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);
252 #define CHECK_COND(cond) \
253 if (unlikely(cond)) { \
254 tracing_disabled = 1; \
260 * check_pages - integrity check of trace buffers
262 * As a safty measure we check to make sure the data pages have not
265 int check_pages(struct trace_array_cpu *data)
267 struct page *page, *tmp;
269 CHECK_COND(data->trace_pages.next->prev != &data->trace_pages);
270 CHECK_COND(data->trace_pages.prev->next != &data->trace_pages);
272 list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
273 CHECK_COND(page->lru.next->prev != &page->lru);
274 CHECK_COND(page->lru.prev->next != &page->lru);
281 * head_page - page address of the first page in per_cpu buffer.
283 * head_page returns the page address of the first page in
284 * a per_cpu buffer. This also preforms various consistency
285 * checks to make sure the buffer has not been corrupted.
287 void *head_page(struct trace_array_cpu *data)
291 if (list_empty(&data->trace_pages))
294 page = list_entry(data->trace_pages.next, struct page, lru);
295 BUG_ON(&page->lru == &data->trace_pages);
297 return page_address(page);
301 * trace_seq_printf - sequence printing of trace information
302 * @s: trace sequence descriptor
303 * @fmt: printf format string
305 * The tracer may use either sequence operations or its own
306 * copy to user routines. To simplify formating of a trace
307 * trace_seq_printf is used to store strings into a special
308 * buffer (@s). Then the output may be either used by
309 * the sequencer or pulled into another buffer.
312 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
314 int len = (PAGE_SIZE - 1) - s->len;
322 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
325 /* If we can't write it all, don't bother writing anything */
335 * trace_seq_puts - trace sequence printing of simple string
336 * @s: trace sequence descriptor
337 * @str: simple string to record
339 * The tracer may use either the sequence operations or its own
340 * copy to user routines. This function records a simple string
341 * into a special buffer (@s) for later retrieval by a sequencer
342 * or other mechanism.
345 trace_seq_puts(struct trace_seq *s, const char *str)
347 int len = strlen(str);
349 if (len > ((PAGE_SIZE - 1) - s->len))
352 memcpy(s->buffer + s->len, str, len);
359 trace_seq_putc(struct trace_seq *s, unsigned char c)
361 if (s->len >= (PAGE_SIZE - 1))
364 s->buffer[s->len++] = c;
370 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
372 if (len > ((PAGE_SIZE - 1) - s->len))
375 memcpy(s->buffer + s->len, mem, len);
382 static const char hex2asc[] = "0123456789abcdef";
385 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
387 unsigned char hex[HEX_CHARS];
388 unsigned char *data = mem;
392 BUG_ON(len >= HEX_CHARS);
395 for (i = 0, j = 0; i < len; i++) {
397 for (i = len-1, j = 0; i >= 0; i--) {
401 hex[j++] = hex2asc[byte & 0x0f];
402 hex[j++] = hex2asc[byte >> 4];
406 return trace_seq_putmem(s, hex, j);
410 trace_seq_reset(struct trace_seq *s)
416 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
421 if (s->len <= s->readpos)
424 len = s->len - s->readpos;
427 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
436 trace_print_seq(struct seq_file *m, struct trace_seq *s)
438 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
441 seq_puts(m, s->buffer);
447 * flip the trace buffers between two trace descriptors.
448 * This usually is the buffers between the global_trace and
449 * the max_tr to record a snapshot of a current trace.
451 * The ftrace_max_lock must be held.
454 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
456 struct list_head flip_pages;
458 INIT_LIST_HEAD(&flip_pages);
460 memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
461 sizeof(struct trace_array_cpu) -
462 offsetof(struct trace_array_cpu, trace_head_idx));
466 list_splice_init(&tr1->trace_pages, &flip_pages);
467 list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
468 list_splice_init(&flip_pages, &tr2->trace_pages);
469 BUG_ON(!list_empty(&flip_pages));
475 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
477 * @tsk: the task with the latency
478 * @cpu: The cpu that initiated the trace.
480 * Flip the buffers between the @tr and the max_tr and record information
481 * about which task was the cause of this latency.
484 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
486 struct trace_array_cpu *data;
489 WARN_ON_ONCE(!irqs_disabled());
490 __raw_spin_lock(&ftrace_max_lock);
491 /* clear out all the previous traces */
492 for_each_tracing_cpu(i) {
494 flip_trace(max_tr.data[i], data);
498 __update_max_tr(tr, tsk, cpu);
499 __raw_spin_unlock(&ftrace_max_lock);
503 * update_max_tr_single - only copy one trace over, and reset the rest
505 * @tsk - task with the latency
506 * @cpu - the cpu of the buffer to copy.
508 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
511 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
513 struct trace_array_cpu *data = tr->data[cpu];
516 WARN_ON_ONCE(!irqs_disabled());
517 __raw_spin_lock(&ftrace_max_lock);
518 for_each_tracing_cpu(i)
519 tracing_reset(max_tr.data[i]);
521 flip_trace(max_tr.data[cpu], data);
524 __update_max_tr(tr, tsk, cpu);
525 __raw_spin_unlock(&ftrace_max_lock);
529 * register_tracer - register a tracer with the ftrace system.
530 * @type - the plugin for the tracer
532 * Register a new plugin tracer.
534 int register_tracer(struct tracer *type)
541 pr_info("Tracer must have a name\n");
545 mutex_lock(&trace_types_lock);
546 for (t = trace_types; t; t = t->next) {
547 if (strcmp(type->name, t->name) == 0) {
549 pr_info("Trace %s already registered\n",
556 #ifdef CONFIG_FTRACE_STARTUP_TEST
557 if (type->selftest) {
558 struct tracer *saved_tracer = current_trace;
559 struct trace_array_cpu *data;
560 struct trace_array *tr = &global_trace;
561 int saved_ctrl = tr->ctrl;
564 * Run a selftest on this tracer.
565 * Here we reset the trace buffer, and set the current
566 * tracer to be this tracer. The tracer can then run some
567 * internal tracing to verify that everything is in order.
568 * If we fail, we do not register this tracer.
570 for_each_tracing_cpu(i) {
572 if (!head_page(data))
576 current_trace = type;
578 /* the test is responsible for initializing and enabling */
579 pr_info("Testing tracer %s: ", type->name);
580 ret = type->selftest(type, tr);
581 /* the test is responsible for resetting too */
582 current_trace = saved_tracer;
583 tr->ctrl = saved_ctrl;
585 printk(KERN_CONT "FAILED!\n");
588 /* Only reset on passing, to avoid touching corrupted buffers */
589 for_each_tracing_cpu(i) {
591 if (!head_page(data))
595 printk(KERN_CONT "PASSED\n");
599 type->next = trace_types;
601 len = strlen(type->name);
602 if (len > max_tracer_type_len)
603 max_tracer_type_len = len;
606 mutex_unlock(&trace_types_lock);
611 void unregister_tracer(struct tracer *type)
616 mutex_lock(&trace_types_lock);
617 for (t = &trace_types; *t; t = &(*t)->next) {
621 pr_info("Trace %s not registered\n", type->name);
626 if (strlen(type->name) != max_tracer_type_len)
629 max_tracer_type_len = 0;
630 for (t = &trace_types; *t; t = &(*t)->next) {
631 len = strlen((*t)->name);
632 if (len > max_tracer_type_len)
633 max_tracer_type_len = len;
636 mutex_unlock(&trace_types_lock);
639 void tracing_reset(struct trace_array_cpu *data)
643 data->trace_head = data->trace_tail = head_page(data);
644 data->trace_head_idx = 0;
645 data->trace_tail_idx = 0;
648 #define SAVED_CMDLINES 128
649 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
650 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
651 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
652 static int cmdline_idx;
653 static DEFINE_SPINLOCK(trace_cmdline_lock);
655 /* temporary disable recording */
656 atomic_t trace_record_cmdline_disabled __read_mostly;
658 static void trace_init_cmdlines(void)
660 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
661 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
665 void trace_stop_cmdline_recording(void);
667 static void trace_save_cmdline(struct task_struct *tsk)
672 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
676 * It's not the end of the world if we don't get
677 * the lock, but we also don't want to spin
678 * nor do we want to disable interrupts,
679 * so if we miss here, then better luck next time.
681 if (!spin_trylock(&trace_cmdline_lock))
684 idx = map_pid_to_cmdline[tsk->pid];
685 if (idx >= SAVED_CMDLINES) {
686 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
688 map = map_cmdline_to_pid[idx];
689 if (map <= PID_MAX_DEFAULT)
690 map_pid_to_cmdline[map] = (unsigned)-1;
692 map_pid_to_cmdline[tsk->pid] = idx;
697 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
699 spin_unlock(&trace_cmdline_lock);
702 static char *trace_find_cmdline(int pid)
704 char *cmdline = "<...>";
710 if (pid > PID_MAX_DEFAULT)
713 map = map_pid_to_cmdline[pid];
714 if (map >= SAVED_CMDLINES)
717 cmdline = saved_cmdlines[map];
723 void tracing_record_cmdline(struct task_struct *tsk)
725 if (atomic_read(&trace_record_cmdline_disabled))
728 trace_save_cmdline(tsk);
731 static inline struct list_head *
732 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
735 * Roundrobin - but skip the head (which is not a real page):
738 if (unlikely(next == &data->trace_pages))
740 BUG_ON(next == &data->trace_pages);
746 trace_next_page(struct trace_array_cpu *data, void *addr)
748 struct list_head *next;
751 page = virt_to_page(addr);
753 next = trace_next_list(data, &page->lru);
754 page = list_entry(next, struct page, lru);
756 return page_address(page);
759 static inline struct trace_entry *
760 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
762 unsigned long idx, idx_next;
763 struct trace_entry *entry;
766 idx = data->trace_head_idx;
769 BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
771 entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
773 if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
774 data->trace_head = trace_next_page(data, data->trace_head);
778 if (data->trace_head == data->trace_tail &&
779 idx_next == data->trace_tail_idx) {
782 data->trace_tail_idx++;
783 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
785 trace_next_page(data, data->trace_tail);
786 data->trace_tail_idx = 0;
790 data->trace_head_idx = idx_next;
796 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
798 struct task_struct *tsk = current;
801 pc = preempt_count();
803 entry->preempt_count = pc & 0xff;
804 entry->pid = (tsk) ? tsk->pid : 0;
805 entry->t = ftrace_now(raw_smp_processor_id());
806 entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
807 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
808 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
809 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
813 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
814 unsigned long ip, unsigned long parent_ip, unsigned long flags)
816 struct trace_entry *entry;
817 unsigned long irq_flags;
819 raw_local_irq_save(irq_flags);
820 __raw_spin_lock(&data->lock);
821 entry = tracing_get_trace_entry(tr, data);
822 tracing_generic_entry_update(entry, flags);
823 entry->type = TRACE_FN;
825 entry->fn.parent_ip = parent_ip;
826 __raw_spin_unlock(&data->lock);
827 raw_local_irq_restore(irq_flags);
831 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
832 unsigned long ip, unsigned long parent_ip, unsigned long flags)
834 if (likely(!atomic_read(&data->disabled)))
835 trace_function(tr, data, ip, parent_ip, flags);
838 void __trace_stack(struct trace_array *tr,
839 struct trace_array_cpu *data,
843 struct trace_entry *entry;
844 struct stack_trace trace;
846 if (!(trace_flags & TRACE_ITER_STACKTRACE))
849 entry = tracing_get_trace_entry(tr, data);
850 tracing_generic_entry_update(entry, flags);
851 entry->type = TRACE_STACK;
853 memset(&entry->stack, 0, sizeof(entry->stack));
855 trace.nr_entries = 0;
856 trace.max_entries = FTRACE_STACK_ENTRIES;
858 trace.entries = entry->stack.caller;
860 save_stack_trace(&trace);
864 __trace_special(void *__tr, void *__data,
865 unsigned long arg1, unsigned long arg2, unsigned long arg3)
867 struct trace_array_cpu *data = __data;
868 struct trace_array *tr = __tr;
869 struct trace_entry *entry;
870 unsigned long irq_flags;
872 raw_local_irq_save(irq_flags);
873 __raw_spin_lock(&data->lock);
874 entry = tracing_get_trace_entry(tr, data);
875 tracing_generic_entry_update(entry, 0);
876 entry->type = TRACE_SPECIAL;
877 entry->special.arg1 = arg1;
878 entry->special.arg2 = arg2;
879 entry->special.arg3 = arg3;
880 __trace_stack(tr, data, irq_flags, 4);
881 __raw_spin_unlock(&data->lock);
882 raw_local_irq_restore(irq_flags);
888 tracing_sched_switch_trace(struct trace_array *tr,
889 struct trace_array_cpu *data,
890 struct task_struct *prev,
891 struct task_struct *next,
894 struct trace_entry *entry;
895 unsigned long irq_flags;
897 raw_local_irq_save(irq_flags);
898 __raw_spin_lock(&data->lock);
899 entry = tracing_get_trace_entry(tr, data);
900 tracing_generic_entry_update(entry, flags);
901 entry->type = TRACE_CTX;
902 entry->ctx.prev_pid = prev->pid;
903 entry->ctx.prev_prio = prev->prio;
904 entry->ctx.prev_state = prev->state;
905 entry->ctx.next_pid = next->pid;
906 entry->ctx.next_prio = next->prio;
907 entry->ctx.next_state = next->state;
908 __trace_stack(tr, data, flags, 5);
909 __raw_spin_unlock(&data->lock);
910 raw_local_irq_restore(irq_flags);
914 tracing_sched_wakeup_trace(struct trace_array *tr,
915 struct trace_array_cpu *data,
916 struct task_struct *wakee,
917 struct task_struct *curr,
920 struct trace_entry *entry;
921 unsigned long irq_flags;
923 raw_local_irq_save(irq_flags);
924 __raw_spin_lock(&data->lock);
925 entry = tracing_get_trace_entry(tr, data);
926 tracing_generic_entry_update(entry, flags);
927 entry->type = TRACE_WAKE;
928 entry->ctx.prev_pid = curr->pid;
929 entry->ctx.prev_prio = curr->prio;
930 entry->ctx.prev_state = curr->state;
931 entry->ctx.next_pid = wakee->pid;
932 entry->ctx.next_prio = wakee->prio;
933 entry->ctx.next_state = wakee->state;
934 __trace_stack(tr, data, flags, 6);
935 __raw_spin_unlock(&data->lock);
936 raw_local_irq_restore(irq_flags);
942 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
944 struct trace_array *tr = &global_trace;
945 struct trace_array_cpu *data;
950 if (tracing_disabled || current_trace == &no_tracer || !tr->ctrl)
953 local_irq_save(flags);
954 cpu = raw_smp_processor_id();
955 data = tr->data[cpu];
956 disabled = atomic_inc_return(&data->disabled);
958 if (likely(disabled == 1))
959 __trace_special(tr, data, arg1, arg2, arg3);
961 atomic_dec(&data->disabled);
962 local_irq_restore(flags);
967 function_trace_call(unsigned long ip, unsigned long parent_ip)
969 struct trace_array *tr = &global_trace;
970 struct trace_array_cpu *data;
975 if (unlikely(!tracer_enabled))
978 local_irq_save(flags);
979 cpu = raw_smp_processor_id();
980 data = tr->data[cpu];
981 disabled = atomic_inc_return(&data->disabled);
983 if (likely(disabled == 1))
984 trace_function(tr, data, ip, parent_ip, flags);
986 atomic_dec(&data->disabled);
987 local_irq_restore(flags);
990 static struct ftrace_ops trace_ops __read_mostly =
992 .func = function_trace_call,
995 void tracing_start_function_trace(void)
997 register_ftrace_function(&trace_ops);
1000 void tracing_stop_function_trace(void)
1002 unregister_ftrace_function(&trace_ops);
1006 enum trace_file_type {
1007 TRACE_FILE_LAT_FMT = 1,
1010 static struct trace_entry *
1011 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
1012 struct trace_iterator *iter, int cpu)
1015 struct trace_entry *array;
1017 if (iter->next_idx[cpu] >= tr->entries ||
1018 iter->next_idx[cpu] >= data->trace_idx ||
1019 (data->trace_head == data->trace_tail &&
1020 data->trace_head_idx == data->trace_tail_idx))
1023 if (!iter->next_page[cpu]) {
1024 /* Initialize the iterator for this cpu trace buffer */
1025 WARN_ON(!data->trace_tail);
1026 page = virt_to_page(data->trace_tail);
1027 iter->next_page[cpu] = &page->lru;
1028 iter->next_page_idx[cpu] = data->trace_tail_idx;
1031 page = list_entry(iter->next_page[cpu], struct page, lru);
1032 BUG_ON(&data->trace_pages == &page->lru);
1034 array = page_address(page);
1036 WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
1037 return &array[iter->next_page_idx[cpu]];
1040 static struct trace_entry *
1041 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
1043 struct trace_array *tr = iter->tr;
1044 struct trace_entry *ent, *next = NULL;
1048 for_each_tracing_cpu(cpu) {
1049 if (!head_page(tr->data[cpu]))
1051 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
1053 * Pick the entry with the smallest timestamp:
1055 if (ent && (!next || ent->t < next->t)) {
1062 *ent_cpu = next_cpu;
1067 static void trace_iterator_increment(struct trace_iterator *iter)
1070 iter->next_idx[iter->cpu]++;
1071 iter->next_page_idx[iter->cpu]++;
1073 if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
1074 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1076 iter->next_page_idx[iter->cpu] = 0;
1077 iter->next_page[iter->cpu] =
1078 trace_next_list(data, iter->next_page[iter->cpu]);
1082 static void trace_consume(struct trace_iterator *iter)
1084 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1086 data->trace_tail_idx++;
1087 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
1088 data->trace_tail = trace_next_page(data, data->trace_tail);
1089 data->trace_tail_idx = 0;
1092 /* Check if we empty it, then reset the index */
1093 if (data->trace_head == data->trace_tail &&
1094 data->trace_head_idx == data->trace_tail_idx)
1095 data->trace_idx = 0;
1098 static void *find_next_entry_inc(struct trace_iterator *iter)
1100 struct trace_entry *next;
1103 next = find_next_entry(iter, &next_cpu);
1105 iter->prev_ent = iter->ent;
1106 iter->prev_cpu = iter->cpu;
1109 iter->cpu = next_cpu;
1112 trace_iterator_increment(iter);
1114 return next ? iter : NULL;
1117 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1119 struct trace_iterator *iter = m->private;
1120 void *last_ent = iter->ent;
1126 /* can't go backwards */
1131 ent = find_next_entry_inc(iter);
1135 while (ent && iter->idx < i)
1136 ent = find_next_entry_inc(iter);
1140 if (last_ent && !ent)
1141 seq_puts(m, "\n\nvim:ft=help\n");
1146 static void *s_start(struct seq_file *m, loff_t *pos)
1148 struct trace_iterator *iter = m->private;
1153 mutex_lock(&trace_types_lock);
1155 if (!current_trace || current_trace != iter->trace) {
1156 mutex_unlock(&trace_types_lock);
1160 atomic_inc(&trace_record_cmdline_disabled);
1162 /* let the tracer grab locks here if needed */
1163 if (current_trace->start)
1164 current_trace->start(iter);
1166 if (*pos != iter->pos) {
1170 iter->prev_ent = NULL;
1171 iter->prev_cpu = -1;
1173 for_each_tracing_cpu(i) {
1174 iter->next_idx[i] = 0;
1175 iter->next_page[i] = NULL;
1178 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1183 p = s_next(m, p, &l);
1189 static void s_stop(struct seq_file *m, void *p)
1191 struct trace_iterator *iter = m->private;
1193 atomic_dec(&trace_record_cmdline_disabled);
1195 /* let the tracer release locks here if needed */
1196 if (current_trace && current_trace == iter->trace && iter->trace->stop)
1197 iter->trace->stop(iter);
1199 mutex_unlock(&trace_types_lock);
1203 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1205 #ifdef CONFIG_KALLSYMS
1206 char str[KSYM_SYMBOL_LEN];
1208 kallsyms_lookup(address, NULL, NULL, NULL, str);
1210 return trace_seq_printf(s, fmt, str);
1216 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1217 unsigned long address)
1219 #ifdef CONFIG_KALLSYMS
1220 char str[KSYM_SYMBOL_LEN];
1222 sprint_symbol(str, address);
1223 return trace_seq_printf(s, fmt, str);
1228 #ifndef CONFIG_64BIT
1229 # define IP_FMT "%08lx"
1231 # define IP_FMT "%016lx"
1235 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1240 return trace_seq_printf(s, "0");
1242 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1243 ret = seq_print_sym_offset(s, "%s", ip);
1245 ret = seq_print_sym_short(s, "%s", ip);
1250 if (sym_flags & TRACE_ITER_SYM_ADDR)
1251 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1255 static void print_lat_help_header(struct seq_file *m)
1257 seq_puts(m, "# _------=> CPU# \n");
1258 seq_puts(m, "# / _-----=> irqs-off \n");
1259 seq_puts(m, "# | / _----=> need-resched \n");
1260 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1261 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1262 seq_puts(m, "# |||| / \n");
1263 seq_puts(m, "# ||||| delay \n");
1264 seq_puts(m, "# cmd pid ||||| time | caller \n");
1265 seq_puts(m, "# \\ / ||||| \\ | / \n");
1268 static void print_func_help_header(struct seq_file *m)
1270 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1271 seq_puts(m, "# | | | | |\n");
1276 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1278 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1279 struct trace_array *tr = iter->tr;
1280 struct trace_array_cpu *data = tr->data[tr->cpu];
1281 struct tracer *type = current_trace;
1282 unsigned long total = 0;
1283 unsigned long entries = 0;
1285 const char *name = "preemption";
1290 for_each_tracing_cpu(cpu) {
1291 if (head_page(tr->data[cpu])) {
1292 total += tr->data[cpu]->trace_idx;
1293 if (tr->data[cpu]->trace_idx > tr->entries)
1294 entries += tr->entries;
1296 entries += tr->data[cpu]->trace_idx;
1300 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1302 seq_puts(m, "-----------------------------------"
1303 "---------------------------------\n");
1304 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1305 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1306 nsecs_to_usecs(data->saved_latency),
1310 #if defined(CONFIG_PREEMPT_NONE)
1312 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1314 #elif defined(CONFIG_PREEMPT_DESKTOP)
1319 /* These are reserved for later use */
1322 seq_printf(m, " #P:%d)\n", num_online_cpus());
1326 seq_puts(m, " -----------------\n");
1327 seq_printf(m, " | task: %.16s-%d "
1328 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1329 data->comm, data->pid, data->uid, data->nice,
1330 data->policy, data->rt_priority);
1331 seq_puts(m, " -----------------\n");
1333 if (data->critical_start) {
1334 seq_puts(m, " => started at: ");
1335 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1336 trace_print_seq(m, &iter->seq);
1337 seq_puts(m, "\n => ended at: ");
1338 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1339 trace_print_seq(m, &iter->seq);
1347 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1349 int hardirq, softirq;
1352 comm = trace_find_cmdline(entry->pid);
1354 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1355 trace_seq_printf(s, "%d", cpu);
1356 trace_seq_printf(s, "%c%c",
1357 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1358 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1360 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1361 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1362 if (hardirq && softirq) {
1363 trace_seq_putc(s, 'H');
1366 trace_seq_putc(s, 'h');
1369 trace_seq_putc(s, 's');
1371 trace_seq_putc(s, '.');
1375 if (entry->preempt_count)
1376 trace_seq_printf(s, "%x", entry->preempt_count);
1378 trace_seq_puts(s, ".");
1381 unsigned long preempt_mark_thresh = 100;
1384 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1385 unsigned long rel_usecs)
1387 trace_seq_printf(s, " %4lldus", abs_usecs);
1388 if (rel_usecs > preempt_mark_thresh)
1389 trace_seq_puts(s, "!: ");
1390 else if (rel_usecs > 1)
1391 trace_seq_puts(s, "+: ");
1393 trace_seq_puts(s, " : ");
1396 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1399 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1401 struct trace_seq *s = &iter->seq;
1402 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1403 struct trace_entry *next_entry = find_next_entry(iter, NULL);
1404 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1405 struct trace_entry *entry = iter->ent;
1406 unsigned long abs_usecs;
1407 unsigned long rel_usecs;
1415 rel_usecs = ns2usecs(next_entry->t - entry->t);
1416 abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1419 comm = trace_find_cmdline(entry->pid);
1420 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1421 " %ld.%03ldms (+%ld.%03ldms): ",
1423 entry->pid, cpu, entry->flags,
1424 entry->preempt_count, trace_idx,
1427 abs_usecs % 1000, rel_usecs/1000,
1430 lat_print_generic(s, entry, cpu);
1431 lat_print_timestamp(s, abs_usecs, rel_usecs);
1433 switch (entry->type) {
1435 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1436 trace_seq_puts(s, " (");
1437 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1438 trace_seq_puts(s, ")\n");
1442 T = entry->ctx.next_state < sizeof(state_to_char) ?
1443 state_to_char[entry->ctx.next_state] : 'X';
1445 state = entry->ctx.prev_state ? __ffs(entry->ctx.prev_state) + 1 : 0;
1446 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1447 comm = trace_find_cmdline(entry->ctx.next_pid);
1448 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1449 entry->ctx.prev_pid,
1450 entry->ctx.prev_prio,
1451 S, entry->type == TRACE_CTX ? "==>" : " +",
1452 entry->ctx.next_pid,
1453 entry->ctx.next_prio,
1457 trace_seq_printf(s, "# %ld %ld %ld\n",
1458 entry->special.arg1,
1459 entry->special.arg2,
1460 entry->special.arg3);
1463 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1465 trace_seq_puts(s, " <= ");
1466 seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1468 trace_seq_puts(s, "\n");
1471 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1476 static int print_trace_fmt(struct trace_iterator *iter)
1478 struct trace_seq *s = &iter->seq;
1479 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1480 struct trace_entry *entry;
1481 unsigned long usec_rem;
1482 unsigned long long t;
1491 comm = trace_find_cmdline(iter->ent->pid);
1493 t = ns2usecs(entry->t);
1494 usec_rem = do_div(t, 1000000ULL);
1495 secs = (unsigned long)t;
1497 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1500 ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1503 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1507 switch (entry->type) {
1509 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1512 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1513 entry->fn.parent_ip) {
1514 ret = trace_seq_printf(s, " <-");
1517 ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1522 ret = trace_seq_printf(s, "\n");
1528 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1529 state_to_char[entry->ctx.prev_state] : 'X';
1530 T = entry->ctx.next_state < sizeof(state_to_char) ?
1531 state_to_char[entry->ctx.next_state] : 'X';
1532 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1533 entry->ctx.prev_pid,
1534 entry->ctx.prev_prio,
1536 entry->type == TRACE_CTX ? "==>" : " +",
1537 entry->ctx.next_pid,
1538 entry->ctx.next_prio,
1544 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1545 entry->special.arg1,
1546 entry->special.arg2,
1547 entry->special.arg3);
1552 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1554 ret = trace_seq_puts(s, " <= ");
1558 ret = seq_print_ip_sym(s, entry->stack.caller[i],
1563 ret = trace_seq_puts(s, "\n");
1571 static int print_raw_fmt(struct trace_iterator *iter)
1573 struct trace_seq *s = &iter->seq;
1574 struct trace_entry *entry;
1580 ret = trace_seq_printf(s, "%d %d %llu ",
1581 entry->pid, iter->cpu, entry->t);
1585 switch (entry->type) {
1587 ret = trace_seq_printf(s, "%x %x\n",
1588 entry->fn.ip, entry->fn.parent_ip);
1594 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1595 state_to_char[entry->ctx.prev_state] : 'X';
1596 T = entry->ctx.next_state < sizeof(state_to_char) ?
1597 state_to_char[entry->ctx.next_state] : 'X';
1598 if (entry->type == TRACE_WAKE)
1600 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1601 entry->ctx.prev_pid,
1602 entry->ctx.prev_prio,
1604 entry->ctx.next_pid,
1605 entry->ctx.next_prio,
1612 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1613 entry->special.arg1,
1614 entry->special.arg2,
1615 entry->special.arg3);
1623 #define SEQ_PUT_FIELD_RET(s, x) \
1625 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1629 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1631 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1635 static int print_hex_fmt(struct trace_iterator *iter)
1637 struct trace_seq *s = &iter->seq;
1638 unsigned char newline = '\n';
1639 struct trace_entry *entry;
1644 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1645 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1646 SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1648 switch (entry->type) {
1650 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1651 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1655 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1656 state_to_char[entry->ctx.prev_state] : 'X';
1657 T = entry->ctx.next_state < sizeof(state_to_char) ?
1658 state_to_char[entry->ctx.next_state] : 'X';
1659 if (entry->type == TRACE_WAKE)
1661 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1662 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1663 SEQ_PUT_HEX_FIELD_RET(s, S);
1664 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1665 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1666 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1667 SEQ_PUT_HEX_FIELD_RET(s, T);
1671 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1672 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1673 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1676 SEQ_PUT_FIELD_RET(s, newline);
1681 static int print_bin_fmt(struct trace_iterator *iter)
1683 struct trace_seq *s = &iter->seq;
1684 struct trace_entry *entry;
1688 SEQ_PUT_FIELD_RET(s, entry->pid);
1689 SEQ_PUT_FIELD_RET(s, entry->cpu);
1690 SEQ_PUT_FIELD_RET(s, entry->t);
1692 switch (entry->type) {
1694 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1695 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1698 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1699 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1700 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1701 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1702 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1703 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1707 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1708 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1709 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1715 static int trace_empty(struct trace_iterator *iter)
1717 struct trace_array_cpu *data;
1720 for_each_tracing_cpu(cpu) {
1721 data = iter->tr->data[cpu];
1723 if (head_page(data) && data->trace_idx &&
1724 (data->trace_tail != data->trace_head ||
1725 data->trace_tail_idx != data->trace_head_idx))
1731 static int print_trace_line(struct trace_iterator *iter)
1733 if (iter->trace && iter->trace->print_line)
1734 return iter->trace->print_line(iter);
1736 if (trace_flags & TRACE_ITER_BIN)
1737 return print_bin_fmt(iter);
1739 if (trace_flags & TRACE_ITER_HEX)
1740 return print_hex_fmt(iter);
1742 if (trace_flags & TRACE_ITER_RAW)
1743 return print_raw_fmt(iter);
1745 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1746 return print_lat_fmt(iter, iter->idx, iter->cpu);
1748 return print_trace_fmt(iter);
1751 static int s_show(struct seq_file *m, void *v)
1753 struct trace_iterator *iter = v;
1755 if (iter->ent == NULL) {
1757 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1760 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1761 /* print nothing if the buffers are empty */
1762 if (trace_empty(iter))
1764 print_trace_header(m, iter);
1765 if (!(trace_flags & TRACE_ITER_VERBOSE))
1766 print_lat_help_header(m);
1768 if (!(trace_flags & TRACE_ITER_VERBOSE))
1769 print_func_help_header(m);
1772 print_trace_line(iter);
1773 trace_print_seq(m, &iter->seq);
1779 static struct seq_operations tracer_seq_ops = {
1786 static struct trace_iterator *
1787 __tracing_open(struct inode *inode, struct file *file, int *ret)
1789 struct trace_iterator *iter;
1791 if (tracing_disabled) {
1796 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1802 mutex_lock(&trace_types_lock);
1803 if (current_trace && current_trace->print_max)
1806 iter->tr = inode->i_private;
1807 iter->trace = current_trace;
1810 /* TODO stop tracer */
1811 *ret = seq_open(file, &tracer_seq_ops);
1813 struct seq_file *m = file->private_data;
1816 /* stop the trace while dumping */
1820 if (iter->trace && iter->trace->open)
1821 iter->trace->open(iter);
1826 mutex_unlock(&trace_types_lock);
1832 int tracing_open_generic(struct inode *inode, struct file *filp)
1834 if (tracing_disabled)
1837 filp->private_data = inode->i_private;
1841 int tracing_release(struct inode *inode, struct file *file)
1843 struct seq_file *m = (struct seq_file *)file->private_data;
1844 struct trace_iterator *iter = m->private;
1846 mutex_lock(&trace_types_lock);
1847 if (iter->trace && iter->trace->close)
1848 iter->trace->close(iter);
1850 /* reenable tracing if it was previously enabled */
1853 mutex_unlock(&trace_types_lock);
1855 seq_release(inode, file);
1860 static int tracing_open(struct inode *inode, struct file *file)
1864 __tracing_open(inode, file, &ret);
1869 static int tracing_lt_open(struct inode *inode, struct file *file)
1871 struct trace_iterator *iter;
1874 iter = __tracing_open(inode, file, &ret);
1877 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1884 t_next(struct seq_file *m, void *v, loff_t *pos)
1886 struct tracer *t = m->private;
1898 static void *t_start(struct seq_file *m, loff_t *pos)
1900 struct tracer *t = m->private;
1903 mutex_lock(&trace_types_lock);
1904 for (; t && l < *pos; t = t_next(m, t, &l))
1910 static void t_stop(struct seq_file *m, void *p)
1912 mutex_unlock(&trace_types_lock);
1915 static int t_show(struct seq_file *m, void *v)
1917 struct tracer *t = v;
1922 seq_printf(m, "%s", t->name);
1931 static struct seq_operations show_traces_seq_ops = {
1938 static int show_traces_open(struct inode *inode, struct file *file)
1942 if (tracing_disabled)
1945 ret = seq_open(file, &show_traces_seq_ops);
1947 struct seq_file *m = file->private_data;
1948 m->private = trace_types;
1954 static struct file_operations tracing_fops = {
1955 .open = tracing_open,
1957 .llseek = seq_lseek,
1958 .release = tracing_release,
1961 static struct file_operations tracing_lt_fops = {
1962 .open = tracing_lt_open,
1964 .llseek = seq_lseek,
1965 .release = tracing_release,
1968 static struct file_operations show_traces_fops = {
1969 .open = show_traces_open,
1971 .release = seq_release,
1975 * Only trace on a CPU if the bitmask is set:
1977 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
1980 * When tracing/tracing_cpu_mask is modified then this holds
1981 * the new bitmask we are about to install:
1983 static cpumask_t tracing_cpumask_new;
1986 * The tracer itself will not take this lock, but still we want
1987 * to provide a consistent cpumask to user-space:
1989 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1992 * Temporary storage for the character representation of the
1993 * CPU bitmask (and one more byte for the newline):
1995 static char mask_str[NR_CPUS + 1];
1998 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1999 size_t count, loff_t *ppos)
2003 mutex_lock(&tracing_cpumask_update_lock);
2005 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2006 if (count - len < 2) {
2010 len += sprintf(mask_str + len, "\n");
2011 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2014 mutex_unlock(&tracing_cpumask_update_lock);
2020 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2021 size_t count, loff_t *ppos)
2025 mutex_lock(&tracing_cpumask_update_lock);
2026 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2030 raw_local_irq_disable();
2031 __raw_spin_lock(&ftrace_max_lock);
2032 for_each_tracing_cpu(cpu) {
2034 * Increase/decrease the disabled counter if we are
2035 * about to flip a bit in the cpumask:
2037 if (cpu_isset(cpu, tracing_cpumask) &&
2038 !cpu_isset(cpu, tracing_cpumask_new)) {
2039 atomic_inc(&global_trace.data[cpu]->disabled);
2041 if (!cpu_isset(cpu, tracing_cpumask) &&
2042 cpu_isset(cpu, tracing_cpumask_new)) {
2043 atomic_dec(&global_trace.data[cpu]->disabled);
2046 __raw_spin_unlock(&ftrace_max_lock);
2047 raw_local_irq_enable();
2049 tracing_cpumask = tracing_cpumask_new;
2051 mutex_unlock(&tracing_cpumask_update_lock);
2056 mutex_unlock(&tracing_cpumask_update_lock);
2061 static struct file_operations tracing_cpumask_fops = {
2062 .open = tracing_open_generic,
2063 .read = tracing_cpumask_read,
2064 .write = tracing_cpumask_write,
2068 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2069 size_t cnt, loff_t *ppos)
2076 /* calulate max size */
2077 for (i = 0; trace_options[i]; i++) {
2078 len += strlen(trace_options[i]);
2079 len += 3; /* "no" and space */
2082 /* +2 for \n and \0 */
2083 buf = kmalloc(len + 2, GFP_KERNEL);
2087 for (i = 0; trace_options[i]; i++) {
2088 if (trace_flags & (1 << i))
2089 r += sprintf(buf + r, "%s ", trace_options[i]);
2091 r += sprintf(buf + r, "no%s ", trace_options[i]);
2094 r += sprintf(buf + r, "\n");
2095 WARN_ON(r >= len + 2);
2097 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2105 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2106 size_t cnt, loff_t *ppos)
2113 if (cnt >= sizeof(buf))
2116 if (copy_from_user(&buf, ubuf, cnt))
2121 if (strncmp(buf, "no", 2) == 0) {
2126 for (i = 0; trace_options[i]; i++) {
2127 int len = strlen(trace_options[i]);
2129 if (strncmp(cmp, trace_options[i], len) == 0) {
2131 trace_flags &= ~(1 << i);
2133 trace_flags |= (1 << i);
2138 * If no option could be set, return an error:
2140 if (!trace_options[i])
2148 static struct file_operations tracing_iter_fops = {
2149 .open = tracing_open_generic,
2150 .read = tracing_iter_ctrl_read,
2151 .write = tracing_iter_ctrl_write,
2154 static const char readme_msg[] =
2155 "tracing mini-HOWTO:\n\n"
2157 "# mount -t debugfs nodev /debug\n\n"
2158 "# cat /debug/tracing/available_tracers\n"
2159 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2160 "# cat /debug/tracing/current_tracer\n"
2162 "# echo sched_switch > /debug/tracing/current_tracer\n"
2163 "# cat /debug/tracing/current_tracer\n"
2165 "# cat /debug/tracing/iter_ctrl\n"
2166 "noprint-parent nosym-offset nosym-addr noverbose\n"
2167 "# echo print-parent > /debug/tracing/iter_ctrl\n"
2168 "# echo 1 > /debug/tracing/tracing_enabled\n"
2169 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2170 "echo 0 > /debug/tracing/tracing_enabled\n"
2174 tracing_readme_read(struct file *filp, char __user *ubuf,
2175 size_t cnt, loff_t *ppos)
2177 return simple_read_from_buffer(ubuf, cnt, ppos,
2178 readme_msg, strlen(readme_msg));
2181 static struct file_operations tracing_readme_fops = {
2182 .open = tracing_open_generic,
2183 .read = tracing_readme_read,
2187 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2188 size_t cnt, loff_t *ppos)
2190 struct trace_array *tr = filp->private_data;
2194 r = sprintf(buf, "%ld\n", tr->ctrl);
2195 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2199 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2200 size_t cnt, loff_t *ppos)
2202 struct trace_array *tr = filp->private_data;
2207 if (cnt >= sizeof(buf))
2210 if (copy_from_user(&buf, ubuf, cnt))
2215 ret = strict_strtoul(buf, 10, &val);
2221 mutex_lock(&trace_types_lock);
2222 if (tr->ctrl ^ val) {
2230 if (current_trace && current_trace->ctrl_update)
2231 current_trace->ctrl_update(tr);
2233 mutex_unlock(&trace_types_lock);
2241 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2242 size_t cnt, loff_t *ppos)
2244 char buf[max_tracer_type_len+2];
2247 mutex_lock(&trace_types_lock);
2249 r = sprintf(buf, "%s\n", current_trace->name);
2251 r = sprintf(buf, "\n");
2252 mutex_unlock(&trace_types_lock);
2254 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2258 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2259 size_t cnt, loff_t *ppos)
2261 struct trace_array *tr = &global_trace;
2263 char buf[max_tracer_type_len+1];
2266 if (cnt > max_tracer_type_len)
2267 cnt = max_tracer_type_len;
2269 if (copy_from_user(&buf, ubuf, cnt))
2274 /* strip ending whitespace. */
2275 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2278 mutex_lock(&trace_types_lock);
2279 for (t = trace_types; t; t = t->next) {
2280 if (strcmp(t->name, buf) == 0)
2283 if (!t || t == current_trace)
2286 if (current_trace && current_trace->reset)
2287 current_trace->reset(tr);
2294 mutex_unlock(&trace_types_lock);
2302 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2303 size_t cnt, loff_t *ppos)
2305 unsigned long *ptr = filp->private_data;
2309 r = snprintf(buf, sizeof(buf), "%ld\n",
2310 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2311 if (r > sizeof(buf))
2313 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2317 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2318 size_t cnt, loff_t *ppos)
2320 long *ptr = filp->private_data;
2325 if (cnt >= sizeof(buf))
2328 if (copy_from_user(&buf, ubuf, cnt))
2333 ret = strict_strtoul(buf, 10, &val);
2342 static atomic_t tracing_reader;
2344 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2346 struct trace_iterator *iter;
2348 if (tracing_disabled)
2351 /* We only allow for reader of the pipe */
2352 if (atomic_inc_return(&tracing_reader) != 1) {
2353 atomic_dec(&tracing_reader);
2357 /* create a buffer to store the information to pass to userspace */
2358 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2362 mutex_lock(&trace_types_lock);
2363 iter->tr = &global_trace;
2364 iter->trace = current_trace;
2365 filp->private_data = iter;
2367 if (iter->trace->pipe_open)
2368 iter->trace->pipe_open(iter);
2369 mutex_unlock(&trace_types_lock);
2374 static int tracing_release_pipe(struct inode *inode, struct file *file)
2376 struct trace_iterator *iter = file->private_data;
2379 atomic_dec(&tracing_reader);
2385 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2387 struct trace_iterator *iter = filp->private_data;
2389 if (trace_flags & TRACE_ITER_BLOCK) {
2391 * Always select as readable when in blocking mode
2393 return POLLIN | POLLRDNORM;
2395 if (!trace_empty(iter))
2396 return POLLIN | POLLRDNORM;
2397 poll_wait(filp, &trace_wait, poll_table);
2398 if (!trace_empty(iter))
2399 return POLLIN | POLLRDNORM;
2409 tracing_read_pipe(struct file *filp, char __user *ubuf,
2410 size_t cnt, loff_t *ppos)
2412 struct trace_iterator *iter = filp->private_data;
2413 struct trace_array_cpu *data;
2414 static cpumask_t mask;
2415 unsigned long flags;
2416 #ifdef CONFIG_FTRACE
2422 /* return any leftover data */
2423 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2428 trace_seq_reset(&iter->seq);
2430 mutex_lock(&trace_types_lock);
2431 if (iter->trace->read) {
2432 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2437 while (trace_empty(iter)) {
2439 if ((filp->f_flags & O_NONBLOCK)) {
2445 * This is a make-shift waitqueue. The reason we don't use
2446 * an actual wait queue is because:
2447 * 1) we only ever have one waiter
2448 * 2) the tracing, traces all functions, we don't want
2449 * the overhead of calling wake_up and friends
2450 * (and tracing them too)
2451 * Anyway, this is really very primitive wakeup.
2453 set_current_state(TASK_INTERRUPTIBLE);
2454 iter->tr->waiter = current;
2456 mutex_unlock(&trace_types_lock);
2458 /* sleep for 100 msecs, and try again. */
2459 schedule_timeout(HZ/10);
2461 mutex_lock(&trace_types_lock);
2463 iter->tr->waiter = NULL;
2465 if (signal_pending(current)) {
2470 if (iter->trace != current_trace)
2474 * We block until we read something and tracing is disabled.
2475 * We still block if tracing is disabled, but we have never
2476 * read anything. This allows a user to cat this file, and
2477 * then enable tracing. But after we have read something,
2478 * we give an EOF when tracing is again disabled.
2480 * iter->pos will be 0 if we haven't read anything.
2482 if (!tracer_enabled && iter->pos)
2488 /* stop when tracing is finished */
2489 if (trace_empty(iter))
2492 if (cnt >= PAGE_SIZE)
2493 cnt = PAGE_SIZE - 1;
2495 /* reset all but tr, trace, and overruns */
2496 memset(&iter->seq, 0,
2497 sizeof(struct trace_iterator) -
2498 offsetof(struct trace_iterator, seq));
2502 * We need to stop all tracing on all CPUS to read the
2503 * the next buffer. This is a bit expensive, but is
2504 * not done often. We fill all what we can read,
2505 * and then release the locks again.
2509 local_irq_save(flags);
2510 #ifdef CONFIG_FTRACE
2511 ftrace_save = ftrace_enabled;
2515 for_each_tracing_cpu(cpu) {
2516 data = iter->tr->data[cpu];
2518 if (!head_page(data) || !data->trace_idx)
2521 atomic_inc(&data->disabled);
2525 for_each_cpu_mask(cpu, mask) {
2526 data = iter->tr->data[cpu];
2527 __raw_spin_lock(&data->lock);
2529 if (data->overrun > iter->last_overrun[cpu])
2530 iter->overrun[cpu] +=
2531 data->overrun - iter->last_overrun[cpu];
2532 iter->last_overrun[cpu] = data->overrun;
2535 while (find_next_entry_inc(iter) != NULL) {
2537 int len = iter->seq.len;
2539 ret = print_trace_line(iter);
2541 /* don't print partial lines */
2542 iter->seq.len = len;
2546 trace_consume(iter);
2548 if (iter->seq.len >= cnt)
2552 for_each_cpu_mask(cpu, mask) {
2553 data = iter->tr->data[cpu];
2554 __raw_spin_unlock(&data->lock);
2557 for_each_cpu_mask(cpu, mask) {
2558 data = iter->tr->data[cpu];
2559 atomic_dec(&data->disabled);
2561 #ifdef CONFIG_FTRACE
2562 ftrace_enabled = ftrace_save;
2564 local_irq_restore(flags);
2566 /* Now copy what we have to the user */
2567 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2568 if (iter->seq.readpos >= iter->seq.len)
2569 trace_seq_reset(&iter->seq);
2574 mutex_unlock(&trace_types_lock);
2580 tracing_entries_read(struct file *filp, char __user *ubuf,
2581 size_t cnt, loff_t *ppos)
2583 struct trace_array *tr = filp->private_data;
2587 r = sprintf(buf, "%lu\n", tr->entries);
2588 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2592 tracing_entries_write(struct file *filp, const char __user *ubuf,
2593 size_t cnt, loff_t *ppos)
2599 if (cnt >= sizeof(buf))
2602 if (copy_from_user(&buf, ubuf, cnt))
2607 ret = strict_strtoul(buf, 10, &val);
2611 /* must have at least 1 entry */
2615 mutex_lock(&trace_types_lock);
2617 if (current_trace != &no_tracer) {
2619 pr_info("ftrace: set current_tracer to none"
2620 " before modifying buffer size\n");
2624 if (val > global_trace.entries) {
2625 long pages_requested;
2626 unsigned long freeable_pages;
2628 /* make sure we have enough memory before mapping */
2630 (val + (ENTRIES_PER_PAGE-1)) / ENTRIES_PER_PAGE;
2632 /* account for each buffer (and max_tr) */
2633 pages_requested *= tracing_nr_buffers * 2;
2635 /* Check for overflow */
2636 if (pages_requested < 0) {
2641 freeable_pages = determine_dirtyable_memory();
2643 /* we only allow to request 1/4 of useable memory */
2644 if (pages_requested >
2645 ((freeable_pages + tracing_pages_allocated) / 4)) {
2650 while (global_trace.entries < val) {
2651 if (trace_alloc_page()) {
2655 /* double check that we don't go over the known pages */
2656 if (tracing_pages_allocated > pages_requested)
2661 /* include the number of entries in val (inc of page entries) */
2662 while (global_trace.entries > val + (ENTRIES_PER_PAGE - 1))
2666 /* check integrity */
2667 for_each_tracing_cpu(i)
2668 check_pages(global_trace.data[i]);
2672 /* If check pages failed, return ENOMEM */
2673 if (tracing_disabled)
2676 max_tr.entries = global_trace.entries;
2677 mutex_unlock(&trace_types_lock);
2682 static struct file_operations tracing_max_lat_fops = {
2683 .open = tracing_open_generic,
2684 .read = tracing_max_lat_read,
2685 .write = tracing_max_lat_write,
2688 static struct file_operations tracing_ctrl_fops = {
2689 .open = tracing_open_generic,
2690 .read = tracing_ctrl_read,
2691 .write = tracing_ctrl_write,
2694 static struct file_operations set_tracer_fops = {
2695 .open = tracing_open_generic,
2696 .read = tracing_set_trace_read,
2697 .write = tracing_set_trace_write,
2700 static struct file_operations tracing_pipe_fops = {
2701 .open = tracing_open_pipe,
2702 .poll = tracing_poll_pipe,
2703 .read = tracing_read_pipe,
2704 .release = tracing_release_pipe,
2707 static struct file_operations tracing_entries_fops = {
2708 .open = tracing_open_generic,
2709 .read = tracing_entries_read,
2710 .write = tracing_entries_write,
2713 #ifdef CONFIG_DYNAMIC_FTRACE
2716 tracing_read_long(struct file *filp, char __user *ubuf,
2717 size_t cnt, loff_t *ppos)
2719 unsigned long *p = filp->private_data;
2723 r = sprintf(buf, "%ld\n", *p);
2725 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2728 static struct file_operations tracing_read_long_fops = {
2729 .open = tracing_open_generic,
2730 .read = tracing_read_long,
2734 static struct dentry *d_tracer;
2736 struct dentry *tracing_init_dentry(void)
2743 d_tracer = debugfs_create_dir("tracing", NULL);
2745 if (!d_tracer && !once) {
2747 pr_warning("Could not create debugfs directory 'tracing'\n");
2754 #ifdef CONFIG_FTRACE_SELFTEST
2755 /* Let selftest have access to static functions in this file */
2756 #include "trace_selftest.c"
2759 static __init void tracer_init_debugfs(void)
2761 struct dentry *d_tracer;
2762 struct dentry *entry;
2764 d_tracer = tracing_init_dentry();
2766 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2767 &global_trace, &tracing_ctrl_fops);
2769 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2771 entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2772 NULL, &tracing_iter_fops);
2774 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2776 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2777 NULL, &tracing_cpumask_fops);
2779 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2781 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2782 &global_trace, &tracing_lt_fops);
2784 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2786 entry = debugfs_create_file("trace", 0444, d_tracer,
2787 &global_trace, &tracing_fops);
2789 pr_warning("Could not create debugfs 'trace' entry\n");
2791 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2792 &global_trace, &show_traces_fops);
2794 pr_warning("Could not create debugfs 'trace' entry\n");
2796 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2797 &global_trace, &set_tracer_fops);
2799 pr_warning("Could not create debugfs 'trace' entry\n");
2801 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2802 &tracing_max_latency,
2803 &tracing_max_lat_fops);
2805 pr_warning("Could not create debugfs "
2806 "'tracing_max_latency' entry\n");
2808 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2809 &tracing_thresh, &tracing_max_lat_fops);
2811 pr_warning("Could not create debugfs "
2812 "'tracing_threash' entry\n");
2813 entry = debugfs_create_file("README", 0644, d_tracer,
2814 NULL, &tracing_readme_fops);
2816 pr_warning("Could not create debugfs 'README' entry\n");
2818 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2819 NULL, &tracing_pipe_fops);
2821 pr_warning("Could not create debugfs "
2822 "'tracing_threash' entry\n");
2824 entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2825 &global_trace, &tracing_entries_fops);
2827 pr_warning("Could not create debugfs "
2828 "'tracing_threash' entry\n");
2830 #ifdef CONFIG_DYNAMIC_FTRACE
2831 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2832 &ftrace_update_tot_cnt,
2833 &tracing_read_long_fops);
2835 pr_warning("Could not create debugfs "
2836 "'dyn_ftrace_total_info' entry\n");
2840 static int trace_alloc_page(void)
2842 struct trace_array_cpu *data;
2843 struct page *page, *tmp;
2846 unsigned pages_allocated = 0;
2849 /* first allocate a page for each CPU */
2850 for_each_tracing_cpu(i) {
2851 array = (void *)__get_free_page(GFP_KERNEL);
2852 if (array == NULL) {
2853 printk(KERN_ERR "tracer: failed to allocate page"
2854 "for trace buffer!\n");
2859 page = virt_to_page(array);
2860 list_add(&page->lru, &pages);
2862 /* Only allocate if we are actually using the max trace */
2863 #ifdef CONFIG_TRACER_MAX_TRACE
2864 array = (void *)__get_free_page(GFP_KERNEL);
2865 if (array == NULL) {
2866 printk(KERN_ERR "tracer: failed to allocate page"
2867 "for trace buffer!\n");
2871 page = virt_to_page(array);
2872 list_add(&page->lru, &pages);
2876 /* Now that we successfully allocate a page per CPU, add them */
2877 for_each_tracing_cpu(i) {
2878 data = global_trace.data[i];
2879 page = list_entry(pages.next, struct page, lru);
2880 list_del_init(&page->lru);
2881 list_add_tail(&page->lru, &data->trace_pages);
2884 #ifdef CONFIG_TRACER_MAX_TRACE
2885 data = max_tr.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);
2892 tracing_pages_allocated += pages_allocated;
2893 global_trace.entries += ENTRIES_PER_PAGE;
2898 list_for_each_entry_safe(page, tmp, &pages, lru) {
2899 list_del_init(&page->lru);
2905 static int trace_free_page(void)
2907 struct trace_array_cpu *data;
2909 struct list_head *p;
2913 /* free one page from each buffer */
2914 for_each_tracing_cpu(i) {
2915 data = global_trace.data[i];
2916 p = data->trace_pages.next;
2917 if (p == &data->trace_pages) {
2918 /* should never happen */
2920 tracing_disabled = 1;
2924 page = list_entry(p, struct page, lru);
2926 list_del(&page->lru);
2927 tracing_pages_allocated--;
2928 tracing_pages_allocated--;
2931 tracing_reset(data);
2933 #ifdef CONFIG_TRACER_MAX_TRACE
2934 data = max_tr.data[i];
2935 p = data->trace_pages.next;
2936 if (p == &data->trace_pages) {
2937 /* should never happen */
2939 tracing_disabled = 1;
2943 page = list_entry(p, struct page, lru);
2945 list_del(&page->lru);
2948 tracing_reset(data);
2951 global_trace.entries -= ENTRIES_PER_PAGE;
2956 __init static int tracer_alloc_buffers(void)
2958 struct trace_array_cpu *data;
2965 /* TODO: make the number of buffers hot pluggable with CPUS */
2966 tracing_nr_buffers = num_possible_cpus();
2967 tracing_buffer_mask = cpu_possible_map;
2969 /* Allocate the first page for all buffers */
2970 for_each_tracing_cpu(i) {
2971 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
2972 max_tr.data[i] = &per_cpu(max_data, i);
2974 array = (void *)__get_free_page(GFP_KERNEL);
2975 if (array == NULL) {
2976 printk(KERN_ERR "tracer: failed to allocate page"
2977 "for trace buffer!\n");
2981 /* set the array to the list */
2982 INIT_LIST_HEAD(&data->trace_pages);
2983 page = virt_to_page(array);
2984 list_add(&page->lru, &data->trace_pages);
2985 /* use the LRU flag to differentiate the two buffers */
2988 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2989 max_tr.data[i]->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
2991 /* Only allocate if we are actually using the max trace */
2992 #ifdef CONFIG_TRACER_MAX_TRACE
2993 array = (void *)__get_free_page(GFP_KERNEL);
2994 if (array == NULL) {
2995 printk(KERN_ERR "tracer: failed to allocate page"
2996 "for trace buffer!\n");
3000 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
3001 page = virt_to_page(array);
3002 list_add(&page->lru, &max_tr.data[i]->trace_pages);
3008 * Since we allocate by orders of pages, we may be able to
3011 global_trace.entries = ENTRIES_PER_PAGE;
3014 while (global_trace.entries < trace_nr_entries) {
3015 if (trace_alloc_page())
3019 max_tr.entries = global_trace.entries;
3021 pr_info("tracer: %d pages allocated for %ld",
3022 pages, trace_nr_entries);
3023 pr_info(" entries of %ld bytes\n", (long)TRACE_ENTRY_SIZE);
3024 pr_info(" actual entries %ld\n", global_trace.entries);
3026 tracer_init_debugfs();
3028 trace_init_cmdlines();
3030 register_tracer(&no_tracer);
3031 current_trace = &no_tracer;
3033 /* All seems OK, enable tracing */
3034 global_trace.ctrl = tracer_enabled;
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);