4 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
6 #include <linux/ring_buffer.h>
7 #include <linux/ftrace_irq.h>
8 #include <linux/spinlock.h>
9 #include <linux/debugfs.h>
10 #include <linux/uaccess.h>
11 #include <linux/module.h>
12 #include <linux/percpu.h>
13 #include <linux/mutex.h>
14 #include <linux/sched.h> /* used for sched_clock() (for now) */
15 #include <linux/init.h>
16 #include <linux/hash.h>
17 #include <linux/list.h>
23 * Since the write to the buffer is still not fully lockless,
24 * we must be careful with NMIs. The locks in the writers
25 * are taken when a write crosses to a new page. The locks
26 * protect against races with the readers (this will soon
27 * be fixed with a lockless solution).
29 * Because we can not protect against NMIs, and we want to
30 * keep traces reentrant, we need to manage what happens
31 * when we are in an NMI.
33 static DEFINE_PER_CPU(int, rb_in_nmi);
35 void ftrace_nmi_enter(void)
37 __get_cpu_var(rb_in_nmi)++;
38 /* call arch specific handler too */
39 arch_ftrace_nmi_enter();
42 void ftrace_nmi_exit(void)
44 arch_ftrace_nmi_exit();
45 __get_cpu_var(rb_in_nmi)--;
46 /* NMIs are not recursive */
47 WARN_ON_ONCE(__get_cpu_var(rb_in_nmi));
52 * A fast way to enable or disable all ring buffers is to
53 * call tracing_on or tracing_off. Turning off the ring buffers
54 * prevents all ring buffers from being recorded to.
55 * Turning this switch on, makes it OK to write to the
56 * ring buffer, if the ring buffer is enabled itself.
58 * There's three layers that must be on in order to write
61 * 1) This global flag must be set.
62 * 2) The ring buffer must be enabled for recording.
63 * 3) The per cpu buffer must be enabled for recording.
65 * In case of an anomaly, this global flag has a bit set that
66 * will permantly disable all ring buffers.
70 * Global flag to disable all recording to ring buffers
71 * This has two bits: ON, DISABLED
75 * 0 0 : ring buffers are off
76 * 1 0 : ring buffers are on
77 * X 1 : ring buffers are permanently disabled
81 RB_BUFFERS_ON_BIT = 0,
82 RB_BUFFERS_DISABLED_BIT = 1,
86 RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT,
87 RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT,
90 static long ring_buffer_flags __read_mostly = RB_BUFFERS_ON;
93 * tracing_on - enable all tracing buffers
95 * This function enables all tracing buffers that may have been
96 * disabled with tracing_off.
100 set_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags);
102 EXPORT_SYMBOL_GPL(tracing_on);
105 * tracing_off - turn off all tracing buffers
107 * This function stops all tracing buffers from recording data.
108 * It does not disable any overhead the tracers themselves may
109 * be causing. This function simply causes all recording to
110 * the ring buffers to fail.
112 void tracing_off(void)
114 clear_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags);
116 EXPORT_SYMBOL_GPL(tracing_off);
119 * tracing_off_permanent - permanently disable ring buffers
121 * This function, once called, will disable all ring buffers
124 void tracing_off_permanent(void)
126 set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags);
131 /* Up this if you want to test the TIME_EXTENTS and normalization */
132 #define DEBUG_SHIFT 0
135 u64 ring_buffer_time_stamp(int cpu)
139 preempt_disable_notrace();
140 /* shift to debug/test normalization and TIME_EXTENTS */
141 time = sched_clock() << DEBUG_SHIFT;
142 preempt_enable_no_resched_notrace();
146 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
148 void ring_buffer_normalize_time_stamp(int cpu, u64 *ts)
150 /* Just stupid testing the normalize function and deltas */
153 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
155 #define RB_EVNT_HDR_SIZE (sizeof(struct ring_buffer_event))
156 #define RB_ALIGNMENT 4U
157 #define RB_MAX_SMALL_DATA 28
160 RB_LEN_TIME_EXTEND = 8,
161 RB_LEN_TIME_STAMP = 16,
164 /* inline for ring buffer fast paths */
166 rb_event_length(struct ring_buffer_event *event)
170 switch (event->type) {
171 case RINGBUF_TYPE_PADDING:
175 case RINGBUF_TYPE_TIME_EXTEND:
176 return RB_LEN_TIME_EXTEND;
178 case RINGBUF_TYPE_TIME_STAMP:
179 return RB_LEN_TIME_STAMP;
181 case RINGBUF_TYPE_DATA:
183 length = event->len * RB_ALIGNMENT;
185 length = event->array[0];
186 return length + RB_EVNT_HDR_SIZE;
195 * ring_buffer_event_length - return the length of the event
196 * @event: the event to get the length of
198 unsigned ring_buffer_event_length(struct ring_buffer_event *event)
200 unsigned length = rb_event_length(event);
201 if (event->type != RINGBUF_TYPE_DATA)
203 length -= RB_EVNT_HDR_SIZE;
204 if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0]))
205 length -= sizeof(event->array[0]);
208 EXPORT_SYMBOL_GPL(ring_buffer_event_length);
210 /* inline for ring buffer fast paths */
212 rb_event_data(struct ring_buffer_event *event)
214 BUG_ON(event->type != RINGBUF_TYPE_DATA);
215 /* If length is in len field, then array[0] has the data */
217 return (void *)&event->array[0];
218 /* Otherwise length is in array[0] and array[1] has the data */
219 return (void *)&event->array[1];
223 * ring_buffer_event_data - return the data of the event
224 * @event: the event to get the data from
226 void *ring_buffer_event_data(struct ring_buffer_event *event)
228 return rb_event_data(event);
230 EXPORT_SYMBOL_GPL(ring_buffer_event_data);
232 #define for_each_buffer_cpu(buffer, cpu) \
233 for_each_cpu(cpu, buffer->cpumask)
236 #define TS_MASK ((1ULL << TS_SHIFT) - 1)
237 #define TS_DELTA_TEST (~TS_MASK)
239 struct buffer_data_page {
240 u64 time_stamp; /* page time stamp */
241 local_t commit; /* write commited index */
242 unsigned char data[]; /* data of buffer page */
246 local_t write; /* index for next write */
247 unsigned read; /* index for next read */
248 struct list_head list; /* list of free pages */
249 struct buffer_data_page *page; /* Actual data page */
252 static void rb_init_page(struct buffer_data_page *bpage)
254 local_set(&bpage->commit, 0);
258 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
261 static void free_buffer_page(struct buffer_page *bpage)
263 free_page((unsigned long)bpage->page);
268 * We need to fit the time_stamp delta into 27 bits.
270 static inline int test_time_stamp(u64 delta)
272 if (delta & TS_DELTA_TEST)
277 #define BUF_PAGE_SIZE (PAGE_SIZE - offsetof(struct buffer_data_page, data))
280 * head_page == tail_page && head == tail then buffer is empty.
282 struct ring_buffer_per_cpu {
284 struct ring_buffer *buffer;
285 spinlock_t reader_lock; /* serialize readers */
287 struct lock_class_key lock_key;
288 struct list_head pages;
289 struct buffer_page *head_page; /* read from head */
290 struct buffer_page *tail_page; /* write to tail */
291 struct buffer_page *commit_page; /* commited pages */
292 struct buffer_page *reader_page;
293 unsigned long overrun;
294 unsigned long entries;
297 atomic_t record_disabled;
304 cpumask_var_t cpumask;
305 atomic_t record_disabled;
309 struct ring_buffer_per_cpu **buffers;
312 struct ring_buffer_iter {
313 struct ring_buffer_per_cpu *cpu_buffer;
315 struct buffer_page *head_page;
319 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
320 #define RB_WARN_ON(buffer, cond) \
322 int _____ret = unlikely(cond); \
324 atomic_inc(&buffer->record_disabled); \
331 * check_pages - integrity check of buffer pages
332 * @cpu_buffer: CPU buffer with pages to test
334 * As a safty measure we check to make sure the data pages have not
337 static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
339 struct list_head *head = &cpu_buffer->pages;
340 struct buffer_page *bpage, *tmp;
342 if (RB_WARN_ON(cpu_buffer, head->next->prev != head))
344 if (RB_WARN_ON(cpu_buffer, head->prev->next != head))
347 list_for_each_entry_safe(bpage, tmp, head, list) {
348 if (RB_WARN_ON(cpu_buffer,
349 bpage->list.next->prev != &bpage->list))
351 if (RB_WARN_ON(cpu_buffer,
352 bpage->list.prev->next != &bpage->list))
359 static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
362 struct list_head *head = &cpu_buffer->pages;
363 struct buffer_page *bpage, *tmp;
368 for (i = 0; i < nr_pages; i++) {
369 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
370 GFP_KERNEL, cpu_to_node(cpu_buffer->cpu));
373 list_add(&bpage->list, &pages);
375 addr = __get_free_page(GFP_KERNEL);
378 bpage->page = (void *)addr;
379 rb_init_page(bpage->page);
382 list_splice(&pages, head);
384 rb_check_pages(cpu_buffer);
389 list_for_each_entry_safe(bpage, tmp, &pages, list) {
390 list_del_init(&bpage->list);
391 free_buffer_page(bpage);
396 static struct ring_buffer_per_cpu *
397 rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu)
399 struct ring_buffer_per_cpu *cpu_buffer;
400 struct buffer_page *bpage;
404 cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
405 GFP_KERNEL, cpu_to_node(cpu));
409 cpu_buffer->cpu = cpu;
410 cpu_buffer->buffer = buffer;
411 spin_lock_init(&cpu_buffer->reader_lock);
412 cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
413 INIT_LIST_HEAD(&cpu_buffer->pages);
415 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
416 GFP_KERNEL, cpu_to_node(cpu));
418 goto fail_free_buffer;
420 cpu_buffer->reader_page = bpage;
421 addr = __get_free_page(GFP_KERNEL);
423 goto fail_free_reader;
424 bpage->page = (void *)addr;
425 rb_init_page(bpage->page);
427 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
429 ret = rb_allocate_pages(cpu_buffer, buffer->pages);
431 goto fail_free_reader;
433 cpu_buffer->head_page
434 = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
435 cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
440 free_buffer_page(cpu_buffer->reader_page);
447 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
449 struct list_head *head = &cpu_buffer->pages;
450 struct buffer_page *bpage, *tmp;
452 list_del_init(&cpu_buffer->reader_page->list);
453 free_buffer_page(cpu_buffer->reader_page);
455 list_for_each_entry_safe(bpage, tmp, head, list) {
456 list_del_init(&bpage->list);
457 free_buffer_page(bpage);
463 * Causes compile errors if the struct buffer_page gets bigger
464 * than the struct page.
466 extern int ring_buffer_page_too_big(void);
469 * ring_buffer_alloc - allocate a new ring_buffer
470 * @size: the size in bytes per cpu that is needed.
471 * @flags: attributes to set for the ring buffer.
473 * Currently the only flag that is available is the RB_FL_OVERWRITE
474 * flag. This flag means that the buffer will overwrite old data
475 * when the buffer wraps. If this flag is not set, the buffer will
476 * drop data when the tail hits the head.
478 struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags)
480 struct ring_buffer *buffer;
484 /* Paranoid! Optimizes out when all is well */
485 if (sizeof(struct buffer_page) > sizeof(struct page))
486 ring_buffer_page_too_big();
489 /* keep it in its own cache line */
490 buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
495 if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL))
496 goto fail_free_buffer;
498 buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
499 buffer->flags = flags;
501 /* need at least two pages */
502 if (buffer->pages == 1)
505 cpumask_copy(buffer->cpumask, cpu_possible_mask);
506 buffer->cpus = nr_cpu_ids;
508 bsize = sizeof(void *) * nr_cpu_ids;
509 buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
511 if (!buffer->buffers)
512 goto fail_free_cpumask;
514 for_each_buffer_cpu(buffer, cpu) {
515 buffer->buffers[cpu] =
516 rb_allocate_cpu_buffer(buffer, cpu);
517 if (!buffer->buffers[cpu])
518 goto fail_free_buffers;
521 mutex_init(&buffer->mutex);
526 for_each_buffer_cpu(buffer, cpu) {
527 if (buffer->buffers[cpu])
528 rb_free_cpu_buffer(buffer->buffers[cpu]);
530 kfree(buffer->buffers);
533 free_cpumask_var(buffer->cpumask);
539 EXPORT_SYMBOL_GPL(ring_buffer_alloc);
542 * ring_buffer_free - free a ring buffer.
543 * @buffer: the buffer to free.
546 ring_buffer_free(struct ring_buffer *buffer)
550 for_each_buffer_cpu(buffer, cpu)
551 rb_free_cpu_buffer(buffer->buffers[cpu]);
553 free_cpumask_var(buffer->cpumask);
557 EXPORT_SYMBOL_GPL(ring_buffer_free);
559 static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
562 rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages)
564 struct buffer_page *bpage;
568 atomic_inc(&cpu_buffer->record_disabled);
571 for (i = 0; i < nr_pages; i++) {
572 if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages)))
574 p = cpu_buffer->pages.next;
575 bpage = list_entry(p, struct buffer_page, list);
576 list_del_init(&bpage->list);
577 free_buffer_page(bpage);
579 if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages)))
582 rb_reset_cpu(cpu_buffer);
584 rb_check_pages(cpu_buffer);
586 atomic_dec(&cpu_buffer->record_disabled);
591 rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer,
592 struct list_head *pages, unsigned nr_pages)
594 struct buffer_page *bpage;
598 atomic_inc(&cpu_buffer->record_disabled);
601 for (i = 0; i < nr_pages; i++) {
602 if (RB_WARN_ON(cpu_buffer, list_empty(pages)))
605 bpage = list_entry(p, struct buffer_page, list);
606 list_del_init(&bpage->list);
607 list_add_tail(&bpage->list, &cpu_buffer->pages);
609 rb_reset_cpu(cpu_buffer);
611 rb_check_pages(cpu_buffer);
613 atomic_dec(&cpu_buffer->record_disabled);
617 * ring_buffer_resize - resize the ring buffer
618 * @buffer: the buffer to resize.
619 * @size: the new size.
621 * The tracer is responsible for making sure that the buffer is
622 * not being used while changing the size.
623 * Note: We may be able to change the above requirement by using
624 * RCU synchronizations.
626 * Minimum size is 2 * BUF_PAGE_SIZE.
628 * Returns -1 on failure.
630 int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
632 struct ring_buffer_per_cpu *cpu_buffer;
633 unsigned nr_pages, rm_pages, new_pages;
634 struct buffer_page *bpage, *tmp;
635 unsigned long buffer_size;
641 * Always succeed at resizing a non-existent buffer:
646 size = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
647 size *= BUF_PAGE_SIZE;
648 buffer_size = buffer->pages * BUF_PAGE_SIZE;
650 /* we need a minimum of two pages */
651 if (size < BUF_PAGE_SIZE * 2)
652 size = BUF_PAGE_SIZE * 2;
654 if (size == buffer_size)
657 mutex_lock(&buffer->mutex);
659 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
661 if (size < buffer_size) {
663 /* easy case, just free pages */
664 if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) {
665 mutex_unlock(&buffer->mutex);
669 rm_pages = buffer->pages - nr_pages;
671 for_each_buffer_cpu(buffer, cpu) {
672 cpu_buffer = buffer->buffers[cpu];
673 rb_remove_pages(cpu_buffer, rm_pages);
679 * This is a bit more difficult. We only want to add pages
680 * when we can allocate enough for all CPUs. We do this
681 * by allocating all the pages and storing them on a local
682 * link list. If we succeed in our allocation, then we
683 * add these pages to the cpu_buffers. Otherwise we just free
684 * them all and return -ENOMEM;
686 if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) {
687 mutex_unlock(&buffer->mutex);
691 new_pages = nr_pages - buffer->pages;
693 for_each_buffer_cpu(buffer, cpu) {
694 for (i = 0; i < new_pages; i++) {
695 bpage = kzalloc_node(ALIGN(sizeof(*bpage),
697 GFP_KERNEL, cpu_to_node(cpu));
700 list_add(&bpage->list, &pages);
701 addr = __get_free_page(GFP_KERNEL);
704 bpage->page = (void *)addr;
705 rb_init_page(bpage->page);
709 for_each_buffer_cpu(buffer, cpu) {
710 cpu_buffer = buffer->buffers[cpu];
711 rb_insert_pages(cpu_buffer, &pages, new_pages);
714 if (RB_WARN_ON(buffer, !list_empty(&pages))) {
715 mutex_unlock(&buffer->mutex);
720 buffer->pages = nr_pages;
721 mutex_unlock(&buffer->mutex);
726 list_for_each_entry_safe(bpage, tmp, &pages, list) {
727 list_del_init(&bpage->list);
728 free_buffer_page(bpage);
730 mutex_unlock(&buffer->mutex);
733 EXPORT_SYMBOL_GPL(ring_buffer_resize);
735 static inline int rb_null_event(struct ring_buffer_event *event)
737 return event->type == RINGBUF_TYPE_PADDING;
741 __rb_data_page_index(struct buffer_data_page *bpage, unsigned index)
743 return bpage->data + index;
746 static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
748 return bpage->page->data + index;
751 static inline struct ring_buffer_event *
752 rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
754 return __rb_page_index(cpu_buffer->reader_page,
755 cpu_buffer->reader_page->read);
758 static inline struct ring_buffer_event *
759 rb_head_event(struct ring_buffer_per_cpu *cpu_buffer)
761 return __rb_page_index(cpu_buffer->head_page,
762 cpu_buffer->head_page->read);
765 static inline struct ring_buffer_event *
766 rb_iter_head_event(struct ring_buffer_iter *iter)
768 return __rb_page_index(iter->head_page, iter->head);
771 static inline unsigned rb_page_write(struct buffer_page *bpage)
773 return local_read(&bpage->write);
776 static inline unsigned rb_page_commit(struct buffer_page *bpage)
778 return local_read(&bpage->page->commit);
781 /* Size is determined by what has been commited */
782 static inline unsigned rb_page_size(struct buffer_page *bpage)
784 return rb_page_commit(bpage);
787 static inline unsigned
788 rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
790 return rb_page_commit(cpu_buffer->commit_page);
793 static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer)
795 return rb_page_commit(cpu_buffer->head_page);
799 * When the tail hits the head and the buffer is in overwrite mode,
800 * the head jumps to the next page and all content on the previous
801 * page is discarded. But before doing so, we update the overrun
802 * variable of the buffer.
804 static void rb_update_overflow(struct ring_buffer_per_cpu *cpu_buffer)
806 struct ring_buffer_event *event;
809 for (head = 0; head < rb_head_size(cpu_buffer);
810 head += rb_event_length(event)) {
812 event = __rb_page_index(cpu_buffer->head_page, head);
813 if (RB_WARN_ON(cpu_buffer, rb_null_event(event)))
815 /* Only count data entries */
816 if (event->type != RINGBUF_TYPE_DATA)
818 cpu_buffer->overrun++;
819 cpu_buffer->entries--;
823 static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer,
824 struct buffer_page **bpage)
826 struct list_head *p = (*bpage)->list.next;
828 if (p == &cpu_buffer->pages)
831 *bpage = list_entry(p, struct buffer_page, list);
834 static inline unsigned
835 rb_event_index(struct ring_buffer_event *event)
837 unsigned long addr = (unsigned long)event;
839 return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE);
843 rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
844 struct ring_buffer_event *event)
846 unsigned long addr = (unsigned long)event;
849 index = rb_event_index(event);
852 return cpu_buffer->commit_page->page == (void *)addr &&
853 rb_commit_index(cpu_buffer) == index;
857 rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer,
858 struct ring_buffer_event *event)
860 unsigned long addr = (unsigned long)event;
863 index = rb_event_index(event);
866 while (cpu_buffer->commit_page->page != (void *)addr) {
867 if (RB_WARN_ON(cpu_buffer,
868 cpu_buffer->commit_page == cpu_buffer->tail_page))
870 cpu_buffer->commit_page->page->commit =
871 cpu_buffer->commit_page->write;
872 rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
873 cpu_buffer->write_stamp =
874 cpu_buffer->commit_page->page->time_stamp;
877 /* Now set the commit to the event's index */
878 local_set(&cpu_buffer->commit_page->page->commit, index);
882 rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
885 * We only race with interrupts and NMIs on this CPU.
886 * If we own the commit event, then we can commit
887 * all others that interrupted us, since the interruptions
888 * are in stack format (they finish before they come
889 * back to us). This allows us to do a simple loop to
890 * assign the commit to the tail.
893 while (cpu_buffer->commit_page != cpu_buffer->tail_page) {
894 cpu_buffer->commit_page->page->commit =
895 cpu_buffer->commit_page->write;
896 rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
897 cpu_buffer->write_stamp =
898 cpu_buffer->commit_page->page->time_stamp;
899 /* add barrier to keep gcc from optimizing too much */
902 while (rb_commit_index(cpu_buffer) !=
903 rb_page_write(cpu_buffer->commit_page)) {
904 cpu_buffer->commit_page->page->commit =
905 cpu_buffer->commit_page->write;
909 /* again, keep gcc from optimizing */
913 * If an interrupt came in just after the first while loop
914 * and pushed the tail page forward, we will be left with
915 * a dangling commit that will never go forward.
917 if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page))
921 static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
923 cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp;
924 cpu_buffer->reader_page->read = 0;
927 static void rb_inc_iter(struct ring_buffer_iter *iter)
929 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
932 * The iterator could be on the reader page (it starts there).
933 * But the head could have moved, since the reader was
934 * found. Check for this case and assign the iterator
935 * to the head page instead of next.
937 if (iter->head_page == cpu_buffer->reader_page)
938 iter->head_page = cpu_buffer->head_page;
940 rb_inc_page(cpu_buffer, &iter->head_page);
942 iter->read_stamp = iter->head_page->page->time_stamp;
947 * ring_buffer_update_event - update event type and data
948 * @event: the even to update
949 * @type: the type of event
950 * @length: the size of the event field in the ring buffer
952 * Update the type and data fields of the event. The length
953 * is the actual size that is written to the ring buffer,
954 * and with this, we can determine what to place into the
958 rb_update_event(struct ring_buffer_event *event,
959 unsigned type, unsigned length)
965 case RINGBUF_TYPE_PADDING:
968 case RINGBUF_TYPE_TIME_EXTEND:
969 event->len = DIV_ROUND_UP(RB_LEN_TIME_EXTEND, RB_ALIGNMENT);
972 case RINGBUF_TYPE_TIME_STAMP:
973 event->len = DIV_ROUND_UP(RB_LEN_TIME_STAMP, RB_ALIGNMENT);
976 case RINGBUF_TYPE_DATA:
977 length -= RB_EVNT_HDR_SIZE;
978 if (length > RB_MAX_SMALL_DATA) {
980 event->array[0] = length;
982 event->len = DIV_ROUND_UP(length, RB_ALIGNMENT);
989 static unsigned rb_calculate_event_length(unsigned length)
991 struct ring_buffer_event event; /* Used only for sizeof array */
993 /* zero length can cause confusions */
997 if (length > RB_MAX_SMALL_DATA)
998 length += sizeof(event.array[0]);
1000 length += RB_EVNT_HDR_SIZE;
1001 length = ALIGN(length, RB_ALIGNMENT);
1006 static struct ring_buffer_event *
1007 __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
1008 unsigned type, unsigned long length, u64 *ts)
1010 struct buffer_page *tail_page, *head_page, *reader_page, *commit_page;
1011 unsigned long tail, write;
1012 struct ring_buffer *buffer = cpu_buffer->buffer;
1013 struct ring_buffer_event *event;
1014 unsigned long flags;
1015 bool lock_taken = false;
1017 commit_page = cpu_buffer->commit_page;
1018 /* we just need to protect against interrupts */
1020 tail_page = cpu_buffer->tail_page;
1021 write = local_add_return(length, &tail_page->write);
1022 tail = write - length;
1024 /* See if we shot pass the end of this buffer page */
1025 if (write > BUF_PAGE_SIZE) {
1026 struct buffer_page *next_page = tail_page;
1028 local_irq_save(flags);
1030 * NMIs can happen after we take the lock.
1031 * If we are in an NMI, only take the lock
1032 * if it is not already taken. Otherwise
1035 if (unlikely(__get_cpu_var(rb_in_nmi))) {
1036 if (!__raw_spin_trylock(&cpu_buffer->lock))
1039 __raw_spin_lock(&cpu_buffer->lock);
1043 rb_inc_page(cpu_buffer, &next_page);
1045 head_page = cpu_buffer->head_page;
1046 reader_page = cpu_buffer->reader_page;
1048 /* we grabbed the lock before incrementing */
1049 if (RB_WARN_ON(cpu_buffer, next_page == reader_page))
1053 * If for some reason, we had an interrupt storm that made
1054 * it all the way around the buffer, bail, and warn
1057 if (unlikely(next_page == commit_page)) {
1062 if (next_page == head_page) {
1063 if (!(buffer->flags & RB_FL_OVERWRITE))
1066 /* tail_page has not moved yet? */
1067 if (tail_page == cpu_buffer->tail_page) {
1068 /* count overflows */
1069 rb_update_overflow(cpu_buffer);
1071 rb_inc_page(cpu_buffer, &head_page);
1072 cpu_buffer->head_page = head_page;
1073 cpu_buffer->head_page->read = 0;
1078 * If the tail page is still the same as what we think
1079 * it is, then it is up to us to update the tail
1082 if (tail_page == cpu_buffer->tail_page) {
1083 local_set(&next_page->write, 0);
1084 local_set(&next_page->page->commit, 0);
1085 cpu_buffer->tail_page = next_page;
1087 /* reread the time stamp */
1088 *ts = ring_buffer_time_stamp(cpu_buffer->cpu);
1089 cpu_buffer->tail_page->page->time_stamp = *ts;
1093 * The actual tail page has moved forward.
1095 if (tail < BUF_PAGE_SIZE) {
1096 /* Mark the rest of the page with padding */
1097 event = __rb_page_index(tail_page, tail);
1098 event->type = RINGBUF_TYPE_PADDING;
1101 if (tail <= BUF_PAGE_SIZE)
1102 /* Set the write back to the previous setting */
1103 local_set(&tail_page->write, tail);
1106 * If this was a commit entry that failed,
1107 * increment that too
1109 if (tail_page == cpu_buffer->commit_page &&
1110 tail == rb_commit_index(cpu_buffer)) {
1111 rb_set_commit_to_write(cpu_buffer);
1114 __raw_spin_unlock(&cpu_buffer->lock);
1115 local_irq_restore(flags);
1117 /* fail and let the caller try again */
1118 return ERR_PTR(-EAGAIN);
1121 /* We reserved something on the buffer */
1123 if (RB_WARN_ON(cpu_buffer, write > BUF_PAGE_SIZE))
1126 event = __rb_page_index(tail_page, tail);
1127 rb_update_event(event, type, length);
1130 * If this is a commit and the tail is zero, then update
1131 * this page's time stamp.
1133 if (!tail && rb_is_commit(cpu_buffer, event))
1134 cpu_buffer->commit_page->page->time_stamp = *ts;
1140 if (tail <= BUF_PAGE_SIZE)
1141 local_set(&tail_page->write, tail);
1143 if (likely(lock_taken))
1144 __raw_spin_unlock(&cpu_buffer->lock);
1145 local_irq_restore(flags);
1150 rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer,
1151 u64 *ts, u64 *delta)
1153 struct ring_buffer_event *event;
1157 if (unlikely(*delta > (1ULL << 59) && !once++)) {
1158 printk(KERN_WARNING "Delta way too big! %llu"
1159 " ts=%llu write stamp = %llu\n",
1160 (unsigned long long)*delta,
1161 (unsigned long long)*ts,
1162 (unsigned long long)cpu_buffer->write_stamp);
1167 * The delta is too big, we to add a
1170 event = __rb_reserve_next(cpu_buffer,
1171 RINGBUF_TYPE_TIME_EXTEND,
1177 if (PTR_ERR(event) == -EAGAIN)
1180 /* Only a commited time event can update the write stamp */
1181 if (rb_is_commit(cpu_buffer, event)) {
1183 * If this is the first on the page, then we need to
1184 * update the page itself, and just put in a zero.
1186 if (rb_event_index(event)) {
1187 event->time_delta = *delta & TS_MASK;
1188 event->array[0] = *delta >> TS_SHIFT;
1190 cpu_buffer->commit_page->page->time_stamp = *ts;
1191 event->time_delta = 0;
1192 event->array[0] = 0;
1194 cpu_buffer->write_stamp = *ts;
1195 /* let the caller know this was the commit */
1198 /* Darn, this is just wasted space */
1199 event->time_delta = 0;
1200 event->array[0] = 0;
1209 static struct ring_buffer_event *
1210 rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer,
1211 unsigned type, unsigned long length)
1213 struct ring_buffer_event *event;
1220 * We allow for interrupts to reenter here and do a trace.
1221 * If one does, it will cause this original code to loop
1222 * back here. Even with heavy interrupts happening, this
1223 * should only happen a few times in a row. If this happens
1224 * 1000 times in a row, there must be either an interrupt
1225 * storm or we have something buggy.
1228 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
1231 ts = ring_buffer_time_stamp(cpu_buffer->cpu);
1234 * Only the first commit can update the timestamp.
1235 * Yes there is a race here. If an interrupt comes in
1236 * just after the conditional and it traces too, then it
1237 * will also check the deltas. More than one timestamp may
1238 * also be made. But only the entry that did the actual
1239 * commit will be something other than zero.
1241 if (cpu_buffer->tail_page == cpu_buffer->commit_page &&
1242 rb_page_write(cpu_buffer->tail_page) ==
1243 rb_commit_index(cpu_buffer)) {
1245 delta = ts - cpu_buffer->write_stamp;
1247 /* make sure this delta is calculated here */
1250 /* Did the write stamp get updated already? */
1251 if (unlikely(ts < cpu_buffer->write_stamp))
1254 if (test_time_stamp(delta)) {
1256 commit = rb_add_time_stamp(cpu_buffer, &ts, &delta);
1258 if (commit == -EBUSY)
1261 if (commit == -EAGAIN)
1264 RB_WARN_ON(cpu_buffer, commit < 0);
1267 /* Non commits have zero deltas */
1270 event = __rb_reserve_next(cpu_buffer, type, length, &ts);
1271 if (PTR_ERR(event) == -EAGAIN)
1275 if (unlikely(commit))
1277 * Ouch! We needed a timestamp and it was commited. But
1278 * we didn't get our event reserved.
1280 rb_set_commit_to_write(cpu_buffer);
1285 * If the timestamp was commited, make the commit our entry
1286 * now so that we will update it when needed.
1289 rb_set_commit_event(cpu_buffer, event);
1290 else if (!rb_is_commit(cpu_buffer, event))
1293 event->time_delta = delta;
1298 static DEFINE_PER_CPU(int, rb_need_resched);
1301 * ring_buffer_lock_reserve - reserve a part of the buffer
1302 * @buffer: the ring buffer to reserve from
1303 * @length: the length of the data to reserve (excluding event header)
1304 * @flags: a pointer to save the interrupt flags
1306 * Returns a reseverd event on the ring buffer to copy directly to.
1307 * The user of this interface will need to get the body to write into
1308 * and can use the ring_buffer_event_data() interface.
1310 * The length is the length of the data needed, not the event length
1311 * which also includes the event header.
1313 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
1314 * If NULL is returned, then nothing has been allocated or locked.
1316 struct ring_buffer_event *
1317 ring_buffer_lock_reserve(struct ring_buffer *buffer,
1318 unsigned long length,
1319 unsigned long *flags)
1321 struct ring_buffer_per_cpu *cpu_buffer;
1322 struct ring_buffer_event *event;
1325 if (ring_buffer_flags != RB_BUFFERS_ON)
1328 if (atomic_read(&buffer->record_disabled))
1331 /* If we are tracing schedule, we don't want to recurse */
1332 resched = ftrace_preempt_disable();
1334 cpu = raw_smp_processor_id();
1336 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1339 cpu_buffer = buffer->buffers[cpu];
1341 if (atomic_read(&cpu_buffer->record_disabled))
1344 length = rb_calculate_event_length(length);
1345 if (length > BUF_PAGE_SIZE)
1348 event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length);
1353 * Need to store resched state on this cpu.
1354 * Only the first needs to.
1357 if (preempt_count() == 1)
1358 per_cpu(rb_need_resched, cpu) = resched;
1363 ftrace_preempt_enable(resched);
1366 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve);
1368 static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
1369 struct ring_buffer_event *event)
1371 cpu_buffer->entries++;
1373 /* Only process further if we own the commit */
1374 if (!rb_is_commit(cpu_buffer, event))
1377 cpu_buffer->write_stamp += event->time_delta;
1379 rb_set_commit_to_write(cpu_buffer);
1383 * ring_buffer_unlock_commit - commit a reserved
1384 * @buffer: The buffer to commit to
1385 * @event: The event pointer to commit.
1386 * @flags: the interrupt flags received from ring_buffer_lock_reserve.
1388 * This commits the data to the ring buffer, and releases any locks held.
1390 * Must be paired with ring_buffer_lock_reserve.
1392 int ring_buffer_unlock_commit(struct ring_buffer *buffer,
1393 struct ring_buffer_event *event,
1394 unsigned long flags)
1396 struct ring_buffer_per_cpu *cpu_buffer;
1397 int cpu = raw_smp_processor_id();
1399 cpu_buffer = buffer->buffers[cpu];
1401 rb_commit(cpu_buffer, event);
1404 * Only the last preempt count needs to restore preemption.
1406 if (preempt_count() == 1)
1407 ftrace_preempt_enable(per_cpu(rb_need_resched, cpu));
1409 preempt_enable_no_resched_notrace();
1413 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit);
1416 * ring_buffer_write - write data to the buffer without reserving
1417 * @buffer: The ring buffer to write to.
1418 * @length: The length of the data being written (excluding the event header)
1419 * @data: The data to write to the buffer.
1421 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
1422 * one function. If you already have the data to write to the buffer, it
1423 * may be easier to simply call this function.
1425 * Note, like ring_buffer_lock_reserve, the length is the length of the data
1426 * and not the length of the event which would hold the header.
1428 int ring_buffer_write(struct ring_buffer *buffer,
1429 unsigned long length,
1432 struct ring_buffer_per_cpu *cpu_buffer;
1433 struct ring_buffer_event *event;
1434 unsigned long event_length;
1439 if (ring_buffer_flags != RB_BUFFERS_ON)
1442 if (atomic_read(&buffer->record_disabled))
1445 resched = ftrace_preempt_disable();
1447 cpu = raw_smp_processor_id();
1449 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1452 cpu_buffer = buffer->buffers[cpu];
1454 if (atomic_read(&cpu_buffer->record_disabled))
1457 event_length = rb_calculate_event_length(length);
1458 event = rb_reserve_next_event(cpu_buffer,
1459 RINGBUF_TYPE_DATA, event_length);
1463 body = rb_event_data(event);
1465 memcpy(body, data, length);
1467 rb_commit(cpu_buffer, event);
1471 ftrace_preempt_enable(resched);
1475 EXPORT_SYMBOL_GPL(ring_buffer_write);
1477 static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
1479 struct buffer_page *reader = cpu_buffer->reader_page;
1480 struct buffer_page *head = cpu_buffer->head_page;
1481 struct buffer_page *commit = cpu_buffer->commit_page;
1483 return reader->read == rb_page_commit(reader) &&
1484 (commit == reader ||
1486 head->read == rb_page_commit(commit)));
1490 * ring_buffer_record_disable - stop all writes into the buffer
1491 * @buffer: The ring buffer to stop writes to.
1493 * This prevents all writes to the buffer. Any attempt to write
1494 * to the buffer after this will fail and return NULL.
1496 * The caller should call synchronize_sched() after this.
1498 void ring_buffer_record_disable(struct ring_buffer *buffer)
1500 atomic_inc(&buffer->record_disabled);
1502 EXPORT_SYMBOL_GPL(ring_buffer_record_disable);
1505 * ring_buffer_record_enable - enable writes to the buffer
1506 * @buffer: The ring buffer to enable writes
1508 * Note, multiple disables will need the same number of enables
1509 * to truely enable the writing (much like preempt_disable).
1511 void ring_buffer_record_enable(struct ring_buffer *buffer)
1513 atomic_dec(&buffer->record_disabled);
1515 EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
1518 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
1519 * @buffer: The ring buffer to stop writes to.
1520 * @cpu: The CPU buffer to stop
1522 * This prevents all writes to the buffer. Any attempt to write
1523 * to the buffer after this will fail and return NULL.
1525 * The caller should call synchronize_sched() after this.
1527 void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu)
1529 struct ring_buffer_per_cpu *cpu_buffer;
1531 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1534 cpu_buffer = buffer->buffers[cpu];
1535 atomic_inc(&cpu_buffer->record_disabled);
1537 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu);
1540 * ring_buffer_record_enable_cpu - enable writes to the buffer
1541 * @buffer: The ring buffer to enable writes
1542 * @cpu: The CPU to enable.
1544 * Note, multiple disables will need the same number of enables
1545 * to truely enable the writing (much like preempt_disable).
1547 void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu)
1549 struct ring_buffer_per_cpu *cpu_buffer;
1551 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1554 cpu_buffer = buffer->buffers[cpu];
1555 atomic_dec(&cpu_buffer->record_disabled);
1557 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
1560 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
1561 * @buffer: The ring buffer
1562 * @cpu: The per CPU buffer to get the entries from.
1564 unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
1566 struct ring_buffer_per_cpu *cpu_buffer;
1568 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1571 cpu_buffer = buffer->buffers[cpu];
1572 return cpu_buffer->entries;
1574 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
1577 * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
1578 * @buffer: The ring buffer
1579 * @cpu: The per CPU buffer to get the number of overruns from
1581 unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu)
1583 struct ring_buffer_per_cpu *cpu_buffer;
1585 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1588 cpu_buffer = buffer->buffers[cpu];
1589 return cpu_buffer->overrun;
1591 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
1594 * ring_buffer_entries - get the number of entries in a buffer
1595 * @buffer: The ring buffer
1597 * Returns the total number of entries in the ring buffer
1600 unsigned long ring_buffer_entries(struct ring_buffer *buffer)
1602 struct ring_buffer_per_cpu *cpu_buffer;
1603 unsigned long entries = 0;
1606 /* if you care about this being correct, lock the buffer */
1607 for_each_buffer_cpu(buffer, cpu) {
1608 cpu_buffer = buffer->buffers[cpu];
1609 entries += cpu_buffer->entries;
1614 EXPORT_SYMBOL_GPL(ring_buffer_entries);
1617 * ring_buffer_overrun_cpu - get the number of overruns in buffer
1618 * @buffer: The ring buffer
1620 * Returns the total number of overruns in the ring buffer
1623 unsigned long ring_buffer_overruns(struct ring_buffer *buffer)
1625 struct ring_buffer_per_cpu *cpu_buffer;
1626 unsigned long overruns = 0;
1629 /* if you care about this being correct, lock the buffer */
1630 for_each_buffer_cpu(buffer, cpu) {
1631 cpu_buffer = buffer->buffers[cpu];
1632 overruns += cpu_buffer->overrun;
1637 EXPORT_SYMBOL_GPL(ring_buffer_overruns);
1639 static void rb_iter_reset(struct ring_buffer_iter *iter)
1641 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1643 /* Iterator usage is expected to have record disabled */
1644 if (list_empty(&cpu_buffer->reader_page->list)) {
1645 iter->head_page = cpu_buffer->head_page;
1646 iter->head = cpu_buffer->head_page->read;
1648 iter->head_page = cpu_buffer->reader_page;
1649 iter->head = cpu_buffer->reader_page->read;
1652 iter->read_stamp = cpu_buffer->read_stamp;
1654 iter->read_stamp = iter->head_page->page->time_stamp;
1658 * ring_buffer_iter_reset - reset an iterator
1659 * @iter: The iterator to reset
1661 * Resets the iterator, so that it will start from the beginning
1664 void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
1666 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1667 unsigned long flags;
1669 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1670 rb_iter_reset(iter);
1671 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1673 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset);
1676 * ring_buffer_iter_empty - check if an iterator has no more to read
1677 * @iter: The iterator to check
1679 int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
1681 struct ring_buffer_per_cpu *cpu_buffer;
1683 cpu_buffer = iter->cpu_buffer;
1685 return iter->head_page == cpu_buffer->commit_page &&
1686 iter->head == rb_commit_index(cpu_buffer);
1688 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty);
1691 rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
1692 struct ring_buffer_event *event)
1696 switch (event->type) {
1697 case RINGBUF_TYPE_PADDING:
1700 case RINGBUF_TYPE_TIME_EXTEND:
1701 delta = event->array[0];
1703 delta += event->time_delta;
1704 cpu_buffer->read_stamp += delta;
1707 case RINGBUF_TYPE_TIME_STAMP:
1708 /* FIXME: not implemented */
1711 case RINGBUF_TYPE_DATA:
1712 cpu_buffer->read_stamp += event->time_delta;
1722 rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
1723 struct ring_buffer_event *event)
1727 switch (event->type) {
1728 case RINGBUF_TYPE_PADDING:
1731 case RINGBUF_TYPE_TIME_EXTEND:
1732 delta = event->array[0];
1734 delta += event->time_delta;
1735 iter->read_stamp += delta;
1738 case RINGBUF_TYPE_TIME_STAMP:
1739 /* FIXME: not implemented */
1742 case RINGBUF_TYPE_DATA:
1743 iter->read_stamp += event->time_delta;
1752 static struct buffer_page *
1753 rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
1755 struct buffer_page *reader = NULL;
1756 unsigned long flags;
1759 local_irq_save(flags);
1760 __raw_spin_lock(&cpu_buffer->lock);
1764 * This should normally only loop twice. But because the
1765 * start of the reader inserts an empty page, it causes
1766 * a case where we will loop three times. There should be no
1767 * reason to loop four times (that I know of).
1769 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
1774 reader = cpu_buffer->reader_page;
1776 /* If there's more to read, return this page */
1777 if (cpu_buffer->reader_page->read < rb_page_size(reader))
1780 /* Never should we have an index greater than the size */
1781 if (RB_WARN_ON(cpu_buffer,
1782 cpu_buffer->reader_page->read > rb_page_size(reader)))
1785 /* check if we caught up to the tail */
1787 if (cpu_buffer->commit_page == cpu_buffer->reader_page)
1791 * Splice the empty reader page into the list around the head.
1792 * Reset the reader page to size zero.
1795 reader = cpu_buffer->head_page;
1796 cpu_buffer->reader_page->list.next = reader->list.next;
1797 cpu_buffer->reader_page->list.prev = reader->list.prev;
1799 local_set(&cpu_buffer->reader_page->write, 0);
1800 local_set(&cpu_buffer->reader_page->page->commit, 0);
1802 /* Make the reader page now replace the head */
1803 reader->list.prev->next = &cpu_buffer->reader_page->list;
1804 reader->list.next->prev = &cpu_buffer->reader_page->list;
1807 * If the tail is on the reader, then we must set the head
1808 * to the inserted page, otherwise we set it one before.
1810 cpu_buffer->head_page = cpu_buffer->reader_page;
1812 if (cpu_buffer->commit_page != reader)
1813 rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
1815 /* Finally update the reader page to the new head */
1816 cpu_buffer->reader_page = reader;
1817 rb_reset_reader_page(cpu_buffer);
1822 __raw_spin_unlock(&cpu_buffer->lock);
1823 local_irq_restore(flags);
1828 static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
1830 struct ring_buffer_event *event;
1831 struct buffer_page *reader;
1834 reader = rb_get_reader_page(cpu_buffer);
1836 /* This function should not be called when buffer is empty */
1837 if (RB_WARN_ON(cpu_buffer, !reader))
1840 event = rb_reader_event(cpu_buffer);
1842 if (event->type == RINGBUF_TYPE_DATA)
1843 cpu_buffer->entries--;
1845 rb_update_read_stamp(cpu_buffer, event);
1847 length = rb_event_length(event);
1848 cpu_buffer->reader_page->read += length;
1851 static void rb_advance_iter(struct ring_buffer_iter *iter)
1853 struct ring_buffer *buffer;
1854 struct ring_buffer_per_cpu *cpu_buffer;
1855 struct ring_buffer_event *event;
1858 cpu_buffer = iter->cpu_buffer;
1859 buffer = cpu_buffer->buffer;
1862 * Check if we are at the end of the buffer.
1864 if (iter->head >= rb_page_size(iter->head_page)) {
1865 if (RB_WARN_ON(buffer,
1866 iter->head_page == cpu_buffer->commit_page))
1872 event = rb_iter_head_event(iter);
1874 length = rb_event_length(event);
1877 * This should not be called to advance the header if we are
1878 * at the tail of the buffer.
1880 if (RB_WARN_ON(cpu_buffer,
1881 (iter->head_page == cpu_buffer->commit_page) &&
1882 (iter->head + length > rb_commit_index(cpu_buffer))))
1885 rb_update_iter_read_stamp(iter, event);
1887 iter->head += length;
1889 /* check for end of page padding */
1890 if ((iter->head >= rb_page_size(iter->head_page)) &&
1891 (iter->head_page != cpu_buffer->commit_page))
1892 rb_advance_iter(iter);
1895 static struct ring_buffer_event *
1896 rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
1898 struct ring_buffer_per_cpu *cpu_buffer;
1899 struct ring_buffer_event *event;
1900 struct buffer_page *reader;
1903 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1906 cpu_buffer = buffer->buffers[cpu];
1910 * We repeat when a timestamp is encountered. It is possible
1911 * to get multiple timestamps from an interrupt entering just
1912 * as one timestamp is about to be written. The max times
1913 * that this can happen is the number of nested interrupts we
1914 * can have. Nesting 10 deep of interrupts is clearly
1917 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10))
1920 reader = rb_get_reader_page(cpu_buffer);
1924 event = rb_reader_event(cpu_buffer);
1926 switch (event->type) {
1927 case RINGBUF_TYPE_PADDING:
1928 RB_WARN_ON(cpu_buffer, 1);
1929 rb_advance_reader(cpu_buffer);
1932 case RINGBUF_TYPE_TIME_EXTEND:
1933 /* Internal data, OK to advance */
1934 rb_advance_reader(cpu_buffer);
1937 case RINGBUF_TYPE_TIME_STAMP:
1938 /* FIXME: not implemented */
1939 rb_advance_reader(cpu_buffer);
1942 case RINGBUF_TYPE_DATA:
1944 *ts = cpu_buffer->read_stamp + event->time_delta;
1945 ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts);
1955 EXPORT_SYMBOL_GPL(ring_buffer_peek);
1957 static struct ring_buffer_event *
1958 rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
1960 struct ring_buffer *buffer;
1961 struct ring_buffer_per_cpu *cpu_buffer;
1962 struct ring_buffer_event *event;
1965 if (ring_buffer_iter_empty(iter))
1968 cpu_buffer = iter->cpu_buffer;
1969 buffer = cpu_buffer->buffer;
1973 * We repeat when a timestamp is encountered. It is possible
1974 * to get multiple timestamps from an interrupt entering just
1975 * as one timestamp is about to be written. The max times
1976 * that this can happen is the number of nested interrupts we
1977 * can have. Nesting 10 deep of interrupts is clearly
1980 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10))
1983 if (rb_per_cpu_empty(cpu_buffer))
1986 event = rb_iter_head_event(iter);
1988 switch (event->type) {
1989 case RINGBUF_TYPE_PADDING:
1993 case RINGBUF_TYPE_TIME_EXTEND:
1994 /* Internal data, OK to advance */
1995 rb_advance_iter(iter);
1998 case RINGBUF_TYPE_TIME_STAMP:
1999 /* FIXME: not implemented */
2000 rb_advance_iter(iter);
2003 case RINGBUF_TYPE_DATA:
2005 *ts = iter->read_stamp + event->time_delta;
2006 ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts);
2016 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek);
2019 * ring_buffer_peek - peek at the next event to be read
2020 * @buffer: The ring buffer to read
2021 * @cpu: The cpu to peak at
2022 * @ts: The timestamp counter of this event.
2024 * This will return the event that will be read next, but does
2025 * not consume the data.
2027 struct ring_buffer_event *
2028 ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
2030 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2031 struct ring_buffer_event *event;
2032 unsigned long flags;
2034 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2035 event = rb_buffer_peek(buffer, cpu, ts);
2036 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2042 * ring_buffer_iter_peek - peek at the next event to be read
2043 * @iter: The ring buffer iterator
2044 * @ts: The timestamp counter of this event.
2046 * This will return the event that will be read next, but does
2047 * not increment the iterator.
2049 struct ring_buffer_event *
2050 ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
2052 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2053 struct ring_buffer_event *event;
2054 unsigned long flags;
2056 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2057 event = rb_iter_peek(iter, ts);
2058 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2064 * ring_buffer_consume - return an event and consume it
2065 * @buffer: The ring buffer to get the next event from
2067 * Returns the next event in the ring buffer, and that event is consumed.
2068 * Meaning, that sequential reads will keep returning a different event,
2069 * and eventually empty the ring buffer if the producer is slower.
2071 struct ring_buffer_event *
2072 ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts)
2074 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2075 struct ring_buffer_event *event;
2076 unsigned long flags;
2078 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2081 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2083 event = rb_buffer_peek(buffer, cpu, ts);
2087 rb_advance_reader(cpu_buffer);
2090 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2094 EXPORT_SYMBOL_GPL(ring_buffer_consume);
2097 * ring_buffer_read_start - start a non consuming read of the buffer
2098 * @buffer: The ring buffer to read from
2099 * @cpu: The cpu buffer to iterate over
2101 * This starts up an iteration through the buffer. It also disables
2102 * the recording to the buffer until the reading is finished.
2103 * This prevents the reading from being corrupted. This is not
2104 * a consuming read, so a producer is not expected.
2106 * Must be paired with ring_buffer_finish.
2108 struct ring_buffer_iter *
2109 ring_buffer_read_start(struct ring_buffer *buffer, int cpu)
2111 struct ring_buffer_per_cpu *cpu_buffer;
2112 struct ring_buffer_iter *iter;
2113 unsigned long flags;
2115 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2118 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
2122 cpu_buffer = buffer->buffers[cpu];
2124 iter->cpu_buffer = cpu_buffer;
2126 atomic_inc(&cpu_buffer->record_disabled);
2127 synchronize_sched();
2129 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2130 __raw_spin_lock(&cpu_buffer->lock);
2131 rb_iter_reset(iter);
2132 __raw_spin_unlock(&cpu_buffer->lock);
2133 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2137 EXPORT_SYMBOL_GPL(ring_buffer_read_start);
2140 * ring_buffer_finish - finish reading the iterator of the buffer
2141 * @iter: The iterator retrieved by ring_buffer_start
2143 * This re-enables the recording to the buffer, and frees the
2147 ring_buffer_read_finish(struct ring_buffer_iter *iter)
2149 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2151 atomic_dec(&cpu_buffer->record_disabled);
2154 EXPORT_SYMBOL_GPL(ring_buffer_read_finish);
2157 * ring_buffer_read - read the next item in the ring buffer by the iterator
2158 * @iter: The ring buffer iterator
2159 * @ts: The time stamp of the event read.
2161 * This reads the next event in the ring buffer and increments the iterator.
2163 struct ring_buffer_event *
2164 ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
2166 struct ring_buffer_event *event;
2167 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2168 unsigned long flags;
2170 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2171 event = rb_iter_peek(iter, ts);
2175 rb_advance_iter(iter);
2177 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2181 EXPORT_SYMBOL_GPL(ring_buffer_read);
2184 * ring_buffer_size - return the size of the ring buffer (in bytes)
2185 * @buffer: The ring buffer.
2187 unsigned long ring_buffer_size(struct ring_buffer *buffer)
2189 return BUF_PAGE_SIZE * buffer->pages;
2191 EXPORT_SYMBOL_GPL(ring_buffer_size);
2194 rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
2196 cpu_buffer->head_page
2197 = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
2198 local_set(&cpu_buffer->head_page->write, 0);
2199 local_set(&cpu_buffer->head_page->page->commit, 0);
2201 cpu_buffer->head_page->read = 0;
2203 cpu_buffer->tail_page = cpu_buffer->head_page;
2204 cpu_buffer->commit_page = cpu_buffer->head_page;
2206 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
2207 local_set(&cpu_buffer->reader_page->write, 0);
2208 local_set(&cpu_buffer->reader_page->page->commit, 0);
2209 cpu_buffer->reader_page->read = 0;
2211 cpu_buffer->overrun = 0;
2212 cpu_buffer->entries = 0;
2214 cpu_buffer->write_stamp = 0;
2215 cpu_buffer->read_stamp = 0;
2219 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
2220 * @buffer: The ring buffer to reset a per cpu buffer of
2221 * @cpu: The CPU buffer to be reset
2223 void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
2225 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2226 unsigned long flags;
2228 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2231 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2233 __raw_spin_lock(&cpu_buffer->lock);
2235 rb_reset_cpu(cpu_buffer);
2237 __raw_spin_unlock(&cpu_buffer->lock);
2239 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2241 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
2244 * ring_buffer_reset - reset a ring buffer
2245 * @buffer: The ring buffer to reset all cpu buffers
2247 void ring_buffer_reset(struct ring_buffer *buffer)
2251 for_each_buffer_cpu(buffer, cpu)
2252 ring_buffer_reset_cpu(buffer, cpu);
2254 EXPORT_SYMBOL_GPL(ring_buffer_reset);
2257 * rind_buffer_empty - is the ring buffer empty?
2258 * @buffer: The ring buffer to test
2260 int ring_buffer_empty(struct ring_buffer *buffer)
2262 struct ring_buffer_per_cpu *cpu_buffer;
2265 /* yes this is racy, but if you don't like the race, lock the buffer */
2266 for_each_buffer_cpu(buffer, cpu) {
2267 cpu_buffer = buffer->buffers[cpu];
2268 if (!rb_per_cpu_empty(cpu_buffer))
2273 EXPORT_SYMBOL_GPL(ring_buffer_empty);
2276 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
2277 * @buffer: The ring buffer
2278 * @cpu: The CPU buffer to test
2280 int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu)
2282 struct ring_buffer_per_cpu *cpu_buffer;
2284 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2287 cpu_buffer = buffer->buffers[cpu];
2288 return rb_per_cpu_empty(cpu_buffer);
2290 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
2293 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
2294 * @buffer_a: One buffer to swap with
2295 * @buffer_b: The other buffer to swap with
2297 * This function is useful for tracers that want to take a "snapshot"
2298 * of a CPU buffer and has another back up buffer lying around.
2299 * it is expected that the tracer handles the cpu buffer not being
2300 * used at the moment.
2302 int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
2303 struct ring_buffer *buffer_b, int cpu)
2305 struct ring_buffer_per_cpu *cpu_buffer_a;
2306 struct ring_buffer_per_cpu *cpu_buffer_b;
2308 if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
2309 !cpumask_test_cpu(cpu, buffer_b->cpumask))
2312 /* At least make sure the two buffers are somewhat the same */
2313 if (buffer_a->pages != buffer_b->pages)
2316 if (ring_buffer_flags != RB_BUFFERS_ON)
2319 if (atomic_read(&buffer_a->record_disabled))
2322 if (atomic_read(&buffer_b->record_disabled))
2325 cpu_buffer_a = buffer_a->buffers[cpu];
2326 cpu_buffer_b = buffer_b->buffers[cpu];
2328 if (atomic_read(&cpu_buffer_a->record_disabled))
2331 if (atomic_read(&cpu_buffer_b->record_disabled))
2335 * We can't do a synchronize_sched here because this
2336 * function can be called in atomic context.
2337 * Normally this will be called from the same CPU as cpu.
2338 * If not it's up to the caller to protect this.
2340 atomic_inc(&cpu_buffer_a->record_disabled);
2341 atomic_inc(&cpu_buffer_b->record_disabled);
2343 buffer_a->buffers[cpu] = cpu_buffer_b;
2344 buffer_b->buffers[cpu] = cpu_buffer_a;
2346 cpu_buffer_b->buffer = buffer_a;
2347 cpu_buffer_a->buffer = buffer_b;
2349 atomic_dec(&cpu_buffer_a->record_disabled);
2350 atomic_dec(&cpu_buffer_b->record_disabled);
2354 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
2356 static void rb_remove_entries(struct ring_buffer_per_cpu *cpu_buffer,
2357 struct buffer_data_page *bpage)
2359 struct ring_buffer_event *event;
2362 __raw_spin_lock(&cpu_buffer->lock);
2363 for (head = 0; head < local_read(&bpage->commit);
2364 head += rb_event_length(event)) {
2366 event = __rb_data_page_index(bpage, head);
2367 if (RB_WARN_ON(cpu_buffer, rb_null_event(event)))
2369 /* Only count data entries */
2370 if (event->type != RINGBUF_TYPE_DATA)
2372 cpu_buffer->entries--;
2374 __raw_spin_unlock(&cpu_buffer->lock);
2378 * ring_buffer_alloc_read_page - allocate a page to read from buffer
2379 * @buffer: the buffer to allocate for.
2381 * This function is used in conjunction with ring_buffer_read_page.
2382 * When reading a full page from the ring buffer, these functions
2383 * can be used to speed up the process. The calling function should
2384 * allocate a few pages first with this function. Then when it
2385 * needs to get pages from the ring buffer, it passes the result
2386 * of this function into ring_buffer_read_page, which will swap
2387 * the page that was allocated, with the read page of the buffer.
2390 * The page allocated, or NULL on error.
2392 void *ring_buffer_alloc_read_page(struct ring_buffer *buffer)
2395 struct buffer_data_page *bpage;
2397 addr = __get_free_page(GFP_KERNEL);
2401 bpage = (void *)addr;
2407 * ring_buffer_free_read_page - free an allocated read page
2408 * @buffer: the buffer the page was allocate for
2409 * @data: the page to free
2411 * Free a page allocated from ring_buffer_alloc_read_page.
2413 void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data)
2415 free_page((unsigned long)data);
2419 * ring_buffer_read_page - extract a page from the ring buffer
2420 * @buffer: buffer to extract from
2421 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
2422 * @cpu: the cpu of the buffer to extract
2423 * @full: should the extraction only happen when the page is full.
2425 * This function will pull out a page from the ring buffer and consume it.
2426 * @data_page must be the address of the variable that was returned
2427 * from ring_buffer_alloc_read_page. This is because the page might be used
2428 * to swap with a page in the ring buffer.
2431 * rpage = ring_buffer_alloc_page(buffer);
2434 * ret = ring_buffer_read_page(buffer, &rpage, cpu, 0);
2436 * process_page(rpage);
2438 * When @full is set, the function will not return true unless
2439 * the writer is off the reader page.
2441 * Note: it is up to the calling functions to handle sleeps and wakeups.
2442 * The ring buffer can be used anywhere in the kernel and can not
2443 * blindly call wake_up. The layer that uses the ring buffer must be
2444 * responsible for that.
2447 * 1 if data has been transferred
2448 * 0 if no data has been transferred.
2450 int ring_buffer_read_page(struct ring_buffer *buffer,
2451 void **data_page, int cpu, int full)
2453 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2454 struct ring_buffer_event *event;
2455 struct buffer_data_page *bpage;
2456 unsigned long flags;
2466 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2469 * rb_buffer_peek will get the next ring buffer if
2470 * the current reader page is empty.
2472 event = rb_buffer_peek(buffer, cpu, NULL);
2476 /* check for data */
2477 if (!local_read(&cpu_buffer->reader_page->page->commit))
2480 * If the writer is already off of the read page, then simply
2481 * switch the read page with the given page. Otherwise
2482 * we need to copy the data from the reader to the writer.
2484 if (cpu_buffer->reader_page == cpu_buffer->commit_page) {
2485 unsigned int read = cpu_buffer->reader_page->read;
2489 /* The writer is still on the reader page, we must copy */
2490 bpage = cpu_buffer->reader_page->page;
2492 cpu_buffer->reader_page->page->data + read,
2493 local_read(&bpage->commit) - read);
2495 /* consume what was read */
2496 cpu_buffer->reader_page += read;
2499 /* swap the pages */
2500 rb_init_page(bpage);
2501 bpage = cpu_buffer->reader_page->page;
2502 cpu_buffer->reader_page->page = *data_page;
2503 cpu_buffer->reader_page->read = 0;
2508 /* update the entry counter */
2509 rb_remove_entries(cpu_buffer, bpage);
2511 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2517 rb_simple_read(struct file *filp, char __user *ubuf,
2518 size_t cnt, loff_t *ppos)
2520 long *p = filp->private_data;
2524 if (test_bit(RB_BUFFERS_DISABLED_BIT, p))
2525 r = sprintf(buf, "permanently disabled\n");
2527 r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p));
2529 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2533 rb_simple_write(struct file *filp, const char __user *ubuf,
2534 size_t cnt, loff_t *ppos)
2536 long *p = filp->private_data;
2541 if (cnt >= sizeof(buf))
2544 if (copy_from_user(&buf, ubuf, cnt))
2549 ret = strict_strtoul(buf, 10, &val);
2554 set_bit(RB_BUFFERS_ON_BIT, p);
2556 clear_bit(RB_BUFFERS_ON_BIT, p);
2563 static struct file_operations rb_simple_fops = {
2564 .open = tracing_open_generic,
2565 .read = rb_simple_read,
2566 .write = rb_simple_write,
2570 static __init int rb_init_debugfs(void)
2572 struct dentry *d_tracer;
2573 struct dentry *entry;
2575 d_tracer = tracing_init_dentry();
2577 entry = debugfs_create_file("tracing_on", 0644, d_tracer,
2578 &ring_buffer_flags, &rb_simple_fops);
2580 pr_warning("Could not create debugfs 'tracing_on' entry\n");
2585 fs_initcall(rb_init_debugfs);