabled)
1809 filp->private_data = inode->i_private;
1813 int tracing_release(struct inode *inode, struct file *file)
1815 struct seq_file *m = (struct seq_file *)file->private_data;
1816 struct trace_iterator *iter = m->private;
1818 mutex_lock(&trace_types_lock);
1819 if (iter->trace && iter->trace->close)
1820 iter->trace->close(iter);
1822 /* reenable tracing if it was previously enabled */
1825 mutex_unlock(&trace_types_lock);
1827 seq_release(inode, file);
1832 static int tracing_open(struct inode *inode, struct file *file)
1836 __tracing_open(inode, file, &ret);
1841 static int tracing_lt_open(struct inode *inode, struct file *file)
1843 struct trace_iterator *iter;
1846 iter = __tracing_open(inode, file, &ret);
1849 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1856 t_next(struct seq_file *m, void *v, loff_t *pos)
1858 struct tracer *t = m->private;
1870 static void *t_start(struct seq_file *m, loff_t *pos)
1872 struct tracer *t = m->private;
1875 mutex_lock(&trace_types_lock);
1876 for (; t && l < *pos; t = t_next(m, t, &l))
1882 static void t_stop(struct seq_file *m, void *p)
1884 mutex_unlock(&trace_types_lock);
1887 static int t_show(struct seq_file *m, void *v)
1889 struct tracer *t = v;
1894 seq_printf(m, "%s", t->name);
1903 static struct seq_operations show_traces_seq_ops = {
1910 static int show_traces_open(struct inode *inode, struct file *file)
1914 if (tracing_disabled)
1917 ret = seq_open(file, &show_traces_seq_ops);
1919 struct seq_file *m = file->private_data;
1920 m->private = trace_types;
1926 static struct file_operations tracing_fops = {
1927 .open = tracing_open,
1929 .llseek = seq_lseek,
1930 .release = tracing_release,
1933 static struct file_operations tracing_lt_fops = {
1934 .open = tracing_lt_open,
1936 .llseek = seq_lseek,
1937 .release = tracing_release,
1940 static struct file_operations show_traces_fops = {
1941 .open = show_traces_open,
1943 .release = seq_release,
1947 * Only trace on a CPU if the bitmask is set:
1949 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1952 * When tracing/tracing_cpu_mask is modified then this holds
1953 * the new bitmask we are about to install:
1955 static cpumask_t tracing_cpumask_new;
1958 * The tracer itself will not take this lock, but still we want
1959 * to provide a consistent cpumask to user-space:
1961 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1964 * Temporary storage for the character representation of the
1965 * CPU bitmask (and one more byte for the newline):
1967 static char mask_str[NR_CPUS + 1];
1970 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1971 size_t count, loff_t *ppos)
1975 mutex_lock(&tracing_cpumask_update_lock);
1977 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1978 if (count - len < 2) {
1982 len += sprintf(mask_str + len, "\n");
1983 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1986 mutex_unlock(&tracing_cpumask_update_lock);
1992 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1993 size_t count, loff_t *ppos)
1997 mutex_lock(&tracing_cpumask_update_lock);
1998 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2002 raw_local_irq_disable();
2003 __raw_spin_lock(&ftrace_max_lock);
2004 for_each_tracing_cpu(cpu) {
2006 * Increase/decrease the disabled counter if we are
2007 * about to flip a bit in the cpumask:
2009 if (cpu_isset(cpu, tracing_cpumask) &&
2010 !cpu_isset(cpu, tracing_cpumask_new)) {
2011 atomic_inc(&global_trace.data[cpu]->disabled);
2013 if (!cpu_isset(cpu, tracing_cpumask) &&
2014 cpu_isset(cpu, tracing_cpumask_new)) {
2015 atomic_dec(&global_trace.data[cpu]->disabled);
2018 __raw_spin_unlock(&ftrace_max_lock);
2019 raw_local_irq_enable();
2021 tracing_cpumask = tracing_cpumask_new;
2023 mutex_unlock(&tracing_cpumask_update_lock);
2028 mutex_unlock(&tracing_cpumask_update_lock);
2033 static struct file_operations tracing_cpumask_fops = {
2034 .open = tracing_open_generic,
2035 .read = tracing_cpumask_read,
2036 .write = tracing_cpumask_write,
2040 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2041 size_t cnt, loff_t *ppos)
2048 /* calulate max size */
2049 for (i = 0; trace_options[i]; i++) {
2050 len += strlen(trace_options[i]);
2051 len += 3; /* "no" and space */
2054 /* +2 for \n and \0 */
2055 buf = kmalloc(len + 2, GFP_KERNEL);
2059 for (i = 0; trace_options[i]; i++) {
2060 if (trace_flags & (1 << i))
2061 r += sprintf(buf + r, "%s ", trace_options[i]);
2063 r += sprintf(buf + r, "no%s ", trace_options[i]);
2066 r += sprintf(buf + r, "\n");
2067 WARN_ON(r >= len + 2);
2069 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2077 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2078 size_t cnt, loff_t *ppos)
2085 if (cnt >= sizeof(buf))
2088 if (copy_from_user(&buf, ubuf, cnt))
2093 if (strncmp(buf, "no", 2) == 0) {
2098 for (i = 0; trace_options[i]; i++) {
2099 int len = strlen(trace_options[i]);
2101 if (strncmp(cmp, trace_options[i], len) == 0) {
2103 trace_flags &= ~(1 << i);
2105 trace_flags |= (1 << i);
2110 * If no option could be set, return an error:
2112 if (!trace_options[i])
2120 static struct file_operations tracing_iter_fops = {
2121 .open = tracing_open_generic,
2122 .read = tracing_iter_ctrl_read,
2123 .write = tracing_iter_ctrl_write,
2126 static const char readme_msg[] =
2127 "tracing mini-HOWTO:\n\n"
2129 "# mount -t debugfs nodev /debug\n\n"
2130 "# cat /debug/tracing/available_tracers\n"
2131 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2132 "# cat /debug/tracing/current_tracer\n"
2134 "# echo sched_switch > /debug/tracing/current_tracer\n"
2135 "# cat /debug/tracing/current_tracer\n"
2137 "# cat /debug/tracing/iter_ctrl\n"
2138 "noprint-parent nosym-offset nosym-addr noverbose\n"
2139 "# echo print-parent > /debug/tracing/iter_ctrl\n"
2140 "# echo 1 > /debug/tracing/tracing_enabled\n"
2141 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2142 "echo 0 > /debug/tracing/tracing_enabled\n"
2146 tracing_readme_read(struct file *filp, char __user *ubuf,
2147 size_t cnt, loff_t *ppos)
2149 return simple_read_from_buffer(ubuf, cnt, ppos,
2150 readme_msg, strlen(readme_msg));
2153 static struct file_operations tracing_readme_fops = {
2154 .open = tracing_open_generic,
2155 .read = tracing_readme_read,
2159 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2160 size_t cnt, loff_t *ppos)
2162 struct trace_array *tr = filp->private_data;
2166 r = sprintf(buf, "%ld\n", tr->ctrl);
2167 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2171 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2172 size_t cnt, loff_t *ppos)
2174 struct trace_array *tr = filp->private_data;
2179 if (cnt >= sizeof(buf))
2182 if (copy_from_user(&buf, ubuf, cnt))
2187 ret = strict_strtoul(buf, 10, &val);
2193 mutex_lock(&trace_types_lock);
2194 if (tr->ctrl ^ val) {
2202 if (current_trace && current_trace->ctrl_update)
2203 current_trace->ctrl_update(tr);
2205 mutex_unlock(&trace_types_lock);
2213 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2214 size_t cnt, loff_t *ppos)
2216 char buf[max_tracer_type_len+2];
2219 mutex_lock(&trace_types_lock);
2221 r = sprintf(buf, "%s\n", current_trace->name);
2223 r = sprintf(buf, "\n");
2224 mutex_unlock(&trace_types_lock);
2226 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2230 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2231 size_t cnt, loff_t *ppos)
2233 struct trace_array *tr = &global_trace;
2235 char buf[max_tracer_type_len+1];
2238 if (cnt > max_tracer_type_len)
2239 cnt = max_tracer_type_len;
2241 if (copy_from_user(&buf, ubuf, cnt))
2246 /* strip ending whitespace. */
2247 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2250 mutex_lock(&trace_types_lock);
2251 for (t = trace_types; t; t = t->next) {
2252 if (strcmp(t->name, buf) == 0)
2255 if (!t || t == current_trace)
2258 if (current_trace && current_trace->reset)
2259 current_trace->reset(tr);
2266 mutex_unlock(&trace_types_lock);
2274 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2275 size_t cnt, loff_t *ppos)
2277 unsigned long *ptr = filp->private_data;
2281 r = snprintf(buf, sizeof(buf), "%ld\n",
2282 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2283 if (r > sizeof(buf))
2285 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2289 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2290 size_t cnt, loff_t *ppos)
2292 long *ptr = filp->private_data;
2297 if (cnt >= sizeof(buf))
2300 if (copy_from_user(&buf, ubuf, cnt))
2305 ret = strict_strtoul(buf, 10, &val);
2314 static atomic_t tracing_reader;
2316 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2318 struct trace_iterator *iter;
2320 if (tracing_disabled)
2323 /* We only allow for reader of the pipe */
2324 if (atomic_inc_return(&tracing_reader) != 1) {
2325 atomic_dec(&tracing_reader);
2329 /* create a buffer to store the information to pass to userspace */
2330 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2334 mutex_lock(&trace_types_lock);
2335 iter->tr = &global_trace;
2336 iter->trace = current_trace;
2337 filp->private_data = iter;
2339 if (iter->trace->pipe_open)
2340 iter->trace->pipe_open(iter);
2341 mutex_unlock(&trace_types_lock);
2346 static int tracing_release_pipe(struct inode *inode, struct file *file)
2348 struct trace_iterator *iter = file->private_data;
2351 atomic_dec(&tracing_reader);
2357 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2359 struct trace_iterator *iter = filp->private_data;
2361 if (trace_flags & TRACE_ITER_BLOCK) {
2363 * Always select as readable when in blocking mode
2365 return POLLIN | POLLRDNORM;
2367 if (!trace_empty(iter))
2368 return POLLIN | POLLRDNORM;
2369 poll_wait(filp, &trace_wait, poll_table);
2370 if (!trace_empty(iter))
2371 return POLLIN | POLLRDNORM;
2381 tracing_read_pipe(struct file *filp, char __user *ubuf,
2382 size_t cnt, loff_t *ppos)
2384 struct trace_iterator *iter = filp->private_data;
2385 struct trace_array_cpu *data;
2386 static cpumask_t mask;
2387 unsigned long flags;
2388 #ifdef CONFIG_FTRACE
2394 /* return any leftover data */
2395 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2400 trace_seq_reset(&iter->seq);
2402 mutex_lock(&trace_types_lock);
2403 if (iter->trace->read) {
2404 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2409 while (trace_empty(iter)) {
2411 if ((filp->f_flags & O_NONBLOCK)) {
2417 * This is a make-shift waitqueue. The reason we don't use
2418 * an actual wait queue is because:
2419 * 1) we only ever have one waiter
2420 * 2) the tracing, traces all functions, we don't want
2421 * the overhead of calling wake_up and friends
2422 * (and tracing them too)
2423 * Anyway, this is really very primitive wakeup.
2425 set_current_state(TASK_INTERRUPTIBLE);
2426 iter->tr->waiter = current;
2428 mutex_unlock(&trace_types_lock);
2430 /* sleep for 100 msecs, and try again. */
2431 schedule_timeout(HZ/10);
2433 mutex_lock(&trace_types_lock);
2435 iter->tr->waiter = NULL;
2437 if (signal_pending(current)) {
2442 if (iter->trace != current_trace)
2446 * We block until we read something and tracing is disabled.
2447 * We still block if tracing is disabled, but we have never
2448 * read anything. This allows a user to cat this file, and
2449 * then enable tracing. But after we have read something,
2450 * we give an EOF when tracing is again disabled.
2452 * iter->pos will be 0 if we haven't read anything.
2454 if (!tracer_enabled && iter->pos)
2460 /* stop when tracing is finished */
2461 if (trace_empty(iter))
2464 if (cnt >= PAGE_SIZE)
2465 cnt = PAGE_SIZE - 1;
2467 /* reset all but tr, trace, and overruns */
2468 memset(&iter->seq, 0,
2469 sizeof(struct trace_iterator) -
2470 offsetof(struct trace_iterator, seq));
2474 * We need to stop all tracing on all CPUS to read the
2475 * the next buffer. This is a bit expensive, but is
2476 * not done often. We fill all what we can read,
2477 * and then release the locks again.
2481 local_irq_save(flags);
2482 #ifdef CONFIG_FTRACE
2483 ftrace_save = ftrace_enabled;
2487 for_each_tracing_cpu(cpu) {
2488 data = iter->tr->data[cpu];
2490 if (!head_page(data) || !data->trace_idx)
2493 atomic_inc(&data->disabled);
2497 for_each_cpu_mask(cpu, mask) {
2498 data = iter->tr->data[cpu];
2499 __raw_spin_lock(&data->lock);
2501 if (data->overrun > iter->last_overrun[cpu])
2502 iter->overrun[cpu] +=
2503 data->overrun - iter->last_overrun[cpu];
2504 iter->last_overrun[cpu] = data->overrun;
2507 while (find_next_entry_inc(iter) != NULL) {
2509 int len = iter->seq.len;
2511 ret = print_trace_line(iter);
2513 /* don't print partial lines */
2514 iter->seq.len = len;
2518 trace_consume(iter);
2520 if (iter->seq.len >= cnt)
2524 for_each_cpu_mask(cpu, mask) {
2525 data = iter->tr->data[cpu];
2526 __raw_spin_unlock(&data->lock);
2529 for_each_cpu_mask(cpu, mask) {
2530 data = iter->tr->data[cpu];
2531 atomic_dec(&data->disabled);
2533 #ifdef CONFIG_FTRACE
2534 ftrace_enabled = ftrace_save;
2536 local_irq_restore(flags);
2538 /* Now copy what we have to the user */
2539 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2540 if (iter->seq.readpos >= iter->seq.len)
2541 trace_seq_reset(&iter->seq);
2546 mutex_unlock(&trace_types_lock);
2552 tracing_entries_read(struct file *filp, char __user *ubuf,
2553 size_t cnt, loff_t *ppos)
2555 struct trace_array *tr = filp->private_data;
2559 r = sprintf(buf, "%lu\n", tr->entries);
2560 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2564 tracing_entries_write(struct file *filp, const char __user *ubuf,
2565 size_t cnt, loff_t *ppos)
2571 if (cnt >= sizeof(buf))
2574 if (copy_from_user(&buf, ubuf, cnt))
2579 ret = strict_strtoul(buf, 10, &val);
2583 /* must have at least 1 entry */
2587 mutex_lock(&trace_types_lock);
2589 if (current_trace != &no_tracer) {
2591 pr_info("ftrace: set current_tracer to none"
2592 " before modifying buffer size\n");
2596 if (val > global_trace.entries) {
2597 long pages_requested;
2598 unsigned long freeable_pages;
2600 /* make sure we have enough memory before mapping */
2602 (val + (ENTRIES_PER_PAGE-1)) / ENTRIES_PER_PAGE;
2604 /* account for each buffer (and max_tr) */
2605 pages_requested *= tracing_nr_buffers * 2;
2607 /* Check for overflow */
2608 if (pages_requested < 0) {
2613 freeable_pages = determine_dirtyable_memory();
2615 /* we only allow to request 1/4 of useable memory */
2616 if (pages_requested >
2617 ((freeable_pages + tracing_pages_allocated) / 4)) {
2622 while (global_trace.entries < val) {
2623 if (trace_alloc_page()) {
2627 /* double check that we don't go over the known pages */
2628 if (tracing_pages_allocated > pages_requested)
2633 /* include the number of entries in val (inc of page entries) */
2634 while (global_trace.entries > val + (ENTRIES_PER_PAGE - 1))
2641 max_tr.entries = global_trace.entries;
2642 mutex_unlock(&trace_types_lock);
2647 static struct file_operations tracing_max_lat_fops = {
2648 .open = tracing_open_generic,
2649 .read = tracing_max_lat_read,
2650 .write = tracing_max_lat_write,
2653 static struct file_operations tracing_ctrl_fops = {
2654 .open = tracing_open_generic,
2655 .read = tracing_ctrl_read,
2656 .write = tracing_ctrl_write,
2659 static struct file_operations set_tracer_fops = {
2660 .open = tracing_open_generic,
2661 .read = tracing_set_trace_read,
2662 .write = tracing_set_trace_write,
2665 static struct file_operations tracing_pipe_fops = {
2666 .open = tracing_open_pipe,
2667 .poll = tracing_poll_pipe,
2668 .read = tracing_read_pipe,
2669 .release = tracing_release_pipe,
2672 static struct file_operations tracing_entries_fops = {
2673 .open = tracing_open_generic,
2674 .read = tracing_entries_read,
2675 .write = tracing_entries_write,
2678 #ifdef CONFIG_DYNAMIC_FTRACE
2681 tracing_read_long(struct file *filp, char __user *ubuf,
2682 size_t cnt, loff_t *ppos)
2684 unsigned long *p = filp->private_data;
2688 r = sprintf(buf, "%ld\n", *p);
2690 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2693 static struct file_operations tracing_read_long_fops = {
2694 .open = tracing_open_generic,
2695 .read = tracing_read_long,
2699 static struct dentry *d_tracer;
2701 struct dentry *tracing_init_dentry(void)
2708 d_tracer = debugfs_create_dir("tracing", NULL);
2710 if (!d_tracer && !once) {
2712 pr_warning("Could not create debugfs directory 'tracing'\n");
2719 #ifdef CONFIG_FTRACE_SELFTEST
2720 /* Let selftest have access to static functions in this file */
2721 #include "trace_selftest.c"
2724 static __init void tracer_init_debugfs(void)
2726 struct dentry *d_tracer;
2727 struct dentry *entry;
2729 d_tracer = tracing_init_dentry();
2731 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2732 &global_trace, &tracing_ctrl_fops);
2734 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2736 entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2737 NULL, &tracing_iter_fops);
2739 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2741 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2742 NULL, &tracing_cpumask_fops);
2744 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2746 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2747 &global_trace, &tracing_lt_fops);
2749 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2751 entry = debugfs_create_file("trace", 0444, d_tracer,
2752 &global_trace, &tracing_fops);
2754 pr_warning("Could not create debugfs 'trace' entry\n");
2756 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2757 &global_trace, &show_traces_fops);
2759 pr_warning("Could not create debugfs 'trace' entry\n");
2761 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2762 &global_trace, &set_tracer_fops);
2764 pr_warning("Could not create debugfs 'trace' entry\n");
2766 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2767 &tracing_max_latency,
2768 &tracing_max_lat_fops);
2770 pr_warning("Could not create debugfs "
2771 "'tracing_max_latency' entry\n");
2773 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2774 &tracing_thresh, &tracing_max_lat_fops);
2776 pr_warning("Could not create debugfs "
2777 "'tracing_threash' entry\n");
2778 entry = debugfs_create_file("README", 0644, d_tracer,
2779 NULL, &tracing_readme_fops);
2781 pr_warning("Could not create debugfs 'README' entry\n");
2783 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2784 NULL, &tracing_pipe_fops);
2786 pr_warning("Could not create debugfs "
2787 "'tracing_threash' entry\n");
2789 entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2790 &global_trace, &tracing_entries_fops);
2792 pr_warning("Could not create debugfs "
2793 "'tracing_threash' entry\n");
2795 #ifdef CONFIG_DYNAMIC_FTRACE
2796 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2797 &ftrace_update_tot_cnt,
2798 &tracing_read_long_fops);
2800 pr_warning("Could not create debugfs "
2801 "'dyn_ftrace_total_info' entry\n");
2805 static int trace_alloc_page(void)
2807 struct trace_array_cpu *data;
2808 struct page *page, *tmp;
2811 unsigned pages_allocated = 0;
2814 /* first allocate a page for each CPU */
2815 for_each_tracing_cpu(i) {
2816 array = (void *)__get_free_page(GFP_KERNEL);
2817 if (array == NULL) {
2818 printk(KERN_ERR "tracer: failed to allocate page"
2819 "for trace buffer!\n");
2824 page = virt_to_page(array);
2825 list_add(&page->lru, &pages);
2827 /* Only allocate if we are actually using the max trace */
2828 #ifdef CONFIG_TRACER_MAX_TRACE
2829 array = (void *)__get_free_page(GFP_KERNEL);
2830 if (array == NULL) {
2831 printk(KERN_ERR "tracer: failed to allocate page"
2832 "for trace buffer!\n");
2836 page = virt_to_page(array);
2837 list_add(&page->lru, &pages);
2841 /* Now that we successfully allocate a page per CPU, add them */
2842 for_each_tracing_cpu(i) {
2843 data = global_trace.data[i];
2844 page = list_entry(pages.next, struct page, lru);
2845 list_del_init(&page->lru);
2846 list_add_tail(&page->lru, &data->trace_pages);
2849 #ifdef CONFIG_TRACER_MAX_TRACE
2850 data = max_tr.data[i];
2851 page = list_entry(pages.next, struct page, lru);
2852 list_del_init(&page->lru);
2853 list_add_tail(&page->lru, &data->trace_pages);
2857 tracing_pages_allocated += pages_allocated;
2858 global_trace.entries += ENTRIES_PER_PAGE;
2863 list_for_each_entry_safe(page, tmp, &pages, lru) {
2864 list_del_init(&page->lru);
2870 static int trace_free_page(void)
2872 struct trace_array_cpu *data;
2874 struct list_head *p;
2878 /* free one page from each buffer */
2879 for_each_tracing_cpu(i) {
2880 data = global_trace.data[i];
2881 p = data->trace_pages.next;
2882 if (p == &data->trace_pages) {
2883 /* should never happen */
2885 tracing_disabled = 1;
2889 page = list_entry(p, struct page, lru);
2891 list_del(&page->lru);
2892 tracing_pages_allocated--;
2893 tracing_pages_allocated--;
2896 tracing_reset(data);
2898 #ifdef CONFIG_TRACER_MAX_TRACE
2899 data = max_tr.data[i];
2900 p = data->trace_pages.next;
2901 if (p == &data->trace_pages) {
2902 /* should never happen */
2904 tracing_disabled = 1;
2908 page = list_entry(p, struct page, lru);
2910 list_del(&page->lru);
2913 tracing_reset(data);
2916 global_trace.entries -= ENTRIES_PER_PAGE;
2921 __init static int tracer_alloc_buffers(void)
2923 struct trace_array_cpu *data;
2930 global_trace.ctrl = tracer_enabled;
2932 /* TODO: make the number of buffers hot pluggable with CPUS */
2933 tracing_nr_buffers = num_possible_cpus();
2934 tracing_buffer_mask = cpu_possible_map;
2936 /* Allocate the first page for all buffers */
2937 for_each_tracing_cpu(i) {
2938 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2939 max_tr.data[i] = &per_cpu(max_data, i);
2941 array = (void *)__get_free_page(GFP_KERNEL);
2942 if (array == NULL) {
2943 printk(KERN_ERR "tracer: failed to allocate page"
2944 "for trace buffer!\n");
2948 /* set the array to the list */
2949 INIT_LIST_HEAD(&data->trace_pages);
2950 page = virt_to_page(array);
2951 list_add(&page->lru, &data->trace_pages);
2952 /* use the LRU flag to differentiate the two buffers */
2955 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2956 max_tr.data[i]->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2958 /* Only allocate if we are actually using the max trace */
2959 #ifdef CONFIG_TRACER_MAX_TRACE
2960 array = (void *)__get_free_page(GFP_KERNEL);
2961 if (array == NULL) {
2962 printk(KERN_ERR "tracer: failed to allocate page"
2963 "for trace buffer!\n");
2967 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
2968 page = virt_to_page(array);
2969 list_add(&page->lru, &max_tr.data[i]->trace_pages);
2975 * Since we allocate by orders of pages, we may be able to
2978 global_trace.entries = ENTRIES_PER_PAGE;
2981 while (global_trace.entries < trace_nr_entries) {
2982 if (trace_alloc_page())
2986 max_tr.entries = global_trace.entries;
2988 pr_info("tracer: %d pages allocated for %ld",
2989 pages, trace_nr_entries);
2990 pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
2991 pr_info(" actual entries %ld\n", global_trace.entries);
2993 tracer_init_debugfs();
2995 trace_init_cmdlines();
2997 register_tracer(&no_tracer);
2998 current_trace = &no_tracer;
3000 /* All seems OK, enable tracing */
3001 tracing_disabled = 0;
3006 for (i-- ; i >= 0; i--) {
3007 struct page *page, *tmp;
3008 struct trace_array_cpu *data = global_trace.data[i];
3011 list_for_each_entry_safe(page, tmp,
3012 &data->trace_pages, lru) {
3013 list_del_init(&page->lru);
3018 #ifdef CONFIG_TRACER_MAX_TRACE
3019 data = max_tr.data[i];
3021 list_for_each_entry_safe(page, tmp,
3022 &data->trace_pages, lru) {
3023 list_del_init(&page->lru);
3031 fs_initcall(tracer_alloc_buffers);