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/kprobes.h>
31 #include <linux/writeback.h>
33 #include <linux/stacktrace.h>
37 unsigned long __read_mostly tracing_max_latency = (cycle_t)ULONG_MAX;
38 unsigned long __read_mostly tracing_thresh;
40 static unsigned long __read_mostly tracing_nr_buffers;
41 static cpumask_t __read_mostly tracing_buffer_mask;
43 #define for_each_tracing_cpu(cpu) \
44 for_each_cpu_mask(cpu, tracing_buffer_mask)
46 static int trace_alloc_page(void);
47 static int trace_free_page(void);
49 static int tracing_disabled = 1;
51 static unsigned long tracing_pages_allocated;
54 ns2usecs(cycle_t nsec)
61 cycle_t ftrace_now(int cpu)
63 return cpu_clock(cpu);
67 * The global_trace is the descriptor that holds the tracing
68 * buffers for the live tracing. For each CPU, it contains
69 * a link list of pages that will store trace entries. The
70 * page descriptor of the pages in the memory is used to hold
71 * the link list by linking the lru item in the page descriptor
72 * to each of the pages in the buffer per CPU.
74 * For each active CPU there is a data field that holds the
75 * pages for the buffer for that CPU. Each CPU has the same number
76 * of pages allocated for its buffer.
78 static struct trace_array global_trace;
80 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
83 * The max_tr is used to snapshot the global_trace when a maximum
84 * latency is reached. Some tracers will use this to store a maximum
85 * trace while it continues examining live traces.
87 * The buffers for the max_tr are set up the same as the global_trace.
88 * When a snapshot is taken, the link list of the max_tr is swapped
89 * with the link list of the global_trace and the buffers are reset for
90 * the global_trace so the tracing can continue.
92 static struct trace_array max_tr;
94 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
96 /* tracer_enabled is used to toggle activation of a tracer */
97 static int tracer_enabled = 1;
100 * trace_nr_entries is the number of entries that is allocated
101 * for a buffer. Note, the number of entries is always rounded
102 * to ENTRIES_PER_PAGE.
104 static unsigned long trace_nr_entries = 65536UL;
106 /* trace_types holds a link list of available tracers. */
107 static struct tracer *trace_types __read_mostly;
109 /* current_trace points to the tracer that is currently active */
110 static struct tracer *current_trace __read_mostly;
113 * max_tracer_type_len is used to simplify the allocating of
114 * buffers to read userspace tracer names. We keep track of
115 * the longest tracer name registered.
117 static int max_tracer_type_len;
120 * trace_types_lock is used to protect the trace_types list.
121 * This lock is also used to keep user access serialized.
122 * Accesses from userspace will grab this lock while userspace
123 * activities happen inside the kernel.
125 static DEFINE_MUTEX(trace_types_lock);
127 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
128 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
130 /* trace_flags holds iter_ctrl options */
131 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT;
133 static notrace void no_trace_init(struct trace_array *tr)
138 for_each_online_cpu(cpu)
139 tracing_reset(tr->data[cpu]);
143 /* dummy trace to disable tracing */
144 static struct tracer no_tracer __read_mostly = {
146 .init = no_trace_init
151 * trace_wake_up - wake up tasks waiting for trace input
153 * Simply wakes up any task that is blocked on the trace_wait
154 * queue. These is used with trace_poll for tasks polling the trace.
156 void trace_wake_up(void)
159 * The runqueue_is_locked() can fail, but this is the best we
162 if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
163 wake_up(&trace_wait);
166 #define ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(struct trace_entry))
168 static int __init set_nr_entries(char *str)
170 unsigned long nr_entries;
175 ret = strict_strtoul(str, 0, &nr_entries);
176 /* nr_entries can not be zero */
177 if (ret < 0 || nr_entries == 0)
179 trace_nr_entries = nr_entries;
182 __setup("trace_entries=", set_nr_entries);
184 unsigned long nsecs_to_usecs(unsigned long nsecs)
190 * trace_flag_type is an enumeration that holds different
191 * states when a trace occurs. These are:
192 * IRQS_OFF - interrupts were disabled
193 * NEED_RESCED - reschedule is requested
194 * HARDIRQ - inside an interrupt handler
195 * SOFTIRQ - inside a softirq handler
197 enum trace_flag_type {
198 TRACE_FLAG_IRQS_OFF = 0x01,
199 TRACE_FLAG_NEED_RESCHED = 0x02,
200 TRACE_FLAG_HARDIRQ = 0x04,
201 TRACE_FLAG_SOFTIRQ = 0x08,
205 * TRACE_ITER_SYM_MASK masks the options in trace_flags that
206 * control the output of kernel symbols.
208 #define TRACE_ITER_SYM_MASK \
209 (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
211 /* These must match the bit postions in trace_iterator_flags */
212 static const char *trace_options[] = {
227 * ftrace_max_lock is used to protect the swapping of buffers
228 * when taking a max snapshot. The buffers themselves are
229 * protected by per_cpu spinlocks. But the action of the swap
230 * needs its own lock.
232 * This is defined as a raw_spinlock_t in order to help
233 * with performance when lockdep debugging is enabled.
235 static raw_spinlock_t ftrace_max_lock =
236 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
239 * Copy the new maximum trace into the separate maximum-trace
240 * structure. (this way the maximum trace is permanently saved,
241 * for later retrieval via /debugfs/tracing/latency_trace)
244 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
246 struct trace_array_cpu *data = tr->data[cpu];
249 max_tr.time_start = data->preempt_timestamp;
251 data = max_tr.data[cpu];
252 data->saved_latency = tracing_max_latency;
254 memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
255 data->pid = tsk->pid;
256 data->uid = tsk->uid;
257 data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
258 data->policy = tsk->policy;
259 data->rt_priority = tsk->rt_priority;
261 /* record this tasks comm */
262 tracing_record_cmdline(current);
265 #define CHECK_COND(cond) \
266 if (unlikely(cond)) { \
267 tracing_disabled = 1; \
273 * check_pages - integrity check of trace buffers
275 * As a safty measure we check to make sure the data pages have not
278 int check_pages(struct trace_array_cpu *data)
280 struct page *page, *tmp;
282 CHECK_COND(data->trace_pages.next->prev != &data->trace_pages);
283 CHECK_COND(data->trace_pages.prev->next != &data->trace_pages);
285 list_for_each_entry_safe(page, tmp, &data->trace_pages, lru) {
286 CHECK_COND(page->lru.next->prev != &page->lru);
287 CHECK_COND(page->lru.prev->next != &page->lru);
294 * head_page - page address of the first page in per_cpu buffer.
296 * head_page returns the page address of the first page in
297 * a per_cpu buffer. This also preforms various consistency
298 * checks to make sure the buffer has not been corrupted.
300 void *head_page(struct trace_array_cpu *data)
304 if (list_empty(&data->trace_pages))
307 page = list_entry(data->trace_pages.next, struct page, lru);
308 BUG_ON(&page->lru == &data->trace_pages);
310 return page_address(page);
314 * trace_seq_printf - sequence printing of trace information
315 * @s: trace sequence descriptor
316 * @fmt: printf format string
318 * The tracer may use either sequence operations or its own
319 * copy to user routines. To simplify formating of a trace
320 * trace_seq_printf is used to store strings into a special
321 * buffer (@s). Then the output may be either used by
322 * the sequencer or pulled into another buffer.
325 trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
327 int len = (PAGE_SIZE - 1) - s->len;
335 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
338 /* If we can't write it all, don't bother writing anything */
348 * trace_seq_puts - trace sequence printing of simple string
349 * @s: trace sequence descriptor
350 * @str: simple string to record
352 * The tracer may use either the sequence operations or its own
353 * copy to user routines. This function records a simple string
354 * into a special buffer (@s) for later retrieval by a sequencer
355 * or other mechanism.
358 trace_seq_puts(struct trace_seq *s, const char *str)
360 int len = strlen(str);
362 if (len > ((PAGE_SIZE - 1) - s->len))
365 memcpy(s->buffer + s->len, str, len);
372 trace_seq_putc(struct trace_seq *s, unsigned char c)
374 if (s->len >= (PAGE_SIZE - 1))
377 s->buffer[s->len++] = c;
383 trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
385 if (len > ((PAGE_SIZE - 1) - s->len))
388 memcpy(s->buffer + s->len, mem, len);
395 static const char hex2asc[] = "0123456789abcdef";
398 trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
400 unsigned char hex[HEX_CHARS];
401 unsigned char *data = mem;
405 BUG_ON(len >= HEX_CHARS);
408 for (i = 0, j = 0; i < len; i++) {
410 for (i = len-1, j = 0; i >= 0; i--) {
414 hex[j++] = hex2asc[byte & 0x0f];
415 hex[j++] = hex2asc[byte >> 4];
419 return trace_seq_putmem(s, hex, j);
423 trace_seq_reset(struct trace_seq *s)
429 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
434 if (s->len <= s->readpos)
437 len = s->len - s->readpos;
440 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
449 trace_print_seq(struct seq_file *m, struct trace_seq *s)
451 int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
454 seq_puts(m, s->buffer);
460 * flip the trace buffers between two trace descriptors.
461 * This usually is the buffers between the global_trace and
462 * the max_tr to record a snapshot of a current trace.
464 * The ftrace_max_lock must be held.
467 flip_trace(struct trace_array_cpu *tr1, struct trace_array_cpu *tr2)
469 struct list_head flip_pages;
471 INIT_LIST_HEAD(&flip_pages);
473 memcpy(&tr1->trace_head_idx, &tr2->trace_head_idx,
474 sizeof(struct trace_array_cpu) -
475 offsetof(struct trace_array_cpu, trace_head_idx));
479 list_splice_init(&tr1->trace_pages, &flip_pages);
480 list_splice_init(&tr2->trace_pages, &tr1->trace_pages);
481 list_splice_init(&flip_pages, &tr2->trace_pages);
482 BUG_ON(!list_empty(&flip_pages));
488 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
490 * @tsk: the task with the latency
491 * @cpu: The cpu that initiated the trace.
493 * Flip the buffers between the @tr and the max_tr and record information
494 * about which task was the cause of this latency.
497 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
499 struct trace_array_cpu *data;
502 WARN_ON_ONCE(!irqs_disabled());
503 __raw_spin_lock(&ftrace_max_lock);
504 /* clear out all the previous traces */
505 for_each_tracing_cpu(i) {
507 flip_trace(max_tr.data[i], data);
511 __update_max_tr(tr, tsk, cpu);
512 __raw_spin_unlock(&ftrace_max_lock);
516 * update_max_tr_single - only copy one trace over, and reset the rest
518 * @tsk - task with the latency
519 * @cpu - the cpu of the buffer to copy.
521 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
524 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
526 struct trace_array_cpu *data = tr->data[cpu];
529 WARN_ON_ONCE(!irqs_disabled());
530 __raw_spin_lock(&ftrace_max_lock);
531 for_each_tracing_cpu(i)
532 tracing_reset(max_tr.data[i]);
534 flip_trace(max_tr.data[cpu], data);
537 __update_max_tr(tr, tsk, cpu);
538 __raw_spin_unlock(&ftrace_max_lock);
542 * register_tracer - register a tracer with the ftrace system.
543 * @type - the plugin for the tracer
545 * Register a new plugin tracer.
547 int register_tracer(struct tracer *type)
554 pr_info("Tracer must have a name\n");
558 mutex_lock(&trace_types_lock);
559 for (t = trace_types; t; t = t->next) {
560 if (strcmp(type->name, t->name) == 0) {
562 pr_info("Trace %s already registered\n",
569 #ifdef CONFIG_FTRACE_STARTUP_TEST
570 if (type->selftest) {
571 struct tracer *saved_tracer = current_trace;
572 struct trace_array_cpu *data;
573 struct trace_array *tr = &global_trace;
574 int saved_ctrl = tr->ctrl;
577 * Run a selftest on this tracer.
578 * Here we reset the trace buffer, and set the current
579 * tracer to be this tracer. The tracer can then run some
580 * internal tracing to verify that everything is in order.
581 * If we fail, we do not register this tracer.
583 for_each_tracing_cpu(i) {
585 if (!head_page(data))
589 current_trace = type;
591 /* the test is responsible for initializing and enabling */
592 pr_info("Testing tracer %s: ", type->name);
593 ret = type->selftest(type, tr);
594 /* the test is responsible for resetting too */
595 current_trace = saved_tracer;
596 tr->ctrl = saved_ctrl;
598 printk(KERN_CONT "FAILED!\n");
601 /* Only reset on passing, to avoid touching corrupted buffers */
602 for_each_tracing_cpu(i) {
604 if (!head_page(data))
608 printk(KERN_CONT "PASSED\n");
612 type->next = trace_types;
614 len = strlen(type->name);
615 if (len > max_tracer_type_len)
616 max_tracer_type_len = len;
619 mutex_unlock(&trace_types_lock);
624 void unregister_tracer(struct tracer *type)
629 mutex_lock(&trace_types_lock);
630 for (t = &trace_types; *t; t = &(*t)->next) {
634 pr_info("Trace %s not registered\n", type->name);
639 if (strlen(type->name) != max_tracer_type_len)
642 max_tracer_type_len = 0;
643 for (t = &trace_types; *t; t = &(*t)->next) {
644 len = strlen((*t)->name);
645 if (len > max_tracer_type_len)
646 max_tracer_type_len = len;
649 mutex_unlock(&trace_types_lock);
652 void tracing_reset(struct trace_array_cpu *data)
656 data->trace_head = data->trace_tail = head_page(data);
657 data->trace_head_idx = 0;
658 data->trace_tail_idx = 0;
661 #define SAVED_CMDLINES 128
662 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
663 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
664 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
665 static int cmdline_idx;
666 static DEFINE_SPINLOCK(trace_cmdline_lock);
668 /* temporary disable recording */
669 atomic_t trace_record_cmdline_disabled __read_mostly;
671 static void trace_init_cmdlines(void)
673 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
674 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
678 void trace_stop_cmdline_recording(void);
680 static void trace_save_cmdline(struct task_struct *tsk)
685 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
689 * It's not the end of the world if we don't get
690 * the lock, but we also don't want to spin
691 * nor do we want to disable interrupts,
692 * so if we miss here, then better luck next time.
694 if (!spin_trylock(&trace_cmdline_lock))
697 idx = map_pid_to_cmdline[tsk->pid];
698 if (idx >= SAVED_CMDLINES) {
699 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
701 map = map_cmdline_to_pid[idx];
702 if (map <= PID_MAX_DEFAULT)
703 map_pid_to_cmdline[map] = (unsigned)-1;
705 map_pid_to_cmdline[tsk->pid] = idx;
710 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
712 spin_unlock(&trace_cmdline_lock);
715 static char *trace_find_cmdline(int pid)
717 char *cmdline = "<...>";
723 if (pid > PID_MAX_DEFAULT)
726 map = map_pid_to_cmdline[pid];
727 if (map >= SAVED_CMDLINES)
730 cmdline = saved_cmdlines[map];
736 void tracing_record_cmdline(struct task_struct *tsk)
738 if (atomic_read(&trace_record_cmdline_disabled))
741 trace_save_cmdline(tsk);
744 static inline struct list_head *
745 trace_next_list(struct trace_array_cpu *data, struct list_head *next)
748 * Roundrobin - but skip the head (which is not a real page):
751 if (unlikely(next == &data->trace_pages))
753 BUG_ON(next == &data->trace_pages);
759 trace_next_page(struct trace_array_cpu *data, void *addr)
761 struct list_head *next;
764 page = virt_to_page(addr);
766 next = trace_next_list(data, &page->lru);
767 page = list_entry(next, struct page, lru);
769 return page_address(page);
772 static inline struct trace_entry *
773 tracing_get_trace_entry(struct trace_array *tr, struct trace_array_cpu *data)
775 unsigned long idx, idx_next;
776 struct trace_entry *entry;
779 idx = data->trace_head_idx;
782 BUG_ON(idx * TRACE_ENTRY_SIZE >= PAGE_SIZE);
784 entry = data->trace_head + idx * TRACE_ENTRY_SIZE;
786 if (unlikely(idx_next >= ENTRIES_PER_PAGE)) {
787 data->trace_head = trace_next_page(data, data->trace_head);
791 if (data->trace_head == data->trace_tail &&
792 idx_next == data->trace_tail_idx) {
795 data->trace_tail_idx++;
796 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
798 trace_next_page(data, data->trace_tail);
799 data->trace_tail_idx = 0;
803 data->trace_head_idx = idx_next;
809 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags)
811 struct task_struct *tsk = current;
814 pc = preempt_count();
816 entry->preempt_count = pc & 0xff;
817 entry->pid = (tsk) ? tsk->pid : 0;
818 entry->t = ftrace_now(raw_smp_processor_id());
819 entry->flags = (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
820 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
821 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
822 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
826 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
827 unsigned long ip, unsigned long parent_ip, unsigned long flags)
829 struct trace_entry *entry;
830 unsigned long irq_flags;
832 raw_local_irq_save(irq_flags);
833 __raw_spin_lock(&data->lock);
834 entry = tracing_get_trace_entry(tr, data);
835 tracing_generic_entry_update(entry, flags);
836 entry->type = TRACE_FN;
838 entry->fn.parent_ip = parent_ip;
839 __raw_spin_unlock(&data->lock);
840 raw_local_irq_restore(irq_flags);
844 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
845 unsigned long ip, unsigned long parent_ip, unsigned long flags)
847 if (likely(!atomic_read(&data->disabled)))
848 trace_function(tr, data, ip, parent_ip, flags);
851 #ifdef CONFIG_MMIOTRACE
852 void __trace_mmiotrace_rw(struct trace_array *tr, struct trace_array_cpu *data,
853 struct mmiotrace_rw *rw)
855 struct trace_entry *entry;
856 unsigned long irq_flags;
858 raw_local_irq_save(irq_flags);
859 __raw_spin_lock(&data->lock);
861 entry = tracing_get_trace_entry(tr, data);
862 tracing_generic_entry_update(entry, 0);
863 entry->type = TRACE_MMIO_RW;
866 __raw_spin_unlock(&data->lock);
867 raw_local_irq_restore(irq_flags);
872 void __trace_mmiotrace_map(struct trace_array *tr, struct trace_array_cpu *data,
873 struct mmiotrace_map *map)
875 struct trace_entry *entry;
876 unsigned long irq_flags;
878 raw_local_irq_save(irq_flags);
879 __raw_spin_lock(&data->lock);
881 entry = tracing_get_trace_entry(tr, data);
882 tracing_generic_entry_update(entry, 0);
883 entry->type = TRACE_MMIO_MAP;
884 entry->mmiomap = *map;
886 __raw_spin_unlock(&data->lock);
887 raw_local_irq_restore(irq_flags);
893 void __trace_stack(struct trace_array *tr,
894 struct trace_array_cpu *data,
898 struct trace_entry *entry;
899 struct stack_trace trace;
901 if (!(trace_flags & TRACE_ITER_STACKTRACE))
904 entry = tracing_get_trace_entry(tr, data);
905 tracing_generic_entry_update(entry, flags);
906 entry->type = TRACE_STACK;
908 memset(&entry->stack, 0, sizeof(entry->stack));
910 trace.nr_entries = 0;
911 trace.max_entries = FTRACE_STACK_ENTRIES;
913 trace.entries = entry->stack.caller;
915 save_stack_trace(&trace);
919 __trace_special(void *__tr, void *__data,
920 unsigned long arg1, unsigned long arg2, unsigned long arg3)
922 struct trace_array_cpu *data = __data;
923 struct trace_array *tr = __tr;
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, 0);
931 entry->type = TRACE_SPECIAL;
932 entry->special.arg1 = arg1;
933 entry->special.arg2 = arg2;
934 entry->special.arg3 = arg3;
935 __trace_stack(tr, data, irq_flags, 4);
936 __raw_spin_unlock(&data->lock);
937 raw_local_irq_restore(irq_flags);
943 tracing_sched_switch_trace(struct trace_array *tr,
944 struct trace_array_cpu *data,
945 struct task_struct *prev,
946 struct task_struct *next,
949 struct trace_entry *entry;
950 unsigned long irq_flags;
952 raw_local_irq_save(irq_flags);
953 __raw_spin_lock(&data->lock);
954 entry = tracing_get_trace_entry(tr, data);
955 tracing_generic_entry_update(entry, flags);
956 entry->type = TRACE_CTX;
957 entry->ctx.prev_pid = prev->pid;
958 entry->ctx.prev_prio = prev->prio;
959 entry->ctx.prev_state = prev->state;
960 entry->ctx.next_pid = next->pid;
961 entry->ctx.next_prio = next->prio;
962 entry->ctx.next_state = next->state;
963 __trace_stack(tr, data, flags, 5);
964 __raw_spin_unlock(&data->lock);
965 raw_local_irq_restore(irq_flags);
969 tracing_sched_wakeup_trace(struct trace_array *tr,
970 struct trace_array_cpu *data,
971 struct task_struct *wakee,
972 struct task_struct *curr,
975 struct trace_entry *entry;
976 unsigned long irq_flags;
978 raw_local_irq_save(irq_flags);
979 __raw_spin_lock(&data->lock);
980 entry = tracing_get_trace_entry(tr, data);
981 tracing_generic_entry_update(entry, flags);
982 entry->type = TRACE_WAKE;
983 entry->ctx.prev_pid = curr->pid;
984 entry->ctx.prev_prio = curr->prio;
985 entry->ctx.prev_state = curr->state;
986 entry->ctx.next_pid = wakee->pid;
987 entry->ctx.next_prio = wakee->prio;
988 entry->ctx.next_state = wakee->state;
989 __trace_stack(tr, data, flags, 6);
990 __raw_spin_unlock(&data->lock);
991 raw_local_irq_restore(irq_flags);
997 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
999 struct trace_array *tr = &global_trace;
1000 struct trace_array_cpu *data;
1001 unsigned long flags;
1005 if (tracing_disabled || current_trace == &no_tracer || !tr->ctrl)
1008 local_irq_save(flags);
1009 cpu = raw_smp_processor_id();
1010 data = tr->data[cpu];
1011 disabled = atomic_inc_return(&data->disabled);
1013 if (likely(disabled == 1))
1014 __trace_special(tr, data, arg1, arg2, arg3);
1016 atomic_dec(&data->disabled);
1017 local_irq_restore(flags);
1020 #ifdef CONFIG_FTRACE
1022 function_trace_call(unsigned long ip, unsigned long parent_ip)
1024 struct trace_array *tr = &global_trace;
1025 struct trace_array_cpu *data;
1026 unsigned long flags;
1030 if (unlikely(!tracer_enabled))
1036 local_irq_save(flags);
1037 cpu = raw_smp_processor_id();
1038 data = tr->data[cpu];
1039 disabled = atomic_inc_return(&data->disabled);
1041 if (likely(disabled == 1))
1042 trace_function(tr, data, ip, parent_ip, flags);
1044 atomic_dec(&data->disabled);
1045 local_irq_restore(flags);
1048 static struct ftrace_ops trace_ops __read_mostly =
1050 .func = function_trace_call,
1053 void tracing_start_function_trace(void)
1055 register_ftrace_function(&trace_ops);
1058 void tracing_stop_function_trace(void)
1060 unregister_ftrace_function(&trace_ops);
1064 enum trace_file_type {
1065 TRACE_FILE_LAT_FMT = 1,
1068 static struct trace_entry *
1069 trace_entry_idx(struct trace_array *tr, struct trace_array_cpu *data,
1070 struct trace_iterator *iter, int cpu)
1073 struct trace_entry *array;
1075 if (iter->next_idx[cpu] >= tr->entries ||
1076 iter->next_idx[cpu] >= data->trace_idx ||
1077 (data->trace_head == data->trace_tail &&
1078 data->trace_head_idx == data->trace_tail_idx))
1081 if (!iter->next_page[cpu]) {
1082 /* Initialize the iterator for this cpu trace buffer */
1083 WARN_ON(!data->trace_tail);
1084 page = virt_to_page(data->trace_tail);
1085 iter->next_page[cpu] = &page->lru;
1086 iter->next_page_idx[cpu] = data->trace_tail_idx;
1089 page = list_entry(iter->next_page[cpu], struct page, lru);
1090 BUG_ON(&data->trace_pages == &page->lru);
1092 array = page_address(page);
1094 WARN_ON(iter->next_page_idx[cpu] >= ENTRIES_PER_PAGE);
1095 return &array[iter->next_page_idx[cpu]];
1098 static struct trace_entry *
1099 find_next_entry(struct trace_iterator *iter, int *ent_cpu)
1101 struct trace_array *tr = iter->tr;
1102 struct trace_entry *ent, *next = NULL;
1106 for_each_tracing_cpu(cpu) {
1107 if (!head_page(tr->data[cpu]))
1109 ent = trace_entry_idx(tr, tr->data[cpu], iter, cpu);
1111 * Pick the entry with the smallest timestamp:
1113 if (ent && (!next || ent->t < next->t)) {
1120 *ent_cpu = next_cpu;
1125 static void trace_iterator_increment(struct trace_iterator *iter)
1128 iter->next_idx[iter->cpu]++;
1129 iter->next_page_idx[iter->cpu]++;
1131 if (iter->next_page_idx[iter->cpu] >= ENTRIES_PER_PAGE) {
1132 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1134 iter->next_page_idx[iter->cpu] = 0;
1135 iter->next_page[iter->cpu] =
1136 trace_next_list(data, iter->next_page[iter->cpu]);
1140 static void trace_consume(struct trace_iterator *iter)
1142 struct trace_array_cpu *data = iter->tr->data[iter->cpu];
1144 data->trace_tail_idx++;
1145 if (data->trace_tail_idx >= ENTRIES_PER_PAGE) {
1146 data->trace_tail = trace_next_page(data, data->trace_tail);
1147 data->trace_tail_idx = 0;
1150 /* Check if we empty it, then reset the index */
1151 if (data->trace_head == data->trace_tail &&
1152 data->trace_head_idx == data->trace_tail_idx)
1153 data->trace_idx = 0;
1156 static void *find_next_entry_inc(struct trace_iterator *iter)
1158 struct trace_entry *next;
1161 next = find_next_entry(iter, &next_cpu);
1163 iter->prev_ent = iter->ent;
1164 iter->prev_cpu = iter->cpu;
1167 iter->cpu = next_cpu;
1170 trace_iterator_increment(iter);
1172 return next ? iter : NULL;
1175 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1177 struct trace_iterator *iter = m->private;
1178 void *last_ent = iter->ent;
1184 /* can't go backwards */
1189 ent = find_next_entry_inc(iter);
1193 while (ent && iter->idx < i)
1194 ent = find_next_entry_inc(iter);
1198 if (last_ent && !ent)
1199 seq_puts(m, "\n\nvim:ft=help\n");
1204 static void *s_start(struct seq_file *m, loff_t *pos)
1206 struct trace_iterator *iter = m->private;
1211 mutex_lock(&trace_types_lock);
1213 if (!current_trace || current_trace != iter->trace) {
1214 mutex_unlock(&trace_types_lock);
1218 atomic_inc(&trace_record_cmdline_disabled);
1220 /* let the tracer grab locks here if needed */
1221 if (current_trace->start)
1222 current_trace->start(iter);
1224 if (*pos != iter->pos) {
1228 iter->prev_ent = NULL;
1229 iter->prev_cpu = -1;
1231 for_each_tracing_cpu(i) {
1232 iter->next_idx[i] = 0;
1233 iter->next_page[i] = NULL;
1236 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1241 p = s_next(m, p, &l);
1247 static void s_stop(struct seq_file *m, void *p)
1249 struct trace_iterator *iter = m->private;
1251 atomic_dec(&trace_record_cmdline_disabled);
1253 /* let the tracer release locks here if needed */
1254 if (current_trace && current_trace == iter->trace && iter->trace->stop)
1255 iter->trace->stop(iter);
1257 mutex_unlock(&trace_types_lock);
1260 #define KRETPROBE_MSG "[unknown/kretprobe'd]"
1262 #ifdef CONFIG_KRETPROBES
1263 static inline int kretprobed(unsigned long addr)
1265 return addr == (unsigned long)kretprobe_trampoline;
1268 static inline int kretprobed(unsigned long addr)
1272 #endif /* CONFIG_KRETPROBES */
1275 seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1277 #ifdef CONFIG_KALLSYMS
1278 char str[KSYM_SYMBOL_LEN];
1280 kallsyms_lookup(address, NULL, NULL, NULL, str);
1282 return trace_seq_printf(s, fmt, str);
1288 seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1289 unsigned long address)
1291 #ifdef CONFIG_KALLSYMS
1292 char str[KSYM_SYMBOL_LEN];
1294 sprint_symbol(str, address);
1295 return trace_seq_printf(s, fmt, str);
1300 #ifndef CONFIG_64BIT
1301 # define IP_FMT "%08lx"
1303 # define IP_FMT "%016lx"
1307 seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1312 return trace_seq_printf(s, "0");
1314 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1315 ret = seq_print_sym_offset(s, "%s", ip);
1317 ret = seq_print_sym_short(s, "%s", ip);
1322 if (sym_flags & TRACE_ITER_SYM_ADDR)
1323 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1327 static void print_lat_help_header(struct seq_file *m)
1329 seq_puts(m, "# _------=> CPU# \n");
1330 seq_puts(m, "# / _-----=> irqs-off \n");
1331 seq_puts(m, "# | / _----=> need-resched \n");
1332 seq_puts(m, "# || / _---=> hardirq/softirq \n");
1333 seq_puts(m, "# ||| / _--=> preempt-depth \n");
1334 seq_puts(m, "# |||| / \n");
1335 seq_puts(m, "# ||||| delay \n");
1336 seq_puts(m, "# cmd pid ||||| time | caller \n");
1337 seq_puts(m, "# \\ / ||||| \\ | / \n");
1340 static void print_func_help_header(struct seq_file *m)
1342 seq_puts(m, "# TASK-PID CPU# TIMESTAMP FUNCTION\n");
1343 seq_puts(m, "# | | | | |\n");
1348 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1350 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1351 struct trace_array *tr = iter->tr;
1352 struct trace_array_cpu *data = tr->data[tr->cpu];
1353 struct tracer *type = current_trace;
1354 unsigned long total = 0;
1355 unsigned long entries = 0;
1357 const char *name = "preemption";
1362 for_each_tracing_cpu(cpu) {
1363 if (head_page(tr->data[cpu])) {
1364 total += tr->data[cpu]->trace_idx;
1365 if (tr->data[cpu]->trace_idx > tr->entries)
1366 entries += tr->entries;
1368 entries += tr->data[cpu]->trace_idx;
1372 seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1374 seq_puts(m, "-----------------------------------"
1375 "---------------------------------\n");
1376 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1377 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1378 nsecs_to_usecs(data->saved_latency),
1382 #if defined(CONFIG_PREEMPT_NONE)
1384 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1386 #elif defined(CONFIG_PREEMPT_DESKTOP)
1391 /* These are reserved for later use */
1394 seq_printf(m, " #P:%d)\n", num_online_cpus());
1398 seq_puts(m, " -----------------\n");
1399 seq_printf(m, " | task: %.16s-%d "
1400 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1401 data->comm, data->pid, data->uid, data->nice,
1402 data->policy, data->rt_priority);
1403 seq_puts(m, " -----------------\n");
1405 if (data->critical_start) {
1406 seq_puts(m, " => started at: ");
1407 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1408 trace_print_seq(m, &iter->seq);
1409 seq_puts(m, "\n => ended at: ");
1410 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1411 trace_print_seq(m, &iter->seq);
1419 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1421 int hardirq, softirq;
1424 comm = trace_find_cmdline(entry->pid);
1426 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1427 trace_seq_printf(s, "%d", cpu);
1428 trace_seq_printf(s, "%c%c",
1429 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' : '.',
1430 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1432 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1433 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1434 if (hardirq && softirq) {
1435 trace_seq_putc(s, 'H');
1438 trace_seq_putc(s, 'h');
1441 trace_seq_putc(s, 's');
1443 trace_seq_putc(s, '.');
1447 if (entry->preempt_count)
1448 trace_seq_printf(s, "%x", entry->preempt_count);
1450 trace_seq_puts(s, ".");
1453 unsigned long preempt_mark_thresh = 100;
1456 lat_print_timestamp(struct trace_seq *s, unsigned long long abs_usecs,
1457 unsigned long rel_usecs)
1459 trace_seq_printf(s, " %4lldus", abs_usecs);
1460 if (rel_usecs > preempt_mark_thresh)
1461 trace_seq_puts(s, "!: ");
1462 else if (rel_usecs > 1)
1463 trace_seq_puts(s, "+: ");
1465 trace_seq_puts(s, " : ");
1468 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1471 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1473 struct trace_seq *s = &iter->seq;
1474 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1475 struct trace_entry *next_entry = find_next_entry(iter, NULL);
1476 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1477 struct trace_entry *entry = iter->ent;
1478 unsigned long abs_usecs;
1479 unsigned long rel_usecs;
1487 rel_usecs = ns2usecs(next_entry->t - entry->t);
1488 abs_usecs = ns2usecs(entry->t - iter->tr->time_start);
1491 comm = trace_find_cmdline(entry->pid);
1492 trace_seq_printf(s, "%16s %5d %d %d %08x %08x [%08lx]"
1493 " %ld.%03ldms (+%ld.%03ldms): ",
1495 entry->pid, cpu, entry->flags,
1496 entry->preempt_count, trace_idx,
1499 abs_usecs % 1000, rel_usecs/1000,
1502 lat_print_generic(s, entry, cpu);
1503 lat_print_timestamp(s, abs_usecs, rel_usecs);
1505 switch (entry->type) {
1507 seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1508 trace_seq_puts(s, " (");
1509 if (kretprobed(entry->fn.parent_ip))
1510 trace_seq_puts(s, KRETPROBE_MSG);
1512 seq_print_ip_sym(s, entry->fn.parent_ip, sym_flags);
1513 trace_seq_puts(s, ")\n");
1517 T = entry->ctx.next_state < sizeof(state_to_char) ?
1518 state_to_char[entry->ctx.next_state] : 'X';
1520 state = entry->ctx.prev_state ? __ffs(entry->ctx.prev_state) + 1 : 0;
1521 S = state < sizeof(state_to_char) - 1 ? state_to_char[state] : 'X';
1522 comm = trace_find_cmdline(entry->ctx.next_pid);
1523 trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c %s\n",
1524 entry->ctx.prev_pid,
1525 entry->ctx.prev_prio,
1526 S, entry->type == TRACE_CTX ? "==>" : " +",
1527 entry->ctx.next_pid,
1528 entry->ctx.next_prio,
1532 trace_seq_printf(s, "# %ld %ld %ld\n",
1533 entry->special.arg1,
1534 entry->special.arg2,
1535 entry->special.arg3);
1538 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1540 trace_seq_puts(s, " <= ");
1541 seq_print_ip_sym(s, entry->stack.caller[i], sym_flags);
1543 trace_seq_puts(s, "\n");
1546 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1551 static int print_trace_fmt(struct trace_iterator *iter)
1553 struct trace_seq *s = &iter->seq;
1554 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1555 struct trace_entry *entry;
1556 unsigned long usec_rem;
1557 unsigned long long t;
1566 comm = trace_find_cmdline(iter->ent->pid);
1568 t = ns2usecs(entry->t);
1569 usec_rem = do_div(t, 1000000ULL);
1570 secs = (unsigned long)t;
1572 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1575 ret = trace_seq_printf(s, "[%02d] ", iter->cpu);
1578 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1582 switch (entry->type) {
1584 ret = seq_print_ip_sym(s, entry->fn.ip, sym_flags);
1587 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
1588 entry->fn.parent_ip) {
1589 ret = trace_seq_printf(s, " <-");
1592 if (kretprobed(entry->fn.parent_ip))
1593 ret = trace_seq_puts(s, KRETPROBE_MSG);
1595 ret = seq_print_ip_sym(s, entry->fn.parent_ip,
1600 ret = trace_seq_printf(s, "\n");
1606 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1607 state_to_char[entry->ctx.prev_state] : 'X';
1608 T = entry->ctx.next_state < sizeof(state_to_char) ?
1609 state_to_char[entry->ctx.next_state] : 'X';
1610 ret = trace_seq_printf(s, " %5d:%3d:%c %s %5d:%3d:%c\n",
1611 entry->ctx.prev_pid,
1612 entry->ctx.prev_prio,
1614 entry->type == TRACE_CTX ? "==>" : " +",
1615 entry->ctx.next_pid,
1616 entry->ctx.next_prio,
1622 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1623 entry->special.arg1,
1624 entry->special.arg2,
1625 entry->special.arg3);
1630 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1632 ret = trace_seq_puts(s, " <= ");
1636 ret = seq_print_ip_sym(s, entry->stack.caller[i],
1641 ret = trace_seq_puts(s, "\n");
1649 static int print_raw_fmt(struct trace_iterator *iter)
1651 struct trace_seq *s = &iter->seq;
1652 struct trace_entry *entry;
1658 ret = trace_seq_printf(s, "%d %d %llu ",
1659 entry->pid, iter->cpu, entry->t);
1663 switch (entry->type) {
1665 ret = trace_seq_printf(s, "%x %x\n",
1666 entry->fn.ip, entry->fn.parent_ip);
1672 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1673 state_to_char[entry->ctx.prev_state] : 'X';
1674 T = entry->ctx.next_state < sizeof(state_to_char) ?
1675 state_to_char[entry->ctx.next_state] : 'X';
1676 if (entry->type == TRACE_WAKE)
1678 ret = trace_seq_printf(s, "%d %d %c %d %d %c\n",
1679 entry->ctx.prev_pid,
1680 entry->ctx.prev_prio,
1682 entry->ctx.next_pid,
1683 entry->ctx.next_prio,
1690 ret = trace_seq_printf(s, "# %ld %ld %ld\n",
1691 entry->special.arg1,
1692 entry->special.arg2,
1693 entry->special.arg3);
1701 #define SEQ_PUT_FIELD_RET(s, x) \
1703 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
1707 #define SEQ_PUT_HEX_FIELD_RET(s, x) \
1709 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
1713 static int print_hex_fmt(struct trace_iterator *iter)
1715 struct trace_seq *s = &iter->seq;
1716 unsigned char newline = '\n';
1717 struct trace_entry *entry;
1722 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1723 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1724 SEQ_PUT_HEX_FIELD_RET(s, entry->t);
1726 switch (entry->type) {
1728 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.ip);
1729 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1733 S = entry->ctx.prev_state < sizeof(state_to_char) ?
1734 state_to_char[entry->ctx.prev_state] : 'X';
1735 T = entry->ctx.next_state < sizeof(state_to_char) ?
1736 state_to_char[entry->ctx.next_state] : 'X';
1737 if (entry->type == TRACE_WAKE)
1739 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_pid);
1740 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.prev_prio);
1741 SEQ_PUT_HEX_FIELD_RET(s, S);
1742 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_pid);
1743 SEQ_PUT_HEX_FIELD_RET(s, entry->ctx.next_prio);
1744 SEQ_PUT_HEX_FIELD_RET(s, entry->fn.parent_ip);
1745 SEQ_PUT_HEX_FIELD_RET(s, T);
1749 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg1);
1750 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg2);
1751 SEQ_PUT_HEX_FIELD_RET(s, entry->special.arg3);
1754 SEQ_PUT_FIELD_RET(s, newline);
1759 static int print_bin_fmt(struct trace_iterator *iter)
1761 struct trace_seq *s = &iter->seq;
1762 struct trace_entry *entry;
1766 SEQ_PUT_FIELD_RET(s, entry->pid);
1767 SEQ_PUT_FIELD_RET(s, entry->cpu);
1768 SEQ_PUT_FIELD_RET(s, entry->t);
1770 switch (entry->type) {
1772 SEQ_PUT_FIELD_RET(s, entry->fn.ip);
1773 SEQ_PUT_FIELD_RET(s, entry->fn.parent_ip);
1776 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_pid);
1777 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_prio);
1778 SEQ_PUT_FIELD_RET(s, entry->ctx.prev_state);
1779 SEQ_PUT_FIELD_RET(s, entry->ctx.next_pid);
1780 SEQ_PUT_FIELD_RET(s, entry->ctx.next_prio);
1781 SEQ_PUT_FIELD_RET(s, entry->ctx.next_state);
1785 SEQ_PUT_FIELD_RET(s, entry->special.arg1);
1786 SEQ_PUT_FIELD_RET(s, entry->special.arg2);
1787 SEQ_PUT_FIELD_RET(s, entry->special.arg3);
1793 static int trace_empty(struct trace_iterator *iter)
1795 struct trace_array_cpu *data;
1798 for_each_tracing_cpu(cpu) {
1799 data = iter->tr->data[cpu];
1801 if (head_page(data) && data->trace_idx &&
1802 (data->trace_tail != data->trace_head ||
1803 data->trace_tail_idx != data->trace_head_idx))
1809 static int print_trace_line(struct trace_iterator *iter)
1811 if (iter->trace && iter->trace->print_line)
1812 return iter->trace->print_line(iter);
1814 if (trace_flags & TRACE_ITER_BIN)
1815 return print_bin_fmt(iter);
1817 if (trace_flags & TRACE_ITER_HEX)
1818 return print_hex_fmt(iter);
1820 if (trace_flags & TRACE_ITER_RAW)
1821 return print_raw_fmt(iter);
1823 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1824 return print_lat_fmt(iter, iter->idx, iter->cpu);
1826 return print_trace_fmt(iter);
1829 static int s_show(struct seq_file *m, void *v)
1831 struct trace_iterator *iter = v;
1833 if (iter->ent == NULL) {
1835 seq_printf(m, "# tracer: %s\n", iter->trace->name);
1838 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1839 /* print nothing if the buffers are empty */
1840 if (trace_empty(iter))
1842 print_trace_header(m, iter);
1843 if (!(trace_flags & TRACE_ITER_VERBOSE))
1844 print_lat_help_header(m);
1846 if (!(trace_flags & TRACE_ITER_VERBOSE))
1847 print_func_help_header(m);
1850 print_trace_line(iter);
1851 trace_print_seq(m, &iter->seq);
1857 static struct seq_operations tracer_seq_ops = {
1864 static struct trace_iterator *
1865 __tracing_open(struct inode *inode, struct file *file, int *ret)
1867 struct trace_iterator *iter;
1869 if (tracing_disabled) {
1874 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1880 mutex_lock(&trace_types_lock);
1881 if (current_trace && current_trace->print_max)
1884 iter->tr = inode->i_private;
1885 iter->trace = current_trace;
1888 /* TODO stop tracer */
1889 *ret = seq_open(file, &tracer_seq_ops);
1891 struct seq_file *m = file->private_data;
1894 /* stop the trace while dumping */
1898 if (iter->trace && iter->trace->open)
1899 iter->trace->open(iter);
1904 mutex_unlock(&trace_types_lock);
1910 int tracing_open_generic(struct inode *inode, struct file *filp)
1912 if (tracing_disabled)
1915 filp->private_data = inode->i_private;
1919 int tracing_release(struct inode *inode, struct file *file)
1921 struct seq_file *m = (struct seq_file *)file->private_data;
1922 struct trace_iterator *iter = m->private;
1924 mutex_lock(&trace_types_lock);
1925 if (iter->trace && iter->trace->close)
1926 iter->trace->close(iter);
1928 /* reenable tracing if it was previously enabled */
1931 mutex_unlock(&trace_types_lock);
1933 seq_release(inode, file);
1938 static int tracing_open(struct inode *inode, struct file *file)
1942 __tracing_open(inode, file, &ret);
1947 static int tracing_lt_open(struct inode *inode, struct file *file)
1949 struct trace_iterator *iter;
1952 iter = __tracing_open(inode, file, &ret);
1955 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1962 t_next(struct seq_file *m, void *v, loff_t *pos)
1964 struct tracer *t = m->private;
1976 static void *t_start(struct seq_file *m, loff_t *pos)
1978 struct tracer *t = m->private;
1981 mutex_lock(&trace_types_lock);
1982 for (; t && l < *pos; t = t_next(m, t, &l))
1988 static void t_stop(struct seq_file *m, void *p)
1990 mutex_unlock(&trace_types_lock);
1993 static int t_show(struct seq_file *m, void *v)
1995 struct tracer *t = v;
2000 seq_printf(m, "%s", t->name);
2009 static struct seq_operations show_traces_seq_ops = {
2016 static int show_traces_open(struct inode *inode, struct file *file)
2020 if (tracing_disabled)
2023 ret = seq_open(file, &show_traces_seq_ops);
2025 struct seq_file *m = file->private_data;
2026 m->private = trace_types;
2032 static struct file_operations tracing_fops = {
2033 .open = tracing_open,
2035 .llseek = seq_lseek,
2036 .release = tracing_release,
2039 static struct file_operations tracing_lt_fops = {
2040 .open = tracing_lt_open,
2042 .llseek = seq_lseek,
2043 .release = tracing_release,
2046 static struct file_operations show_traces_fops = {
2047 .open = show_traces_open,
2049 .release = seq_release,
2053 * Only trace on a CPU if the bitmask is set:
2055 static cpumask_t tracing_cpumask = CPU_MASK_ALL;
2058 * When tracing/tracing_cpu_mask is modified then this holds
2059 * the new bitmask we are about to install:
2061 static cpumask_t tracing_cpumask_new;
2064 * The tracer itself will not take this lock, but still we want
2065 * to provide a consistent cpumask to user-space:
2067 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2070 * Temporary storage for the character representation of the
2071 * CPU bitmask (and one more byte for the newline):
2073 static char mask_str[NR_CPUS + 1];
2076 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2077 size_t count, loff_t *ppos)
2081 mutex_lock(&tracing_cpumask_update_lock);
2083 len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2084 if (count - len < 2) {
2088 len += sprintf(mask_str + len, "\n");
2089 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2092 mutex_unlock(&tracing_cpumask_update_lock);
2098 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2099 size_t count, loff_t *ppos)
2103 mutex_lock(&tracing_cpumask_update_lock);
2104 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2108 raw_local_irq_disable();
2109 __raw_spin_lock(&ftrace_max_lock);
2110 for_each_tracing_cpu(cpu) {
2112 * Increase/decrease the disabled counter if we are
2113 * about to flip a bit in the cpumask:
2115 if (cpu_isset(cpu, tracing_cpumask) &&
2116 !cpu_isset(cpu, tracing_cpumask_new)) {
2117 atomic_inc(&global_trace.data[cpu]->disabled);
2119 if (!cpu_isset(cpu, tracing_cpumask) &&
2120 cpu_isset(cpu, tracing_cpumask_new)) {
2121 atomic_dec(&global_trace.data[cpu]->disabled);
2124 __raw_spin_unlock(&ftrace_max_lock);
2125 raw_local_irq_enable();
2127 tracing_cpumask = tracing_cpumask_new;
2129 mutex_unlock(&tracing_cpumask_update_lock);
2134 mutex_unlock(&tracing_cpumask_update_lock);
2139 static struct file_operations tracing_cpumask_fops = {
2140 .open = tracing_open_generic,
2141 .read = tracing_cpumask_read,
2142 .write = tracing_cpumask_write,
2146 tracing_iter_ctrl_read(struct file *filp, char __user *ubuf,
2147 size_t cnt, loff_t *ppos)
2154 /* calulate max size */
2155 for (i = 0; trace_options[i]; i++) {
2156 len += strlen(trace_options[i]);
2157 len += 3; /* "no" and space */
2160 /* +2 for \n and \0 */
2161 buf = kmalloc(len + 2, GFP_KERNEL);
2165 for (i = 0; trace_options[i]; i++) {
2166 if (trace_flags & (1 << i))
2167 r += sprintf(buf + r, "%s ", trace_options[i]);
2169 r += sprintf(buf + r, "no%s ", trace_options[i]);
2172 r += sprintf(buf + r, "\n");
2173 WARN_ON(r >= len + 2);
2175 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2183 tracing_iter_ctrl_write(struct file *filp, const char __user *ubuf,
2184 size_t cnt, loff_t *ppos)
2191 if (cnt >= sizeof(buf))
2194 if (copy_from_user(&buf, ubuf, cnt))
2199 if (strncmp(buf, "no", 2) == 0) {
2204 for (i = 0; trace_options[i]; i++) {
2205 int len = strlen(trace_options[i]);
2207 if (strncmp(cmp, trace_options[i], len) == 0) {
2209 trace_flags &= ~(1 << i);
2211 trace_flags |= (1 << i);
2216 * If no option could be set, return an error:
2218 if (!trace_options[i])
2226 static struct file_operations tracing_iter_fops = {
2227 .open = tracing_open_generic,
2228 .read = tracing_iter_ctrl_read,
2229 .write = tracing_iter_ctrl_write,
2232 static const char readme_msg[] =
2233 "tracing mini-HOWTO:\n\n"
2235 "# mount -t debugfs nodev /debug\n\n"
2236 "# cat /debug/tracing/available_tracers\n"
2237 "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2238 "# cat /debug/tracing/current_tracer\n"
2240 "# echo sched_switch > /debug/tracing/current_tracer\n"
2241 "# cat /debug/tracing/current_tracer\n"
2243 "# cat /debug/tracing/iter_ctrl\n"
2244 "noprint-parent nosym-offset nosym-addr noverbose\n"
2245 "# echo print-parent > /debug/tracing/iter_ctrl\n"
2246 "# echo 1 > /debug/tracing/tracing_enabled\n"
2247 "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2248 "echo 0 > /debug/tracing/tracing_enabled\n"
2252 tracing_readme_read(struct file *filp, char __user *ubuf,
2253 size_t cnt, loff_t *ppos)
2255 return simple_read_from_buffer(ubuf, cnt, ppos,
2256 readme_msg, strlen(readme_msg));
2259 static struct file_operations tracing_readme_fops = {
2260 .open = tracing_open_generic,
2261 .read = tracing_readme_read,
2265 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2266 size_t cnt, loff_t *ppos)
2268 struct trace_array *tr = filp->private_data;
2272 r = sprintf(buf, "%ld\n", tr->ctrl);
2273 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2277 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2278 size_t cnt, loff_t *ppos)
2280 struct trace_array *tr = filp->private_data;
2285 if (cnt >= sizeof(buf))
2288 if (copy_from_user(&buf, ubuf, cnt))
2293 ret = strict_strtoul(buf, 10, &val);
2299 mutex_lock(&trace_types_lock);
2300 if (tr->ctrl ^ val) {
2308 if (current_trace && current_trace->ctrl_update)
2309 current_trace->ctrl_update(tr);
2311 mutex_unlock(&trace_types_lock);
2319 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2320 size_t cnt, loff_t *ppos)
2322 char buf[max_tracer_type_len+2];
2325 mutex_lock(&trace_types_lock);
2327 r = sprintf(buf, "%s\n", current_trace->name);
2329 r = sprintf(buf, "\n");
2330 mutex_unlock(&trace_types_lock);
2332 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2336 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2337 size_t cnt, loff_t *ppos)
2339 struct trace_array *tr = &global_trace;
2341 char buf[max_tracer_type_len+1];
2344 if (cnt > max_tracer_type_len)
2345 cnt = max_tracer_type_len;
2347 if (copy_from_user(&buf, ubuf, cnt))
2352 /* strip ending whitespace. */
2353 for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2356 mutex_lock(&trace_types_lock);
2357 for (t = trace_types; t; t = t->next) {
2358 if (strcmp(t->name, buf) == 0)
2361 if (!t || t == current_trace)
2364 if (current_trace && current_trace->reset)
2365 current_trace->reset(tr);
2372 mutex_unlock(&trace_types_lock);
2380 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2381 size_t cnt, loff_t *ppos)
2383 unsigned long *ptr = filp->private_data;
2387 r = snprintf(buf, sizeof(buf), "%ld\n",
2388 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2389 if (r > sizeof(buf))
2391 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2395 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2396 size_t cnt, loff_t *ppos)
2398 long *ptr = filp->private_data;
2403 if (cnt >= sizeof(buf))
2406 if (copy_from_user(&buf, ubuf, cnt))
2411 ret = strict_strtoul(buf, 10, &val);
2420 static atomic_t tracing_reader;
2422 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2424 struct trace_iterator *iter;
2426 if (tracing_disabled)
2429 /* We only allow for reader of the pipe */
2430 if (atomic_inc_return(&tracing_reader) != 1) {
2431 atomic_dec(&tracing_reader);
2435 /* create a buffer to store the information to pass to userspace */
2436 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2440 mutex_lock(&trace_types_lock);
2441 iter->tr = &global_trace;
2442 iter->trace = current_trace;
2443 filp->private_data = iter;
2445 if (iter->trace->pipe_open)
2446 iter->trace->pipe_open(iter);
2447 mutex_unlock(&trace_types_lock);
2452 static int tracing_release_pipe(struct inode *inode, struct file *file)
2454 struct trace_iterator *iter = file->private_data;
2457 atomic_dec(&tracing_reader);
2463 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2465 struct trace_iterator *iter = filp->private_data;
2467 if (trace_flags & TRACE_ITER_BLOCK) {
2469 * Always select as readable when in blocking mode
2471 return POLLIN | POLLRDNORM;
2473 if (!trace_empty(iter))
2474 return POLLIN | POLLRDNORM;
2475 poll_wait(filp, &trace_wait, poll_table);
2476 if (!trace_empty(iter))
2477 return POLLIN | POLLRDNORM;
2487 tracing_read_pipe(struct file *filp, char __user *ubuf,
2488 size_t cnt, loff_t *ppos)
2490 struct trace_iterator *iter = filp->private_data;
2491 struct trace_array_cpu *data;
2492 static cpumask_t mask;
2493 unsigned long flags;
2494 #ifdef CONFIG_FTRACE
2500 /* return any leftover data */
2501 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2506 trace_seq_reset(&iter->seq);
2508 mutex_lock(&trace_types_lock);
2509 if (iter->trace->read) {
2510 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2515 while (trace_empty(iter)) {
2517 if ((filp->f_flags & O_NONBLOCK)) {
2523 * This is a make-shift waitqueue. The reason we don't use
2524 * an actual wait queue is because:
2525 * 1) we only ever have one waiter
2526 * 2) the tracing, traces all functions, we don't want
2527 * the overhead of calling wake_up and friends
2528 * (and tracing them too)
2529 * Anyway, this is really very primitive wakeup.
2531 set_current_state(TASK_INTERRUPTIBLE);
2532 iter->tr->waiter = current;
2534 mutex_unlock(&trace_types_lock);
2536 /* sleep for 100 msecs, and try again. */
2537 schedule_timeout(HZ/10);
2539 mutex_lock(&trace_types_lock);
2541 iter->tr->waiter = NULL;
2543 if (signal_pending(current)) {
2548 if (iter->trace != current_trace)
2552 * We block until we read something and tracing is disabled.
2553 * We still block if tracing is disabled, but we have never
2554 * read anything. This allows a user to cat this file, and
2555 * then enable tracing. But after we have read something,
2556 * we give an EOF when tracing is again disabled.
2558 * iter->pos will be 0 if we haven't read anything.
2560 if (!tracer_enabled && iter->pos)
2566 /* stop when tracing is finished */
2567 if (trace_empty(iter))
2570 if (cnt >= PAGE_SIZE)
2571 cnt = PAGE_SIZE - 1;
2573 /* reset all but tr, trace, and overruns */
2574 memset(&iter->seq, 0,
2575 sizeof(struct trace_iterator) -
2576 offsetof(struct trace_iterator, seq));
2580 * We need to stop all tracing on all CPUS to read the
2581 * the next buffer. This is a bit expensive, but is
2582 * not done often. We fill all what we can read,
2583 * and then release the locks again.
2587 local_irq_save(flags);
2588 #ifdef CONFIG_FTRACE
2589 ftrace_save = ftrace_enabled;
2593 for_each_tracing_cpu(cpu) {
2594 data = iter->tr->data[cpu];
2596 if (!head_page(data) || !data->trace_idx)
2599 atomic_inc(&data->disabled);
2603 for_each_cpu_mask(cpu, mask) {
2604 data = iter->tr->data[cpu];
2605 __raw_spin_lock(&data->lock);
2607 if (data->overrun > iter->last_overrun[cpu])
2608 iter->overrun[cpu] +=
2609 data->overrun - iter->last_overrun[cpu];
2610 iter->last_overrun[cpu] = data->overrun;
2613 while (find_next_entry_inc(iter) != NULL) {
2615 int len = iter->seq.len;
2617 ret = print_trace_line(iter);
2619 /* don't print partial lines */
2620 iter->seq.len = len;
2624 trace_consume(iter);
2626 if (iter->seq.len >= cnt)
2630 for_each_cpu_mask(cpu, mask) {
2631 data = iter->tr->data[cpu];
2632 __raw_spin_unlock(&data->lock);
2635 for_each_cpu_mask(cpu, mask) {
2636 data = iter->tr->data[cpu];
2637 atomic_dec(&data->disabled);
2639 #ifdef CONFIG_FTRACE
2640 ftrace_enabled = ftrace_save;
2642 local_irq_restore(flags);
2644 /* Now copy what we have to the user */
2645 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2646 if (iter->seq.readpos >= iter->seq.len)
2647 trace_seq_reset(&iter->seq);
2652 mutex_unlock(&trace_types_lock);
2658 tracing_entries_read(struct file *filp, char __user *ubuf,
2659 size_t cnt, loff_t *ppos)
2661 struct trace_array *tr = filp->private_data;
2665 r = sprintf(buf, "%lu\n", tr->entries);
2666 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2670 tracing_entries_write(struct file *filp, const char __user *ubuf,
2671 size_t cnt, loff_t *ppos)
2677 if (cnt >= sizeof(buf))
2680 if (copy_from_user(&buf, ubuf, cnt))
2685 ret = strict_strtoul(buf, 10, &val);
2689 /* must have at least 1 entry */
2693 mutex_lock(&trace_types_lock);
2695 if (current_trace != &no_tracer) {
2697 pr_info("ftrace: set current_tracer to none"
2698 " before modifying buffer size\n");
2702 if (val > global_trace.entries) {
2703 long pages_requested;
2704 unsigned long freeable_pages;
2706 /* make sure we have enough memory before mapping */
2708 (val + (ENTRIES_PER_PAGE-1)) / ENTRIES_PER_PAGE;
2710 /* account for each buffer (and max_tr) */
2711 pages_requested *= tracing_nr_buffers * 2;
2713 /* Check for overflow */
2714 if (pages_requested < 0) {
2719 freeable_pages = determine_dirtyable_memory();
2721 /* we only allow to request 1/4 of useable memory */
2722 if (pages_requested >
2723 ((freeable_pages + tracing_pages_allocated) / 4)) {
2728 while (global_trace.entries < val) {
2729 if (trace_alloc_page()) {
2733 /* double check that we don't go over the known pages */
2734 if (tracing_pages_allocated > pages_requested)
2739 /* include the number of entries in val (inc of page entries) */
2740 while (global_trace.entries > val + (ENTRIES_PER_PAGE - 1))
2744 /* check integrity */
2745 for_each_tracing_cpu(i)
2746 check_pages(global_trace.data[i]);
2750 /* If check pages failed, return ENOMEM */
2751 if (tracing_disabled)
2754 max_tr.entries = global_trace.entries;
2755 mutex_unlock(&trace_types_lock);
2760 static struct file_operations tracing_max_lat_fops = {
2761 .open = tracing_open_generic,
2762 .read = tracing_max_lat_read,
2763 .write = tracing_max_lat_write,
2766 static struct file_operations tracing_ctrl_fops = {
2767 .open = tracing_open_generic,
2768 .read = tracing_ctrl_read,
2769 .write = tracing_ctrl_write,
2772 static struct file_operations set_tracer_fops = {
2773 .open = tracing_open_generic,
2774 .read = tracing_set_trace_read,
2775 .write = tracing_set_trace_write,
2778 static struct file_operations tracing_pipe_fops = {
2779 .open = tracing_open_pipe,
2780 .poll = tracing_poll_pipe,
2781 .read = tracing_read_pipe,
2782 .release = tracing_release_pipe,
2785 static struct file_operations tracing_entries_fops = {
2786 .open = tracing_open_generic,
2787 .read = tracing_entries_read,
2788 .write = tracing_entries_write,
2791 #ifdef CONFIG_DYNAMIC_FTRACE
2794 tracing_read_long(struct file *filp, char __user *ubuf,
2795 size_t cnt, loff_t *ppos)
2797 unsigned long *p = filp->private_data;
2801 r = sprintf(buf, "%ld\n", *p);
2803 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2806 static struct file_operations tracing_read_long_fops = {
2807 .open = tracing_open_generic,
2808 .read = tracing_read_long,
2812 static struct dentry *d_tracer;
2814 struct dentry *tracing_init_dentry(void)
2821 d_tracer = debugfs_create_dir("tracing", NULL);
2823 if (!d_tracer && !once) {
2825 pr_warning("Could not create debugfs directory 'tracing'\n");
2832 #ifdef CONFIG_FTRACE_SELFTEST
2833 /* Let selftest have access to static functions in this file */
2834 #include "trace_selftest.c"
2837 static __init void tracer_init_debugfs(void)
2839 struct dentry *d_tracer;
2840 struct dentry *entry;
2842 d_tracer = tracing_init_dentry();
2844 entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2845 &global_trace, &tracing_ctrl_fops);
2847 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2849 entry = debugfs_create_file("iter_ctrl", 0644, d_tracer,
2850 NULL, &tracing_iter_fops);
2852 pr_warning("Could not create debugfs 'iter_ctrl' entry\n");
2854 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2855 NULL, &tracing_cpumask_fops);
2857 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2859 entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2860 &global_trace, &tracing_lt_fops);
2862 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2864 entry = debugfs_create_file("trace", 0444, d_tracer,
2865 &global_trace, &tracing_fops);
2867 pr_warning("Could not create debugfs 'trace' entry\n");
2869 entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2870 &global_trace, &show_traces_fops);
2872 pr_warning("Could not create debugfs 'trace' entry\n");
2874 entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2875 &global_trace, &set_tracer_fops);
2877 pr_warning("Could not create debugfs 'trace' entry\n");
2879 entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2880 &tracing_max_latency,
2881 &tracing_max_lat_fops);
2883 pr_warning("Could not create debugfs "
2884 "'tracing_max_latency' entry\n");
2886 entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2887 &tracing_thresh, &tracing_max_lat_fops);
2889 pr_warning("Could not create debugfs "
2890 "'tracing_threash' entry\n");
2891 entry = debugfs_create_file("README", 0644, d_tracer,
2892 NULL, &tracing_readme_fops);
2894 pr_warning("Could not create debugfs 'README' entry\n");
2896 entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2897 NULL, &tracing_pipe_fops);
2899 pr_warning("Could not create debugfs "
2900 "'tracing_threash' entry\n");
2902 entry = debugfs_create_file("trace_entries", 0644, d_tracer,
2903 &global_trace, &tracing_entries_fops);
2905 pr_warning("Could not create debugfs "
2906 "'tracing_threash' entry\n");
2908 #ifdef CONFIG_DYNAMIC_FTRACE
2909 entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2910 &ftrace_update_tot_cnt,
2911 &tracing_read_long_fops);
2913 pr_warning("Could not create debugfs "
2914 "'dyn_ftrace_total_info' entry\n");
2916 #ifdef CONFIG_SYSPROF_TRACER
2917 init_tracer_sysprof_debugfs(d_tracer);
2921 static int trace_alloc_page(void)
2923 struct trace_array_cpu *data;
2924 struct page *page, *tmp;
2927 unsigned pages_allocated = 0;
2930 /* first allocate a page for each CPU */
2931 for_each_tracing_cpu(i) {
2932 array = (void *)__get_free_page(GFP_KERNEL);
2933 if (array == NULL) {
2934 printk(KERN_ERR "tracer: failed to allocate page"
2935 "for trace buffer!\n");
2940 page = virt_to_page(array);
2941 list_add(&page->lru, &pages);
2943 /* Only allocate if we are actually using the max trace */
2944 #ifdef CONFIG_TRACER_MAX_TRACE
2945 array = (void *)__get_free_page(GFP_KERNEL);
2946 if (array == NULL) {
2947 printk(KERN_ERR "tracer: failed to allocate page"
2948 "for trace buffer!\n");
2952 page = virt_to_page(array);
2953 list_add(&page->lru, &pages);
2957 /* Now that we successfully allocate a page per CPU, add them */
2958 for_each_tracing_cpu(i) {
2959 data = global_trace.data[i];
2960 page = list_entry(pages.next, struct page, lru);
2961 list_del_init(&page->lru);
2962 list_add_tail(&page->lru, &data->trace_pages);
2965 #ifdef CONFIG_TRACER_MAX_TRACE
2966 data = max_tr.data[i];
2967 page = list_entry(pages.next, struct page, lru);
2968 list_del_init(&page->lru);
2969 list_add_tail(&page->lru, &data->trace_pages);
2973 tracing_pages_allocated += pages_allocated;
2974 global_trace.entries += ENTRIES_PER_PAGE;
2979 list_for_each_entry_safe(page, tmp, &pages, lru) {
2980 list_del_init(&page->lru);
2986 static int trace_free_page(void)
2988 struct trace_array_cpu *data;
2990 struct list_head *p;
2994 /* free one page from each buffer */
2995 for_each_tracing_cpu(i) {
2996 data = global_trace.data[i];
2997 p = data->trace_pages.next;
2998 if (p == &data->trace_pages) {
2999 /* should never happen */
3001 tracing_disabled = 1;
3005 page = list_entry(p, struct page, lru);
3007 list_del(&page->lru);
3008 tracing_pages_allocated--;
3009 tracing_pages_allocated--;
3012 tracing_reset(data);
3014 #ifdef CONFIG_TRACER_MAX_TRACE
3015 data = max_tr.data[i];
3016 p = data->trace_pages.next;
3017 if (p == &data->trace_pages) {
3018 /* should never happen */
3020 tracing_disabled = 1;
3024 page = list_entry(p, struct page, lru);
3026 list_del(&page->lru);
3029 tracing_reset(data);
3032 global_trace.entries -= ENTRIES_PER_PAGE;
3037 __init static int tracer_alloc_buffers(void)
3039 struct trace_array_cpu *data;
3046 /* TODO: make the number of buffers hot pluggable with CPUS */
3047 tracing_nr_buffers = num_possible_cpus();
3048 tracing_buffer_mask = cpu_possible_map;
3050 /* Allocate the first page for all buffers */
3051 for_each_tracing_cpu(i) {
3052 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3053 max_tr.data[i] = &per_cpu(max_data, i);
3055 array = (void *)__get_free_page(GFP_KERNEL);
3056 if (array == NULL) {
3057 printk(KERN_ERR "tracer: failed to allocate page"
3058 "for trace buffer!\n");
3062 /* set the array to the list */
3063 INIT_LIST_HEAD(&data->trace_pages);
3064 page = virt_to_page(array);
3065 list_add(&page->lru, &data->trace_pages);
3066 /* use the LRU flag to differentiate the two buffers */
3069 data->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
3070 max_tr.data[i]->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
3072 /* Only allocate if we are actually using the max trace */
3073 #ifdef CONFIG_TRACER_MAX_TRACE
3074 array = (void *)__get_free_page(GFP_KERNEL);
3075 if (array == NULL) {
3076 printk(KERN_ERR "tracer: failed to allocate page"
3077 "for trace buffer!\n");
3081 INIT_LIST_HEAD(&max_tr.data[i]->trace_pages);
3082 page = virt_to_page(array);
3083 list_add(&page->lru, &max_tr.data[i]->trace_pages);
3089 * Since we allocate by orders of pages, we may be able to
3092 global_trace.entries = ENTRIES_PER_PAGE;
3095 while (global_trace.entries < trace_nr_entries) {
3096 if (trace_alloc_page())
3100 max_tr.entries = global_trace.entries;
3102 pr_info("tracer: %d pages allocated for %ld entries of %ld bytes\n",
3103 pages, trace_nr_entries, (long)TRACE_ENTRY_SIZE);
3104 pr_info(" actual entries %ld\n", global_trace.entries);
3106 tracer_init_debugfs();
3108 trace_init_cmdlines();
3110 register_tracer(&no_tracer);
3111 current_trace = &no_tracer;
3113 /* All seems OK, enable tracing */
3114 global_trace.ctrl = tracer_enabled;
3115 tracing_disabled = 0;
3120 for (i-- ; i >= 0; i--) {
3121 struct page *page, *tmp;
3122 struct trace_array_cpu *data = global_trace.data[i];
3125 list_for_each_entry_safe(page, tmp,
3126 &data->trace_pages, lru) {
3127 list_del_init(&page->lru);
3132 #ifdef CONFIG_TRACER_MAX_TRACE
3133 data = max_tr.data[i];
3135 list_for_each_entry_safe(page, tmp,
3136 &data->trace_pages, lru) {
3137 list_del_init(&page->lru);
3145 fs_initcall(tracer_alloc_buffers);