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
);
72 EXPORT_SYMBOL_GPL(tracing_on
);
75 * tracing_off - turn off all tracing buffers
77 * This function stops all tracing buffers from recording data.
78 * It does not disable any overhead the tracers themselves may
79 * be causing. This function simply causes all recording to
80 * the ring buffers to fail.
82 void tracing_off(void)
84 clear_bit(RB_BUFFERS_ON_BIT
, &ring_buffer_flags
);
86 EXPORT_SYMBOL_GPL(tracing_off
);
89 * tracing_off_permanent - permanently disable ring buffers
91 * This function, once called, will disable all ring buffers
94 void tracing_off_permanent(void)
96 set_bit(RB_BUFFERS_DISABLED_BIT
, &ring_buffer_flags
);
101 /* Up this if you want to test the TIME_EXTENTS and normalization */
102 #define DEBUG_SHIFT 0
105 u64
ring_buffer_time_stamp(int cpu
)
109 preempt_disable_notrace();
110 /* shift to debug/test normalization and TIME_EXTENTS */
111 time
= sched_clock() << DEBUG_SHIFT
;
112 preempt_enable_no_resched_notrace();
116 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp
);
118 void ring_buffer_normalize_time_stamp(int cpu
, u64
*ts
)
120 /* Just stupid testing the normalize function and deltas */
123 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp
);
125 #define RB_EVNT_HDR_SIZE (sizeof(struct ring_buffer_event))
126 #define RB_ALIGNMENT_SHIFT 2
127 #define RB_ALIGNMENT (1 << RB_ALIGNMENT_SHIFT)
128 #define RB_MAX_SMALL_DATA 28
131 RB_LEN_TIME_EXTEND
= 8,
132 RB_LEN_TIME_STAMP
= 16,
135 /* inline for ring buffer fast paths */
136 static inline unsigned
137 rb_event_length(struct ring_buffer_event
*event
)
141 switch (event
->type
) {
142 case RINGBUF_TYPE_PADDING
:
146 case RINGBUF_TYPE_TIME_EXTEND
:
147 return RB_LEN_TIME_EXTEND
;
149 case RINGBUF_TYPE_TIME_STAMP
:
150 return RB_LEN_TIME_STAMP
;
152 case RINGBUF_TYPE_DATA
:
154 length
= event
->len
<< RB_ALIGNMENT_SHIFT
;
156 length
= event
->array
[0];
157 return length
+ RB_EVNT_HDR_SIZE
;
166 * ring_buffer_event_length - return the length of the event
167 * @event: the event to get the length of
169 unsigned ring_buffer_event_length(struct ring_buffer_event
*event
)
171 unsigned length
= rb_event_length(event
);
172 if (event
->type
!= RINGBUF_TYPE_DATA
)
174 length
-= RB_EVNT_HDR_SIZE
;
175 if (length
> RB_MAX_SMALL_DATA
+ sizeof(event
->array
[0]))
176 length
-= sizeof(event
->array
[0]);
179 EXPORT_SYMBOL_GPL(ring_buffer_event_length
);
181 /* inline for ring buffer fast paths */
183 rb_event_data(struct ring_buffer_event
*event
)
185 BUG_ON(event
->type
!= RINGBUF_TYPE_DATA
);
186 /* If length is in len field, then array[0] has the data */
188 return (void *)&event
->array
[0];
189 /* Otherwise length is in array[0] and array[1] has the data */
190 return (void *)&event
->array
[1];
194 * ring_buffer_event_data - return the data of the event
195 * @event: the event to get the data from
197 void *ring_buffer_event_data(struct ring_buffer_event
*event
)
199 return rb_event_data(event
);
201 EXPORT_SYMBOL_GPL(ring_buffer_event_data
);
203 #define for_each_buffer_cpu(buffer, cpu) \
204 for_each_cpu(cpu, buffer->cpumask)
207 #define TS_MASK ((1ULL << TS_SHIFT) - 1)
208 #define TS_DELTA_TEST (~TS_MASK)
210 struct buffer_data_page
{
211 u64 time_stamp
; /* page time stamp */
212 local_t commit
; /* write commited index */
213 unsigned char data
[]; /* data of buffer page */
217 local_t write
; /* index for next write */
218 unsigned read
; /* index for next read */
219 struct list_head list
; /* list of free pages */
220 struct buffer_data_page
*page
; /* Actual data page */
223 static void rb_init_page(struct buffer_data_page
*bpage
)
225 local_set(&bpage
->commit
, 0);
229 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
232 static inline void free_buffer_page(struct buffer_page
*bpage
)
235 free_page((unsigned long)bpage
->page
);
240 * We need to fit the time_stamp delta into 27 bits.
242 static inline int test_time_stamp(u64 delta
)
244 if (delta
& TS_DELTA_TEST
)
249 #define BUF_PAGE_SIZE (PAGE_SIZE - offsetof(struct buffer_data_page, data))
252 * head_page == tail_page && head == tail then buffer is empty.
254 struct ring_buffer_per_cpu
{
256 struct ring_buffer
*buffer
;
257 spinlock_t reader_lock
; /* serialize readers */
259 struct lock_class_key lock_key
;
260 struct list_head pages
;
261 struct buffer_page
*head_page
; /* read from head */
262 struct buffer_page
*tail_page
; /* write to tail */
263 struct buffer_page
*commit_page
; /* commited pages */
264 struct buffer_page
*reader_page
;
265 unsigned long overrun
;
266 unsigned long entries
;
269 atomic_t record_disabled
;
276 cpumask_var_t cpumask
;
277 atomic_t record_disabled
;
281 struct ring_buffer_per_cpu
**buffers
;
284 struct ring_buffer_iter
{
285 struct ring_buffer_per_cpu
*cpu_buffer
;
287 struct buffer_page
*head_page
;
291 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
292 #define RB_WARN_ON(buffer, cond) \
294 int _____ret = unlikely(cond); \
296 atomic_inc(&buffer->record_disabled); \
303 * check_pages - integrity check of buffer pages
304 * @cpu_buffer: CPU buffer with pages to test
306 * As a safty measure we check to make sure the data pages have not
309 static int rb_check_pages(struct ring_buffer_per_cpu
*cpu_buffer
)
311 struct list_head
*head
= &cpu_buffer
->pages
;
312 struct buffer_page
*bpage
, *tmp
;
314 if (RB_WARN_ON(cpu_buffer
, head
->next
->prev
!= head
))
316 if (RB_WARN_ON(cpu_buffer
, head
->prev
->next
!= head
))
319 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
320 if (RB_WARN_ON(cpu_buffer
,
321 bpage
->list
.next
->prev
!= &bpage
->list
))
323 if (RB_WARN_ON(cpu_buffer
,
324 bpage
->list
.prev
->next
!= &bpage
->list
))
331 static int rb_allocate_pages(struct ring_buffer_per_cpu
*cpu_buffer
,
334 struct list_head
*head
= &cpu_buffer
->pages
;
335 struct buffer_page
*bpage
, *tmp
;
340 for (i
= 0; i
< nr_pages
; i
++) {
341 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
342 GFP_KERNEL
, cpu_to_node(cpu_buffer
->cpu
));
345 list_add(&bpage
->list
, &pages
);
347 addr
= __get_free_page(GFP_KERNEL
);
350 bpage
->page
= (void *)addr
;
351 rb_init_page(bpage
->page
);
354 list_splice(&pages
, head
);
356 rb_check_pages(cpu_buffer
);
361 list_for_each_entry_safe(bpage
, tmp
, &pages
, list
) {
362 list_del_init(&bpage
->list
);
363 free_buffer_page(bpage
);
368 static struct ring_buffer_per_cpu
*
369 rb_allocate_cpu_buffer(struct ring_buffer
*buffer
, int cpu
)
371 struct ring_buffer_per_cpu
*cpu_buffer
;
372 struct buffer_page
*bpage
;
376 cpu_buffer
= kzalloc_node(ALIGN(sizeof(*cpu_buffer
), cache_line_size()),
377 GFP_KERNEL
, cpu_to_node(cpu
));
381 cpu_buffer
->cpu
= cpu
;
382 cpu_buffer
->buffer
= buffer
;
383 spin_lock_init(&cpu_buffer
->reader_lock
);
384 cpu_buffer
->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
385 INIT_LIST_HEAD(&cpu_buffer
->pages
);
387 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
388 GFP_KERNEL
, cpu_to_node(cpu
));
390 goto fail_free_buffer
;
392 cpu_buffer
->reader_page
= bpage
;
393 addr
= __get_free_page(GFP_KERNEL
);
395 goto fail_free_reader
;
396 bpage
->page
= (void *)addr
;
397 rb_init_page(bpage
->page
);
399 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
401 ret
= rb_allocate_pages(cpu_buffer
, buffer
->pages
);
403 goto fail_free_reader
;
405 cpu_buffer
->head_page
406 = list_entry(cpu_buffer
->pages
.next
, struct buffer_page
, list
);
407 cpu_buffer
->tail_page
= cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
412 free_buffer_page(cpu_buffer
->reader_page
);
419 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu
*cpu_buffer
)
421 struct list_head
*head
= &cpu_buffer
->pages
;
422 struct buffer_page
*bpage
, *tmp
;
424 list_del_init(&cpu_buffer
->reader_page
->list
);
425 free_buffer_page(cpu_buffer
->reader_page
);
427 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
428 list_del_init(&bpage
->list
);
429 free_buffer_page(bpage
);
435 * Causes compile errors if the struct buffer_page gets bigger
436 * than the struct page.
438 extern int ring_buffer_page_too_big(void);
441 * ring_buffer_alloc - allocate a new ring_buffer
442 * @size: the size in bytes per cpu that is needed.
443 * @flags: attributes to set for the ring buffer.
445 * Currently the only flag that is available is the RB_FL_OVERWRITE
446 * flag. This flag means that the buffer will overwrite old data
447 * when the buffer wraps. If this flag is not set, the buffer will
448 * drop data when the tail hits the head.
450 struct ring_buffer
*ring_buffer_alloc(unsigned long size
, unsigned flags
)
452 struct ring_buffer
*buffer
;
456 /* Paranoid! Optimizes out when all is well */
457 if (sizeof(struct buffer_page
) > sizeof(struct page
))
458 ring_buffer_page_too_big();
461 /* keep it in its own cache line */
462 buffer
= kzalloc(ALIGN(sizeof(*buffer
), cache_line_size()),
467 if (!alloc_cpumask_var(&buffer
->cpumask
, GFP_KERNEL
))
468 goto fail_free_buffer
;
470 buffer
->pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
471 buffer
->flags
= flags
;
473 /* need at least two pages */
474 if (buffer
->pages
== 1)
477 cpumask_copy(buffer
->cpumask
, cpu_possible_mask
);
478 buffer
->cpus
= nr_cpu_ids
;
480 bsize
= sizeof(void *) * nr_cpu_ids
;
481 buffer
->buffers
= kzalloc(ALIGN(bsize
, cache_line_size()),
483 if (!buffer
->buffers
)
484 goto fail_free_cpumask
;
486 for_each_buffer_cpu(buffer
, cpu
) {
487 buffer
->buffers
[cpu
] =
488 rb_allocate_cpu_buffer(buffer
, cpu
);
489 if (!buffer
->buffers
[cpu
])
490 goto fail_free_buffers
;
493 mutex_init(&buffer
->mutex
);
498 for_each_buffer_cpu(buffer
, cpu
) {
499 if (buffer
->buffers
[cpu
])
500 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
502 kfree(buffer
->buffers
);
505 free_cpumask_var(buffer
->cpumask
);
511 EXPORT_SYMBOL_GPL(ring_buffer_alloc
);
514 * ring_buffer_free - free a ring buffer.
515 * @buffer: the buffer to free.
518 ring_buffer_free(struct ring_buffer
*buffer
)
522 for_each_buffer_cpu(buffer
, cpu
)
523 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
525 free_cpumask_var(buffer
->cpumask
);
529 EXPORT_SYMBOL_GPL(ring_buffer_free
);
531 static void rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
);
534 rb_remove_pages(struct ring_buffer_per_cpu
*cpu_buffer
, unsigned nr_pages
)
536 struct buffer_page
*bpage
;
540 atomic_inc(&cpu_buffer
->record_disabled
);
543 for (i
= 0; i
< nr_pages
; i
++) {
544 if (RB_WARN_ON(cpu_buffer
, list_empty(&cpu_buffer
->pages
)))
546 p
= cpu_buffer
->pages
.next
;
547 bpage
= list_entry(p
, struct buffer_page
, list
);
548 list_del_init(&bpage
->list
);
549 free_buffer_page(bpage
);
551 if (RB_WARN_ON(cpu_buffer
, list_empty(&cpu_buffer
->pages
)))
554 rb_reset_cpu(cpu_buffer
);
556 rb_check_pages(cpu_buffer
);
558 atomic_dec(&cpu_buffer
->record_disabled
);
563 rb_insert_pages(struct ring_buffer_per_cpu
*cpu_buffer
,
564 struct list_head
*pages
, unsigned nr_pages
)
566 struct buffer_page
*bpage
;
570 atomic_inc(&cpu_buffer
->record_disabled
);
573 for (i
= 0; i
< nr_pages
; i
++) {
574 if (RB_WARN_ON(cpu_buffer
, list_empty(pages
)))
577 bpage
= list_entry(p
, struct buffer_page
, list
);
578 list_del_init(&bpage
->list
);
579 list_add_tail(&bpage
->list
, &cpu_buffer
->pages
);
581 rb_reset_cpu(cpu_buffer
);
583 rb_check_pages(cpu_buffer
);
585 atomic_dec(&cpu_buffer
->record_disabled
);
589 * ring_buffer_resize - resize the ring buffer
590 * @buffer: the buffer to resize.
591 * @size: the new size.
593 * The tracer is responsible for making sure that the buffer is
594 * not being used while changing the size.
595 * Note: We may be able to change the above requirement by using
596 * RCU synchronizations.
598 * Minimum size is 2 * BUF_PAGE_SIZE.
600 * Returns -1 on failure.
602 int ring_buffer_resize(struct ring_buffer
*buffer
, unsigned long size
)
604 struct ring_buffer_per_cpu
*cpu_buffer
;
605 unsigned nr_pages
, rm_pages
, new_pages
;
606 struct buffer_page
*bpage
, *tmp
;
607 unsigned long buffer_size
;
613 * Always succeed at resizing a non-existent buffer:
618 size
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
619 size
*= BUF_PAGE_SIZE
;
620 buffer_size
= buffer
->pages
* BUF_PAGE_SIZE
;
622 /* we need a minimum of two pages */
623 if (size
< BUF_PAGE_SIZE
* 2)
624 size
= BUF_PAGE_SIZE
* 2;
626 if (size
== buffer_size
)
629 mutex_lock(&buffer
->mutex
);
631 nr_pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
633 if (size
< buffer_size
) {
635 /* easy case, just free pages */
636 if (RB_WARN_ON(buffer
, nr_pages
>= buffer
->pages
)) {
637 mutex_unlock(&buffer
->mutex
);
641 rm_pages
= buffer
->pages
- nr_pages
;
643 for_each_buffer_cpu(buffer
, cpu
) {
644 cpu_buffer
= buffer
->buffers
[cpu
];
645 rb_remove_pages(cpu_buffer
, rm_pages
);
651 * This is a bit more difficult. We only want to add pages
652 * when we can allocate enough for all CPUs. We do this
653 * by allocating all the pages and storing them on a local
654 * link list. If we succeed in our allocation, then we
655 * add these pages to the cpu_buffers. Otherwise we just free
656 * them all and return -ENOMEM;
658 if (RB_WARN_ON(buffer
, nr_pages
<= buffer
->pages
)) {
659 mutex_unlock(&buffer
->mutex
);
663 new_pages
= nr_pages
- buffer
->pages
;
665 for_each_buffer_cpu(buffer
, cpu
) {
666 for (i
= 0; i
< new_pages
; i
++) {
667 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
),
669 GFP_KERNEL
, cpu_to_node(cpu
));
672 list_add(&bpage
->list
, &pages
);
673 addr
= __get_free_page(GFP_KERNEL
);
676 bpage
->page
= (void *)addr
;
677 rb_init_page(bpage
->page
);
681 for_each_buffer_cpu(buffer
, cpu
) {
682 cpu_buffer
= buffer
->buffers
[cpu
];
683 rb_insert_pages(cpu_buffer
, &pages
, new_pages
);
686 if (RB_WARN_ON(buffer
, !list_empty(&pages
))) {
687 mutex_unlock(&buffer
->mutex
);
692 buffer
->pages
= nr_pages
;
693 mutex_unlock(&buffer
->mutex
);
698 list_for_each_entry_safe(bpage
, tmp
, &pages
, list
) {
699 list_del_init(&bpage
->list
);
700 free_buffer_page(bpage
);
702 mutex_unlock(&buffer
->mutex
);
705 EXPORT_SYMBOL_GPL(ring_buffer_resize
);
707 static inline int rb_null_event(struct ring_buffer_event
*event
)
709 return event
->type
== RINGBUF_TYPE_PADDING
;
713 __rb_data_page_index(struct buffer_data_page
*bpage
, unsigned index
)
715 return bpage
->data
+ index
;
718 static inline void *__rb_page_index(struct buffer_page
*bpage
, unsigned index
)
720 return bpage
->page
->data
+ index
;
723 static inline struct ring_buffer_event
*
724 rb_reader_event(struct ring_buffer_per_cpu
*cpu_buffer
)
726 return __rb_page_index(cpu_buffer
->reader_page
,
727 cpu_buffer
->reader_page
->read
);
730 static inline struct ring_buffer_event
*
731 rb_head_event(struct ring_buffer_per_cpu
*cpu_buffer
)
733 return __rb_page_index(cpu_buffer
->head_page
,
734 cpu_buffer
->head_page
->read
);
737 static inline struct ring_buffer_event
*
738 rb_iter_head_event(struct ring_buffer_iter
*iter
)
740 return __rb_page_index(iter
->head_page
, iter
->head
);
743 static inline unsigned rb_page_write(struct buffer_page
*bpage
)
745 return local_read(&bpage
->write
);
748 static inline unsigned rb_page_commit(struct buffer_page
*bpage
)
750 return local_read(&bpage
->page
->commit
);
753 /* Size is determined by what has been commited */
754 static inline unsigned rb_page_size(struct buffer_page
*bpage
)
756 return rb_page_commit(bpage
);
759 static inline unsigned
760 rb_commit_index(struct ring_buffer_per_cpu
*cpu_buffer
)
762 return rb_page_commit(cpu_buffer
->commit_page
);
765 static inline unsigned rb_head_size(struct ring_buffer_per_cpu
*cpu_buffer
)
767 return rb_page_commit(cpu_buffer
->head_page
);
771 * When the tail hits the head and the buffer is in overwrite mode,
772 * the head jumps to the next page and all content on the previous
773 * page is discarded. But before doing so, we update the overrun
774 * variable of the buffer.
776 static void rb_update_overflow(struct ring_buffer_per_cpu
*cpu_buffer
)
778 struct ring_buffer_event
*event
;
781 for (head
= 0; head
< rb_head_size(cpu_buffer
);
782 head
+= rb_event_length(event
)) {
784 event
= __rb_page_index(cpu_buffer
->head_page
, head
);
785 if (RB_WARN_ON(cpu_buffer
, rb_null_event(event
)))
787 /* Only count data entries */
788 if (event
->type
!= RINGBUF_TYPE_DATA
)
790 cpu_buffer
->overrun
++;
791 cpu_buffer
->entries
--;
795 static inline void rb_inc_page(struct ring_buffer_per_cpu
*cpu_buffer
,
796 struct buffer_page
**bpage
)
798 struct list_head
*p
= (*bpage
)->list
.next
;
800 if (p
== &cpu_buffer
->pages
)
803 *bpage
= list_entry(p
, struct buffer_page
, list
);
806 static inline unsigned
807 rb_event_index(struct ring_buffer_event
*event
)
809 unsigned long addr
= (unsigned long)event
;
811 return (addr
& ~PAGE_MASK
) - (PAGE_SIZE
- BUF_PAGE_SIZE
);
815 rb_is_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
816 struct ring_buffer_event
*event
)
818 unsigned long addr
= (unsigned long)event
;
821 index
= rb_event_index(event
);
824 return cpu_buffer
->commit_page
->page
== (void *)addr
&&
825 rb_commit_index(cpu_buffer
) == index
;
829 rb_set_commit_event(struct ring_buffer_per_cpu
*cpu_buffer
,
830 struct ring_buffer_event
*event
)
832 unsigned long addr
= (unsigned long)event
;
835 index
= rb_event_index(event
);
838 while (cpu_buffer
->commit_page
->page
!= (void *)addr
) {
839 if (RB_WARN_ON(cpu_buffer
,
840 cpu_buffer
->commit_page
== cpu_buffer
->tail_page
))
842 cpu_buffer
->commit_page
->page
->commit
=
843 cpu_buffer
->commit_page
->write
;
844 rb_inc_page(cpu_buffer
, &cpu_buffer
->commit_page
);
845 cpu_buffer
->write_stamp
=
846 cpu_buffer
->commit_page
->page
->time_stamp
;
849 /* Now set the commit to the event's index */
850 local_set(&cpu_buffer
->commit_page
->page
->commit
, index
);
854 rb_set_commit_to_write(struct ring_buffer_per_cpu
*cpu_buffer
)
857 * We only race with interrupts and NMIs on this CPU.
858 * If we own the commit event, then we can commit
859 * all others that interrupted us, since the interruptions
860 * are in stack format (they finish before they come
861 * back to us). This allows us to do a simple loop to
862 * assign the commit to the tail.
865 while (cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
) {
866 cpu_buffer
->commit_page
->page
->commit
=
867 cpu_buffer
->commit_page
->write
;
868 rb_inc_page(cpu_buffer
, &cpu_buffer
->commit_page
);
869 cpu_buffer
->write_stamp
=
870 cpu_buffer
->commit_page
->page
->time_stamp
;
871 /* add barrier to keep gcc from optimizing too much */
874 while (rb_commit_index(cpu_buffer
) !=
875 rb_page_write(cpu_buffer
->commit_page
)) {
876 cpu_buffer
->commit_page
->page
->commit
=
877 cpu_buffer
->commit_page
->write
;
881 /* again, keep gcc from optimizing */
885 * If an interrupt came in just after the first while loop
886 * and pushed the tail page forward, we will be left with
887 * a dangling commit that will never go forward.
889 if (unlikely(cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
))
893 static void rb_reset_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
895 cpu_buffer
->read_stamp
= cpu_buffer
->reader_page
->page
->time_stamp
;
896 cpu_buffer
->reader_page
->read
= 0;
899 static inline void rb_inc_iter(struct ring_buffer_iter
*iter
)
901 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
904 * The iterator could be on the reader page (it starts there).
905 * But the head could have moved, since the reader was
906 * found. Check for this case and assign the iterator
907 * to the head page instead of next.
909 if (iter
->head_page
== cpu_buffer
->reader_page
)
910 iter
->head_page
= cpu_buffer
->head_page
;
912 rb_inc_page(cpu_buffer
, &iter
->head_page
);
914 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
919 * ring_buffer_update_event - update event type and data
920 * @event: the even to update
921 * @type: the type of event
922 * @length: the size of the event field in the ring buffer
924 * Update the type and data fields of the event. The length
925 * is the actual size that is written to the ring buffer,
926 * and with this, we can determine what to place into the
930 rb_update_event(struct ring_buffer_event
*event
,
931 unsigned type
, unsigned length
)
937 case RINGBUF_TYPE_PADDING
:
940 case RINGBUF_TYPE_TIME_EXTEND
:
942 (RB_LEN_TIME_EXTEND
+ (RB_ALIGNMENT
-1))
943 >> RB_ALIGNMENT_SHIFT
;
946 case RINGBUF_TYPE_TIME_STAMP
:
948 (RB_LEN_TIME_STAMP
+ (RB_ALIGNMENT
-1))
949 >> RB_ALIGNMENT_SHIFT
;
952 case RINGBUF_TYPE_DATA
:
953 length
-= RB_EVNT_HDR_SIZE
;
954 if (length
> RB_MAX_SMALL_DATA
) {
956 event
->array
[0] = length
;
959 (length
+ (RB_ALIGNMENT
-1))
960 >> RB_ALIGNMENT_SHIFT
;
967 static inline unsigned rb_calculate_event_length(unsigned length
)
969 struct ring_buffer_event event
; /* Used only for sizeof array */
971 /* zero length can cause confusions */
975 if (length
> RB_MAX_SMALL_DATA
)
976 length
+= sizeof(event
.array
[0]);
978 length
+= RB_EVNT_HDR_SIZE
;
979 length
= ALIGN(length
, RB_ALIGNMENT
);
984 static struct ring_buffer_event
*
985 __rb_reserve_next(struct ring_buffer_per_cpu
*cpu_buffer
,
986 unsigned type
, unsigned long length
, u64
*ts
)
988 struct buffer_page
*tail_page
, *head_page
, *reader_page
, *commit_page
;
989 unsigned long tail
, write
;
990 struct ring_buffer
*buffer
= cpu_buffer
->buffer
;
991 struct ring_buffer_event
*event
;
994 commit_page
= cpu_buffer
->commit_page
;
995 /* we just need to protect against interrupts */
997 tail_page
= cpu_buffer
->tail_page
;
998 write
= local_add_return(length
, &tail_page
->write
);
999 tail
= write
- length
;
1001 /* See if we shot pass the end of this buffer page */
1002 if (write
> BUF_PAGE_SIZE
) {
1003 struct buffer_page
*next_page
= tail_page
;
1005 local_irq_save(flags
);
1006 __raw_spin_lock(&cpu_buffer
->lock
);
1008 rb_inc_page(cpu_buffer
, &next_page
);
1010 head_page
= cpu_buffer
->head_page
;
1011 reader_page
= cpu_buffer
->reader_page
;
1013 /* we grabbed the lock before incrementing */
1014 if (RB_WARN_ON(cpu_buffer
, next_page
== reader_page
))
1018 * If for some reason, we had an interrupt storm that made
1019 * it all the way around the buffer, bail, and warn
1022 if (unlikely(next_page
== commit_page
)) {
1027 if (next_page
== head_page
) {
1028 if (!(buffer
->flags
& RB_FL_OVERWRITE
))
1031 /* tail_page has not moved yet? */
1032 if (tail_page
== cpu_buffer
->tail_page
) {
1033 /* count overflows */
1034 rb_update_overflow(cpu_buffer
);
1036 rb_inc_page(cpu_buffer
, &head_page
);
1037 cpu_buffer
->head_page
= head_page
;
1038 cpu_buffer
->head_page
->read
= 0;
1043 * If the tail page is still the same as what we think
1044 * it is, then it is up to us to update the tail
1047 if (tail_page
== cpu_buffer
->tail_page
) {
1048 local_set(&next_page
->write
, 0);
1049 local_set(&next_page
->page
->commit
, 0);
1050 cpu_buffer
->tail_page
= next_page
;
1052 /* reread the time stamp */
1053 *ts
= ring_buffer_time_stamp(cpu_buffer
->cpu
);
1054 cpu_buffer
->tail_page
->page
->time_stamp
= *ts
;
1058 * The actual tail page has moved forward.
1060 if (tail
< BUF_PAGE_SIZE
) {
1061 /* Mark the rest of the page with padding */
1062 event
= __rb_page_index(tail_page
, tail
);
1063 event
->type
= RINGBUF_TYPE_PADDING
;
1066 if (tail
<= BUF_PAGE_SIZE
)
1067 /* Set the write back to the previous setting */
1068 local_set(&tail_page
->write
, tail
);
1071 * If this was a commit entry that failed,
1072 * increment that too
1074 if (tail_page
== cpu_buffer
->commit_page
&&
1075 tail
== rb_commit_index(cpu_buffer
)) {
1076 rb_set_commit_to_write(cpu_buffer
);
1079 __raw_spin_unlock(&cpu_buffer
->lock
);
1080 local_irq_restore(flags
);
1082 /* fail and let the caller try again */
1083 return ERR_PTR(-EAGAIN
);
1086 /* We reserved something on the buffer */
1088 if (RB_WARN_ON(cpu_buffer
, write
> BUF_PAGE_SIZE
))
1091 event
= __rb_page_index(tail_page
, tail
);
1092 rb_update_event(event
, type
, length
);
1095 * If this is a commit and the tail is zero, then update
1096 * this page's time stamp.
1098 if (!tail
&& rb_is_commit(cpu_buffer
, event
))
1099 cpu_buffer
->commit_page
->page
->time_stamp
= *ts
;
1105 if (tail
<= BUF_PAGE_SIZE
)
1106 local_set(&tail_page
->write
, tail
);
1108 __raw_spin_unlock(&cpu_buffer
->lock
);
1109 local_irq_restore(flags
);
1114 rb_add_time_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
1115 u64
*ts
, u64
*delta
)
1117 struct ring_buffer_event
*event
;
1121 if (unlikely(*delta
> (1ULL << 59) && !once
++)) {
1122 printk(KERN_WARNING
"Delta way too big! %llu"
1123 " ts=%llu write stamp = %llu\n",
1124 (unsigned long long)*delta
,
1125 (unsigned long long)*ts
,
1126 (unsigned long long)cpu_buffer
->write_stamp
);
1131 * The delta is too big, we to add a
1134 event
= __rb_reserve_next(cpu_buffer
,
1135 RINGBUF_TYPE_TIME_EXTEND
,
1141 if (PTR_ERR(event
) == -EAGAIN
)
1144 /* Only a commited time event can update the write stamp */
1145 if (rb_is_commit(cpu_buffer
, event
)) {
1147 * If this is the first on the page, then we need to
1148 * update the page itself, and just put in a zero.
1150 if (rb_event_index(event
)) {
1151 event
->time_delta
= *delta
& TS_MASK
;
1152 event
->array
[0] = *delta
>> TS_SHIFT
;
1154 cpu_buffer
->commit_page
->page
->time_stamp
= *ts
;
1155 event
->time_delta
= 0;
1156 event
->array
[0] = 0;
1158 cpu_buffer
->write_stamp
= *ts
;
1159 /* let the caller know this was the commit */
1162 /* Darn, this is just wasted space */
1163 event
->time_delta
= 0;
1164 event
->array
[0] = 0;
1173 static struct ring_buffer_event
*
1174 rb_reserve_next_event(struct ring_buffer_per_cpu
*cpu_buffer
,
1175 unsigned type
, unsigned long length
)
1177 struct ring_buffer_event
*event
;
1184 * We allow for interrupts to reenter here and do a trace.
1185 * If one does, it will cause this original code to loop
1186 * back here. Even with heavy interrupts happening, this
1187 * should only happen a few times in a row. If this happens
1188 * 1000 times in a row, there must be either an interrupt
1189 * storm or we have something buggy.
1192 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 1000))
1195 ts
= ring_buffer_time_stamp(cpu_buffer
->cpu
);
1198 * Only the first commit can update the timestamp.
1199 * Yes there is a race here. If an interrupt comes in
1200 * just after the conditional and it traces too, then it
1201 * will also check the deltas. More than one timestamp may
1202 * also be made. But only the entry that did the actual
1203 * commit will be something other than zero.
1205 if (cpu_buffer
->tail_page
== cpu_buffer
->commit_page
&&
1206 rb_page_write(cpu_buffer
->tail_page
) ==
1207 rb_commit_index(cpu_buffer
)) {
1209 delta
= ts
- cpu_buffer
->write_stamp
;
1211 /* make sure this delta is calculated here */
1214 /* Did the write stamp get updated already? */
1215 if (unlikely(ts
< cpu_buffer
->write_stamp
))
1218 if (test_time_stamp(delta
)) {
1220 commit
= rb_add_time_stamp(cpu_buffer
, &ts
, &delta
);
1222 if (commit
== -EBUSY
)
1225 if (commit
== -EAGAIN
)
1228 RB_WARN_ON(cpu_buffer
, commit
< 0);
1231 /* Non commits have zero deltas */
1234 event
= __rb_reserve_next(cpu_buffer
, type
, length
, &ts
);
1235 if (PTR_ERR(event
) == -EAGAIN
)
1239 if (unlikely(commit
))
1241 * Ouch! We needed a timestamp and it was commited. But
1242 * we didn't get our event reserved.
1244 rb_set_commit_to_write(cpu_buffer
);
1249 * If the timestamp was commited, make the commit our entry
1250 * now so that we will update it when needed.
1253 rb_set_commit_event(cpu_buffer
, event
);
1254 else if (!rb_is_commit(cpu_buffer
, event
))
1257 event
->time_delta
= delta
;
1262 static DEFINE_PER_CPU(int, rb_need_resched
);
1265 * ring_buffer_lock_reserve - reserve a part of the buffer
1266 * @buffer: the ring buffer to reserve from
1267 * @length: the length of the data to reserve (excluding event header)
1268 * @flags: a pointer to save the interrupt flags
1270 * Returns a reseverd event on the ring buffer to copy directly to.
1271 * The user of this interface will need to get the body to write into
1272 * and can use the ring_buffer_event_data() interface.
1274 * The length is the length of the data needed, not the event length
1275 * which also includes the event header.
1277 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
1278 * If NULL is returned, then nothing has been allocated or locked.
1280 struct ring_buffer_event
*
1281 ring_buffer_lock_reserve(struct ring_buffer
*buffer
,
1282 unsigned long length
,
1283 unsigned long *flags
)
1285 struct ring_buffer_per_cpu
*cpu_buffer
;
1286 struct ring_buffer_event
*event
;
1289 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
1292 if (atomic_read(&buffer
->record_disabled
))
1295 /* If we are tracing schedule, we don't want to recurse */
1296 resched
= ftrace_preempt_disable();
1298 cpu
= raw_smp_processor_id();
1300 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1303 cpu_buffer
= buffer
->buffers
[cpu
];
1305 if (atomic_read(&cpu_buffer
->record_disabled
))
1308 length
= rb_calculate_event_length(length
);
1309 if (length
> BUF_PAGE_SIZE
)
1312 event
= rb_reserve_next_event(cpu_buffer
, RINGBUF_TYPE_DATA
, length
);
1317 * Need to store resched state on this cpu.
1318 * Only the first needs to.
1321 if (preempt_count() == 1)
1322 per_cpu(rb_need_resched
, cpu
) = resched
;
1327 ftrace_preempt_enable(resched
);
1330 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve
);
1332 static void rb_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
1333 struct ring_buffer_event
*event
)
1335 cpu_buffer
->entries
++;
1337 /* Only process further if we own the commit */
1338 if (!rb_is_commit(cpu_buffer
, event
))
1341 cpu_buffer
->write_stamp
+= event
->time_delta
;
1343 rb_set_commit_to_write(cpu_buffer
);
1347 * ring_buffer_unlock_commit - commit a reserved
1348 * @buffer: The buffer to commit to
1349 * @event: The event pointer to commit.
1350 * @flags: the interrupt flags received from ring_buffer_lock_reserve.
1352 * This commits the data to the ring buffer, and releases any locks held.
1354 * Must be paired with ring_buffer_lock_reserve.
1356 int ring_buffer_unlock_commit(struct ring_buffer
*buffer
,
1357 struct ring_buffer_event
*event
,
1358 unsigned long flags
)
1360 struct ring_buffer_per_cpu
*cpu_buffer
;
1361 int cpu
= raw_smp_processor_id();
1363 cpu_buffer
= buffer
->buffers
[cpu
];
1365 rb_commit(cpu_buffer
, event
);
1368 * Only the last preempt count needs to restore preemption.
1370 if (preempt_count() == 1)
1371 ftrace_preempt_enable(per_cpu(rb_need_resched
, cpu
));
1373 preempt_enable_no_resched_notrace();
1377 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit
);
1380 * ring_buffer_write - write data to the buffer without reserving
1381 * @buffer: The ring buffer to write to.
1382 * @length: The length of the data being written (excluding the event header)
1383 * @data: The data to write to the buffer.
1385 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
1386 * one function. If you already have the data to write to the buffer, it
1387 * may be easier to simply call this function.
1389 * Note, like ring_buffer_lock_reserve, the length is the length of the data
1390 * and not the length of the event which would hold the header.
1392 int ring_buffer_write(struct ring_buffer
*buffer
,
1393 unsigned long length
,
1396 struct ring_buffer_per_cpu
*cpu_buffer
;
1397 struct ring_buffer_event
*event
;
1398 unsigned long event_length
;
1403 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
1406 if (atomic_read(&buffer
->record_disabled
))
1409 resched
= ftrace_preempt_disable();
1411 cpu
= raw_smp_processor_id();
1413 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1416 cpu_buffer
= buffer
->buffers
[cpu
];
1418 if (atomic_read(&cpu_buffer
->record_disabled
))
1421 event_length
= rb_calculate_event_length(length
);
1422 event
= rb_reserve_next_event(cpu_buffer
,
1423 RINGBUF_TYPE_DATA
, event_length
);
1427 body
= rb_event_data(event
);
1429 memcpy(body
, data
, length
);
1431 rb_commit(cpu_buffer
, event
);
1435 ftrace_preempt_enable(resched
);
1439 EXPORT_SYMBOL_GPL(ring_buffer_write
);
1441 static inline int rb_per_cpu_empty(struct ring_buffer_per_cpu
*cpu_buffer
)
1443 struct buffer_page
*reader
= cpu_buffer
->reader_page
;
1444 struct buffer_page
*head
= cpu_buffer
->head_page
;
1445 struct buffer_page
*commit
= cpu_buffer
->commit_page
;
1447 return reader
->read
== rb_page_commit(reader
) &&
1448 (commit
== reader
||
1450 head
->read
== rb_page_commit(commit
)));
1454 * ring_buffer_record_disable - stop all writes into the buffer
1455 * @buffer: The ring buffer to stop writes to.
1457 * This prevents all writes to the buffer. Any attempt to write
1458 * to the buffer after this will fail and return NULL.
1460 * The caller should call synchronize_sched() after this.
1462 void ring_buffer_record_disable(struct ring_buffer
*buffer
)
1464 atomic_inc(&buffer
->record_disabled
);
1466 EXPORT_SYMBOL_GPL(ring_buffer_record_disable
);
1469 * ring_buffer_record_enable - enable writes to the buffer
1470 * @buffer: The ring buffer to enable writes
1472 * Note, multiple disables will need the same number of enables
1473 * to truely enable the writing (much like preempt_disable).
1475 void ring_buffer_record_enable(struct ring_buffer
*buffer
)
1477 atomic_dec(&buffer
->record_disabled
);
1479 EXPORT_SYMBOL_GPL(ring_buffer_record_enable
);
1482 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
1483 * @buffer: The ring buffer to stop writes to.
1484 * @cpu: The CPU buffer to stop
1486 * This prevents all writes to the buffer. Any attempt to write
1487 * to the buffer after this will fail and return NULL.
1489 * The caller should call synchronize_sched() after this.
1491 void ring_buffer_record_disable_cpu(struct ring_buffer
*buffer
, int cpu
)
1493 struct ring_buffer_per_cpu
*cpu_buffer
;
1495 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1498 cpu_buffer
= buffer
->buffers
[cpu
];
1499 atomic_inc(&cpu_buffer
->record_disabled
);
1501 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu
);
1504 * ring_buffer_record_enable_cpu - enable writes to the buffer
1505 * @buffer: The ring buffer to enable writes
1506 * @cpu: The CPU to enable.
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_cpu(struct ring_buffer
*buffer
, int cpu
)
1513 struct ring_buffer_per_cpu
*cpu_buffer
;
1515 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1518 cpu_buffer
= buffer
->buffers
[cpu
];
1519 atomic_dec(&cpu_buffer
->record_disabled
);
1521 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu
);
1524 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
1525 * @buffer: The ring buffer
1526 * @cpu: The per CPU buffer to get the entries from.
1528 unsigned long ring_buffer_entries_cpu(struct ring_buffer
*buffer
, int cpu
)
1530 struct ring_buffer_per_cpu
*cpu_buffer
;
1532 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1535 cpu_buffer
= buffer
->buffers
[cpu
];
1536 return cpu_buffer
->entries
;
1538 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu
);
1541 * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
1542 * @buffer: The ring buffer
1543 * @cpu: The per CPU buffer to get the number of overruns from
1545 unsigned long ring_buffer_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
1547 struct ring_buffer_per_cpu
*cpu_buffer
;
1549 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1552 cpu_buffer
= buffer
->buffers
[cpu
];
1553 return cpu_buffer
->overrun
;
1555 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu
);
1558 * ring_buffer_entries - get the number of entries in a buffer
1559 * @buffer: The ring buffer
1561 * Returns the total number of entries in the ring buffer
1564 unsigned long ring_buffer_entries(struct ring_buffer
*buffer
)
1566 struct ring_buffer_per_cpu
*cpu_buffer
;
1567 unsigned long entries
= 0;
1570 /* if you care about this being correct, lock the buffer */
1571 for_each_buffer_cpu(buffer
, cpu
) {
1572 cpu_buffer
= buffer
->buffers
[cpu
];
1573 entries
+= cpu_buffer
->entries
;
1578 EXPORT_SYMBOL_GPL(ring_buffer_entries
);
1581 * ring_buffer_overrun_cpu - get the number of overruns in buffer
1582 * @buffer: The ring buffer
1584 * Returns the total number of overruns in the ring buffer
1587 unsigned long ring_buffer_overruns(struct ring_buffer
*buffer
)
1589 struct ring_buffer_per_cpu
*cpu_buffer
;
1590 unsigned long overruns
= 0;
1593 /* if you care about this being correct, lock the buffer */
1594 for_each_buffer_cpu(buffer
, cpu
) {
1595 cpu_buffer
= buffer
->buffers
[cpu
];
1596 overruns
+= cpu_buffer
->overrun
;
1601 EXPORT_SYMBOL_GPL(ring_buffer_overruns
);
1603 static void rb_iter_reset(struct ring_buffer_iter
*iter
)
1605 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
1607 /* Iterator usage is expected to have record disabled */
1608 if (list_empty(&cpu_buffer
->reader_page
->list
)) {
1609 iter
->head_page
= cpu_buffer
->head_page
;
1610 iter
->head
= cpu_buffer
->head_page
->read
;
1612 iter
->head_page
= cpu_buffer
->reader_page
;
1613 iter
->head
= cpu_buffer
->reader_page
->read
;
1616 iter
->read_stamp
= cpu_buffer
->read_stamp
;
1618 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
1622 * ring_buffer_iter_reset - reset an iterator
1623 * @iter: The iterator to reset
1625 * Resets the iterator, so that it will start from the beginning
1628 void ring_buffer_iter_reset(struct ring_buffer_iter
*iter
)
1630 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
1631 unsigned long flags
;
1633 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
1634 rb_iter_reset(iter
);
1635 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
1637 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset
);
1640 * ring_buffer_iter_empty - check if an iterator has no more to read
1641 * @iter: The iterator to check
1643 int ring_buffer_iter_empty(struct ring_buffer_iter
*iter
)
1645 struct ring_buffer_per_cpu
*cpu_buffer
;
1647 cpu_buffer
= iter
->cpu_buffer
;
1649 return iter
->head_page
== cpu_buffer
->commit_page
&&
1650 iter
->head
== rb_commit_index(cpu_buffer
);
1652 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty
);
1655 rb_update_read_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
1656 struct ring_buffer_event
*event
)
1660 switch (event
->type
) {
1661 case RINGBUF_TYPE_PADDING
:
1664 case RINGBUF_TYPE_TIME_EXTEND
:
1665 delta
= event
->array
[0];
1667 delta
+= event
->time_delta
;
1668 cpu_buffer
->read_stamp
+= delta
;
1671 case RINGBUF_TYPE_TIME_STAMP
:
1672 /* FIXME: not implemented */
1675 case RINGBUF_TYPE_DATA
:
1676 cpu_buffer
->read_stamp
+= event
->time_delta
;
1686 rb_update_iter_read_stamp(struct ring_buffer_iter
*iter
,
1687 struct ring_buffer_event
*event
)
1691 switch (event
->type
) {
1692 case RINGBUF_TYPE_PADDING
:
1695 case RINGBUF_TYPE_TIME_EXTEND
:
1696 delta
= event
->array
[0];
1698 delta
+= event
->time_delta
;
1699 iter
->read_stamp
+= delta
;
1702 case RINGBUF_TYPE_TIME_STAMP
:
1703 /* FIXME: not implemented */
1706 case RINGBUF_TYPE_DATA
:
1707 iter
->read_stamp
+= event
->time_delta
;
1716 static struct buffer_page
*
1717 rb_get_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
1719 struct buffer_page
*reader
= NULL
;
1720 unsigned long flags
;
1723 local_irq_save(flags
);
1724 __raw_spin_lock(&cpu_buffer
->lock
);
1728 * This should normally only loop twice. But because the
1729 * start of the reader inserts an empty page, it causes
1730 * a case where we will loop three times. There should be no
1731 * reason to loop four times (that I know of).
1733 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 3)) {
1738 reader
= cpu_buffer
->reader_page
;
1740 /* If there's more to read, return this page */
1741 if (cpu_buffer
->reader_page
->read
< rb_page_size(reader
))
1744 /* Never should we have an index greater than the size */
1745 if (RB_WARN_ON(cpu_buffer
,
1746 cpu_buffer
->reader_page
->read
> rb_page_size(reader
)))
1749 /* check if we caught up to the tail */
1751 if (cpu_buffer
->commit_page
== cpu_buffer
->reader_page
)
1755 * Splice the empty reader page into the list around the head.
1756 * Reset the reader page to size zero.
1759 reader
= cpu_buffer
->head_page
;
1760 cpu_buffer
->reader_page
->list
.next
= reader
->list
.next
;
1761 cpu_buffer
->reader_page
->list
.prev
= reader
->list
.prev
;
1763 local_set(&cpu_buffer
->reader_page
->write
, 0);
1764 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
1766 /* Make the reader page now replace the head */
1767 reader
->list
.prev
->next
= &cpu_buffer
->reader_page
->list
;
1768 reader
->list
.next
->prev
= &cpu_buffer
->reader_page
->list
;
1771 * If the tail is on the reader, then we must set the head
1772 * to the inserted page, otherwise we set it one before.
1774 cpu_buffer
->head_page
= cpu_buffer
->reader_page
;
1776 if (cpu_buffer
->commit_page
!= reader
)
1777 rb_inc_page(cpu_buffer
, &cpu_buffer
->head_page
);
1779 /* Finally update the reader page to the new head */
1780 cpu_buffer
->reader_page
= reader
;
1781 rb_reset_reader_page(cpu_buffer
);
1786 __raw_spin_unlock(&cpu_buffer
->lock
);
1787 local_irq_restore(flags
);
1792 static void rb_advance_reader(struct ring_buffer_per_cpu
*cpu_buffer
)
1794 struct ring_buffer_event
*event
;
1795 struct buffer_page
*reader
;
1798 reader
= rb_get_reader_page(cpu_buffer
);
1800 /* This function should not be called when buffer is empty */
1801 if (RB_WARN_ON(cpu_buffer
, !reader
))
1804 event
= rb_reader_event(cpu_buffer
);
1806 if (event
->type
== RINGBUF_TYPE_DATA
)
1807 cpu_buffer
->entries
--;
1809 rb_update_read_stamp(cpu_buffer
, event
);
1811 length
= rb_event_length(event
);
1812 cpu_buffer
->reader_page
->read
+= length
;
1815 static void rb_advance_iter(struct ring_buffer_iter
*iter
)
1817 struct ring_buffer
*buffer
;
1818 struct ring_buffer_per_cpu
*cpu_buffer
;
1819 struct ring_buffer_event
*event
;
1822 cpu_buffer
= iter
->cpu_buffer
;
1823 buffer
= cpu_buffer
->buffer
;
1826 * Check if we are at the end of the buffer.
1828 if (iter
->head
>= rb_page_size(iter
->head_page
)) {
1829 if (RB_WARN_ON(buffer
,
1830 iter
->head_page
== cpu_buffer
->commit_page
))
1836 event
= rb_iter_head_event(iter
);
1838 length
= rb_event_length(event
);
1841 * This should not be called to advance the header if we are
1842 * at the tail of the buffer.
1844 if (RB_WARN_ON(cpu_buffer
,
1845 (iter
->head_page
== cpu_buffer
->commit_page
) &&
1846 (iter
->head
+ length
> rb_commit_index(cpu_buffer
))))
1849 rb_update_iter_read_stamp(iter
, event
);
1851 iter
->head
+= length
;
1853 /* check for end of page padding */
1854 if ((iter
->head
>= rb_page_size(iter
->head_page
)) &&
1855 (iter
->head_page
!= cpu_buffer
->commit_page
))
1856 rb_advance_iter(iter
);
1859 static struct ring_buffer_event
*
1860 rb_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
1862 struct ring_buffer_per_cpu
*cpu_buffer
;
1863 struct ring_buffer_event
*event
;
1864 struct buffer_page
*reader
;
1867 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1870 cpu_buffer
= buffer
->buffers
[cpu
];
1874 * We repeat when a timestamp is encountered. It is possible
1875 * to get multiple timestamps from an interrupt entering just
1876 * as one timestamp is about to be written. The max times
1877 * that this can happen is the number of nested interrupts we
1878 * can have. Nesting 10 deep of interrupts is clearly
1881 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 10))
1884 reader
= rb_get_reader_page(cpu_buffer
);
1888 event
= rb_reader_event(cpu_buffer
);
1890 switch (event
->type
) {
1891 case RINGBUF_TYPE_PADDING
:
1892 RB_WARN_ON(cpu_buffer
, 1);
1893 rb_advance_reader(cpu_buffer
);
1896 case RINGBUF_TYPE_TIME_EXTEND
:
1897 /* Internal data, OK to advance */
1898 rb_advance_reader(cpu_buffer
);
1901 case RINGBUF_TYPE_TIME_STAMP
:
1902 /* FIXME: not implemented */
1903 rb_advance_reader(cpu_buffer
);
1906 case RINGBUF_TYPE_DATA
:
1908 *ts
= cpu_buffer
->read_stamp
+ event
->time_delta
;
1909 ring_buffer_normalize_time_stamp(cpu_buffer
->cpu
, ts
);
1919 EXPORT_SYMBOL_GPL(ring_buffer_peek
);
1921 static struct ring_buffer_event
*
1922 rb_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
1924 struct ring_buffer
*buffer
;
1925 struct ring_buffer_per_cpu
*cpu_buffer
;
1926 struct ring_buffer_event
*event
;
1929 if (ring_buffer_iter_empty(iter
))
1932 cpu_buffer
= iter
->cpu_buffer
;
1933 buffer
= cpu_buffer
->buffer
;
1937 * We repeat when a timestamp is encountered. It is possible
1938 * to get multiple timestamps from an interrupt entering just
1939 * as one timestamp is about to be written. The max times
1940 * that this can happen is the number of nested interrupts we
1941 * can have. Nesting 10 deep of interrupts is clearly
1944 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 10))
1947 if (rb_per_cpu_empty(cpu_buffer
))
1950 event
= rb_iter_head_event(iter
);
1952 switch (event
->type
) {
1953 case RINGBUF_TYPE_PADDING
:
1957 case RINGBUF_TYPE_TIME_EXTEND
:
1958 /* Internal data, OK to advance */
1959 rb_advance_iter(iter
);
1962 case RINGBUF_TYPE_TIME_STAMP
:
1963 /* FIXME: not implemented */
1964 rb_advance_iter(iter
);
1967 case RINGBUF_TYPE_DATA
:
1969 *ts
= iter
->read_stamp
+ event
->time_delta
;
1970 ring_buffer_normalize_time_stamp(cpu_buffer
->cpu
, ts
);
1980 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek
);
1983 * ring_buffer_peek - peek at the next event to be read
1984 * @buffer: The ring buffer to read
1985 * @cpu: The cpu to peak at
1986 * @ts: The timestamp counter of this event.
1988 * This will return the event that will be read next, but does
1989 * not consume the data.
1991 struct ring_buffer_event
*
1992 ring_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
1994 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
1995 struct ring_buffer_event
*event
;
1996 unsigned long flags
;
1998 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
1999 event
= rb_buffer_peek(buffer
, cpu
, ts
);
2000 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2006 * ring_buffer_iter_peek - peek at the next event to be read
2007 * @iter: The ring buffer iterator
2008 * @ts: The timestamp counter of this event.
2010 * This will return the event that will be read next, but does
2011 * not increment the iterator.
2013 struct ring_buffer_event
*
2014 ring_buffer_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
2016 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2017 struct ring_buffer_event
*event
;
2018 unsigned long flags
;
2020 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2021 event
= rb_iter_peek(iter
, ts
);
2022 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2028 * ring_buffer_consume - return an event and consume it
2029 * @buffer: The ring buffer to get the next event from
2031 * Returns the next event in the ring buffer, and that event is consumed.
2032 * Meaning, that sequential reads will keep returning a different event,
2033 * and eventually empty the ring buffer if the producer is slower.
2035 struct ring_buffer_event
*
2036 ring_buffer_consume(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2038 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2039 struct ring_buffer_event
*event
;
2040 unsigned long flags
;
2042 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2045 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2047 event
= rb_buffer_peek(buffer
, cpu
, ts
);
2051 rb_advance_reader(cpu_buffer
);
2054 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2058 EXPORT_SYMBOL_GPL(ring_buffer_consume
);
2061 * ring_buffer_read_start - start a non consuming read of the buffer
2062 * @buffer: The ring buffer to read from
2063 * @cpu: The cpu buffer to iterate over
2065 * This starts up an iteration through the buffer. It also disables
2066 * the recording to the buffer until the reading is finished.
2067 * This prevents the reading from being corrupted. This is not
2068 * a consuming read, so a producer is not expected.
2070 * Must be paired with ring_buffer_finish.
2072 struct ring_buffer_iter
*
2073 ring_buffer_read_start(struct ring_buffer
*buffer
, int cpu
)
2075 struct ring_buffer_per_cpu
*cpu_buffer
;
2076 struct ring_buffer_iter
*iter
;
2077 unsigned long flags
;
2079 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2082 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
2086 cpu_buffer
= buffer
->buffers
[cpu
];
2088 iter
->cpu_buffer
= cpu_buffer
;
2090 atomic_inc(&cpu_buffer
->record_disabled
);
2091 synchronize_sched();
2093 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2094 __raw_spin_lock(&cpu_buffer
->lock
);
2095 rb_iter_reset(iter
);
2096 __raw_spin_unlock(&cpu_buffer
->lock
);
2097 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2101 EXPORT_SYMBOL_GPL(ring_buffer_read_start
);
2104 * ring_buffer_finish - finish reading the iterator of the buffer
2105 * @iter: The iterator retrieved by ring_buffer_start
2107 * This re-enables the recording to the buffer, and frees the
2111 ring_buffer_read_finish(struct ring_buffer_iter
*iter
)
2113 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2115 atomic_dec(&cpu_buffer
->record_disabled
);
2118 EXPORT_SYMBOL_GPL(ring_buffer_read_finish
);
2121 * ring_buffer_read - read the next item in the ring buffer by the iterator
2122 * @iter: The ring buffer iterator
2123 * @ts: The time stamp of the event read.
2125 * This reads the next event in the ring buffer and increments the iterator.
2127 struct ring_buffer_event
*
2128 ring_buffer_read(struct ring_buffer_iter
*iter
, u64
*ts
)
2130 struct ring_buffer_event
*event
;
2131 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2132 unsigned long flags
;
2134 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2135 event
= rb_iter_peek(iter
, ts
);
2139 rb_advance_iter(iter
);
2141 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2145 EXPORT_SYMBOL_GPL(ring_buffer_read
);
2148 * ring_buffer_size - return the size of the ring buffer (in bytes)
2149 * @buffer: The ring buffer.
2151 unsigned long ring_buffer_size(struct ring_buffer
*buffer
)
2153 return BUF_PAGE_SIZE
* buffer
->pages
;
2155 EXPORT_SYMBOL_GPL(ring_buffer_size
);
2158 rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
)
2160 cpu_buffer
->head_page
2161 = list_entry(cpu_buffer
->pages
.next
, struct buffer_page
, list
);
2162 local_set(&cpu_buffer
->head_page
->write
, 0);
2163 local_set(&cpu_buffer
->head_page
->page
->commit
, 0);
2165 cpu_buffer
->head_page
->read
= 0;
2167 cpu_buffer
->tail_page
= cpu_buffer
->head_page
;
2168 cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
2170 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
2171 local_set(&cpu_buffer
->reader_page
->write
, 0);
2172 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
2173 cpu_buffer
->reader_page
->read
= 0;
2175 cpu_buffer
->overrun
= 0;
2176 cpu_buffer
->entries
= 0;
2178 cpu_buffer
->write_stamp
= 0;
2179 cpu_buffer
->read_stamp
= 0;
2183 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
2184 * @buffer: The ring buffer to reset a per cpu buffer of
2185 * @cpu: The CPU buffer to be reset
2187 void ring_buffer_reset_cpu(struct ring_buffer
*buffer
, int cpu
)
2189 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2190 unsigned long flags
;
2192 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2195 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2197 __raw_spin_lock(&cpu_buffer
->lock
);
2199 rb_reset_cpu(cpu_buffer
);
2201 __raw_spin_unlock(&cpu_buffer
->lock
);
2203 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2205 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu
);
2208 * ring_buffer_reset - reset a ring buffer
2209 * @buffer: The ring buffer to reset all cpu buffers
2211 void ring_buffer_reset(struct ring_buffer
*buffer
)
2215 for_each_buffer_cpu(buffer
, cpu
)
2216 ring_buffer_reset_cpu(buffer
, cpu
);
2218 EXPORT_SYMBOL_GPL(ring_buffer_reset
);
2221 * rind_buffer_empty - is the ring buffer empty?
2222 * @buffer: The ring buffer to test
2224 int ring_buffer_empty(struct ring_buffer
*buffer
)
2226 struct ring_buffer_per_cpu
*cpu_buffer
;
2229 /* yes this is racy, but if you don't like the race, lock the buffer */
2230 for_each_buffer_cpu(buffer
, cpu
) {
2231 cpu_buffer
= buffer
->buffers
[cpu
];
2232 if (!rb_per_cpu_empty(cpu_buffer
))
2237 EXPORT_SYMBOL_GPL(ring_buffer_empty
);
2240 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
2241 * @buffer: The ring buffer
2242 * @cpu: The CPU buffer to test
2244 int ring_buffer_empty_cpu(struct ring_buffer
*buffer
, int cpu
)
2246 struct ring_buffer_per_cpu
*cpu_buffer
;
2248 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2251 cpu_buffer
= buffer
->buffers
[cpu
];
2252 return rb_per_cpu_empty(cpu_buffer
);
2254 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu
);
2257 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
2258 * @buffer_a: One buffer to swap with
2259 * @buffer_b: The other buffer to swap with
2261 * This function is useful for tracers that want to take a "snapshot"
2262 * of a CPU buffer and has another back up buffer lying around.
2263 * it is expected that the tracer handles the cpu buffer not being
2264 * used at the moment.
2266 int ring_buffer_swap_cpu(struct ring_buffer
*buffer_a
,
2267 struct ring_buffer
*buffer_b
, int cpu
)
2269 struct ring_buffer_per_cpu
*cpu_buffer_a
;
2270 struct ring_buffer_per_cpu
*cpu_buffer_b
;
2272 if (!cpumask_test_cpu(cpu
, buffer_a
->cpumask
) ||
2273 !cpumask_test_cpu(cpu
, buffer_b
->cpumask
))
2276 /* At least make sure the two buffers are somewhat the same */
2277 if (buffer_a
->pages
!= buffer_b
->pages
)
2280 cpu_buffer_a
= buffer_a
->buffers
[cpu
];
2281 cpu_buffer_b
= buffer_b
->buffers
[cpu
];
2284 * We can't do a synchronize_sched here because this
2285 * function can be called in atomic context.
2286 * Normally this will be called from the same CPU as cpu.
2287 * If not it's up to the caller to protect this.
2289 atomic_inc(&cpu_buffer_a
->record_disabled
);
2290 atomic_inc(&cpu_buffer_b
->record_disabled
);
2292 buffer_a
->buffers
[cpu
] = cpu_buffer_b
;
2293 buffer_b
->buffers
[cpu
] = cpu_buffer_a
;
2295 cpu_buffer_b
->buffer
= buffer_a
;
2296 cpu_buffer_a
->buffer
= buffer_b
;
2298 atomic_dec(&cpu_buffer_a
->record_disabled
);
2299 atomic_dec(&cpu_buffer_b
->record_disabled
);
2303 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu
);
2305 static void rb_remove_entries(struct ring_buffer_per_cpu
*cpu_buffer
,
2306 struct buffer_data_page
*bpage
)
2308 struct ring_buffer_event
*event
;
2311 __raw_spin_lock(&cpu_buffer
->lock
);
2312 for (head
= 0; head
< local_read(&bpage
->commit
);
2313 head
+= rb_event_length(event
)) {
2315 event
= __rb_data_page_index(bpage
, head
);
2316 if (RB_WARN_ON(cpu_buffer
, rb_null_event(event
)))
2318 /* Only count data entries */
2319 if (event
->type
!= RINGBUF_TYPE_DATA
)
2321 cpu_buffer
->entries
--;
2323 __raw_spin_unlock(&cpu_buffer
->lock
);
2327 * ring_buffer_alloc_read_page - allocate a page to read from buffer
2328 * @buffer: the buffer to allocate for.
2330 * This function is used in conjunction with ring_buffer_read_page.
2331 * When reading a full page from the ring buffer, these functions
2332 * can be used to speed up the process. The calling function should
2333 * allocate a few pages first with this function. Then when it
2334 * needs to get pages from the ring buffer, it passes the result
2335 * of this function into ring_buffer_read_page, which will swap
2336 * the page that was allocated, with the read page of the buffer.
2339 * The page allocated, or NULL on error.
2341 void *ring_buffer_alloc_read_page(struct ring_buffer
*buffer
)
2344 struct buffer_data_page
*bpage
;
2346 addr
= __get_free_page(GFP_KERNEL
);
2350 bpage
= (void *)addr
;
2356 * ring_buffer_free_read_page - free an allocated read page
2357 * @buffer: the buffer the page was allocate for
2358 * @data: the page to free
2360 * Free a page allocated from ring_buffer_alloc_read_page.
2362 void ring_buffer_free_read_page(struct ring_buffer
*buffer
, void *data
)
2364 free_page((unsigned long)data
);
2368 * ring_buffer_read_page - extract a page from the ring buffer
2369 * @buffer: buffer to extract from
2370 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
2371 * @cpu: the cpu of the buffer to extract
2372 * @full: should the extraction only happen when the page is full.
2374 * This function will pull out a page from the ring buffer and consume it.
2375 * @data_page must be the address of the variable that was returned
2376 * from ring_buffer_alloc_read_page. This is because the page might be used
2377 * to swap with a page in the ring buffer.
2380 * rpage = ring_buffer_alloc_page(buffer);
2383 * ret = ring_buffer_read_page(buffer, &rpage, cpu, 0);
2385 * process_page(rpage);
2387 * When @full is set, the function will not return true unless
2388 * the writer is off the reader page.
2390 * Note: it is up to the calling functions to handle sleeps and wakeups.
2391 * The ring buffer can be used anywhere in the kernel and can not
2392 * blindly call wake_up. The layer that uses the ring buffer must be
2393 * responsible for that.
2396 * 1 if data has been transferred
2397 * 0 if no data has been transferred.
2399 int ring_buffer_read_page(struct ring_buffer
*buffer
,
2400 void **data_page
, int cpu
, int full
)
2402 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2403 struct ring_buffer_event
*event
;
2404 struct buffer_data_page
*bpage
;
2405 unsigned long flags
;
2415 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2418 * rb_buffer_peek will get the next ring buffer if
2419 * the current reader page is empty.
2421 event
= rb_buffer_peek(buffer
, cpu
, NULL
);
2425 /* check for data */
2426 if (!local_read(&cpu_buffer
->reader_page
->page
->commit
))
2429 * If the writer is already off of the read page, then simply
2430 * switch the read page with the given page. Otherwise
2431 * we need to copy the data from the reader to the writer.
2433 if (cpu_buffer
->reader_page
== cpu_buffer
->commit_page
) {
2434 unsigned int read
= cpu_buffer
->reader_page
->read
;
2438 /* The writer is still on the reader page, we must copy */
2439 bpage
= cpu_buffer
->reader_page
->page
;
2441 cpu_buffer
->reader_page
->page
->data
+ read
,
2442 local_read(&bpage
->commit
) - read
);
2444 /* consume what was read */
2445 cpu_buffer
->reader_page
+= read
;
2448 /* swap the pages */
2449 rb_init_page(bpage
);
2450 bpage
= cpu_buffer
->reader_page
->page
;
2451 cpu_buffer
->reader_page
->page
= *data_page
;
2452 cpu_buffer
->reader_page
->read
= 0;
2457 /* update the entry counter */
2458 rb_remove_entries(cpu_buffer
, bpage
);
2460 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2466 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
2467 size_t cnt
, loff_t
*ppos
)
2469 long *p
= filp
->private_data
;
2473 if (test_bit(RB_BUFFERS_DISABLED_BIT
, p
))
2474 r
= sprintf(buf
, "permanently disabled\n");
2476 r
= sprintf(buf
, "%d\n", test_bit(RB_BUFFERS_ON_BIT
, p
));
2478 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2482 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
2483 size_t cnt
, loff_t
*ppos
)
2485 long *p
= filp
->private_data
;
2490 if (cnt
>= sizeof(buf
))
2493 if (copy_from_user(&buf
, ubuf
, cnt
))
2498 ret
= strict_strtoul(buf
, 10, &val
);
2503 set_bit(RB_BUFFERS_ON_BIT
, p
);
2505 clear_bit(RB_BUFFERS_ON_BIT
, p
);
2512 static struct file_operations rb_simple_fops
= {
2513 .open
= tracing_open_generic
,
2514 .read
= rb_simple_read
,
2515 .write
= rb_simple_write
,
2519 static __init
int rb_init_debugfs(void)
2521 struct dentry
*d_tracer
;
2522 struct dentry
*entry
;
2524 d_tracer
= tracing_init_dentry();
2526 entry
= debugfs_create_file("tracing_on", 0644, d_tracer
,
2527 &ring_buffer_flags
, &rb_simple_fops
);
2529 pr_warning("Could not create debugfs 'tracing_on' entry\n");
2534 fs_initcall(rb_init_debugfs
);