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 #ifdef CONFIG_MMIOTRACE
835 void __trace_mmiotrace_rw(struct trace_array *tr, struct trace_array_cpu *data,
836 struct mmiotrace_rw *rw)
838 struct trace_entry *entry;
839 unsigned long irq_flags;
841 raw_local_irq_save(irq_flags);
842 __raw_spin_lock(&data->lock);
844 entry = tracing_get_trace_entry(tr, data);
845 tracing_generic_entry_update(entry, 0);
846 entry->type = TRACE_MMIO_RW;
849 __raw_spin_unlock(&data->lock);
850 raw_local_irq_restore(irq_flags);
855 void __trace_mmiotrace_map(struct trace_array *tr, struct trace_array_cpu *data,
856 struct mmiotrace_map *map)
858 struct trace_entry *entry;
859 unsigned long irq_flags;
861 raw_local_irq_save(irq_flags);
862 __raw_spin_lock(&data->lock);
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;
869 __raw_spin_unlock(&data->lock);
870 raw_local_irq_restore(irq_flags);
876 void __trace_stack(struct trace_array *tr,
877 struct trace_array_cpu *data,
881 struct trace_entry *entry;
882 struct stack_trace trace;
884 if (!(trace_flags & TRACE_ITER_STACKTRACE))
887 entry = tracing_get_trace_entry(tr, data);
888 tracing_generic_entry_update(entry, flags);
889 entry->type = TRACE_STACK;
891 memset(&entry->stack, 0, sizeof(entry->stack));
893 trace.nr_entries = 0;
894 trace.max_entries = FTRACE_STACK_ENTRIES;
896 trace.entries = entry->stack.caller;
898 save_stack_trace(&trace);
902 __trace_special(void *__tr, void *__data,
903 unsigned long arg1, unsigned long arg2, unsigned long arg3)
905 struct trace_array_cpu *data = __data;
906 struct trace_array *tr = __tr;
907 struct trace_entry *entry;
908 unsigned long irq_flags;
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);
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,
932 struct trace_entry *entry;
933 unsigned long irq_flags;
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);
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,
958 struct trace_entry *entry;
959 unsigned long irq_flags;
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);
981 function_trace_call(unsigned long ip, unsigned long parent_ip)
983 struct trace_array *tr = &global_trace;
984 struct trace_array_cpu *data;
989 if (unlikely(!tracer_enabled))
992 local_irq_save(flags);
993 cpu = raw_smp_processor_id();
994 data = tr->data[cpu];
995 disabled = atomic_inc_return(&data->disabled);
997 if (likely(disabled == 1))
998 trace_function(tr, data, ip, parent_ip, flags);
1000 atomic_dec(&data->disabled);
1001 local_irq_restore(flags);
1004 static struct ftrace_ops trace_ops __read_mostly =
1006 .func = function_trace_call,
1009 void tracing_start_function_trace(void)
1011 register_ftrace_function(&trace_ops);
1014 void tracing_stop_function_trace(void)
1016 unregister_ftrace_function(&trace_ops);
1020 enum trace_file_type {
1021 TRACE_FILE_LAT_FMT = 1,
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)
1029 struct trace_entry *array;
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))
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;
1045 page = list_entry(iter->next_page[cpu], struct page, lru);
1046 BUG_ON(&data->trace_pages == &page->lru);
1048 array = page_address(page);
1050 WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
1051 return &array[iter->next_page_idx[cpu]];
1054 static struct trace_entry *
1055 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
1057 struct trace_array *tr = iter->tr;
1058 struct trace_entry *ent, *next = NULL;
1062 for_each_tracing_cpu(cpu) {
1063 if (!head_page(tr->data[cpu]))
1065 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
1067 * Pick the entry with the smallest timestamp:
1069 if (ent && (!next || ent->t < next->t)) {
1076 *ent_cpu = next_cpu;
1081 static void trace_iterator_increment(struct trace_iterator *iter)
1084 iter->next_idx[iter->cpu]++;
1085 iter->next_page_idx[iter->cpu]++;
1087 if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
1088 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1090 iter->next_page_idx[iter->cpu] = 0;
1091 iter->next_page[iter->cpu] =
1092 trace_next_list(data, iter->next_page[iter->cpu]);
1096 static void trace_consume(struct trace_iterator *iter)
1098 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
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;
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;
1112 static void *find_next_entry_inc(struct trace_iterator *iter)
1114 struct trace_entry *next;
1117 next = find_next_entry(iter, &next_cpu);
1119 iter->prev_ent = iter->ent;
1120 iter->prev_cpu = iter->cpu;
1123 iter->cpu = next_cpu;
1126 trace_iterator_increment(iter);
1128 return next ? iter : NULL;
1131 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1133 struct trace_iterator *iter = m->private;
1134 void *last_ent = iter->ent;
1140 /* can't go backwards */
1145 ent = find_next_entry_inc(iter);
1149 while (ent && iter->idx < i)
1150 ent = find_next_entry_inc(iter);
1154 if (last_ent && !ent)
1155 seq_puts(m, "\n\nvim:ft=help\n");
1160 static void *s_start(struct seq_file *m, loff_t *pos)
1162 struct trace_iterator *iter = m->private;
1167 mutex_lock(&trace_types_lock);
1169 if (!current_trace || current_trace != iter->trace) {
1170 mutex_unlock(&trace_types_lock);
1174 atomic_inc(&trace_record_cmdline_disabled);
1176 /* let the tracer grab locks here if needed */
1177 if (current_trace->start)
1178 current_trace->start(iter);
1180 if (*pos != iter->pos) {
1184 iter->prev_ent = NULL;
1185 iter->prev_cpu = -1;
1187 for_each_tracing_cpu(i) {
1188 iter->next_idx[i] = 0;
1189 iter->next_page[i] = NULL;
1192 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1197 p = s_next(m, p, &l);
1203 static void s_stop(struct seq_file *m, void *p)
1205 struct trace_iterator *iter = m->private;
1207 atomic_dec(&trace_record_cmdline_disabled);
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);
1213 mutex_unlock(&trace_types_lock);
1217 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1219 #ifdef CONFIG_KALLSYMS
1220 char str[KSYM_SYMBOL_LEN];
1222 kallsyms_lookup(address, NULL, NULL, NULL, str);
1224 return trace_seq_printf(s, fmt, str);
1230 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1231 unsigned long address)
1233 #ifdef CONFIG_KALLSYMS
1234 char str[KSYM_SYMBOL_LEN];
1236 sprint_symbol(str, address);
1237 return trace_seq_printf(s, fmt, str);
1242 #ifndef CONFIG_64BIT
1243 # define IP_FMT "%08lx"
1245 # define IP_FMT "%016lx"
1249 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1254 return trace_seq_printf(s, "0");
1256 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1257 ret = seq_print_sym_offset(s, "%s", ip);
1259 ret = seq_print_sym_short(s, "%s", ip);
1264 if (sym_flags & TRACE_ITER_SYM_ADDR)
1265 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1269 static void print_lat_help_header(struct seq_file *m)
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");
1282 static void print_func_help_header(struct seq_file *m)
1284 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1285 seq_puts(m, "# | | | | |\n");
1290 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
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;
1299 const char *name = "preemption";
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;
1310 entries += tr->data[cpu]->trace_idx;
1314 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
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),
1324 #if defined(CONFIG_PREEMPT_NONE)
1326 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1328 #elif defined(CONFIG_PREEMPT_DESKTOP)
1333 /* These are reserved for later use */
1336 seq_printf(m, " #P:%d)\n", num_online_cpus());
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");
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);
1361 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1363 int hardirq, softirq;
1366 comm = trace_find_cmdline(entry->pid);
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' : '.'));
1374 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1375 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1376 if (hardirq && softirq) {
1377 trace_seq_putc(s, 'H');
1380 trace_seq_putc(s, 'h');
1383 trace_seq_putc(s, 's');
1385 trace_seq_putc(s, '.');
1389 if (entry->preempt_count)
1390 trace_seq_printf(s, "%x", entry->preempt_count);
1392 trace_seq_puts(s, ".");
1395 unsigned long preempt_mark_thresh = 100;
1398 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1399 unsigned long rel_usecs)
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, "+: ");
1407 trace_seq_puts(s, " : ");
1410 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1413 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
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;
1429 rel_usecs = ns2usecs(next_entry->t - entry->t);
1430 abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1433 comm = trace_find_cmdline(entry->pid);
1434 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1435 " %ld.%03ldms (+%ld.%03ldms): ",
1437 entry->pid, cpu, entry->flags,
1438 entry->preempt_count, trace_idx,
1441 abs_usecs % 1000, rel_usecs/1000,
1444 lat_print_generic(s, entry, cpu);
1445 lat_print_timestamp(s, abs_usecs, rel_usecs);
1447 switch (entry->type) {
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");
1456 T = entry->ctx.next_state < sizeof(state_to_char) ?
1457 state_to_char[entry->ctx.next_state] : 'X';
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,
1471 trace_seq_printf(s, "# %ld %ld %ld\n",
1472 entry->special.arg1,
1473 entry->special.arg2,
1474 entry->special.arg3);
1477 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1479 trace_seq_puts(s, " <= ");
1480 seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1482 trace_seq_puts(s, "\n");
1485 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1490 static int print_trace_fmt(struct trace_iterator *iter)
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;
1505 comm = trace_find_cmdline(iter->ent->pid);
1507 t = ns2usecs(entry->t);
1508 usec_rem = do_div(t, 1000000ULL);
1509 secs = (unsigned long)t;
1511 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1514 ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1517 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1521 switch (entry->type) {
1523 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1526 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1527 entry->fn.parent_ip) {
1528 ret = trace_seq_printf(s, " <-");
1531 ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1536 ret = trace_seq_printf(s, "\n");
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,
1550 entry->type == TRACE_CTX ? "==>" : " +",
1551 entry->ctx.next_pid,
1552 entry->ctx.next_prio,
1558 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1559 entry->special.arg1,
1560 entry->special.arg2,
1561 entry->special.arg3);
1566 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1568 ret = trace_seq_puts(s, " <= ");
1572 ret = seq_print_ip_sym(s, entry->stack.caller[i],
1577 ret = trace_seq_puts(s, "\n");
1585 static int print_raw_fmt(struct trace_iterator *iter)
1587 struct trace_seq *s = &iter->seq;
1588 struct trace_entry *entry;
1594 ret = trace_seq_printf(s, "%d %d %llu ",
1595 entry->pid, iter->cpu, entry->t);
1599 switch (entry->type) {
1601 ret = trace_seq_printf(s, "%x %x\n",
1602 entry->fn.ip, entry->fn.parent_ip);
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)
1614 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1615 entry->ctx.prev_pid,
1616 entry->ctx.prev_prio,
1618 entry->ctx.next_pid,
1619 entry->ctx.next_prio,
1626 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1627 entry->special.arg1,
1628 entry->special.arg2,
1629 entry->special.arg3);
1637 #define SEQ_PUT_FIELD_RET(s, x) \
1639 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1643 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1645 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1649 static int print_hex_fmt(struct trace_iterator *iter)
1651 struct trace_seq *s = &iter->seq;
1652 unsigned char newline = '\n';
1653 struct trace_entry *entry;
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);
1662 switch (entry->type) {
1664 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1665 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
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)
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);
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);
1690 SEQ_PUT_FIELD_RET(s, newline);
1695 static int print_bin_fmt(struct trace_iterator *iter)
1697 struct trace_seq *s = &iter->seq;
1698 struct trace_entry *entry;
1702 SEQ_PUT_FIELD_RET(s, entry->pid);
1703 SEQ_PUT_FIELD_RET(s, entry->cpu);
1704 SEQ_PUT_FIELD_RET(s, entry->t);
1706 switch (entry->type) {
1708 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1709 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
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);
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);
1729 static int trace_empty(struct trace_iterator *iter)
1731 struct trace_array_cpu *data;
1734 for_each_tracing_cpu(cpu) {
1735 data = iter->tr->data[cpu];
1737 if (head_page(data) && data->trace_idx &&
1738 (data->trace_tail != data->trace_head ||
1739 data->trace_tail_idx != data->trace_head_idx))
1745 static int print_trace_line(struct trace_iterator *iter)
1747 if (iter->trace && iter->trace->print_line)
1748 return iter->trace->print_line(iter);
1750 if (trace_flags & TRACE_ITER_BIN)
1751 return print_bin_fmt(iter);
1753 if (trace_flags & TRACE_ITER_HEX)
1754 return print_hex_fmt(iter);
1756 if (trace_flags & TRACE_ITER_RAW)
1757 return print_raw_fmt(iter);
1759 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1760 return print_lat_fmt(iter, iter->idx, iter->cpu);
1762 return print_trace_fmt(iter);
1765 static int s_show(struct seq_file *m, void *v)
1767 struct trace_iterator *iter = v;
1769 if (iter->ent == NULL) {
1771 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1774 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1775 /* print nothing if the buffers are empty */
1776 if (trace_empty(iter))
1778 print_trace_header(m, iter);
1779 if (!(trace_flags & TRACE_ITER_VERBOSE))
1780 print_lat_help_header(m);
1782 if (!(trace_flags & TRACE_ITER_VERBOSE))
1783 print_func_help_header(m);
1786 print_trace_line(iter);
1787 trace_print_seq(m, &iter->seq);
1793 static struct seq_operations tracer_seq_ops = {
1800 static struct trace_iterator *
1801 __tracing_open(struct inode *inode, struct file *file, int *ret)
1803 struct trace_iterator *iter;
1805 if (tracing_disabled) {
1810 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1816 mutex_lock(&trace_types_lock);
1817 if (current_trace && current_trace->print_max)
1820 iter->tr = inode->i_private;
1821 iter->trace = current_trace;
1824 /* TODO stop tracer */
1825 *ret = seq_open(file, &tracer_seq_ops);
1827 struct seq_file *m = file->private_data;
1830 /* stop the trace while dumping */
1834 if (iter->trace && iter->trace->open)
1835 iter->trace->open(iter);
1840 mutex_unlock(&trace_types_lock);
1846 int tracing_open_generic(struct inode *inode, struct file *filp)
1848 if (tracing_disabled)
1851 filp->private_data = inode->i_private;
1855 int tracing_release(struct inode *inode, struct file *file)
1857 struct seq_file *m = (struct seq_file *)file->private_data;
1858 struct trace_iterator *iter = m->private;
1860 mutex_lock(&trace_types_lock);
1861 if (iter->trace && iter->trace->close)
1862 iter->trace->close(iter);
1864 /* reenable tracing if it was previously enabled */
1867 mutex_unlock(&trace_types_lock);
1869 seq_release(inode, file);
1874 static int tracing_open(struct inode *inode, struct file *file)
1878 __tracing_open(inode, file, &ret);
1883 static int tracing_lt_open(struct inode *inode, struct file *file)
1885 struct trace_iterator *iter;
1888 iter = __tracing_open(inode, file, &ret);
1891 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1898 t_next(struct seq_file *m, void *v, loff_t *pos)
1900 struct tracer *t = m->private;
1912 static void *t_start(struct seq_file *m, loff_t *pos)
1914 struct tracer *t = m->private;
1917 mutex_lock(&trace_types_lock);
1918 for (; t && l < *pos; t = t_next(m, t, &l))
1924 static void t_stop(struct seq_file *m, void *p)
1926 mutex_unlock(&trace_types_lock);
1929 static int t_show(struct seq_file *m, void *v)
1931 struct tracer *t = v;
1936 seq_printf(m, "%s", t->name);
1945 static struct seq_operations show_traces_seq_ops = {
1952 static int show_traces_open(struct inode *inode, struct file *file)
1956 if (tracing_disabled)
1959 ret = seq_open(file, &show_traces_seq_ops);
1961 struct seq_file *m = file->private_data;
1962 m->private = trace_types;
1968 static struct file_operations tracing_fops = {
1969 .open = tracing_open,
1971 .llseek = seq_lseek,
1972 .release = tracing_release,
1975 static struct file_operations tracing_lt_fops = {
1976 .open = tracing_lt_open,
1978 .llseek = seq_lseek,
1979 .release = tracing_release,
1982 static struct file_operations show_traces_fops = {
1983 .open = show_traces_open,
1985 .release = seq_release,
1989 * Only trace on a CPU if the bitmask is set:
1991 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1994 * When tracing/tracing_cpu_mask is modified then this holds
1995 * the new bitmask we are about to install:
1997 static cpumask_t tracing_cpumask_new;
2000 * The tracer itself will not take this lock, but still we want
2001 * to provide a consistent cpumask to user-space:
2003 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2006 * Temporary storage for the character representation of the
2007 * CPU bitmask (and one more byte for the newline):
2009 static char mask_str[NR_CPUS + 1];
2012 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2013 size_t count, loff_t *ppos)
2017 mutex_lock(&tracing_cpumask_update_lock);
2019 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2020 if (count - len < 2) {
2024 len += sprintf(mask_str + len, "\n");
2025 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2028 mutex_unlock(&tracing_cpumask_update_lock);
2034 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2035 size_t count, loff_t *ppos)
2039 mutex_lock(&tracing_cpumask_update_lock);
2040 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2044 raw_local_irq_disable();
2045 __raw_spin_lock(&ftrace_max_lock);
2046 for_each_tracing_cpu(cpu) {
2048 * Increase/decrease the disabled counter if we are
2049 * about to flip a bit in the cpumask:
2051 if (cpu_isset(cpu, tracing_cpumask) &&
2052 !cpu_isset(cpu, tracing_cpumask_new)) {
2053 atomic_inc(&global_trace.data[cpu]->disabled);
2055 if (!cpu_isset(cpu, tracing_cpumask) &&
2056 cpu_isset(cpu, tracing_cpumask_new)) {
2057 atomic_dec(&global_trace.data[cpu]->disabled);
2060 __raw_spin_unlock(&ftrace_max_lock);
2061 raw_local_irq_enable();
2063 tracing_cpumask = tracing_cpumask_new;
2065 mutex_unlock(&tracing_cpumask_update_lock);
2070 mutex_unlock(&tracing_cpumask_update_lock);
2075 static struct file_operations tracing_cpumask_fops = {
2076 .open = tracing_open_generic,
2077 .read = tracing_cpumask_read,
2078 .write = tracing_cpumask_write,
2082 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2083 size_t cnt, loff_t *ppos)
2090 /* calulate max size */
2091 for (i = 0; trace_options[i]; i++) {
2092 len += strlen(trace_options[i]);
2093 len += 3; /* "no" and space */
2096 /* +2 for \n and \0 */
2097 buf = kmalloc(len + 2, GFP_KERNEL);
2101 for (i = 0; trace_options[i]; i++) {
2102 if (trace_flags & (1 << i))
2103 r += sprintf(buf + r, "%s ", trace_options[i]);
2105 r += sprintf(buf + r, "no%s ", trace_options[i]);
2108 r += sprintf(buf + r, "\n");
2109 WARN_ON(r >= len + 2);
2111 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2119 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2120 size_t cnt, loff_t *ppos)
2127 if (cnt >= sizeof(buf))
2130 if (copy_from_user(&buf, ubuf, cnt))
2135 if (strncmp(buf, "no", 2) == 0) {
2140 for (i = 0; trace_options[i]; i++) {
2141 int len = strlen(trace_options[i]);
2143 if (strncmp(cmp, trace_options[i], len) == 0) {
2145 trace_flags &= ~(1 << i);
2147 trace_flags |= (1 << i);
2152 * If no option could be set, return an error:
2154 if (!trace_options[i])
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,
2168 static const char readme_msg[] =
2169 "tracing mini-HOWTO:\n\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"
2176 "# echo sched_switch > /debug/tracing/current_tracer\n"
2177 "# cat /debug/tracing/current_tracer\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"
2188 tracing_readme_read(struct file *filp, char __user *ubuf,
2189 size_t cnt, loff_t *ppos)
2191 return simple_read_from_buffer(ubuf, cnt, ppos,
2192 readme_msg, strlen(readme_msg));
2195 static struct file_operations tracing_readme_fops = {
2196 .open = tracing_open_generic,
2197 .read = tracing_readme_read,
2201 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2202 size_t cnt, loff_t *ppos)
2204 struct trace_array *tr = filp->private_data;
2208 r = sprintf(buf, "%ld\n", tr->ctrl);
2209 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2213 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2214 size_t cnt, loff_t *ppos)
2216 struct trace_array *tr = filp->private_data;
2221 if (cnt >= sizeof(buf))
2224 if (copy_from_user(&buf, ubuf, cnt))
2229 ret = strict_strtoul(buf, 10, &val);
2235 mutex_lock(&trace_types_lock);
2236 if (tr->ctrl ^ val) {
2244 if (current_trace && current_trace->ctrl_update)
2245 current_trace->ctrl_update(tr);
2247 mutex_unlock(&trace_types_lock);
2255 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2256 size_t cnt, loff_t *ppos)
2258 char buf[max_tracer_type_len+2];
2261 mutex_lock(&trace_types_lock);
2263 r = sprintf(buf, "%s\n", current_trace->name);
2265 r = sprintf(buf, "\n");
2266 mutex_unlock(&trace_types_lock);
2268 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2272 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2273 size_t cnt, loff_t *ppos)
2275 struct trace_array *tr = &global_trace;
2277 char buf[max_tracer_type_len+1];
2280 if (cnt > max_tracer_type_len)
2281 cnt = max_tracer_type_len;
2283 if (copy_from_user(&buf, ubuf, cnt))
2288 /* strip ending whitespace. */
2289 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2292 mutex_lock(&trace_types_lock);
2293 for (t = trace_types; t; t = t->next) {
2294 if (strcmp(t->name, buf) == 0)
2297 if (!t || t == current_trace)
2300 if (current_trace && current_trace->reset)
2301 current_trace->reset(tr);
2308 mutex_unlock(&trace_types_lock);
2316 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2317 size_t cnt, loff_t *ppos)
2319 unsigned long *ptr = filp->private_data;
2323 r = snprintf(buf, sizeof(buf), "%ld\n",
2324 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2325 if (r > sizeof(buf))
2327 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2331 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2332 size_t cnt, loff_t *ppos)
2334 long *ptr = filp->private_data;
2339 if (cnt >= sizeof(buf))
2342 if (copy_from_user(&buf, ubuf, cnt))
2347 ret = strict_strtoul(buf, 10, &val);
2356 static atomic_t tracing_reader;
2358 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2360 struct trace_iterator *iter;
2362 if (tracing_disabled)
2365 /* We only allow for reader of the pipe */
2366 if (atomic_inc_return(&tracing_reader) != 1) {
2367 atomic_dec(&tracing_reader);
2371 /* create a buffer to store the information to pass to userspace */
2372 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2376 mutex_lock(&trace_types_lock);
2377 iter->tr = &global_trace;
2378 iter->trace = current_trace;
2379 filp->private_data = iter;
2381 if (iter->trace->pipe_open)
2382 iter->trace->pipe_open(iter);
2383 mutex_unlock(&trace_types_lock);
2388 static int tracing_release_pipe(struct inode *inode, struct file *file)
2390 struct trace_iterator *iter = file->private_data;
2393 atomic_dec(&tracing_reader);
2399 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2401 struct trace_iterator *iter = filp->private_data;
2403 if (trace_flags & TRACE_ITER_BLOCK) {
2405 * Always select as readable when in blocking mode
2407 return POLLIN | POLLRDNORM;
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;
2423 tracing_read_pipe(struct file *filp, char __user *ubuf,
2424 size_t cnt, loff_t *ppos)
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
2436 /* return any leftover data */
2437 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2442 trace_seq_reset(&iter->seq);
2444 mutex_lock(&trace_types_lock);
2445 if (iter->trace->read) {
2446 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2451 while (trace_empty(iter)) {
2453 if ((filp->f_flags & O_NONBLOCK)) {
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.
2467 set_current_state(TASK_INTERRUPTIBLE);
2468 iter->tr->waiter = current;
2470 mutex_unlock(&trace_types_lock);
2472 /* sleep for 100 msecs, and try again. */
2473 schedule_timeout(HZ/10);
2475 mutex_lock(&trace_types_lock);
2477 iter->tr->waiter = NULL;
2479 if (signal_pending(current)) {
2484 if (iter->trace != current_trace)
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.
2494 * iter->pos will be 0 if we haven't read anything.
2496 if (!tracer_enabled && iter->pos)
2502 /* stop when tracing is finished */
2503 if (trace_empty(iter))
2506 if (cnt >= PAGE_SIZE)
2507 cnt = PAGE_SIZE - 1;
2509 /* reset all but tr, trace, and overruns */
2510 memset(&iter->seq, 0,
2511 sizeof(struct trace_iterator) -
2512 offsetof(struct trace_iterator, seq));
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.
2523 local_irq_save(flags);
2524 #ifdef CONFIG_FTRACE
2525 ftrace_save = ftrace_enabled;
2529 for_each_tracing_cpu(cpu) {
2530 data = iter->tr->data[cpu];
2532 if (!head_page(data) || !data->trace_idx)
2535 atomic_inc(&data->disabled);
2539 for_each_cpu_mask(cpu, mask) {
2540 data = iter->tr->data[cpu];
2541 __raw_spin_lock(&data->lock);
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;
2549 while (find_next_entry_inc(iter) != NULL) {
2551 int len = iter->seq.len;
2553 ret = print_trace_line(iter);
2555 /* don't print partial lines */
2556 iter->seq.len = len;
2560 trace_consume(iter);
2562 if (iter->seq.len >= cnt)
2566 for_each_cpu_mask(cpu, mask) {
2567 data = iter->tr->data[cpu];
2568 __raw_spin_unlock(&data->lock);
2571 for_each_cpu_mask(cpu, mask) {
2572 data = iter->tr->data[cpu];
2573 atomic_dec(&data->disabled);
2575 #ifdef CONFIG_FTRACE
2576 ftrace_enabled = ftrace_save;
2578 local_irq_restore(flags);
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);
2588 mutex_unlock(&trace_types_lock);
2594 tracing_entries_read(struct file *filp, char __user *ubuf,
2595 size_t cnt, loff_t *ppos)
2597 struct trace_array *tr = filp->private_data;
2601 r = sprintf(buf, "%lu\n", tr->entries);
2602 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2606 tracing_entries_write(struct file *filp, const char __user *ubuf,
2607 size_t cnt, loff_t *ppos)
2613 if (cnt >= sizeof(buf))
2616 if (copy_from_user(&buf, ubuf, cnt))
2621 ret = strict_strtoul(buf, 10, &val);
2625 /* must have at least 1 entry */
2629 mutex_lock(&trace_types_lock);
2631 if (current_trace != &no_tracer) {
2633 pr_info("ftrace: set current_tracer to none"
2634 " before modifying buffer size\n");
2638 if (val > global_trace.entries) {
2639 long pages_requested;
2640 unsigned long freeable_pages;
2642 /* make sure we have enough memory before mapping */
2644 (val + (ENTRIES_PER_PAGE-1)) / ENTRIES_PER_PAGE;
2646 /* account for each buffer (and max_tr) */
2647 pages_requested *= tracing_nr_buffers * 2;
2649 /* Check for overflow */
2650 if (pages_requested < 0) {
2655 freeable_pages = determine_dirtyable_memory();
2657 /* we only allow to request 1/4 of useable memory */
2658 if (pages_requested >
2659 ((freeable_pages + tracing_pages_allocated) / 4)) {
2664 while (global_trace.entries < val) {
2665 if (trace_alloc_page()) {
2669 /* double check that we don't go over the known pages */
2670 if (tracing_pages_allocated > pages_requested)
2675 /* include the number of entries in val (inc of page entries) */
2676 while (global_trace.entries > val + (ENTRIES_PER_PAGE - 1))
2683 max_tr.entries = global_trace.entries;
2684 mutex_unlock(&trace_types_lock);
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,
2695 static struct file_operations tracing_ctrl_fops = {
2696 .open = tracing_open_generic,
2697 .read = tracing_ctrl_read,
2698 .write = tracing_ctrl_write,
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,
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,
2714 static struct file_operations tracing_entries_fops = {
2715 .open = tracing_open_generic,
2716 .read = tracing_entries_read,
2717 .write = tracing_entries_write,
2720 #ifdef CONFIG_DYNAMIC_FTRACE
2723 tracing_read_long(struct file *filp, char __user *ubuf,
2724 size_t cnt, loff_t *ppos)
2726 unsigned long *p = filp->private_data;
2730 r = sprintf(buf, "%ld\n", *p);
2732 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2735 static struct file_operations tracing_read_long_fops = {
2736 .open = tracing_open_generic,
2737 .read = tracing_read_long,
2741 static struct dentry *d_tracer;
2743 struct dentry *tracing_init_dentry(void)
2750 d_tracer = debugfs_create_dir("tracing", NULL);
2752 if (!d_tracer && !once) {
2754 pr_warning("Could not create debugfs directory 'tracing'\n");
2761 #ifdef CONFIG_FTRACE_SELFTEST
2762 /* Let selftest have access to static functions in this file */
2763 #include "trace_selftest.c"
2766 static __init void tracer_init_debugfs(void)
2768 struct dentry *d_tracer;
2769 struct dentry *entry;
2771 d_tracer = tracing_init_dentry();
2773 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2774 &global_trace, &tracing_ctrl_fops);
2776 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2778 entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2779 NULL, &tracing_iter_fops);
2781 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2783 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2784 NULL, &tracing_cpumask_fops);
2786 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2788 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2789 &global_trace, &tracing_lt_fops);
2791 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2793 entry = debugfs_create_file("trace", 0444, d_tracer,
2794 &global_trace, &tracing_fops);
2796 pr_warning("Could not create debugfs 'trace' entry\n");
2798 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2799 &global_trace, &show_traces_fops);
2801 pr_warning("Could not create debugfs 'trace' entry\n");
2803 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2804 &global_trace, &set_tracer_fops);
2806 pr_warning("Could not create debugfs 'trace' entry\n");
2808 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2809 &tracing_max_latency,
2810 &tracing_max_lat_fops);
2812 pr_warning("Could not create debugfs "
2813 "'tracing_max_latency' entry\n");
2815 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2816 &tracing_thresh, &tracing_max_lat_fops);
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);
2823 pr_warning("Could not create debugfs 'README' entry\n");
2825 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2826 NULL, &tracing_pipe_fops);
2828 pr_warning("Could not create debugfs "
2829 "'tracing_threash' entry\n");
2831 entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2832 &global_trace, &tracing_entries_fops);
2834 pr_warning("Could not create debugfs "
2835 "'tracing_threash' entry\n");
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);
2842 pr_warning("Could not create debugfs "
2843 "'dyn_ftrace_total_info' entry\n");
2847 static int trace_alloc_page(void)
2849 struct trace_array_cpu *data;
2850 struct page *page, *tmp;
2853 unsigned pages_allocated = 0;
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");
2866 page = virt_to_page(array);
2867 list_add(&page->lru, &pages);
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");
2878 page = virt_to_page(array);
2879 list_add(&page->lru, &pages);
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);
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);
2899 tracing_pages_allocated += pages_allocated;
2900 global_trace.entries += ENTRIES_PER_PAGE;
2905 list_for_each_entry_safe(page, tmp, &pages, lru) {
2906 list_del_init(&page->lru);
2912 static int trace_free_page(void)
2914 struct trace_array_cpu *data;
2916 struct list_head *p;
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 */
2927 tracing_disabled = 1;
2931 page = list_entry(p, struct page, lru);
2933 list_del(&page->lru);
2934 tracing_pages_allocated--;
2935 tracing_pages_allocated--;
2938 tracing_reset(data);
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 */
2946 tracing_disabled = 1;
2950 page = list_entry(p, struct page, lru);
2952 list_del(&page->lru);
2955 tracing_reset(data);
2958 global_trace.entries -= ENTRIES_PER_PAGE;
2963 __init static int tracer_alloc_buffers(void)
2965 struct trace_array_cpu *data;
2972 global_trace.ctrl = tracer_enabled;
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;
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);
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");
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 */
2997 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2998 max_tr.data[i]->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
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");
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);
3017 * Since we allocate by orders of pages, we may be able to
3020 global_trace.entries = ENTRIES_PER_PAGE;
3023 while (global_trace.entries < trace_nr_entries) {
3024 if (trace_alloc_page())
3028 max_tr.entries = global_trace.entries;
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);
3035 tracer_init_debugfs();
3037 trace_init_cmdlines();
3039 register_tracer(&no_tracer);
3040 current_trace = &no_tracer;
3042 /* All seems OK, enable tracing */
3043 tracing_disabled = 0;
3048 for (i-- ; i >= 0; i--) {
3049 struct page *page, *tmp;
3050 struct trace_array_cpu *data = global_trace.data[i];
3053 list_for_each_entry_safe(page, tmp,
3054 &data->trace_pages, lru) {
3055 list_del_init(&page->lru);
3060 #ifdef CONFIG_TRACER_MAX_TRACE
3061 data = max_tr.data[i];
3063 list_for_each_entry_safe(page, tmp,
3064 &data->trace_pages, lru) {
3065 list_del_init(&page->lru);
3073 fs_initcall(tracer_alloc_buffers);