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 spin_lock_irqsave(&data->lock, irq_flags);
842 entry = tracing_get_trace_entry(tr, data);
843 tracing_generic_entry_update(entry, 0);
844 entry->type = TRACE_MMIO_RW;
846 spin_unlock_irqrestore(&data->lock, irq_flags);
851 void __trace_mmiotrace_map(struct trace_array *tr, struct trace_array_cpu *data,
852 struct mmiotrace_map *map)
854 struct trace_entry *entry;
855 unsigned long irq_flags;
857 spin_lock_irqsave(&data->lock, irq_flags);
858 entry = tracing_get_trace_entry(tr, data);
859 tracing_generic_entry_update(entry, 0);
860 entry->type = TRACE_MMIO_MAP;
861 entry->mmiomap = *map;
862 spin_unlock_irqrestore(&data->lock, irq_flags);
868 void __trace_stack(struct trace_array *tr,
869 struct trace_array_cpu *data,
873 struct trace_entry *entry;
874 struct stack_trace trace;
876 if (!(trace_flags & TRACE_ITER_STACKTRACE))
879 entry = tracing_get_trace_entry(tr, data);
880 tracing_generic_entry_update(entry, flags);
881 entry->type = TRACE_STACK;
883 memset(&entry->stack, 0, sizeof(entry->stack));
885 trace.nr_entries = 0;
886 trace.max_entries = FTRACE_STACK_ENTRIES;
888 trace.entries = entry->stack.caller;
890 save_stack_trace(&trace);
894 __trace_special(void *__tr, void *__data,
895 unsigned long arg1, unsigned long arg2, unsigned long arg3)
897 struct trace_array_cpu *data = __data;
898 struct trace_array *tr = __tr;
899 struct trace_entry *entry;
900 unsigned long irq_flags;
902 raw_local_irq_save(irq_flags);
903 __raw_spin_lock(&data->lock);
904 entry = tracing_get_trace_entry(tr, data);
905 tracing_generic_entry_update(entry, 0);
906 entry->type = TRACE_SPECIAL;
907 entry->special.arg1 = arg1;
908 entry->special.arg2 = arg2;
909 entry->special.arg3 = arg3;
910 __trace_stack(tr, data, irq_flags, 4);
911 __raw_spin_unlock(&data->lock);
912 raw_local_irq_restore(irq_flags);
918 tracing_sched_switch_trace(struct trace_array *tr,
919 struct trace_array_cpu *data,
920 struct task_struct *prev,
921 struct task_struct *next,
924 struct trace_entry *entry;
925 unsigned long irq_flags;
927 raw_local_irq_save(irq_flags);
928 __raw_spin_lock(&data->lock);
929 entry = tracing_get_trace_entry(tr, data);
930 tracing_generic_entry_update(entry, flags);
931 entry->type = TRACE_CTX;
932 entry->ctx.prev_pid = prev->pid;
933 entry->ctx.prev_prio = prev->prio;
934 entry->ctx.prev_state = prev->state;
935 entry->ctx.next_pid = next->pid;
936 entry->ctx.next_prio = next->prio;
937 entry->ctx.next_state = next->state;
938 __trace_stack(tr, data, flags, 5);
939 __raw_spin_unlock(&data->lock);
940 raw_local_irq_restore(irq_flags);
944 tracing_sched_wakeup_trace(struct trace_array *tr,
945 struct trace_array_cpu *data,
946 struct task_struct *wakee,
947 struct task_struct *curr,
950 struct trace_entry *entry;
951 unsigned long irq_flags;
953 raw_local_irq_save(irq_flags);
954 __raw_spin_lock(&data->lock);
955 entry = tracing_get_trace_entry(tr, data);
956 tracing_generic_entry_update(entry, flags);
957 entry->type = TRACE_WAKE;
958 entry->ctx.prev_pid = curr->pid;
959 entry->ctx.prev_prio = curr->prio;
960 entry->ctx.prev_state = curr->state;
961 entry->ctx.next_pid = wakee->pid;
962 entry->ctx.next_prio = wakee->prio;
963 entry->ctx.next_state = wakee->state;
964 __trace_stack(tr, data, flags, 6);
965 __raw_spin_unlock(&data->lock);
966 raw_local_irq_restore(irq_flags);
973 function_trace_call(unsigned long ip, unsigned long parent_ip)
975 struct trace_array *tr = &global_trace;
976 struct trace_array_cpu *data;
981 if (unlikely(!tracer_enabled))
984 local_irq_save(flags);
985 cpu = raw_smp_processor_id();
986 data = tr->data[cpu];
987 disabled = atomic_inc_return(&data->disabled);
989 if (likely(disabled == 1))
990 trace_function(tr, data, ip, parent_ip, flags);
992 atomic_dec(&data->disabled);
993 local_irq_restore(flags);
996 static struct ftrace_ops trace_ops __read_mostly =
998 .func = function_trace_call,
1001 void tracing_start_function_trace(void)
1003 register_ftrace_function(&trace_ops);
1006 void tracing_stop_function_trace(void)
1008 unregister_ftrace_function(&trace_ops);
1012 enum trace_file_type {
1013 TRACE_FILE_LAT_FMT = 1,
1016 static struct trace_entry *
1017 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
1018 struct trace_iterator *iter, int cpu)
1021 struct trace_entry *array;
1023 if (iter->next_idx[cpu] >= tr->entries ||
1024 iter->next_idx[cpu] >= data->trace_idx ||
1025 (data->trace_head == data->trace_tail &&
1026 data->trace_head_idx == data->trace_tail_idx))
1029 if (!iter->next_page[cpu]) {
1030 /* Initialize the iterator for this cpu trace buffer */
1031 WARN_ON(!data->trace_tail);
1032 page = virt_to_page(data->trace_tail);
1033 iter->next_page[cpu] = &page->lru;
1034 iter->next_page_idx[cpu] = data->trace_tail_idx;
1037 page = list_entry(iter->next_page[cpu], struct page, lru);
1038 BUG_ON(&data->trace_pages == &page->lru);
1040 array = page_address(page);
1042 WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
1043 return &array[iter->next_page_idx[cpu]];
1046 static struct trace_entry *
1047 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
1049 struct trace_array *tr = iter->tr;
1050 struct trace_entry *ent, *next = NULL;
1054 for_each_tracing_cpu(cpu) {
1055 if (!head_page(tr->data[cpu]))
1057 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
1059 * Pick the entry with the smallest timestamp:
1061 if (ent && (!next || ent->t < next->t)) {
1068 *ent_cpu = next_cpu;
1073 static void trace_iterator_increment(struct trace_iterator *iter)
1076 iter->next_idx[iter->cpu]++;
1077 iter->next_page_idx[iter->cpu]++;
1079 if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
1080 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1082 iter->next_page_idx[iter->cpu] = 0;
1083 iter->next_page[iter->cpu] =
1084 trace_next_list(data, iter->next_page[iter->cpu]);
1088 static void trace_consume(struct trace_iterator *iter)
1090 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1092 data->trace_tail_idx++;
1093 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
1094 data->trace_tail = trace_next_page(data, data->trace_tail);
1095 data->trace_tail_idx = 0;
1098 /* Check if we empty it, then reset the index */
1099 if (data->trace_head == data->trace_tail &&
1100 data->trace_head_idx == data->trace_tail_idx)
1101 data->trace_idx = 0;
1104 static void *find_next_entry_inc(struct trace_iterator *iter)
1106 struct trace_entry *next;
1109 next = find_next_entry(iter, &next_cpu);
1111 iter->prev_ent = iter->ent;
1112 iter->prev_cpu = iter->cpu;
1115 iter->cpu = next_cpu;
1118 trace_iterator_increment(iter);
1120 return next ? iter : NULL;
1123 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1125 struct trace_iterator *iter = m->private;
1126 void *last_ent = iter->ent;
1132 /* can't go backwards */
1137 ent = find_next_entry_inc(iter);
1141 while (ent && iter->idx < i)
1142 ent = find_next_entry_inc(iter);
1146 if (last_ent && !ent)
1147 seq_puts(m, "\n\nvim:ft=help\n");
1152 static void *s_start(struct seq_file *m, loff_t *pos)
1154 struct trace_iterator *iter = m->private;
1159 mutex_lock(&trace_types_lock);
1161 if (!current_trace || current_trace != iter->trace) {
1162 mutex_unlock(&trace_types_lock);
1166 atomic_inc(&trace_record_cmdline_disabled);
1168 /* let the tracer grab locks here if needed */
1169 if (current_trace->start)
1170 current_trace->start(iter);
1172 if (*pos != iter->pos) {
1176 iter->prev_ent = NULL;
1177 iter->prev_cpu = -1;
1179 for_each_tracing_cpu(i) {
1180 iter->next_idx[i] = 0;
1181 iter->next_page[i] = NULL;
1184 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1189 p = s_next(m, p, &l);
1195 static void s_stop(struct seq_file *m, void *p)
1197 struct trace_iterator *iter = m->private;
1199 atomic_dec(&trace_record_cmdline_disabled);
1201 /* let the tracer release locks here if needed */
1202 if (current_trace && current_trace == iter->trace && iter->trace->stop)
1203 iter->trace->stop(iter);
1205 mutex_unlock(&trace_types_lock);
1209 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1211 #ifdef CONFIG_KALLSYMS
1212 char str[KSYM_SYMBOL_LEN];
1214 kallsyms_lookup(address, NULL, NULL, NULL, str);
1216 return trace_seq_printf(s, fmt, str);
1222 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1223 unsigned long address)
1225 #ifdef CONFIG_KALLSYMS
1226 char str[KSYM_SYMBOL_LEN];
1228 sprint_symbol(str, address);
1229 return trace_seq_printf(s, fmt, str);
1234 #ifndef CONFIG_64BIT
1235 # define IP_FMT "%08lx"
1237 # define IP_FMT "%016lx"
1241 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1246 return trace_seq_printf(s, "0");
1248 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1249 ret = seq_print_sym_offset(s, "%s", ip);
1251 ret = seq_print_sym_short(s, "%s", ip);
1256 if (sym_flags & TRACE_ITER_SYM_ADDR)
1257 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1261 static void print_lat_help_header(struct seq_file *m)
1263 seq_puts(m, "# _------=> CPU# \n");
1264 seq_puts(m, "# / _-----=> irqs-off \n");
1265 seq_puts(m, "# | / _----=> need-resched \n");
1266 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1267 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1268 seq_puts(m, "# |||| / \n");
1269 seq_puts(m, "# ||||| delay \n");
1270 seq_puts(m, "# cmd pid ||||| time | caller \n");
1271 seq_puts(m, "# \\ / ||||| \\ | / \n");
1274 static void print_func_help_header(struct seq_file *m)
1276 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1277 seq_puts(m, "# | | | | |\n");
1282 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1284 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1285 struct trace_array *tr = iter->tr;
1286 struct trace_array_cpu *data = tr->data[tr->cpu];
1287 struct tracer *type = current_trace;
1288 unsigned long total = 0;
1289 unsigned long entries = 0;
1291 const char *name = "preemption";
1296 for_each_tracing_cpu(cpu) {
1297 if (head_page(tr->data[cpu])) {
1298 total += tr->data[cpu]->trace_idx;
1299 if (tr->data[cpu]->trace_idx > tr->entries)
1300 entries += tr->entries;
1302 entries += tr->data[cpu]->trace_idx;
1306 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1308 seq_puts(m, "-----------------------------------"
1309 "---------------------------------\n");
1310 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1311 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1312 nsecs_to_usecs(data->saved_latency),
1316 #if defined(CONFIG_PREEMPT_NONE)
1318 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1320 #elif defined(CONFIG_PREEMPT_DESKTOP)
1325 /* These are reserved for later use */
1328 seq_printf(m, " #P:%d)\n", num_online_cpus());
1332 seq_puts(m, " -----------------\n");
1333 seq_printf(m, " | task: %.16s-%d "
1334 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1335 data->comm, data->pid, data->uid, data->nice,
1336 data->policy, data->rt_priority);
1337 seq_puts(m, " -----------------\n");
1339 if (data->critical_start) {
1340 seq_puts(m, " => started at: ");
1341 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1342 trace_print_seq(m, &iter->seq);
1343 seq_puts(m, "\n => ended at: ");
1344 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1345 trace_print_seq(m, &iter->seq);
1353 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1355 int hardirq, softirq;
1358 comm = trace_find_cmdline(entry->pid);
1360 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1361 trace_seq_printf(s, "%d", cpu);
1362 trace_seq_printf(s, "%c%c",
1363 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1364 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1366 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1367 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1368 if (hardirq && softirq) {
1369 trace_seq_putc(s, 'H');
1372 trace_seq_putc(s, 'h');
1375 trace_seq_putc(s, 's');
1377 trace_seq_putc(s, '.');
1381 if (entry->preempt_count)
1382 trace_seq_printf(s, "%x", entry->preempt_count);
1384 trace_seq_puts(s, ".");
1387 unsigned long preempt_mark_thresh = 100;
1390 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1391 unsigned long rel_usecs)
1393 trace_seq_printf(s, " %4lldus", abs_usecs);
1394 if (rel_usecs > preempt_mark_thresh)
1395 trace_seq_puts(s, "!: ");
1396 else if (rel_usecs > 1)
1397 trace_seq_puts(s, "+: ");
1399 trace_seq_puts(s, " : ");
1402 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1405 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1407 struct trace_seq *s = &iter->seq;
1408 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1409 struct trace_entry *next_entry = find_next_entry(iter, NULL);
1410 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1411 struct trace_entry *entry = iter->ent;
1412 unsigned long abs_usecs;
1413 unsigned long rel_usecs;
1421 rel_usecs = ns2usecs(next_entry->t - entry->t);
1422 abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1425 comm = trace_find_cmdline(entry->pid);
1426 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1427 " %ld.%03ldms (+%ld.%03ldms): ",
1429 entry->pid, cpu, entry->flags,
1430 entry->preempt_count, trace_idx,
1433 abs_usecs % 1000, rel_usecs/1000,
1436 lat_print_generic(s, entry, cpu);
1437 lat_print_timestamp(s, abs_usecs, rel_usecs);
1439 switch (entry->type) {
1441 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1442 trace_seq_puts(s, " (");
1443 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1444 trace_seq_puts(s, ")\n");
1448 T = entry->ctx.next_state < sizeof(state_to_char) ?
1449 state_to_char[entry->ctx.next_state] : 'X';
1451 state = entry->ctx.prev_state ? __ffs(entry->ctx.prev_state) + 1 : 0;
1452 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1453 comm = trace_find_cmdline(entry->ctx.next_pid);
1454 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1455 entry->ctx.prev_pid,
1456 entry->ctx.prev_prio,
1457 S, entry->type == TRACE_CTX ? "==>" : " +",
1458 entry->ctx.next_pid,
1459 entry->ctx.next_prio,
1463 trace_seq_printf(s, "# %ld %ld %ld\n",
1464 entry->special.arg1,
1465 entry->special.arg2,
1466 entry->special.arg3);
1469 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1471 trace_seq_puts(s, " <= ");
1472 seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1474 trace_seq_puts(s, "\n");
1477 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1482 static int print_trace_fmt(struct trace_iterator *iter)
1484 struct trace_seq *s = &iter->seq;
1485 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1486 struct trace_entry *entry;
1487 unsigned long usec_rem;
1488 unsigned long long t;
1497 comm = trace_find_cmdline(iter->ent->pid);
1499 t = ns2usecs(entry->t);
1500 usec_rem = do_div(t, 1000000ULL);
1501 secs = (unsigned long)t;
1503 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1506 ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1509 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1513 switch (entry->type) {
1515 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1518 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1519 entry->fn.parent_ip) {
1520 ret = trace_seq_printf(s, " <-");
1523 ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1528 ret = trace_seq_printf(s, "\n");
1534 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1535 state_to_char[entry->ctx.prev_state] : 'X';
1536 T = entry->ctx.next_state < sizeof(state_to_char) ?
1537 state_to_char[entry->ctx.next_state] : 'X';
1538 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1539 entry->ctx.prev_pid,
1540 entry->ctx.prev_prio,
1542 entry->type == TRACE_CTX ? "==>" : " +",
1543 entry->ctx.next_pid,
1544 entry->ctx.next_prio,
1550 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1551 entry->special.arg1,
1552 entry->special.arg2,
1553 entry->special.arg3);
1558 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1560 ret = trace_seq_puts(s, " <= ");
1564 ret = seq_print_ip_sym(s, entry->stack.caller[i],
1569 ret = trace_seq_puts(s, "\n");
1577 static int print_raw_fmt(struct trace_iterator *iter)
1579 struct trace_seq *s = &iter->seq;
1580 struct trace_entry *entry;
1586 ret = trace_seq_printf(s, "%d %d %llu ",
1587 entry->pid, iter->cpu, entry->t);
1591 switch (entry->type) {
1593 ret = trace_seq_printf(s, "%x %x\n",
1594 entry->fn.ip, entry->fn.parent_ip);
1600 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1601 state_to_char[entry->ctx.prev_state] : 'X';
1602 T = entry->ctx.next_state < sizeof(state_to_char) ?
1603 state_to_char[entry->ctx.next_state] : 'X';
1604 if (entry->type == TRACE_WAKE)
1606 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1607 entry->ctx.prev_pid,
1608 entry->ctx.prev_prio,
1610 entry->ctx.next_pid,
1611 entry->ctx.next_prio,
1618 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1619 entry->special.arg1,
1620 entry->special.arg2,
1621 entry->special.arg3);
1629 #define SEQ_PUT_FIELD_RET(s, x) \
1631 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1635 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1637 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1641 static int print_hex_fmt(struct trace_iterator *iter)
1643 struct trace_seq *s = &iter->seq;
1644 unsigned char newline = '\n';
1645 struct trace_entry *entry;
1650 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1651 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1652 SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1654 switch (entry->type) {
1656 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1657 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1661 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1662 state_to_char[entry->ctx.prev_state] : 'X';
1663 T = entry->ctx.next_state < sizeof(state_to_char) ?
1664 state_to_char[entry->ctx.next_state] : 'X';
1665 if (entry->type == TRACE_WAKE)
1667 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1668 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1669 SEQ_PUT_HEX_FIELD_RET(s, S);
1670 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1671 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1672 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1673 SEQ_PUT_HEX_FIELD_RET(s, T);
1677 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1678 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1679 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1682 SEQ_PUT_FIELD_RET(s, newline);
1687 static int print_bin_fmt(struct trace_iterator *iter)
1689 struct trace_seq *s = &iter->seq;
1690 struct trace_entry *entry;
1694 SEQ_PUT_FIELD_RET(s, entry->pid);
1695 SEQ_PUT_FIELD_RET(s, entry->cpu);
1696 SEQ_PUT_FIELD_RET(s, entry->t);
1698 switch (entry->type) {
1700 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1701 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1704 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1705 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1706 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1707 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1708 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1709 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1713 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1714 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1715 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1721 static int trace_empty(struct trace_iterator *iter)
1723 struct trace_array_cpu *data;
1726 for_each_tracing_cpu(cpu) {
1727 data = iter->tr->data[cpu];
1729 if (head_page(data) && data->trace_idx &&
1730 (data->trace_tail != data->trace_head ||
1731 data->trace_tail_idx != data->trace_head_idx))
1737 static int print_trace_line(struct trace_iterator *iter)
1739 if (iter->trace && iter->trace->print_line)
1740 return iter->trace->print_line(iter);
1742 if (trace_flags & TRACE_ITER_BIN)
1743 return print_bin_fmt(iter);
1745 if (trace_flags & TRACE_ITER_HEX)
1746 return print_hex_fmt(iter);
1748 if (trace_flags & TRACE_ITER_RAW)
1749 return print_raw_fmt(iter);
1751 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1752 return print_lat_fmt(iter, iter->idx, iter->cpu);
1754 return print_trace_fmt(iter);
1757 static int s_show(struct seq_file *m, void *v)
1759 struct trace_iterator *iter = v;
1761 if (iter->ent == NULL) {
1763 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1766 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1767 /* print nothing if the buffers are empty */
1768 if (trace_empty(iter))
1770 print_trace_header(m, iter);
1771 if (!(trace_flags & TRACE_ITER_VERBOSE))
1772 print_lat_help_header(m);
1774 if (!(trace_flags & TRACE_ITER_VERBOSE))
1775 print_func_help_header(m);
1778 print_trace_line(iter);
1779 trace_print_seq(m, &iter->seq);
1785 static struct seq_operations tracer_seq_ops = {
1792 static struct trace_iterator *
1793 __tracing_open(struct inode *inode, struct file *file, int *ret)
1795 struct trace_iterator *iter;
1797 if (tracing_disabled) {
1802 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1808 mutex_lock(&trace_types_lock);
1809 if (current_trace && current_trace->print_max)
1812 iter->tr = inode->i_private;
1813 iter->trace = current_trace;
1816 /* TODO stop tracer */
1817 *ret = seq_open(file, &tracer_seq_ops);
1819 struct seq_file *m = file->private_data;
1822 /* stop the trace while dumping */
1826 if (iter->trace && iter->trace->open)
1827 iter->trace->open(iter);
1832 mutex_unlock(&trace_types_lock);
1838 int tracing_open_generic(struct inode *inode, struct file *filp)
1840 if (tracing_disabled)
1843 filp->private_data = inode->i_private;
1847 int tracing_release(struct inode *inode, struct file *file)
1849 struct seq_file *m = (struct seq_file *)file->private_data;
1850 struct trace_iterator *iter = m->private;
1852 mutex_lock(&trace_types_lock);
1853 if (iter->trace && iter->trace->close)
1854 iter->trace->close(iter);
1856 /* reenable tracing if it was previously enabled */
1859 mutex_unlock(&trace_types_lock);
1861 seq_release(inode, file);
1866 static int tracing_open(struct inode *inode, struct file *file)
1870 __tracing_open(inode, file, &ret);
1875 static int tracing_lt_open(struct inode *inode, struct file *file)
1877 struct trace_iterator *iter;
1880 iter = __tracing_open(inode, file, &ret);
1883 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1890 t_next(struct seq_file *m, void *v, loff_t *pos)
1892 struct tracer *t = m->private;
1904 static void *t_start(struct seq_file *m, loff_t *pos)
1906 struct tracer *t = m->private;
1909 mutex_lock(&trace_types_lock);
1910 for (; t && l < *pos; t = t_next(m, t, &l))
1916 static void t_stop(struct seq_file *m, void *p)
1918 mutex_unlock(&trace_types_lock);
1921 static int t_show(struct seq_file *m, void *v)
1923 struct tracer *t = v;
1928 seq_printf(m, "%s", t->name);
1937 static struct seq_operations show_traces_seq_ops = {
1944 static int show_traces_open(struct inode *inode, struct file *file)
1948 if (tracing_disabled)
1951 ret = seq_open(file, &show_traces_seq_ops);
1953 struct seq_file *m = file->private_data;
1954 m->private = trace_types;
1960 static struct file_operations tracing_fops = {
1961 .open = tracing_open,
1963 .llseek = seq_lseek,
1964 .release = tracing_release,
1967 static struct file_operations tracing_lt_fops = {
1968 .open = tracing_lt_open,
1970 .llseek = seq_lseek,
1971 .release = tracing_release,
1974 static struct file_operations show_traces_fops = {
1975 .open = show_traces_open,
1977 .release = seq_release,
1981 * Only trace on a CPU if the bitmask is set:
1983 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1986 * When tracing/tracing_cpu_mask is modified then this holds
1987 * the new bitmask we are about to install:
1989 static cpumask_t tracing_cpumask_new;
1992 * The tracer itself will not take this lock, but still we want
1993 * to provide a consistent cpumask to user-space:
1995 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1998 * Temporary storage for the character representation of the
1999 * CPU bitmask (and one more byte for the newline):
2001 static char mask_str[NR_CPUS + 1];
2004 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2005 size_t count, loff_t *ppos)
2009 mutex_lock(&tracing_cpumask_update_lock);
2011 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2012 if (count - len < 2) {
2016 len += sprintf(mask_str + len, "\n");
2017 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2020 mutex_unlock(&tracing_cpumask_update_lock);
2026 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2027 size_t count, loff_t *ppos)
2031 mutex_lock(&tracing_cpumask_update_lock);
2032 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2036 raw_local_irq_disable();
2037 __raw_spin_lock(&ftrace_max_lock);
2038 for_each_tracing_cpu(cpu) {
2040 * Increase/decrease the disabled counter if we are
2041 * about to flip a bit in the cpumask:
2043 if (cpu_isset(cpu, tracing_cpumask) &&
2044 !cpu_isset(cpu, tracing_cpumask_new)) {
2045 atomic_inc(&global_trace.data[cpu]->disabled);
2047 if (!cpu_isset(cpu, tracing_cpumask) &&
2048 cpu_isset(cpu, tracing_cpumask_new)) {
2049 atomic_dec(&global_trace.data[cpu]->disabled);
2052 __raw_spin_unlock(&ftrace_max_lock);
2053 raw_local_irq_enable();
2055 tracing_cpumask = tracing_cpumask_new;
2057 mutex_unlock(&tracing_cpumask_update_lock);
2062 mutex_unlock(&tracing_cpumask_update_lock);
2067 static struct file_operations tracing_cpumask_fops = {
2068 .open = tracing_open_generic,
2069 .read = tracing_cpumask_read,
2070 .write = tracing_cpumask_write,
2074 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2075 size_t cnt, loff_t *ppos)
2082 /* calulate max size */
2083 for (i = 0; trace_options[i]; i++) {
2084 len += strlen(trace_options[i]);
2085 len += 3; /* "no" and space */
2088 /* +2 for \n and \0 */
2089 buf = kmalloc(len + 2, GFP_KERNEL);
2093 for (i = 0; trace_options[i]; i++) {
2094 if (trace_flags & (1 << i))
2095 r += sprintf(buf + r, "%s ", trace_options[i]);
2097 r += sprintf(buf + r, "no%s ", trace_options[i]);
2100 r += sprintf(buf + r, "\n");
2101 WARN_ON(r >= len + 2);
2103 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2111 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2112 size_t cnt, loff_t *ppos)
2119 if (cnt >= sizeof(buf))
2122 if (copy_from_user(&buf, ubuf, cnt))
2127 if (strncmp(buf, "no", 2) == 0) {
2132 for (i = 0; trace_options[i]; i++) {
2133 int len = strlen(trace_options[i]);
2135 if (strncmp(cmp, trace_options[i], len) == 0) {
2137 trace_flags &= ~(1 << i);
2139 trace_flags |= (1 << i);
2144 * If no option could be set, return an error:
2146 if (!trace_options[i])
2154 static struct file_operations tracing_iter_fops = {
2155 .open = tracing_open_generic,
2156 .read = tracing_iter_ctrl_read,
2157 .write = tracing_iter_ctrl_write,
2160 static const char readme_msg[] =
2161 "tracing mini-HOWTO:\n\n"
2163 "# mount -t debugfs nodev /debug\n\n"
2164 "# cat /debug/tracing/available_tracers\n"
2165 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2166 "# cat /debug/tracing/current_tracer\n"
2168 "# echo sched_switch > /debug/tracing/current_tracer\n"
2169 "# cat /debug/tracing/current_tracer\n"
2171 "# cat /debug/tracing/iter_ctrl\n"
2172 "noprint-parent nosym-offset nosym-addr noverbose\n"
2173 "# echo print-parent > /debug/tracing/iter_ctrl\n"
2174 "# echo 1 > /debug/tracing/tracing_enabled\n"
2175 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2176 "echo 0 > /debug/tracing/tracing_enabled\n"
2180 tracing_readme_read(struct file *filp, char __user *ubuf,
2181 size_t cnt, loff_t *ppos)
2183 return simple_read_from_buffer(ubuf, cnt, ppos,
2184 readme_msg, strlen(readme_msg));
2187 static struct file_operations tracing_readme_fops = {
2188 .open = tracing_open_generic,
2189 .read = tracing_readme_read,
2193 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2194 size_t cnt, loff_t *ppos)
2196 struct trace_array *tr = filp->private_data;
2200 r = sprintf(buf, "%ld\n", tr->ctrl);
2201 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2205 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2206 size_t cnt, loff_t *ppos)
2208 struct trace_array *tr = filp->private_data;
2213 if (cnt >= sizeof(buf))
2216 if (copy_from_user(&buf, ubuf, cnt))
2221 ret = strict_strtoul(buf, 10, &val);
2227 mutex_lock(&trace_types_lock);
2228 if (tr->ctrl ^ val) {
2236 if (current_trace && current_trace->ctrl_update)
2237 current_trace->ctrl_update(tr);
2239 mutex_unlock(&trace_types_lock);
2247 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2248 size_t cnt, loff_t *ppos)
2250 char buf[max_tracer_type_len+2];
2253 mutex_lock(&trace_types_lock);
2255 r = sprintf(buf, "%s\n", current_trace->name);
2257 r = sprintf(buf, "\n");
2258 mutex_unlock(&trace_types_lock);
2260 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2264 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2265 size_t cnt, loff_t *ppos)
2267 struct trace_array *tr = &global_trace;
2269 char buf[max_tracer_type_len+1];
2272 if (cnt > max_tracer_type_len)
2273 cnt = max_tracer_type_len;
2275 if (copy_from_user(&buf, ubuf, cnt))
2280 /* strip ending whitespace. */
2281 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2284 mutex_lock(&trace_types_lock);
2285 for (t = trace_types; t; t = t->next) {
2286 if (strcmp(t->name, buf) == 0)
2289 if (!t || t == current_trace)
2292 if (current_trace && current_trace->reset)
2293 current_trace->reset(tr);
2300 mutex_unlock(&trace_types_lock);
2308 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2309 size_t cnt, loff_t *ppos)
2311 unsigned long *ptr = filp->private_data;
2315 r = snprintf(buf, sizeof(buf), "%ld\n",
2316 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2317 if (r > sizeof(buf))
2319 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2323 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2324 size_t cnt, loff_t *ppos)
2326 long *ptr = filp->private_data;
2331 if (cnt >= sizeof(buf))
2334 if (copy_from_user(&buf, ubuf, cnt))
2339 ret = strict_strtoul(buf, 10, &val);
2348 static atomic_t tracing_reader;
2350 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2352 struct trace_iterator *iter;
2354 if (tracing_disabled)
2357 /* We only allow for reader of the pipe */
2358 if (atomic_inc_return(&tracing_reader) != 1) {
2359 atomic_dec(&tracing_reader);
2363 /* create a buffer to store the information to pass to userspace */
2364 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2368 mutex_lock(&trace_types_lock);
2369 iter->tr = &global_trace;
2370 iter->trace = current_trace;
2371 filp->private_data = iter;
2373 if (iter->trace->pipe_open)
2374 iter->trace->pipe_open(iter);
2375 mutex_unlock(&trace_types_lock);
2380 static int tracing_release_pipe(struct inode *inode, struct file *file)
2382 struct trace_iterator *iter = file->private_data;
2385 atomic_dec(&tracing_reader);
2391 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2393 struct trace_iterator *iter = filp->private_data;
2395 if (trace_flags & TRACE_ITER_BLOCK) {
2397 * Always select as readable when in blocking mode
2399 return POLLIN | POLLRDNORM;
2401 if (!trace_empty(iter))
2402 return POLLIN | POLLRDNORM;
2403 poll_wait(filp, &trace_wait, poll_table);
2404 if (!trace_empty(iter))
2405 return POLLIN | POLLRDNORM;
2415 tracing_read_pipe(struct file *filp, char __user *ubuf,
2416 size_t cnt, loff_t *ppos)
2418 struct trace_iterator *iter = filp->private_data;
2419 struct trace_array_cpu *data;
2420 static cpumask_t mask;
2421 unsigned long flags;
2422 #ifdef CONFIG_FTRACE
2428 /* return any leftover data */
2429 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2434 trace_seq_reset(&iter->seq);
2436 mutex_lock(&trace_types_lock);
2437 if (iter->trace->read) {
2438 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2443 while (trace_empty(iter)) {
2445 if ((filp->f_flags & O_NONBLOCK)) {
2451 * This is a make-shift waitqueue. The reason we don't use
2452 * an actual wait queue is because:
2453 * 1) we only ever have one waiter
2454 * 2) the tracing, traces all functions, we don't want
2455 * the overhead of calling wake_up and friends
2456 * (and tracing them too)
2457 * Anyway, this is really very primitive wakeup.
2459 set_current_state(TASK_INTERRUPTIBLE);
2460 iter->tr->waiter = current;
2462 mutex_unlock(&trace_types_lock);
2464 /* sleep for 100 msecs, and try again. */
2465 schedule_timeout(HZ/10);
2467 mutex_lock(&trace_types_lock);
2469 iter->tr->waiter = NULL;
2471 if (signal_pending(current)) {
2476 if (iter->trace != current_trace)
2480 * We block until we read something and tracing is disabled.
2481 * We still block if tracing is disabled, but we have never
2482 * read anything. This allows a user to cat this file, and
2483 * then enable tracing. But after we have read something,
2484 * we give an EOF when tracing is again disabled.
2486 * iter->pos will be 0 if we haven't read anything.
2488 if (!tracer_enabled && iter->pos)
2494 /* stop when tracing is finished */
2495 if (trace_empty(iter))
2498 if (cnt >= PAGE_SIZE)
2499 cnt = PAGE_SIZE - 1;
2501 /* reset all but tr, trace, and overruns */
2502 memset(&iter->seq, 0,
2503 sizeof(struct trace_iterator) -
2504 offsetof(struct trace_iterator, seq));
2508 * We need to stop all tracing on all CPUS to read the
2509 * the next buffer. This is a bit expensive, but is
2510 * not done often. We fill all what we can read,
2511 * and then release the locks again.
2515 local_irq_save(flags);
2516 #ifdef CONFIG_FTRACE
2517 ftrace_save = ftrace_enabled;
2521 for_each_tracing_cpu(cpu) {
2522 data = iter->tr->data[cpu];
2524 if (!head_page(data) || !data->trace_idx)
2527 atomic_inc(&data->disabled);
2531 for_each_cpu_mask(cpu, mask) {
2532 data = iter->tr->data[cpu];
2533 __raw_spin_lock(&data->lock);
2535 if (data->overrun > iter->last_overrun[cpu])
2536 iter->overrun[cpu] +=
2537 data->overrun - iter->last_overrun[cpu];
2538 iter->last_overrun[cpu] = data->overrun;
2541 while (find_next_entry_inc(iter) != NULL) {
2543 int len = iter->seq.len;
2545 ret = print_trace_line(iter);
2547 /* don't print partial lines */
2548 iter->seq.len = len;
2552 trace_consume(iter);
2554 if (iter->seq.len >= cnt)
2558 for_each_cpu_mask(cpu, mask) {
2559 data = iter->tr->data[cpu];
2560 __raw_spin_unlock(&data->lock);
2563 for_each_cpu_mask(cpu, mask) {
2564 data = iter->tr->data[cpu];
2565 atomic_dec(&data->disabled);
2567 #ifdef CONFIG_FTRACE
2568 ftrace_enabled = ftrace_save;
2570 local_irq_restore(flags);
2572 /* Now copy what we have to the user */
2573 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2574 if (iter->seq.readpos >= iter->seq.len)
2575 trace_seq_reset(&iter->seq);
2580 mutex_unlock(&trace_types_lock);
2586 tracing_entries_read(struct file *filp, char __user *ubuf,
2587 size_t cnt, loff_t *ppos)
2589 struct trace_array *tr = filp->private_data;
2593 r = sprintf(buf, "%lu\n", tr->entries);
2594 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2598 tracing_entries_write(struct file *filp, const char __user *ubuf,
2599 size_t cnt, loff_t *ppos)
2605 if (cnt >= sizeof(buf))
2608 if (copy_from_user(&buf, ubuf, cnt))
2613 ret = strict_strtoul(buf, 10, &val);
2617 /* must have at least 1 entry */
2621 mutex_lock(&trace_types_lock);
2623 if (current_trace != &no_tracer) {
2625 pr_info("ftrace: set current_tracer to none"
2626 " before modifying buffer size\n");
2630 if (val > global_trace.entries) {
2631 long pages_requested;
2632 unsigned long freeable_pages;
2634 /* make sure we have enough memory before mapping */
2636 (val + (ENTRIES_PER_PAGE-1)) / ENTRIES_PER_PAGE;
2638 /* account for each buffer (and max_tr) */
2639 pages_requested *= tracing_nr_buffers * 2;
2641 /* Check for overflow */
2642 if (pages_requested < 0) {
2647 freeable_pages = determine_dirtyable_memory();
2649 /* we only allow to request 1/4 of useable memory */
2650 if (pages_requested >
2651 ((freeable_pages + tracing_pages_allocated) / 4)) {
2656 while (global_trace.entries < val) {
2657 if (trace_alloc_page()) {
2661 /* double check that we don't go over the known pages */
2662 if (tracing_pages_allocated > pages_requested)
2667 /* include the number of entries in val (inc of page entries) */
2668 while (global_trace.entries > val + (ENTRIES_PER_PAGE - 1))
2675 max_tr.entries = global_trace.entries;
2676 mutex_unlock(&trace_types_lock);
2681 static struct file_operations tracing_max_lat_fops = {
2682 .open = tracing_open_generic,
2683 .read = tracing_max_lat_read,
2684 .write = tracing_max_lat_write,
2687 static struct file_operations tracing_ctrl_fops = {
2688 .open = tracing_open_generic,
2689 .read = tracing_ctrl_read,
2690 .write = tracing_ctrl_write,
2693 static struct file_operations set_tracer_fops = {
2694 .open = tracing_open_generic,
2695 .read = tracing_set_trace_read,
2696 .write = tracing_set_trace_write,
2699 static struct file_operations tracing_pipe_fops = {
2700 .open = tracing_open_pipe,
2701 .poll = tracing_poll_pipe,
2702 .read = tracing_read_pipe,
2703 .release = tracing_release_pipe,
2706 static struct file_operations tracing_entries_fops = {
2707 .open = tracing_open_generic,
2708 .read = tracing_entries_read,
2709 .write = tracing_entries_write,
2712 #ifdef CONFIG_DYNAMIC_FTRACE
2715 tracing_read_long(struct file *filp, char __user *ubuf,
2716 size_t cnt, loff_t *ppos)
2718 unsigned long *p = filp->private_data;
2722 r = sprintf(buf, "%ld\n", *p);
2724 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2727 static struct file_operations tracing_read_long_fops = {
2728 .open = tracing_open_generic,
2729 .read = tracing_read_long,
2733 static struct dentry *d_tracer;
2735 struct dentry *tracing_init_dentry(void)
2742 d_tracer = debugfs_create_dir("tracing", NULL);
2744 if (!d_tracer && !once) {
2746 pr_warning("Could not create debugfs directory 'tracing'\n");
2753 #ifdef CONFIG_FTRACE_SELFTEST
2754 /* Let selftest have access to static functions in this file */
2755 #include "trace_selftest.c"
2758 static __init void tracer_init_debugfs(void)
2760 struct dentry *d_tracer;
2761 struct dentry *entry;
2763 d_tracer = tracing_init_dentry();
2765 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2766 &global_trace, &tracing_ctrl_fops);
2768 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2770 entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2771 NULL, &tracing_iter_fops);
2773 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2775 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2776 NULL, &tracing_cpumask_fops);
2778 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2780 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2781 &global_trace, &tracing_lt_fops);
2783 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2785 entry = debugfs_create_file("trace", 0444, d_tracer,
2786 &global_trace, &tracing_fops);
2788 pr_warning("Could not create debugfs 'trace' entry\n");
2790 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2791 &global_trace, &show_traces_fops);
2793 pr_warning("Could not create debugfs 'trace' entry\n");
2795 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2796 &global_trace, &set_tracer_fops);
2798 pr_warning("Could not create debugfs 'trace' entry\n");
2800 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2801 &tracing_max_latency,
2802 &tracing_max_lat_fops);
2804 pr_warning("Could not create debugfs "
2805 "'tracing_max_latency' entry\n");
2807 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2808 &tracing_thresh, &tracing_max_lat_fops);
2810 pr_warning("Could not create debugfs "
2811 "'tracing_threash' entry\n");
2812 entry = debugfs_create_file("README", 0644, d_tracer,
2813 NULL, &tracing_readme_fops);
2815 pr_warning("Could not create debugfs 'README' entry\n");
2817 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2818 NULL, &tracing_pipe_fops);
2820 pr_warning("Could not create debugfs "
2821 "'tracing_threash' entry\n");
2823 entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2824 &global_trace, &tracing_entries_fops);
2826 pr_warning("Could not create debugfs "
2827 "'tracing_threash' entry\n");
2829 #ifdef CONFIG_DYNAMIC_FTRACE
2830 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2831 &ftrace_update_tot_cnt,
2832 &tracing_read_long_fops);
2834 pr_warning("Could not create debugfs "
2835 "'dyn_ftrace_total_info' entry\n");
2839 static int trace_alloc_page(void)
2841 struct trace_array_cpu *data;
2842 struct page *page, *tmp;
2845 unsigned pages_allocated = 0;
2848 /* first allocate a page for each CPU */
2849 for_each_tracing_cpu(i) {
2850 array = (void *)__get_free_page(GFP_KERNEL);
2851 if (array == NULL) {
2852 printk(KERN_ERR "tracer: failed to allocate page"
2853 "for trace buffer!\n");
2858 page = virt_to_page(array);
2859 list_add(&page->lru, &pages);
2861 /* Only allocate if we are actually using the max trace */
2862 #ifdef CONFIG_TRACER_MAX_TRACE
2863 array = (void *)__get_free_page(GFP_KERNEL);
2864 if (array == NULL) {
2865 printk(KERN_ERR "tracer: failed to allocate page"
2866 "for trace buffer!\n");
2870 page = virt_to_page(array);
2871 list_add(&page->lru, &pages);
2875 /* Now that we successfully allocate a page per CPU, add them */
2876 for_each_tracing_cpu(i) {
2877 data = global_trace.data[i];
2878 page = list_entry(pages.next, struct page, lru);
2879 list_del_init(&page->lru);
2880 list_add_tail(&page->lru, &data->trace_pages);
2883 #ifdef CONFIG_TRACER_MAX_TRACE
2884 data = max_tr.data[i];
2885 page = list_entry(pages.next, struct page, lru);
2886 list_del_init(&page->lru);
2887 list_add_tail(&page->lru, &data->trace_pages);
2891 tracing_pages_allocated += pages_allocated;
2892 global_trace.entries += ENTRIES_PER_PAGE;
2897 list_for_each_entry_safe(page, tmp, &pages, lru) {
2898 list_del_init(&page->lru);
2904 static int trace_free_page(void)
2906 struct trace_array_cpu *data;
2908 struct list_head *p;
2912 /* free one page from each buffer */
2913 for_each_tracing_cpu(i) {
2914 data = global_trace.data[i];
2915 p = data->trace_pages.next;
2916 if (p == &data->trace_pages) {
2917 /* should never happen */
2919 tracing_disabled = 1;
2923 page = list_entry(p, struct page, lru);
2925 list_del(&page->lru);
2926 tracing_pages_allocated--;
2927 tracing_pages_allocated--;
2930 tracing_reset(data);
2932 #ifdef CONFIG_TRACER_MAX_TRACE
2933 data = max_tr.data[i];
2934 p = data->trace_pages.next;
2935 if (p == &data->trace_pages) {
2936 /* should never happen */
2938 tracing_disabled = 1;
2942 page = list_entry(p, struct page, lru);
2944 list_del(&page->lru);
2947 tracing_reset(data);
2950 global_trace.entries -= ENTRIES_PER_PAGE;
2955 __init static int tracer_alloc_buffers(void)
2957 struct trace_array_cpu *data;
2964 global_trace.ctrl = tracer_enabled;
2966 /* TODO: make the number of buffers hot pluggable with CPUS */
2967 tracing_nr_buffers = num_possible_cpus();
2968 tracing_buffer_mask = cpu_possible_map;
2970 /* Allocate the first page for all buffers */
2971 for_each_tracing_cpu(i) {
2972 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2973 max_tr.data[i] = &per_cpu(max_data, i);
2975 array = (void *)__get_free_page(GFP_KERNEL);
2976 if (array == NULL) {
2977 printk(KERN_ERR "tracer: failed to allocate page"
2978 "for trace buffer!\n");
2982 /* set the array to the list */
2983 INIT_LIST_HEAD(&data->trace_pages);
2984 page = virt_to_page(array);
2985 list_add(&page->lru, &data->trace_pages);
2986 /* use the LRU flag to differentiate the two buffers */
2989 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2990 max_tr.data[i]->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2992 /* Only allocate if we are actually using the max trace */
2993 #ifdef CONFIG_TRACER_MAX_TRACE
2994 array = (void *)__get_free_page(GFP_KERNEL);
2995 if (array == NULL) {
2996 printk(KERN_ERR "tracer: failed to allocate page"
2997 "for trace buffer!\n");
3001 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
3002 page = virt_to_page(array);
3003 list_add(&page->lru, &max_tr.data[i]->trace_pages);
3009 * Since we allocate by orders of pages, we may be able to
3012 global_trace.entries = ENTRIES_PER_PAGE;
3015 while (global_trace.entries < trace_nr_entries) {
3016 if (trace_alloc_page())
3020 max_tr.entries = global_trace.entries;
3022 pr_info("tracer: %d pages allocated for %ld",
3023 pages, trace_nr_entries);
3024 pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
3025 pr_info(" actual entries %ld\n", global_trace.entries);
3027 tracer_init_debugfs();
3029 trace_init_cmdlines();
3031 register_tracer(&no_tracer);
3032 current_trace = &no_tracer;
3034 /* All seems OK, enable tracing */
3035 tracing_disabled = 0;
3040 for (i-- ; i >= 0; i--) {
3041 struct page *page, *tmp;
3042 struct trace_array_cpu *data = global_trace.data[i];
3045 list_for_each_entry_safe(page, tmp,
3046 &data->trace_pages, lru) {
3047 list_del_init(&page->lru);
3052 #ifdef CONFIG_TRACER_MAX_TRACE
3053 data = max_tr.data[i];
3055 list_for_each_entry_safe(page, tmp,
3056 &data->trace_pages, lru) {
3057 list_del_init(&page->lru);
3065 fs_initcall(tracer_alloc_buffers);