4 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
6 #include <linux/ring_buffer.h>
7 #include <linux/spinlock.h>
8 #include <linux/debugfs.h>
9 #include <linux/uaccess.h>
10 #include <linux/module.h>
11 #include <linux/percpu.h>
12 #include <linux/mutex.h>
13 #include <linux/sched.h> /* used for sched_clock() (for now) */
14 #include <linux/init.h>
15 #include <linux/hash.h>
16 #include <linux/list.h>
22 * A fast way to enable or disable all ring buffers is to
23 * call tracing_on or tracing_off. Turning off the ring buffers
24 * prevents all ring buffers from being recorded to.
25 * Turning this switch on, makes it OK to write to the
26 * ring buffer, if the ring buffer is enabled itself.
28 * There's three layers that must be on in order to write
31 * 1) This global flag must be set.
32 * 2) The ring buffer must be enabled for recording.
33 * 3) The per cpu buffer must be enabled for recording.
35 * In case of an anomaly, this global flag has a bit set that
36 * will permantly disable all ring buffers.
40 * Global flag to disable all recording to ring buffers
41 * This has two bits: ON, DISABLED
45 * 0 0 : ring buffers are off
46 * 1 0 : ring buffers are on
47 * X 1 : ring buffers are permanently disabled
51 RB_BUFFERS_ON_BIT = 0,
52 RB_BUFFERS_DISABLED_BIT = 1,
56 RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT,
57 RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT,
60 static long ring_buffer_flags __read_mostly = RB_BUFFERS_ON;
63 * tracing_on - enable all tracing buffers
65 * This function enables all tracing buffers that may have been
66 * disabled with tracing_off.
70 set_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags);
74 * tracing_off - turn off all tracing buffers
76 * This function stops all tracing buffers from recording data.
77 * It does not disable any overhead the tracers themselves may
78 * be causing. This function simply causes all recording to
79 * the ring buffers to fail.
81 void tracing_off(void)
83 clear_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags);
87 * tracing_off_permanent - permanently disable ring buffers
89 * This function, once called, will disable all ring buffers
92 void tracing_off_permanent(void)
94 set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags);
99 /* Up this if you want to test the TIME_EXTENTS and normalization */
100 #define DEBUG_SHIFT 0
103 u64 ring_buffer_time_stamp(int cpu)
107 preempt_disable_notrace();
108 /* shift to debug/test normalization and TIME_EXTENTS */
109 time = sched_clock() << DEBUG_SHIFT;
110 preempt_enable_notrace();
115 void ring_buffer_normalize_time_stamp(int cpu, u64 *ts)
117 /* Just stupid testing the normalize function and deltas */
121 #define RB_EVNT_HDR_SIZE (sizeof(struct ring_buffer_event))
122 #define RB_ALIGNMENT_SHIFT 2
123 #define RB_ALIGNMENT (1 << RB_ALIGNMENT_SHIFT)
124 #define RB_MAX_SMALL_DATA 28
127 RB_LEN_TIME_EXTEND = 8,
128 RB_LEN_TIME_STAMP = 16,
131 /* inline for ring buffer fast paths */
132 static inline unsigned
133 rb_event_length(struct ring_buffer_event *event)
137 switch (event->type) {
138 case RINGBUF_TYPE_PADDING:
142 case RINGBUF_TYPE_TIME_EXTEND:
143 return RB_LEN_TIME_EXTEND;
145 case RINGBUF_TYPE_TIME_STAMP:
146 return RB_LEN_TIME_STAMP;
148 case RINGBUF_TYPE_DATA:
150 length = event->len << RB_ALIGNMENT_SHIFT;
152 length = event->array[0];
153 return length + RB_EVNT_HDR_SIZE;
162 * ring_buffer_event_length - return the length of the event
163 * @event: the event to get the length of
165 unsigned ring_buffer_event_length(struct ring_buffer_event *event)
167 return rb_event_length(event);
170 /* inline for ring buffer fast paths */
172 rb_event_data(struct ring_buffer_event *event)
174 BUG_ON(event->type != RINGBUF_TYPE_DATA);
175 /* If length is in len field, then array[0] has the data */
177 return (void *)&event->array[0];
178 /* Otherwise length is in array[0] and array[1] has the data */
179 return (void *)&event->array[1];
183 * ring_buffer_event_data - return the data of the event
184 * @event: the event to get the data from
186 void *ring_buffer_event_data(struct ring_buffer_event *event)
188 return rb_event_data(event);
191 #define for_each_buffer_cpu(buffer, cpu) \
192 for_each_cpu_mask(cpu, buffer->cpumask)
195 #define TS_MASK ((1ULL << TS_SHIFT) - 1)
196 #define TS_DELTA_TEST (~TS_MASK)
198 struct buffer_data_page {
199 u64 time_stamp; /* page time stamp */
200 local_t commit; /* write commited index */
201 unsigned char data[]; /* data of buffer page */
205 local_t write; /* index for next write */
206 unsigned read; /* index for next read */
207 struct list_head list; /* list of free pages */
208 struct buffer_data_page *page; /* Actual data page */
211 static void rb_init_page(struct buffer_data_page *bpage)
213 local_set(&bpage->commit, 0);
217 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
220 static inline void free_buffer_page(struct buffer_page *bpage)
223 free_page((unsigned long)bpage->page);
228 * We need to fit the time_stamp delta into 27 bits.
230 static inline int test_time_stamp(u64 delta)
232 if (delta & TS_DELTA_TEST)
237 #define BUF_PAGE_SIZE (PAGE_SIZE - sizeof(struct buffer_data_page))
240 * head_page == tail_page && head == tail then buffer is empty.
242 struct ring_buffer_per_cpu {
244 struct ring_buffer *buffer;
245 spinlock_t reader_lock; /* serialize readers */
247 struct lock_class_key lock_key;
248 struct list_head pages;
249 struct buffer_page *head_page; /* read from head */
250 struct buffer_page *tail_page; /* write to tail */
251 struct buffer_page *commit_page; /* commited pages */
252 struct buffer_page *reader_page;
253 unsigned long overrun;
254 unsigned long entries;
257 atomic_t record_disabled;
266 atomic_t record_disabled;
270 struct ring_buffer_per_cpu **buffers;
273 struct ring_buffer_iter {
274 struct ring_buffer_per_cpu *cpu_buffer;
276 struct buffer_page *head_page;
280 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
281 #define RB_WARN_ON(buffer, cond) \
283 int _____ret = unlikely(cond); \
285 atomic_inc(&buffer->record_disabled); \
292 * check_pages - integrity check of buffer pages
293 * @cpu_buffer: CPU buffer with pages to test
295 * As a safty measure we check to make sure the data pages have not
298 static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
300 struct list_head *head = &cpu_buffer->pages;
301 struct buffer_page *bpage, *tmp;
303 if (RB_WARN_ON(cpu_buffer, head->next->prev != head))
305 if (RB_WARN_ON(cpu_buffer, head->prev->next != head))
308 list_for_each_entry_safe(bpage, tmp, head, list) {
309 if (RB_WARN_ON(cpu_buffer,
310 bpage->list.next->prev != &bpage->list))
312 if (RB_WARN_ON(cpu_buffer,
313 bpage->list.prev->next != &bpage->list))
320 static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
323 struct list_head *head = &cpu_buffer->pages;
324 struct buffer_page *bpage, *tmp;
329 for (i = 0; i < nr_pages; i++) {
330 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
331 GFP_KERNEL, cpu_to_node(cpu_buffer->cpu));
334 list_add(&bpage->list, &pages);
336 addr = __get_free_page(GFP_KERNEL);
339 bpage->page = (void *)addr;
340 rb_init_page(bpage->page);
343 list_splice(&pages, head);
345 rb_check_pages(cpu_buffer);
350 list_for_each_entry_safe(bpage, tmp, &pages, list) {
351 list_del_init(&bpage->list);
352 free_buffer_page(bpage);
357 static struct ring_buffer_per_cpu *
358 rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu)
360 struct ring_buffer_per_cpu *cpu_buffer;
361 struct buffer_page *bpage;
365 cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
366 GFP_KERNEL, cpu_to_node(cpu));
370 cpu_buffer->cpu = cpu;
371 cpu_buffer->buffer = buffer;
372 spin_lock_init(&cpu_buffer->reader_lock);
373 cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
374 INIT_LIST_HEAD(&cpu_buffer->pages);
376 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
377 GFP_KERNEL, cpu_to_node(cpu));
379 goto fail_free_buffer;
381 cpu_buffer->reader_page = bpage;
382 addr = __get_free_page(GFP_KERNEL);
384 goto fail_free_reader;
385 bpage->page = (void *)addr;
386 rb_init_page(bpage->page);
388 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
390 ret = rb_allocate_pages(cpu_buffer, buffer->pages);
392 goto fail_free_reader;
394 cpu_buffer->head_page
395 = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
396 cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
401 free_buffer_page(cpu_buffer->reader_page);
408 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
410 struct list_head *head = &cpu_buffer->pages;
411 struct buffer_page *bpage, *tmp;
413 list_del_init(&cpu_buffer->reader_page->list);
414 free_buffer_page(cpu_buffer->reader_page);
416 list_for_each_entry_safe(bpage, tmp, head, list) {
417 list_del_init(&bpage->list);
418 free_buffer_page(bpage);
424 * Causes compile errors if the struct buffer_page gets bigger
425 * than the struct page.
427 extern int ring_buffer_page_too_big(void);
430 * ring_buffer_alloc - allocate a new ring_buffer
431 * @size: the size in bytes that is needed.
432 * @flags: attributes to set for the ring buffer.
434 * Currently the only flag that is available is the RB_FL_OVERWRITE
435 * flag. This flag means that the buffer will overwrite old data
436 * when the buffer wraps. If this flag is not set, the buffer will
437 * drop data when the tail hits the head.
439 struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags)
441 struct ring_buffer *buffer;
445 /* Paranoid! Optimizes out when all is well */
446 if (sizeof(struct buffer_page) > sizeof(struct page))
447 ring_buffer_page_too_big();
450 /* keep it in its own cache line */
451 buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
456 buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
457 buffer->flags = flags;
459 /* need at least two pages */
460 if (buffer->pages == 1)
463 buffer->cpumask = cpu_possible_map;
464 buffer->cpus = nr_cpu_ids;
466 bsize = sizeof(void *) * nr_cpu_ids;
467 buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
469 if (!buffer->buffers)
470 goto fail_free_buffer;
472 for_each_buffer_cpu(buffer, cpu) {
473 buffer->buffers[cpu] =
474 rb_allocate_cpu_buffer(buffer, cpu);
475 if (!buffer->buffers[cpu])
476 goto fail_free_buffers;
479 mutex_init(&buffer->mutex);
484 for_each_buffer_cpu(buffer, cpu) {
485 if (buffer->buffers[cpu])
486 rb_free_cpu_buffer(buffer->buffers[cpu]);
488 kfree(buffer->buffers);
496 * ring_buffer_free - free a ring buffer.
497 * @buffer: the buffer to free.
500 ring_buffer_free(struct ring_buffer *buffer)
504 for_each_buffer_cpu(buffer, cpu)
505 rb_free_cpu_buffer(buffer->buffers[cpu]);
510 static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
513 rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages)
515 struct buffer_page *bpage;
519 atomic_inc(&cpu_buffer->record_disabled);
522 for (i = 0; i < nr_pages; i++) {
523 if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages)))
525 p = cpu_buffer->pages.next;
526 bpage = list_entry(p, struct buffer_page, list);
527 list_del_init(&bpage->list);
528 free_buffer_page(bpage);
530 if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages)))
533 rb_reset_cpu(cpu_buffer);
535 rb_check_pages(cpu_buffer);
537 atomic_dec(&cpu_buffer->record_disabled);
542 rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer,
543 struct list_head *pages, unsigned nr_pages)
545 struct buffer_page *bpage;
549 atomic_inc(&cpu_buffer->record_disabled);
552 for (i = 0; i < nr_pages; i++) {
553 if (RB_WARN_ON(cpu_buffer, list_empty(pages)))
556 bpage = list_entry(p, struct buffer_page, list);
557 list_del_init(&bpage->list);
558 list_add_tail(&bpage->list, &cpu_buffer->pages);
560 rb_reset_cpu(cpu_buffer);
562 rb_check_pages(cpu_buffer);
564 atomic_dec(&cpu_buffer->record_disabled);
568 * ring_buffer_resize - resize the ring buffer
569 * @buffer: the buffer to resize.
570 * @size: the new size.
572 * The tracer is responsible for making sure that the buffer is
573 * not being used while changing the size.
574 * Note: We may be able to change the above requirement by using
575 * RCU synchronizations.
577 * Minimum size is 2 * BUF_PAGE_SIZE.
579 * Returns -1 on failure.
581 int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
583 struct ring_buffer_per_cpu *cpu_buffer;
584 unsigned nr_pages, rm_pages, new_pages;
585 struct buffer_page *bpage, *tmp;
586 unsigned long buffer_size;
592 * Always succeed at resizing a non-existent buffer:
597 size = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
598 size *= BUF_PAGE_SIZE;
599 buffer_size = buffer->pages * BUF_PAGE_SIZE;
601 /* we need a minimum of two pages */
602 if (size < BUF_PAGE_SIZE * 2)
603 size = BUF_PAGE_SIZE * 2;
605 if (size == buffer_size)
608 mutex_lock(&buffer->mutex);
610 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
612 if (size < buffer_size) {
614 /* easy case, just free pages */
615 if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) {
616 mutex_unlock(&buffer->mutex);
620 rm_pages = buffer->pages - nr_pages;
622 for_each_buffer_cpu(buffer, cpu) {
623 cpu_buffer = buffer->buffers[cpu];
624 rb_remove_pages(cpu_buffer, rm_pages);
630 * This is a bit more difficult. We only want to add pages
631 * when we can allocate enough for all CPUs. We do this
632 * by allocating all the pages and storing them on a local
633 * link list. If we succeed in our allocation, then we
634 * add these pages to the cpu_buffers. Otherwise we just free
635 * them all and return -ENOMEM;
637 if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) {
638 mutex_unlock(&buffer->mutex);
642 new_pages = nr_pages - buffer->pages;
644 for_each_buffer_cpu(buffer, cpu) {
645 for (i = 0; i < new_pages; i++) {
646 bpage = kzalloc_node(ALIGN(sizeof(*bpage),
648 GFP_KERNEL, cpu_to_node(cpu));
651 list_add(&bpage->list, &pages);
652 addr = __get_free_page(GFP_KERNEL);
655 bpage->page = (void *)addr;
656 rb_init_page(bpage->page);
660 for_each_buffer_cpu(buffer, cpu) {
661 cpu_buffer = buffer->buffers[cpu];
662 rb_insert_pages(cpu_buffer, &pages, new_pages);
665 if (RB_WARN_ON(buffer, !list_empty(&pages))) {
666 mutex_unlock(&buffer->mutex);
671 buffer->pages = nr_pages;
672 mutex_unlock(&buffer->mutex);
677 list_for_each_entry_safe(bpage, tmp, &pages, list) {
678 list_del_init(&bpage->list);
679 free_buffer_page(bpage);
681 mutex_unlock(&buffer->mutex);
685 static inline int rb_null_event(struct ring_buffer_event *event)
687 return event->type == RINGBUF_TYPE_PADDING;
691 __rb_data_page_index(struct buffer_data_page *bpage, unsigned index)
693 return bpage->data + index;
696 static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
698 return bpage->page->data + index;
701 static inline struct ring_buffer_event *
702 rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
704 return __rb_page_index(cpu_buffer->reader_page,
705 cpu_buffer->reader_page->read);
708 static inline struct ring_buffer_event *
709 rb_head_event(struct ring_buffer_per_cpu *cpu_buffer)
711 return __rb_page_index(cpu_buffer->head_page,
712 cpu_buffer->head_page->read);
715 static inline struct ring_buffer_event *
716 rb_iter_head_event(struct ring_buffer_iter *iter)
718 return __rb_page_index(iter->head_page, iter->head);
721 static inline unsigned rb_page_write(struct buffer_page *bpage)
723 return local_read(&bpage->write);
726 static inline unsigned rb_page_commit(struct buffer_page *bpage)
728 return local_read(&bpage->page->commit);
731 /* Size is determined by what has been commited */
732 static inline unsigned rb_page_size(struct buffer_page *bpage)
734 return rb_page_commit(bpage);
737 static inline unsigned
738 rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
740 return rb_page_commit(cpu_buffer->commit_page);
743 static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer)
745 return rb_page_commit(cpu_buffer->head_page);
749 * When the tail hits the head and the buffer is in overwrite mode,
750 * the head jumps to the next page and all content on the previous
751 * page is discarded. But before doing so, we update the overrun
752 * variable of the buffer.
754 static void rb_update_overflow(struct ring_buffer_per_cpu *cpu_buffer)
756 struct ring_buffer_event *event;
759 for (head = 0; head < rb_head_size(cpu_buffer);
760 head += rb_event_length(event)) {
762 event = __rb_page_index(cpu_buffer->head_page, head);
763 if (RB_WARN_ON(cpu_buffer, rb_null_event(event)))
765 /* Only count data entries */
766 if (event->type != RINGBUF_TYPE_DATA)
768 cpu_buffer->overrun++;
769 cpu_buffer->entries--;
773 static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer,
774 struct buffer_page **bpage)
776 struct list_head *p = (*bpage)->list.next;
778 if (p == &cpu_buffer->pages)
781 *bpage = list_entry(p, struct buffer_page, list);
784 static inline unsigned
785 rb_event_index(struct ring_buffer_event *event)
787 unsigned long addr = (unsigned long)event;
789 return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE);
793 rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
794 struct ring_buffer_event *event)
796 unsigned long addr = (unsigned long)event;
799 index = rb_event_index(event);
802 return cpu_buffer->commit_page->page == (void *)addr &&
803 rb_commit_index(cpu_buffer) == index;
807 rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer,
808 struct ring_buffer_event *event)
810 unsigned long addr = (unsigned long)event;
813 index = rb_event_index(event);
816 while (cpu_buffer->commit_page->page != (void *)addr) {
817 if (RB_WARN_ON(cpu_buffer,
818 cpu_buffer->commit_page == cpu_buffer->tail_page))
820 cpu_buffer->commit_page->page->commit =
821 cpu_buffer->commit_page->write;
822 rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
823 cpu_buffer->write_stamp =
824 cpu_buffer->commit_page->page->time_stamp;
827 /* Now set the commit to the event's index */
828 local_set(&cpu_buffer->commit_page->page->commit, index);
832 rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
835 * We only race with interrupts and NMIs on this CPU.
836 * If we own the commit event, then we can commit
837 * all others that interrupted us, since the interruptions
838 * are in stack format (they finish before they come
839 * back to us). This allows us to do a simple loop to
840 * assign the commit to the tail.
842 while (cpu_buffer->commit_page != cpu_buffer->tail_page) {
843 cpu_buffer->commit_page->page->commit =
844 cpu_buffer->commit_page->write;
845 rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
846 cpu_buffer->write_stamp =
847 cpu_buffer->commit_page->page->time_stamp;
848 /* add barrier to keep gcc from optimizing too much */
851 while (rb_commit_index(cpu_buffer) !=
852 rb_page_write(cpu_buffer->commit_page)) {
853 cpu_buffer->commit_page->page->commit =
854 cpu_buffer->commit_page->write;
859 static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
861 cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp;
862 cpu_buffer->reader_page->read = 0;
865 static inline void rb_inc_iter(struct ring_buffer_iter *iter)
867 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
870 * The iterator could be on the reader page (it starts there).
871 * But the head could have moved, since the reader was
872 * found. Check for this case and assign the iterator
873 * to the head page instead of next.
875 if (iter->head_page == cpu_buffer->reader_page)
876 iter->head_page = cpu_buffer->head_page;
878 rb_inc_page(cpu_buffer, &iter->head_page);
880 iter->read_stamp = iter->head_page->page->time_stamp;
885 * ring_buffer_update_event - update event type and data
886 * @event: the even to update
887 * @type: the type of event
888 * @length: the size of the event field in the ring buffer
890 * Update the type and data fields of the event. The length
891 * is the actual size that is written to the ring buffer,
892 * and with this, we can determine what to place into the
896 rb_update_event(struct ring_buffer_event *event,
897 unsigned type, unsigned length)
903 case RINGBUF_TYPE_PADDING:
906 case RINGBUF_TYPE_TIME_EXTEND:
908 (RB_LEN_TIME_EXTEND + (RB_ALIGNMENT-1))
909 >> RB_ALIGNMENT_SHIFT;
912 case RINGBUF_TYPE_TIME_STAMP:
914 (RB_LEN_TIME_STAMP + (RB_ALIGNMENT-1))
915 >> RB_ALIGNMENT_SHIFT;
918 case RINGBUF_TYPE_DATA:
919 length -= RB_EVNT_HDR_SIZE;
920 if (length > RB_MAX_SMALL_DATA) {
922 event->array[0] = length;
925 (length + (RB_ALIGNMENT-1))
926 >> RB_ALIGNMENT_SHIFT;
933 static inline unsigned rb_calculate_event_length(unsigned length)
935 struct ring_buffer_event event; /* Used only for sizeof array */
937 /* zero length can cause confusions */
941 if (length > RB_MAX_SMALL_DATA)
942 length += sizeof(event.array[0]);
944 length += RB_EVNT_HDR_SIZE;
945 length = ALIGN(length, RB_ALIGNMENT);
950 static struct ring_buffer_event *
951 __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
952 unsigned type, unsigned long length, u64 *ts)
954 struct buffer_page *tail_page, *head_page, *reader_page;
955 unsigned long tail, write;
956 struct ring_buffer *buffer = cpu_buffer->buffer;
957 struct ring_buffer_event *event;
960 tail_page = cpu_buffer->tail_page;
961 write = local_add_return(length, &tail_page->write);
962 tail = write - length;
964 /* See if we shot pass the end of this buffer page */
965 if (write > BUF_PAGE_SIZE) {
966 struct buffer_page *next_page = tail_page;
968 local_irq_save(flags);
969 __raw_spin_lock(&cpu_buffer->lock);
971 rb_inc_page(cpu_buffer, &next_page);
973 head_page = cpu_buffer->head_page;
974 reader_page = cpu_buffer->reader_page;
976 /* we grabbed the lock before incrementing */
977 if (RB_WARN_ON(cpu_buffer, next_page == reader_page))
981 * If for some reason, we had an interrupt storm that made
982 * it all the way around the buffer, bail, and warn
985 if (unlikely(next_page == cpu_buffer->commit_page)) {
990 if (next_page == head_page) {
991 if (!(buffer->flags & RB_FL_OVERWRITE)) {
993 if (tail <= BUF_PAGE_SIZE)
994 local_set(&tail_page->write, tail);
998 /* tail_page has not moved yet? */
999 if (tail_page == cpu_buffer->tail_page) {
1000 /* count overflows */
1001 rb_update_overflow(cpu_buffer);
1003 rb_inc_page(cpu_buffer, &head_page);
1004 cpu_buffer->head_page = head_page;
1005 cpu_buffer->head_page->read = 0;
1010 * If the tail page is still the same as what we think
1011 * it is, then it is up to us to update the tail
1014 if (tail_page == cpu_buffer->tail_page) {
1015 local_set(&next_page->write, 0);
1016 local_set(&next_page->page->commit, 0);
1017 cpu_buffer->tail_page = next_page;
1019 /* reread the time stamp */
1020 *ts = ring_buffer_time_stamp(cpu_buffer->cpu);
1021 cpu_buffer->tail_page->page->time_stamp = *ts;
1025 * The actual tail page has moved forward.
1027 if (tail < BUF_PAGE_SIZE) {
1028 /* Mark the rest of the page with padding */
1029 event = __rb_page_index(tail_page, tail);
1030 event->type = RINGBUF_TYPE_PADDING;
1033 if (tail <= BUF_PAGE_SIZE)
1034 /* Set the write back to the previous setting */
1035 local_set(&tail_page->write, tail);
1038 * If this was a commit entry that failed,
1039 * increment that too
1041 if (tail_page == cpu_buffer->commit_page &&
1042 tail == rb_commit_index(cpu_buffer)) {
1043 rb_set_commit_to_write(cpu_buffer);
1046 __raw_spin_unlock(&cpu_buffer->lock);
1047 local_irq_restore(flags);
1049 /* fail and let the caller try again */
1050 return ERR_PTR(-EAGAIN);
1053 /* We reserved something on the buffer */
1055 if (RB_WARN_ON(cpu_buffer, write > BUF_PAGE_SIZE))
1058 event = __rb_page_index(tail_page, tail);
1059 rb_update_event(event, type, length);
1062 * If this is a commit and the tail is zero, then update
1063 * this page's time stamp.
1065 if (!tail && rb_is_commit(cpu_buffer, event))
1066 cpu_buffer->commit_page->page->time_stamp = *ts;
1071 __raw_spin_unlock(&cpu_buffer->lock);
1072 local_irq_restore(flags);
1077 rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer,
1078 u64 *ts, u64 *delta)
1080 struct ring_buffer_event *event;
1084 if (unlikely(*delta > (1ULL << 59) && !once++)) {
1085 printk(KERN_WARNING "Delta way too big! %llu"
1086 " ts=%llu write stamp = %llu\n",
1087 (unsigned long long)*delta,
1088 (unsigned long long)*ts,
1089 (unsigned long long)cpu_buffer->write_stamp);
1094 * The delta is too big, we to add a
1097 event = __rb_reserve_next(cpu_buffer,
1098 RINGBUF_TYPE_TIME_EXTEND,
1104 if (PTR_ERR(event) == -EAGAIN)
1107 /* Only a commited time event can update the write stamp */
1108 if (rb_is_commit(cpu_buffer, event)) {
1110 * If this is the first on the page, then we need to
1111 * update the page itself, and just put in a zero.
1113 if (rb_event_index(event)) {
1114 event->time_delta = *delta & TS_MASK;
1115 event->array[0] = *delta >> TS_SHIFT;
1117 cpu_buffer->commit_page->page->time_stamp = *ts;
1118 event->time_delta = 0;
1119 event->array[0] = 0;
1121 cpu_buffer->write_stamp = *ts;
1122 /* let the caller know this was the commit */
1125 /* Darn, this is just wasted space */
1126 event->time_delta = 0;
1127 event->array[0] = 0;
1136 static struct ring_buffer_event *
1137 rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer,
1138 unsigned type, unsigned long length)
1140 struct ring_buffer_event *event;
1147 * We allow for interrupts to reenter here and do a trace.
1148 * If one does, it will cause this original code to loop
1149 * back here. Even with heavy interrupts happening, this
1150 * should only happen a few times in a row. If this happens
1151 * 1000 times in a row, there must be either an interrupt
1152 * storm or we have something buggy.
1155 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
1158 ts = ring_buffer_time_stamp(cpu_buffer->cpu);
1161 * Only the first commit can update the timestamp.
1162 * Yes there is a race here. If an interrupt comes in
1163 * just after the conditional and it traces too, then it
1164 * will also check the deltas. More than one timestamp may
1165 * also be made. But only the entry that did the actual
1166 * commit will be something other than zero.
1168 if (cpu_buffer->tail_page == cpu_buffer->commit_page &&
1169 rb_page_write(cpu_buffer->tail_page) ==
1170 rb_commit_index(cpu_buffer)) {
1172 delta = ts - cpu_buffer->write_stamp;
1174 /* make sure this delta is calculated here */
1177 /* Did the write stamp get updated already? */
1178 if (unlikely(ts < cpu_buffer->write_stamp))
1181 if (test_time_stamp(delta)) {
1183 commit = rb_add_time_stamp(cpu_buffer, &ts, &delta);
1185 if (commit == -EBUSY)
1188 if (commit == -EAGAIN)
1191 RB_WARN_ON(cpu_buffer, commit < 0);
1194 /* Non commits have zero deltas */
1197 event = __rb_reserve_next(cpu_buffer, type, length, &ts);
1198 if (PTR_ERR(event) == -EAGAIN)
1202 if (unlikely(commit))
1204 * Ouch! We needed a timestamp and it was commited. But
1205 * we didn't get our event reserved.
1207 rb_set_commit_to_write(cpu_buffer);
1212 * If the timestamp was commited, make the commit our entry
1213 * now so that we will update it when needed.
1216 rb_set_commit_event(cpu_buffer, event);
1217 else if (!rb_is_commit(cpu_buffer, event))
1220 event->time_delta = delta;
1225 static DEFINE_PER_CPU(int, rb_need_resched);
1228 * ring_buffer_lock_reserve - reserve a part of the buffer
1229 * @buffer: the ring buffer to reserve from
1230 * @length: the length of the data to reserve (excluding event header)
1231 * @flags: a pointer to save the interrupt flags
1233 * Returns a reseverd event on the ring buffer to copy directly to.
1234 * The user of this interface will need to get the body to write into
1235 * and can use the ring_buffer_event_data() interface.
1237 * The length is the length of the data needed, not the event length
1238 * which also includes the event header.
1240 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
1241 * If NULL is returned, then nothing has been allocated or locked.
1243 struct ring_buffer_event *
1244 ring_buffer_lock_reserve(struct ring_buffer *buffer,
1245 unsigned long length,
1246 unsigned long *flags)
1248 struct ring_buffer_per_cpu *cpu_buffer;
1249 struct ring_buffer_event *event;
1252 if (ring_buffer_flags != RB_BUFFERS_ON)
1255 if (atomic_read(&buffer->record_disabled))
1258 /* If we are tracing schedule, we don't want to recurse */
1259 resched = ftrace_preempt_disable();
1261 cpu = raw_smp_processor_id();
1263 if (!cpu_isset(cpu, buffer->cpumask))
1266 cpu_buffer = buffer->buffers[cpu];
1268 if (atomic_read(&cpu_buffer->record_disabled))
1271 length = rb_calculate_event_length(length);
1272 if (length > BUF_PAGE_SIZE)
1275 event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length);
1280 * Need to store resched state on this cpu.
1281 * Only the first needs to.
1284 if (preempt_count() == 1)
1285 per_cpu(rb_need_resched, cpu) = resched;
1290 ftrace_preempt_enable(resched);
1294 static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
1295 struct ring_buffer_event *event)
1297 cpu_buffer->entries++;
1299 /* Only process further if we own the commit */
1300 if (!rb_is_commit(cpu_buffer, event))
1303 cpu_buffer->write_stamp += event->time_delta;
1305 rb_set_commit_to_write(cpu_buffer);
1309 * ring_buffer_unlock_commit - commit a reserved
1310 * @buffer: The buffer to commit to
1311 * @event: The event pointer to commit.
1312 * @flags: the interrupt flags received from ring_buffer_lock_reserve.
1314 * This commits the data to the ring buffer, and releases any locks held.
1316 * Must be paired with ring_buffer_lock_reserve.
1318 int ring_buffer_unlock_commit(struct ring_buffer *buffer,
1319 struct ring_buffer_event *event,
1320 unsigned long flags)
1322 struct ring_buffer_per_cpu *cpu_buffer;
1323 int cpu = raw_smp_processor_id();
1325 cpu_buffer = buffer->buffers[cpu];
1327 rb_commit(cpu_buffer, event);
1330 * Only the last preempt count needs to restore preemption.
1332 if (preempt_count() == 1)
1333 ftrace_preempt_enable(per_cpu(rb_need_resched, cpu));
1335 preempt_enable_no_resched_notrace();
1341 * ring_buffer_write - write data to the buffer without reserving
1342 * @buffer: The ring buffer to write to.
1343 * @length: The length of the data being written (excluding the event header)
1344 * @data: The data to write to the buffer.
1346 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
1347 * one function. If you already have the data to write to the buffer, it
1348 * may be easier to simply call this function.
1350 * Note, like ring_buffer_lock_reserve, the length is the length of the data
1351 * and not the length of the event which would hold the header.
1353 int ring_buffer_write(struct ring_buffer *buffer,
1354 unsigned long length,
1357 struct ring_buffer_per_cpu *cpu_buffer;
1358 struct ring_buffer_event *event;
1359 unsigned long event_length;
1364 if (ring_buffer_flags != RB_BUFFERS_ON)
1367 if (atomic_read(&buffer->record_disabled))
1370 resched = ftrace_preempt_disable();
1372 cpu = raw_smp_processor_id();
1374 if (!cpu_isset(cpu, buffer->cpumask))
1377 cpu_buffer = buffer->buffers[cpu];
1379 if (atomic_read(&cpu_buffer->record_disabled))
1382 event_length = rb_calculate_event_length(length);
1383 event = rb_reserve_next_event(cpu_buffer,
1384 RINGBUF_TYPE_DATA, event_length);
1388 body = rb_event_data(event);
1390 memcpy(body, data, length);
1392 rb_commit(cpu_buffer, event);
1396 ftrace_preempt_enable(resched);
1401 static inline int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
1403 struct buffer_page *reader = cpu_buffer->reader_page;
1404 struct buffer_page *head = cpu_buffer->head_page;
1405 struct buffer_page *commit = cpu_buffer->commit_page;
1407 return reader->read == rb_page_commit(reader) &&
1408 (commit == reader ||
1410 head->read == rb_page_commit(commit)));
1414 * ring_buffer_record_disable - stop all writes into the buffer
1415 * @buffer: The ring buffer to stop writes to.
1417 * This prevents all writes to the buffer. Any attempt to write
1418 * to the buffer after this will fail and return NULL.
1420 * The caller should call synchronize_sched() after this.
1422 void ring_buffer_record_disable(struct ring_buffer *buffer)
1424 atomic_inc(&buffer->record_disabled);
1428 * ring_buffer_record_enable - enable writes to the buffer
1429 * @buffer: The ring buffer to enable writes
1431 * Note, multiple disables will need the same number of enables
1432 * to truely enable the writing (much like preempt_disable).
1434 void ring_buffer_record_enable(struct ring_buffer *buffer)
1436 atomic_dec(&buffer->record_disabled);
1440 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
1441 * @buffer: The ring buffer to stop writes to.
1442 * @cpu: The CPU buffer to stop
1444 * This prevents all writes to the buffer. Any attempt to write
1445 * to the buffer after this will fail and return NULL.
1447 * The caller should call synchronize_sched() after this.
1449 void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu)
1451 struct ring_buffer_per_cpu *cpu_buffer;
1453 if (!cpu_isset(cpu, buffer->cpumask))
1456 cpu_buffer = buffer->buffers[cpu];
1457 atomic_inc(&cpu_buffer->record_disabled);
1461 * ring_buffer_record_enable_cpu - enable writes to the buffer
1462 * @buffer: The ring buffer to enable writes
1463 * @cpu: The CPU to enable.
1465 * Note, multiple disables will need the same number of enables
1466 * to truely enable the writing (much like preempt_disable).
1468 void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu)
1470 struct ring_buffer_per_cpu *cpu_buffer;
1472 if (!cpu_isset(cpu, buffer->cpumask))
1475 cpu_buffer = buffer->buffers[cpu];
1476 atomic_dec(&cpu_buffer->record_disabled);
1480 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
1481 * @buffer: The ring buffer
1482 * @cpu: The per CPU buffer to get the entries from.
1484 unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
1486 struct ring_buffer_per_cpu *cpu_buffer;
1488 if (!cpu_isset(cpu, buffer->cpumask))
1491 cpu_buffer = buffer->buffers[cpu];
1492 return cpu_buffer->entries;
1496 * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
1497 * @buffer: The ring buffer
1498 * @cpu: The per CPU buffer to get the number of overruns from
1500 unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu)
1502 struct ring_buffer_per_cpu *cpu_buffer;
1504 if (!cpu_isset(cpu, buffer->cpumask))
1507 cpu_buffer = buffer->buffers[cpu];
1508 return cpu_buffer->overrun;
1512 * ring_buffer_entries - get the number of entries in a buffer
1513 * @buffer: The ring buffer
1515 * Returns the total number of entries in the ring buffer
1518 unsigned long ring_buffer_entries(struct ring_buffer *buffer)
1520 struct ring_buffer_per_cpu *cpu_buffer;
1521 unsigned long entries = 0;
1524 /* if you care about this being correct, lock the buffer */
1525 for_each_buffer_cpu(buffer, cpu) {
1526 cpu_buffer = buffer->buffers[cpu];
1527 entries += cpu_buffer->entries;
1534 * ring_buffer_overrun_cpu - get the number of overruns in buffer
1535 * @buffer: The ring buffer
1537 * Returns the total number of overruns in the ring buffer
1540 unsigned long ring_buffer_overruns(struct ring_buffer *buffer)
1542 struct ring_buffer_per_cpu *cpu_buffer;
1543 unsigned long overruns = 0;
1546 /* if you care about this being correct, lock the buffer */
1547 for_each_buffer_cpu(buffer, cpu) {
1548 cpu_buffer = buffer->buffers[cpu];
1549 overruns += cpu_buffer->overrun;
1555 static void rb_iter_reset(struct ring_buffer_iter *iter)
1557 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1559 /* Iterator usage is expected to have record disabled */
1560 if (list_empty(&cpu_buffer->reader_page->list)) {
1561 iter->head_page = cpu_buffer->head_page;
1562 iter->head = cpu_buffer->head_page->read;
1564 iter->head_page = cpu_buffer->reader_page;
1565 iter->head = cpu_buffer->reader_page->read;
1568 iter->read_stamp = cpu_buffer->read_stamp;
1570 iter->read_stamp = iter->head_page->page->time_stamp;
1574 * ring_buffer_iter_reset - reset an iterator
1575 * @iter: The iterator to reset
1577 * Resets the iterator, so that it will start from the beginning
1580 void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
1582 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1583 unsigned long flags;
1585 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1586 rb_iter_reset(iter);
1587 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1591 * ring_buffer_iter_empty - check if an iterator has no more to read
1592 * @iter: The iterator to check
1594 int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
1596 struct ring_buffer_per_cpu *cpu_buffer;
1598 cpu_buffer = iter->cpu_buffer;
1600 return iter->head_page == cpu_buffer->commit_page &&
1601 iter->head == rb_commit_index(cpu_buffer);
1605 rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
1606 struct ring_buffer_event *event)
1610 switch (event->type) {
1611 case RINGBUF_TYPE_PADDING:
1614 case RINGBUF_TYPE_TIME_EXTEND:
1615 delta = event->array[0];
1617 delta += event->time_delta;
1618 cpu_buffer->read_stamp += delta;
1621 case RINGBUF_TYPE_TIME_STAMP:
1622 /* FIXME: not implemented */
1625 case RINGBUF_TYPE_DATA:
1626 cpu_buffer->read_stamp += event->time_delta;
1636 rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
1637 struct ring_buffer_event *event)
1641 switch (event->type) {
1642 case RINGBUF_TYPE_PADDING:
1645 case RINGBUF_TYPE_TIME_EXTEND:
1646 delta = event->array[0];
1648 delta += event->time_delta;
1649 iter->read_stamp += delta;
1652 case RINGBUF_TYPE_TIME_STAMP:
1653 /* FIXME: not implemented */
1656 case RINGBUF_TYPE_DATA:
1657 iter->read_stamp += event->time_delta;
1666 static struct buffer_page *
1667 rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
1669 struct buffer_page *reader = NULL;
1670 unsigned long flags;
1673 local_irq_save(flags);
1674 __raw_spin_lock(&cpu_buffer->lock);
1678 * This should normally only loop twice. But because the
1679 * start of the reader inserts an empty page, it causes
1680 * a case where we will loop three times. There should be no
1681 * reason to loop four times (that I know of).
1683 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
1688 reader = cpu_buffer->reader_page;
1690 /* If there's more to read, return this page */
1691 if (cpu_buffer->reader_page->read < rb_page_size(reader))
1694 /* Never should we have an index greater than the size */
1695 if (RB_WARN_ON(cpu_buffer,
1696 cpu_buffer->reader_page->read > rb_page_size(reader)))
1699 /* check if we caught up to the tail */
1701 if (cpu_buffer->commit_page == cpu_buffer->reader_page)
1705 * Splice the empty reader page into the list around the head.
1706 * Reset the reader page to size zero.
1709 reader = cpu_buffer->head_page;
1710 cpu_buffer->reader_page->list.next = reader->list.next;
1711 cpu_buffer->reader_page->list.prev = reader->list.prev;
1713 local_set(&cpu_buffer->reader_page->write, 0);
1714 local_set(&cpu_buffer->reader_page->page->commit, 0);
1716 /* Make the reader page now replace the head */
1717 reader->list.prev->next = &cpu_buffer->reader_page->list;
1718 reader->list.next->prev = &cpu_buffer->reader_page->list;
1721 * If the tail is on the reader, then we must set the head
1722 * to the inserted page, otherwise we set it one before.
1724 cpu_buffer->head_page = cpu_buffer->reader_page;
1726 if (cpu_buffer->commit_page != reader)
1727 rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
1729 /* Finally update the reader page to the new head */
1730 cpu_buffer->reader_page = reader;
1731 rb_reset_reader_page(cpu_buffer);
1736 __raw_spin_unlock(&cpu_buffer->lock);
1737 local_irq_restore(flags);
1742 static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
1744 struct ring_buffer_event *event;
1745 struct buffer_page *reader;
1748 reader = rb_get_reader_page(cpu_buffer);
1750 /* This function should not be called when buffer is empty */
1751 if (RB_WARN_ON(cpu_buffer, !reader))
1754 event = rb_reader_event(cpu_buffer);
1756 if (event->type == RINGBUF_TYPE_DATA)
1757 cpu_buffer->entries--;
1759 rb_update_read_stamp(cpu_buffer, event);
1761 length = rb_event_length(event);
1762 cpu_buffer->reader_page->read += length;
1765 static void rb_advance_iter(struct ring_buffer_iter *iter)
1767 struct ring_buffer *buffer;
1768 struct ring_buffer_per_cpu *cpu_buffer;
1769 struct ring_buffer_event *event;
1772 cpu_buffer = iter->cpu_buffer;
1773 buffer = cpu_buffer->buffer;
1776 * Check if we are at the end of the buffer.
1778 if (iter->head >= rb_page_size(iter->head_page)) {
1779 if (RB_WARN_ON(buffer,
1780 iter->head_page == cpu_buffer->commit_page))
1786 event = rb_iter_head_event(iter);
1788 length = rb_event_length(event);
1791 * This should not be called to advance the header if we are
1792 * at the tail of the buffer.
1794 if (RB_WARN_ON(cpu_buffer,
1795 (iter->head_page == cpu_buffer->commit_page) &&
1796 (iter->head + length > rb_commit_index(cpu_buffer))))
1799 rb_update_iter_read_stamp(iter, event);
1801 iter->head += length;
1803 /* check for end of page padding */
1804 if ((iter->head >= rb_page_size(iter->head_page)) &&
1805 (iter->head_page != cpu_buffer->commit_page))
1806 rb_advance_iter(iter);
1809 static struct ring_buffer_event *
1810 rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
1812 struct ring_buffer_per_cpu *cpu_buffer;
1813 struct ring_buffer_event *event;
1814 struct buffer_page *reader;
1817 if (!cpu_isset(cpu, buffer->cpumask))
1820 cpu_buffer = buffer->buffers[cpu];
1824 * We repeat when a timestamp is encountered. It is possible
1825 * to get multiple timestamps from an interrupt entering just
1826 * as one timestamp is about to be written. The max times
1827 * that this can happen is the number of nested interrupts we
1828 * can have. Nesting 10 deep of interrupts is clearly
1831 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10))
1834 reader = rb_get_reader_page(cpu_buffer);
1838 event = rb_reader_event(cpu_buffer);
1840 switch (event->type) {
1841 case RINGBUF_TYPE_PADDING:
1842 RB_WARN_ON(cpu_buffer, 1);
1843 rb_advance_reader(cpu_buffer);
1846 case RINGBUF_TYPE_TIME_EXTEND:
1847 /* Internal data, OK to advance */
1848 rb_advance_reader(cpu_buffer);
1851 case RINGBUF_TYPE_TIME_STAMP:
1852 /* FIXME: not implemented */
1853 rb_advance_reader(cpu_buffer);
1856 case RINGBUF_TYPE_DATA:
1858 *ts = cpu_buffer->read_stamp + event->time_delta;
1859 ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts);
1870 static struct ring_buffer_event *
1871 rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
1873 struct ring_buffer *buffer;
1874 struct ring_buffer_per_cpu *cpu_buffer;
1875 struct ring_buffer_event *event;
1878 if (ring_buffer_iter_empty(iter))
1881 cpu_buffer = iter->cpu_buffer;
1882 buffer = cpu_buffer->buffer;
1886 * We repeat when a timestamp is encountered. It is possible
1887 * to get multiple timestamps from an interrupt entering just
1888 * as one timestamp is about to be written. The max times
1889 * that this can happen is the number of nested interrupts we
1890 * can have. Nesting 10 deep of interrupts is clearly
1893 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10))
1896 if (rb_per_cpu_empty(cpu_buffer))
1899 event = rb_iter_head_event(iter);
1901 switch (event->type) {
1902 case RINGBUF_TYPE_PADDING:
1906 case RINGBUF_TYPE_TIME_EXTEND:
1907 /* Internal data, OK to advance */
1908 rb_advance_iter(iter);
1911 case RINGBUF_TYPE_TIME_STAMP:
1912 /* FIXME: not implemented */
1913 rb_advance_iter(iter);
1916 case RINGBUF_TYPE_DATA:
1918 *ts = iter->read_stamp + event->time_delta;
1919 ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts);
1931 * ring_buffer_peek - peek at the next event to be read
1932 * @buffer: The ring buffer to read
1933 * @cpu: The cpu to peak at
1934 * @ts: The timestamp counter of this event.
1936 * This will return the event that will be read next, but does
1937 * not consume the data.
1939 struct ring_buffer_event *
1940 ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
1942 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
1943 struct ring_buffer_event *event;
1944 unsigned long flags;
1946 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1947 event = rb_buffer_peek(buffer, cpu, ts);
1948 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1954 * ring_buffer_iter_peek - peek at the next event to be read
1955 * @iter: The ring buffer iterator
1956 * @ts: The timestamp counter of this event.
1958 * This will return the event that will be read next, but does
1959 * not increment the iterator.
1961 struct ring_buffer_event *
1962 ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
1964 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1965 struct ring_buffer_event *event;
1966 unsigned long flags;
1968 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1969 event = rb_iter_peek(iter, ts);
1970 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1976 * ring_buffer_consume - return an event and consume it
1977 * @buffer: The ring buffer to get the next event from
1979 * Returns the next event in the ring buffer, and that event is consumed.
1980 * Meaning, that sequential reads will keep returning a different event,
1981 * and eventually empty the ring buffer if the producer is slower.
1983 struct ring_buffer_event *
1984 ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts)
1986 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
1987 struct ring_buffer_event *event;
1988 unsigned long flags;
1990 if (!cpu_isset(cpu, buffer->cpumask))
1993 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1995 event = rb_buffer_peek(buffer, cpu, ts);
1999 rb_advance_reader(cpu_buffer);
2002 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2008 * ring_buffer_read_start - start a non consuming read of the buffer
2009 * @buffer: The ring buffer to read from
2010 * @cpu: The cpu buffer to iterate over
2012 * This starts up an iteration through the buffer. It also disables
2013 * the recording to the buffer until the reading is finished.
2014 * This prevents the reading from being corrupted. This is not
2015 * a consuming read, so a producer is not expected.
2017 * Must be paired with ring_buffer_finish.
2019 struct ring_buffer_iter *
2020 ring_buffer_read_start(struct ring_buffer *buffer, int cpu)
2022 struct ring_buffer_per_cpu *cpu_buffer;
2023 struct ring_buffer_iter *iter;
2024 unsigned long flags;
2026 if (!cpu_isset(cpu, buffer->cpumask))
2029 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
2033 cpu_buffer = buffer->buffers[cpu];
2035 iter->cpu_buffer = cpu_buffer;
2037 atomic_inc(&cpu_buffer->record_disabled);
2038 synchronize_sched();
2040 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2041 __raw_spin_lock(&cpu_buffer->lock);
2042 rb_iter_reset(iter);
2043 __raw_spin_unlock(&cpu_buffer->lock);
2044 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2050 * ring_buffer_finish - finish reading the iterator of the buffer
2051 * @iter: The iterator retrieved by ring_buffer_start
2053 * This re-enables the recording to the buffer, and frees the
2057 ring_buffer_read_finish(struct ring_buffer_iter *iter)
2059 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2061 atomic_dec(&cpu_buffer->record_disabled);
2066 * ring_buffer_read - read the next item in the ring buffer by the iterator
2067 * @iter: The ring buffer iterator
2068 * @ts: The time stamp of the event read.
2070 * This reads the next event in the ring buffer and increments the iterator.
2072 struct ring_buffer_event *
2073 ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
2075 struct ring_buffer_event *event;
2076 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2077 unsigned long flags;
2079 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2080 event = rb_iter_peek(iter, ts);
2084 rb_advance_iter(iter);
2086 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2092 * ring_buffer_size - return the size of the ring buffer (in bytes)
2093 * @buffer: The ring buffer.
2095 unsigned long ring_buffer_size(struct ring_buffer *buffer)
2097 return BUF_PAGE_SIZE * buffer->pages;
2101 rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
2103 cpu_buffer->head_page
2104 = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
2105 local_set(&cpu_buffer->head_page->write, 0);
2106 local_set(&cpu_buffer->head_page->page->commit, 0);
2108 cpu_buffer->head_page->read = 0;
2110 cpu_buffer->tail_page = cpu_buffer->head_page;
2111 cpu_buffer->commit_page = cpu_buffer->head_page;
2113 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
2114 local_set(&cpu_buffer->reader_page->write, 0);
2115 local_set(&cpu_buffer->reader_page->page->commit, 0);
2116 cpu_buffer->reader_page->read = 0;
2118 cpu_buffer->overrun = 0;
2119 cpu_buffer->entries = 0;
2123 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
2124 * @buffer: The ring buffer to reset a per cpu buffer of
2125 * @cpu: The CPU buffer to be reset
2127 void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
2129 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2130 unsigned long flags;
2132 if (!cpu_isset(cpu, buffer->cpumask))
2135 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2137 __raw_spin_lock(&cpu_buffer->lock);
2139 rb_reset_cpu(cpu_buffer);
2141 __raw_spin_unlock(&cpu_buffer->lock);
2143 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2147 * ring_buffer_reset - reset a ring buffer
2148 * @buffer: The ring buffer to reset all cpu buffers
2150 void ring_buffer_reset(struct ring_buffer *buffer)
2154 for_each_buffer_cpu(buffer, cpu)
2155 ring_buffer_reset_cpu(buffer, cpu);
2159 * rind_buffer_empty - is the ring buffer empty?
2160 * @buffer: The ring buffer to test
2162 int ring_buffer_empty(struct ring_buffer *buffer)
2164 struct ring_buffer_per_cpu *cpu_buffer;
2167 /* yes this is racy, but if you don't like the race, lock the buffer */
2168 for_each_buffer_cpu(buffer, cpu) {
2169 cpu_buffer = buffer->buffers[cpu];
2170 if (!rb_per_cpu_empty(cpu_buffer))
2177 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
2178 * @buffer: The ring buffer
2179 * @cpu: The CPU buffer to test
2181 int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu)
2183 struct ring_buffer_per_cpu *cpu_buffer;
2185 if (!cpu_isset(cpu, buffer->cpumask))
2188 cpu_buffer = buffer->buffers[cpu];
2189 return rb_per_cpu_empty(cpu_buffer);
2193 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
2194 * @buffer_a: One buffer to swap with
2195 * @buffer_b: The other buffer to swap with
2197 * This function is useful for tracers that want to take a "snapshot"
2198 * of a CPU buffer and has another back up buffer lying around.
2199 * it is expected that the tracer handles the cpu buffer not being
2200 * used at the moment.
2202 int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
2203 struct ring_buffer *buffer_b, int cpu)
2205 struct ring_buffer_per_cpu *cpu_buffer_a;
2206 struct ring_buffer_per_cpu *cpu_buffer_b;
2208 if (!cpu_isset(cpu, buffer_a->cpumask) ||
2209 !cpu_isset(cpu, buffer_b->cpumask))
2212 /* At least make sure the two buffers are somewhat the same */
2213 if (buffer_a->size != buffer_b->size ||
2214 buffer_a->pages != buffer_b->pages)
2217 cpu_buffer_a = buffer_a->buffers[cpu];
2218 cpu_buffer_b = buffer_b->buffers[cpu];
2221 * We can't do a synchronize_sched here because this
2222 * function can be called in atomic context.
2223 * Normally this will be called from the same CPU as cpu.
2224 * If not it's up to the caller to protect this.
2226 atomic_inc(&cpu_buffer_a->record_disabled);
2227 atomic_inc(&cpu_buffer_b->record_disabled);
2229 buffer_a->buffers[cpu] = cpu_buffer_b;
2230 buffer_b->buffers[cpu] = cpu_buffer_a;
2232 cpu_buffer_b->buffer = buffer_a;
2233 cpu_buffer_a->buffer = buffer_b;
2235 atomic_dec(&cpu_buffer_a->record_disabled);
2236 atomic_dec(&cpu_buffer_b->record_disabled);
2241 static void rb_remove_entries(struct ring_buffer_per_cpu *cpu_buffer,
2242 struct buffer_data_page *bpage)
2244 struct ring_buffer_event *event;
2247 __raw_spin_lock(&cpu_buffer->lock);
2248 for (head = 0; head < local_read(&bpage->commit);
2249 head += rb_event_length(event)) {
2251 event = __rb_data_page_index(bpage, head);
2252 if (RB_WARN_ON(cpu_buffer, rb_null_event(event)))
2254 /* Only count data entries */
2255 if (event->type != RINGBUF_TYPE_DATA)
2257 cpu_buffer->entries--;
2259 __raw_spin_unlock(&cpu_buffer->lock);
2263 * ring_buffer_alloc_read_page - allocate a page to read from buffer
2264 * @buffer: the buffer to allocate for.
2266 * This function is used in conjunction with ring_buffer_read_page.
2267 * When reading a full page from the ring buffer, these functions
2268 * can be used to speed up the process. The calling function should
2269 * allocate a few pages first with this function. Then when it
2270 * needs to get pages from the ring buffer, it passes the result
2271 * of this function into ring_buffer_read_page, which will swap
2272 * the page that was allocated, with the read page of the buffer.
2275 * The page allocated, or NULL on error.
2277 void *ring_buffer_alloc_read_page(struct ring_buffer *buffer)
2280 struct buffer_data_page *bpage;
2282 addr = __get_free_page(GFP_KERNEL);
2286 bpage = (void *)addr;
2292 * ring_buffer_free_read_page - free an allocated read page
2293 * @buffer: the buffer the page was allocate for
2294 * @data: the page to free
2296 * Free a page allocated from ring_buffer_alloc_read_page.
2298 void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data)
2300 free_page((unsigned long)data);
2304 * ring_buffer_read_page - extract a page from the ring buffer
2305 * @buffer: buffer to extract from
2306 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
2307 * @cpu: the cpu of the buffer to extract
2308 * @full: should the extraction only happen when the page is full.
2310 * This function will pull out a page from the ring buffer and consume it.
2311 * @data_page must be the address of the variable that was returned
2312 * from ring_buffer_alloc_read_page. This is because the page might be used
2313 * to swap with a page in the ring buffer.
2316 * rpage = ring_buffer_alloc_page(buffer);
2319 * ret = ring_buffer_read_page(buffer, &rpage, cpu, 0);
2321 * process_page(rpage);
2323 * When @full is set, the function will not return true unless
2324 * the writer is off the reader page.
2326 * Note: it is up to the calling functions to handle sleeps and wakeups.
2327 * The ring buffer can be used anywhere in the kernel and can not
2328 * blindly call wake_up. The layer that uses the ring buffer must be
2329 * responsible for that.
2332 * 1 if data has been transferred
2333 * 0 if no data has been transferred.
2335 int ring_buffer_read_page(struct ring_buffer *buffer,
2336 void **data_page, int cpu, int full)
2338 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2339 struct ring_buffer_event *event;
2340 struct buffer_data_page *bpage;
2341 unsigned long flags;
2351 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2354 * rb_buffer_peek will get the next ring buffer if
2355 * the current reader page is empty.
2357 event = rb_buffer_peek(buffer, cpu, NULL);
2361 /* check for data */
2362 if (!local_read(&cpu_buffer->reader_page->page->commit))
2365 * If the writer is already off of the read page, then simply
2366 * switch the read page with the given page. Otherwise
2367 * we need to copy the data from the reader to the writer.
2369 if (cpu_buffer->reader_page == cpu_buffer->commit_page) {
2370 unsigned int read = cpu_buffer->reader_page->read;
2374 /* The writer is still on the reader page, we must copy */
2375 bpage = cpu_buffer->reader_page->page;
2377 cpu_buffer->reader_page->page->data + read,
2378 local_read(&bpage->commit) - read);
2380 /* consume what was read */
2381 cpu_buffer->reader_page += read;
2384 /* swap the pages */
2385 rb_init_page(bpage);
2386 bpage = cpu_buffer->reader_page->page;
2387 cpu_buffer->reader_page->page = *data_page;
2388 cpu_buffer->reader_page->read = 0;
2393 /* update the entry counter */
2394 rb_remove_entries(cpu_buffer, bpage);
2396 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2402 rb_simple_read(struct file *filp, char __user *ubuf,
2403 size_t cnt, loff_t *ppos)
2405 long *p = filp->private_data;
2409 if (test_bit(RB_BUFFERS_DISABLED_BIT, p))
2410 r = sprintf(buf, "permanently disabled\n");
2412 r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p));
2414 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2418 rb_simple_write(struct file *filp, const char __user *ubuf,
2419 size_t cnt, loff_t *ppos)
2421 long *p = filp->private_data;
2426 if (cnt >= sizeof(buf))
2429 if (copy_from_user(&buf, ubuf, cnt))
2434 ret = strict_strtoul(buf, 10, &val);
2439 set_bit(RB_BUFFERS_ON_BIT, p);
2441 clear_bit(RB_BUFFERS_ON_BIT, p);
2448 static struct file_operations rb_simple_fops = {
2449 .open = tracing_open_generic,
2450 .read = rb_simple_read,
2451 .write = rb_simple_write,
2455 static __init int rb_init_debugfs(void)
2457 struct dentry *d_tracer;
2458 struct dentry *entry;
2460 d_tracer = tracing_init_dentry();
2462 entry = debugfs_create_file("tracing_on", 0644, d_tracer,
2463 &ring_buffer_flags, &rb_simple_fops);
2465 pr_warning("Could not create debugfs 'tracing_on' entry\n");
2470 fs_initcall(rb_init_debugfs);