4 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
6 #include <linux/ring_buffer.h>
7 #include <linux/trace_clock.h>
8 #include <linux/ftrace_irq.h>
9 #include <linux/spinlock.h>
10 #include <linux/debugfs.h>
11 #include <linux/uaccess.h>
12 #include <linux/hardirq.h>
13 #include <linux/module.h>
14 #include <linux/percpu.h>
15 #include <linux/mutex.h>
16 #include <linux/init.h>
17 #include <linux/hash.h>
18 #include <linux/list.h>
19 #include <linux/cpu.h>
25 * The ring buffer header is special. We must manually up keep it.
27 int ring_buffer_print_entry_header(struct trace_seq
*s
)
31 ret
= trace_seq_printf(s
, "# compressed entry header\n");
32 ret
= trace_seq_printf(s
, "\ttype_len : 5 bits\n");
33 ret
= trace_seq_printf(s
, "\ttime_delta : 27 bits\n");
34 ret
= trace_seq_printf(s
, "\tarray : 32 bits\n");
35 ret
= trace_seq_printf(s
, "\n");
36 ret
= trace_seq_printf(s
, "\tpadding : type == %d\n",
37 RINGBUF_TYPE_PADDING
);
38 ret
= trace_seq_printf(s
, "\ttime_extend : type == %d\n",
39 RINGBUF_TYPE_TIME_EXTEND
);
40 ret
= trace_seq_printf(s
, "\tdata max type_len == %d\n",
41 RINGBUF_TYPE_DATA_TYPE_LEN_MAX
);
47 * The ring buffer is made up of a list of pages. A separate list of pages is
48 * allocated for each CPU. A writer may only write to a buffer that is
49 * associated with the CPU it is currently executing on. A reader may read
50 * from any per cpu buffer.
52 * The reader is special. For each per cpu buffer, the reader has its own
53 * reader page. When a reader has read the entire reader page, this reader
54 * page is swapped with another page in the ring buffer.
56 * Now, as long as the writer is off the reader page, the reader can do what
57 * ever it wants with that page. The writer will never write to that page
58 * again (as long as it is out of the ring buffer).
60 * Here's some silly ASCII art.
63 * |reader| RING BUFFER
65 * +------+ +---+ +---+ +---+
74 * |reader| RING BUFFER
75 * |page |------------------v
76 * +------+ +---+ +---+ +---+
85 * |reader| RING BUFFER
86 * |page |------------------v
87 * +------+ +---+ +---+ +---+
92 * +------------------------------+
96 * |buffer| RING BUFFER
97 * |page |------------------v
98 * +------+ +---+ +---+ +---+
100 * | New +---+ +---+ +---+
103 * +------------------------------+
106 * After we make this swap, the reader can hand this page off to the splice
107 * code and be done with it. It can even allocate a new page if it needs to
108 * and swap that into the ring buffer.
110 * We will be using cmpxchg soon to make all this lockless.
115 * A fast way to enable or disable all ring buffers is to
116 * call tracing_on or tracing_off. Turning off the ring buffers
117 * prevents all ring buffers from being recorded to.
118 * Turning this switch on, makes it OK to write to the
119 * ring buffer, if the ring buffer is enabled itself.
121 * There's three layers that must be on in order to write
122 * to the ring buffer.
124 * 1) This global flag must be set.
125 * 2) The ring buffer must be enabled for recording.
126 * 3) The per cpu buffer must be enabled for recording.
128 * In case of an anomaly, this global flag has a bit set that
129 * will permantly disable all ring buffers.
133 * Global flag to disable all recording to ring buffers
134 * This has two bits: ON, DISABLED
138 * 0 0 : ring buffers are off
139 * 1 0 : ring buffers are on
140 * X 1 : ring buffers are permanently disabled
144 RB_BUFFERS_ON_BIT
= 0,
145 RB_BUFFERS_DISABLED_BIT
= 1,
149 RB_BUFFERS_ON
= 1 << RB_BUFFERS_ON_BIT
,
150 RB_BUFFERS_DISABLED
= 1 << RB_BUFFERS_DISABLED_BIT
,
153 static unsigned long ring_buffer_flags __read_mostly
= RB_BUFFERS_ON
;
155 #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
158 * tracing_on - enable all tracing buffers
160 * This function enables all tracing buffers that may have been
161 * disabled with tracing_off.
163 void tracing_on(void)
165 set_bit(RB_BUFFERS_ON_BIT
, &ring_buffer_flags
);
167 EXPORT_SYMBOL_GPL(tracing_on
);
170 * tracing_off - turn off all tracing buffers
172 * This function stops all tracing buffers from recording data.
173 * It does not disable any overhead the tracers themselves may
174 * be causing. This function simply causes all recording to
175 * the ring buffers to fail.
177 void tracing_off(void)
179 clear_bit(RB_BUFFERS_ON_BIT
, &ring_buffer_flags
);
181 EXPORT_SYMBOL_GPL(tracing_off
);
184 * tracing_off_permanent - permanently disable ring buffers
186 * This function, once called, will disable all ring buffers
189 void tracing_off_permanent(void)
191 set_bit(RB_BUFFERS_DISABLED_BIT
, &ring_buffer_flags
);
195 * tracing_is_on - show state of ring buffers enabled
197 int tracing_is_on(void)
199 return ring_buffer_flags
== RB_BUFFERS_ON
;
201 EXPORT_SYMBOL_GPL(tracing_is_on
);
205 #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
206 #define RB_ALIGNMENT 4U
207 #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
209 /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
210 #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
213 RB_LEN_TIME_EXTEND
= 8,
214 RB_LEN_TIME_STAMP
= 16,
217 static inline int rb_null_event(struct ring_buffer_event
*event
)
219 return event
->type_len
== RINGBUF_TYPE_PADDING
220 && event
->time_delta
== 0;
223 static inline int rb_discarded_event(struct ring_buffer_event
*event
)
225 return event
->type_len
== RINGBUF_TYPE_PADDING
&& event
->time_delta
;
228 static void rb_event_set_padding(struct ring_buffer_event
*event
)
230 event
->type_len
= RINGBUF_TYPE_PADDING
;
231 event
->time_delta
= 0;
235 rb_event_data_length(struct ring_buffer_event
*event
)
240 length
= event
->type_len
* RB_ALIGNMENT
;
242 length
= event
->array
[0];
243 return length
+ RB_EVNT_HDR_SIZE
;
246 /* inline for ring buffer fast paths */
248 rb_event_length(struct ring_buffer_event
*event
)
250 switch (event
->type_len
) {
251 case RINGBUF_TYPE_PADDING
:
252 if (rb_null_event(event
))
255 return event
->array
[0] + RB_EVNT_HDR_SIZE
;
257 case RINGBUF_TYPE_TIME_EXTEND
:
258 return RB_LEN_TIME_EXTEND
;
260 case RINGBUF_TYPE_TIME_STAMP
:
261 return RB_LEN_TIME_STAMP
;
263 case RINGBUF_TYPE_DATA
:
264 return rb_event_data_length(event
);
273 * ring_buffer_event_length - return the length of the event
274 * @event: the event to get the length of
276 unsigned ring_buffer_event_length(struct ring_buffer_event
*event
)
278 unsigned length
= rb_event_length(event
);
279 if (event
->type_len
> RINGBUF_TYPE_DATA_TYPE_LEN_MAX
)
281 length
-= RB_EVNT_HDR_SIZE
;
282 if (length
> RB_MAX_SMALL_DATA
+ sizeof(event
->array
[0]))
283 length
-= sizeof(event
->array
[0]);
286 EXPORT_SYMBOL_GPL(ring_buffer_event_length
);
288 /* inline for ring buffer fast paths */
290 rb_event_data(struct ring_buffer_event
*event
)
292 BUG_ON(event
->type_len
> RINGBUF_TYPE_DATA_TYPE_LEN_MAX
);
293 /* If length is in len field, then array[0] has the data */
295 return (void *)&event
->array
[0];
296 /* Otherwise length is in array[0] and array[1] has the data */
297 return (void *)&event
->array
[1];
301 * ring_buffer_event_data - return the data of the event
302 * @event: the event to get the data from
304 void *ring_buffer_event_data(struct ring_buffer_event
*event
)
306 return rb_event_data(event
);
308 EXPORT_SYMBOL_GPL(ring_buffer_event_data
);
310 #define for_each_buffer_cpu(buffer, cpu) \
311 for_each_cpu(cpu, buffer->cpumask)
314 #define TS_MASK ((1ULL << TS_SHIFT) - 1)
315 #define TS_DELTA_TEST (~TS_MASK)
317 struct buffer_data_page
{
318 u64 time_stamp
; /* page time stamp */
319 local_t commit
; /* write committed index */
320 unsigned char data
[]; /* data of buffer page */
324 local_t write
; /* index for next write */
325 unsigned read
; /* index for next read */
326 struct list_head list
; /* list of free pages */
327 struct buffer_data_page
*page
; /* Actual data page */
330 static void rb_init_page(struct buffer_data_page
*bpage
)
332 local_set(&bpage
->commit
, 0);
336 * ring_buffer_page_len - the size of data on the page.
337 * @page: The page to read
339 * Returns the amount of data on the page, including buffer page header.
341 size_t ring_buffer_page_len(void *page
)
343 return local_read(&((struct buffer_data_page
*)page
)->commit
)
348 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
351 static void free_buffer_page(struct buffer_page
*bpage
)
353 free_page((unsigned long)bpage
->page
);
358 * We need to fit the time_stamp delta into 27 bits.
360 static inline int test_time_stamp(u64 delta
)
362 if (delta
& TS_DELTA_TEST
)
367 #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
369 int ring_buffer_print_page_header(struct trace_seq
*s
)
371 struct buffer_data_page field
;
374 ret
= trace_seq_printf(s
, "\tfield: u64 timestamp;\t"
375 "offset:0;\tsize:%u;\n",
376 (unsigned int)sizeof(field
.time_stamp
));
378 ret
= trace_seq_printf(s
, "\tfield: local_t commit;\t"
379 "offset:%u;\tsize:%u;\n",
380 (unsigned int)offsetof(typeof(field
), commit
),
381 (unsigned int)sizeof(field
.commit
));
383 ret
= trace_seq_printf(s
, "\tfield: char data;\t"
384 "offset:%u;\tsize:%u;\n",
385 (unsigned int)offsetof(typeof(field
), data
),
386 (unsigned int)BUF_PAGE_SIZE
);
392 * head_page == tail_page && head == tail then buffer is empty.
394 struct ring_buffer_per_cpu
{
396 struct ring_buffer
*buffer
;
397 spinlock_t reader_lock
; /* serialize readers */
399 struct lock_class_key lock_key
;
400 struct list_head pages
;
401 struct buffer_page
*head_page
; /* read from head */
402 struct buffer_page
*tail_page
; /* write to tail */
403 struct buffer_page
*commit_page
; /* committed pages */
404 struct buffer_page
*reader_page
;
405 unsigned long overrun
;
406 unsigned long entries
;
409 atomic_t record_disabled
;
416 atomic_t record_disabled
;
417 cpumask_var_t cpumask
;
421 struct ring_buffer_per_cpu
**buffers
;
423 #ifdef CONFIG_HOTPLUG_CPU
424 struct notifier_block cpu_notify
;
429 struct ring_buffer_iter
{
430 struct ring_buffer_per_cpu
*cpu_buffer
;
432 struct buffer_page
*head_page
;
436 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
437 #define RB_WARN_ON(buffer, cond) \
439 int _____ret = unlikely(cond); \
441 atomic_inc(&buffer->record_disabled); \
447 /* Up this if you want to test the TIME_EXTENTS and normalization */
448 #define DEBUG_SHIFT 0
450 u64
ring_buffer_time_stamp(struct ring_buffer
*buffer
, int cpu
)
454 preempt_disable_notrace();
455 /* shift to debug/test normalization and TIME_EXTENTS */
456 time
= buffer
->clock() << DEBUG_SHIFT
;
457 preempt_enable_no_resched_notrace();
461 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp
);
463 void ring_buffer_normalize_time_stamp(struct ring_buffer
*buffer
,
466 /* Just stupid testing the normalize function and deltas */
469 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp
);
472 * check_pages - integrity check of buffer pages
473 * @cpu_buffer: CPU buffer with pages to test
475 * As a safety measure we check to make sure the data pages have not
478 static int rb_check_pages(struct ring_buffer_per_cpu
*cpu_buffer
)
480 struct list_head
*head
= &cpu_buffer
->pages
;
481 struct buffer_page
*bpage
, *tmp
;
483 if (RB_WARN_ON(cpu_buffer
, head
->next
->prev
!= head
))
485 if (RB_WARN_ON(cpu_buffer
, head
->prev
->next
!= head
))
488 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
489 if (RB_WARN_ON(cpu_buffer
,
490 bpage
->list
.next
->prev
!= &bpage
->list
))
492 if (RB_WARN_ON(cpu_buffer
,
493 bpage
->list
.prev
->next
!= &bpage
->list
))
500 static int rb_allocate_pages(struct ring_buffer_per_cpu
*cpu_buffer
,
503 struct list_head
*head
= &cpu_buffer
->pages
;
504 struct buffer_page
*bpage
, *tmp
;
509 for (i
= 0; i
< nr_pages
; i
++) {
510 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
511 GFP_KERNEL
, cpu_to_node(cpu_buffer
->cpu
));
514 list_add(&bpage
->list
, &pages
);
516 addr
= __get_free_page(GFP_KERNEL
);
519 bpage
->page
= (void *)addr
;
520 rb_init_page(bpage
->page
);
523 list_splice(&pages
, head
);
525 rb_check_pages(cpu_buffer
);
530 list_for_each_entry_safe(bpage
, tmp
, &pages
, list
) {
531 list_del_init(&bpage
->list
);
532 free_buffer_page(bpage
);
537 static struct ring_buffer_per_cpu
*
538 rb_allocate_cpu_buffer(struct ring_buffer
*buffer
, int cpu
)
540 struct ring_buffer_per_cpu
*cpu_buffer
;
541 struct buffer_page
*bpage
;
545 cpu_buffer
= kzalloc_node(ALIGN(sizeof(*cpu_buffer
), cache_line_size()),
546 GFP_KERNEL
, cpu_to_node(cpu
));
550 cpu_buffer
->cpu
= cpu
;
551 cpu_buffer
->buffer
= buffer
;
552 spin_lock_init(&cpu_buffer
->reader_lock
);
553 cpu_buffer
->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
554 INIT_LIST_HEAD(&cpu_buffer
->pages
);
556 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
557 GFP_KERNEL
, cpu_to_node(cpu
));
559 goto fail_free_buffer
;
561 cpu_buffer
->reader_page
= bpage
;
562 addr
= __get_free_page(GFP_KERNEL
);
564 goto fail_free_reader
;
565 bpage
->page
= (void *)addr
;
566 rb_init_page(bpage
->page
);
568 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
570 ret
= rb_allocate_pages(cpu_buffer
, buffer
->pages
);
572 goto fail_free_reader
;
574 cpu_buffer
->head_page
575 = list_entry(cpu_buffer
->pages
.next
, struct buffer_page
, list
);
576 cpu_buffer
->tail_page
= cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
581 free_buffer_page(cpu_buffer
->reader_page
);
588 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu
*cpu_buffer
)
590 struct list_head
*head
= &cpu_buffer
->pages
;
591 struct buffer_page
*bpage
, *tmp
;
593 free_buffer_page(cpu_buffer
->reader_page
);
595 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
596 list_del_init(&bpage
->list
);
597 free_buffer_page(bpage
);
603 * Causes compile errors if the struct buffer_page gets bigger
604 * than the struct page.
606 extern int ring_buffer_page_too_big(void);
608 #ifdef CONFIG_HOTPLUG_CPU
609 static int rb_cpu_notify(struct notifier_block
*self
,
610 unsigned long action
, void *hcpu
);
614 * ring_buffer_alloc - allocate a new ring_buffer
615 * @size: the size in bytes per cpu that is needed.
616 * @flags: attributes to set for the ring buffer.
618 * Currently the only flag that is available is the RB_FL_OVERWRITE
619 * flag. This flag means that the buffer will overwrite old data
620 * when the buffer wraps. If this flag is not set, the buffer will
621 * drop data when the tail hits the head.
623 struct ring_buffer
*ring_buffer_alloc(unsigned long size
, unsigned flags
)
625 struct ring_buffer
*buffer
;
629 /* Paranoid! Optimizes out when all is well */
630 if (sizeof(struct buffer_page
) > sizeof(struct page
))
631 ring_buffer_page_too_big();
634 /* keep it in its own cache line */
635 buffer
= kzalloc(ALIGN(sizeof(*buffer
), cache_line_size()),
640 if (!alloc_cpumask_var(&buffer
->cpumask
, GFP_KERNEL
))
641 goto fail_free_buffer
;
643 buffer
->pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
644 buffer
->flags
= flags
;
645 buffer
->clock
= trace_clock_local
;
647 /* need at least two pages */
648 if (buffer
->pages
== 1)
652 * In case of non-hotplug cpu, if the ring-buffer is allocated
653 * in early initcall, it will not be notified of secondary cpus.
654 * In that off case, we need to allocate for all possible cpus.
656 #ifdef CONFIG_HOTPLUG_CPU
658 cpumask_copy(buffer
->cpumask
, cpu_online_mask
);
660 cpumask_copy(buffer
->cpumask
, cpu_possible_mask
);
662 buffer
->cpus
= nr_cpu_ids
;
664 bsize
= sizeof(void *) * nr_cpu_ids
;
665 buffer
->buffers
= kzalloc(ALIGN(bsize
, cache_line_size()),
667 if (!buffer
->buffers
)
668 goto fail_free_cpumask
;
670 for_each_buffer_cpu(buffer
, cpu
) {
671 buffer
->buffers
[cpu
] =
672 rb_allocate_cpu_buffer(buffer
, cpu
);
673 if (!buffer
->buffers
[cpu
])
674 goto fail_free_buffers
;
677 #ifdef CONFIG_HOTPLUG_CPU
678 buffer
->cpu_notify
.notifier_call
= rb_cpu_notify
;
679 buffer
->cpu_notify
.priority
= 0;
680 register_cpu_notifier(&buffer
->cpu_notify
);
684 mutex_init(&buffer
->mutex
);
689 for_each_buffer_cpu(buffer
, cpu
) {
690 if (buffer
->buffers
[cpu
])
691 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
693 kfree(buffer
->buffers
);
696 free_cpumask_var(buffer
->cpumask
);
703 EXPORT_SYMBOL_GPL(ring_buffer_alloc
);
706 * ring_buffer_free - free a ring buffer.
707 * @buffer: the buffer to free.
710 ring_buffer_free(struct ring_buffer
*buffer
)
716 #ifdef CONFIG_HOTPLUG_CPU
717 unregister_cpu_notifier(&buffer
->cpu_notify
);
720 for_each_buffer_cpu(buffer
, cpu
)
721 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
725 free_cpumask_var(buffer
->cpumask
);
729 EXPORT_SYMBOL_GPL(ring_buffer_free
);
731 void ring_buffer_set_clock(struct ring_buffer
*buffer
,
734 buffer
->clock
= clock
;
737 static void rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
);
740 rb_remove_pages(struct ring_buffer_per_cpu
*cpu_buffer
, unsigned nr_pages
)
742 struct buffer_page
*bpage
;
746 atomic_inc(&cpu_buffer
->record_disabled
);
749 for (i
= 0; i
< nr_pages
; i
++) {
750 if (RB_WARN_ON(cpu_buffer
, list_empty(&cpu_buffer
->pages
)))
752 p
= cpu_buffer
->pages
.next
;
753 bpage
= list_entry(p
, struct buffer_page
, list
);
754 list_del_init(&bpage
->list
);
755 free_buffer_page(bpage
);
757 if (RB_WARN_ON(cpu_buffer
, list_empty(&cpu_buffer
->pages
)))
760 rb_reset_cpu(cpu_buffer
);
762 rb_check_pages(cpu_buffer
);
764 atomic_dec(&cpu_buffer
->record_disabled
);
769 rb_insert_pages(struct ring_buffer_per_cpu
*cpu_buffer
,
770 struct list_head
*pages
, unsigned nr_pages
)
772 struct buffer_page
*bpage
;
776 atomic_inc(&cpu_buffer
->record_disabled
);
779 for (i
= 0; i
< nr_pages
; i
++) {
780 if (RB_WARN_ON(cpu_buffer
, list_empty(pages
)))
783 bpage
= list_entry(p
, struct buffer_page
, list
);
784 list_del_init(&bpage
->list
);
785 list_add_tail(&bpage
->list
, &cpu_buffer
->pages
);
787 rb_reset_cpu(cpu_buffer
);
789 rb_check_pages(cpu_buffer
);
791 atomic_dec(&cpu_buffer
->record_disabled
);
795 * ring_buffer_resize - resize the ring buffer
796 * @buffer: the buffer to resize.
797 * @size: the new size.
799 * The tracer is responsible for making sure that the buffer is
800 * not being used while changing the size.
801 * Note: We may be able to change the above requirement by using
802 * RCU synchronizations.
804 * Minimum size is 2 * BUF_PAGE_SIZE.
806 * Returns -1 on failure.
808 int ring_buffer_resize(struct ring_buffer
*buffer
, unsigned long size
)
810 struct ring_buffer_per_cpu
*cpu_buffer
;
811 unsigned nr_pages
, rm_pages
, new_pages
;
812 struct buffer_page
*bpage
, *tmp
;
813 unsigned long buffer_size
;
819 * Always succeed at resizing a non-existent buffer:
824 size
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
825 size
*= BUF_PAGE_SIZE
;
826 buffer_size
= buffer
->pages
* BUF_PAGE_SIZE
;
828 /* we need a minimum of two pages */
829 if (size
< BUF_PAGE_SIZE
* 2)
830 size
= BUF_PAGE_SIZE
* 2;
832 if (size
== buffer_size
)
835 mutex_lock(&buffer
->mutex
);
838 nr_pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
840 if (size
< buffer_size
) {
842 /* easy case, just free pages */
843 if (RB_WARN_ON(buffer
, nr_pages
>= buffer
->pages
))
846 rm_pages
= buffer
->pages
- nr_pages
;
848 for_each_buffer_cpu(buffer
, cpu
) {
849 cpu_buffer
= buffer
->buffers
[cpu
];
850 rb_remove_pages(cpu_buffer
, rm_pages
);
856 * This is a bit more difficult. We only want to add pages
857 * when we can allocate enough for all CPUs. We do this
858 * by allocating all the pages and storing them on a local
859 * link list. If we succeed in our allocation, then we
860 * add these pages to the cpu_buffers. Otherwise we just free
861 * them all and return -ENOMEM;
863 if (RB_WARN_ON(buffer
, nr_pages
<= buffer
->pages
))
866 new_pages
= nr_pages
- buffer
->pages
;
868 for_each_buffer_cpu(buffer
, cpu
) {
869 for (i
= 0; i
< new_pages
; i
++) {
870 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
),
872 GFP_KERNEL
, cpu_to_node(cpu
));
875 list_add(&bpage
->list
, &pages
);
876 addr
= __get_free_page(GFP_KERNEL
);
879 bpage
->page
= (void *)addr
;
880 rb_init_page(bpage
->page
);
884 for_each_buffer_cpu(buffer
, cpu
) {
885 cpu_buffer
= buffer
->buffers
[cpu
];
886 rb_insert_pages(cpu_buffer
, &pages
, new_pages
);
889 if (RB_WARN_ON(buffer
, !list_empty(&pages
)))
893 buffer
->pages
= nr_pages
;
895 mutex_unlock(&buffer
->mutex
);
900 list_for_each_entry_safe(bpage
, tmp
, &pages
, list
) {
901 list_del_init(&bpage
->list
);
902 free_buffer_page(bpage
);
905 mutex_unlock(&buffer
->mutex
);
909 * Something went totally wrong, and we are too paranoid
910 * to even clean up the mess.
914 mutex_unlock(&buffer
->mutex
);
917 EXPORT_SYMBOL_GPL(ring_buffer_resize
);
920 __rb_data_page_index(struct buffer_data_page
*bpage
, unsigned index
)
922 return bpage
->data
+ index
;
925 static inline void *__rb_page_index(struct buffer_page
*bpage
, unsigned index
)
927 return bpage
->page
->data
+ index
;
930 static inline struct ring_buffer_event
*
931 rb_reader_event(struct ring_buffer_per_cpu
*cpu_buffer
)
933 return __rb_page_index(cpu_buffer
->reader_page
,
934 cpu_buffer
->reader_page
->read
);
937 static inline struct ring_buffer_event
*
938 rb_head_event(struct ring_buffer_per_cpu
*cpu_buffer
)
940 return __rb_page_index(cpu_buffer
->head_page
,
941 cpu_buffer
->head_page
->read
);
944 static inline struct ring_buffer_event
*
945 rb_iter_head_event(struct ring_buffer_iter
*iter
)
947 return __rb_page_index(iter
->head_page
, iter
->head
);
950 static inline unsigned rb_page_write(struct buffer_page
*bpage
)
952 return local_read(&bpage
->write
);
955 static inline unsigned rb_page_commit(struct buffer_page
*bpage
)
957 return local_read(&bpage
->page
->commit
);
960 /* Size is determined by what has been commited */
961 static inline unsigned rb_page_size(struct buffer_page
*bpage
)
963 return rb_page_commit(bpage
);
966 static inline unsigned
967 rb_commit_index(struct ring_buffer_per_cpu
*cpu_buffer
)
969 return rb_page_commit(cpu_buffer
->commit_page
);
972 static inline unsigned rb_head_size(struct ring_buffer_per_cpu
*cpu_buffer
)
974 return rb_page_commit(cpu_buffer
->head_page
);
978 * When the tail hits the head and the buffer is in overwrite mode,
979 * the head jumps to the next page and all content on the previous
980 * page is discarded. But before doing so, we update the overrun
981 * variable of the buffer.
983 static void rb_update_overflow(struct ring_buffer_per_cpu
*cpu_buffer
)
985 struct ring_buffer_event
*event
;
988 for (head
= 0; head
< rb_head_size(cpu_buffer
);
989 head
+= rb_event_length(event
)) {
991 event
= __rb_page_index(cpu_buffer
->head_page
, head
);
992 if (RB_WARN_ON(cpu_buffer
, rb_null_event(event
)))
994 /* Only count data entries */
995 if (event
->type_len
> RINGBUF_TYPE_DATA_TYPE_LEN_MAX
)
997 cpu_buffer
->overrun
++;
998 cpu_buffer
->entries
--;
1002 static inline void rb_inc_page(struct ring_buffer_per_cpu
*cpu_buffer
,
1003 struct buffer_page
**bpage
)
1005 struct list_head
*p
= (*bpage
)->list
.next
;
1007 if (p
== &cpu_buffer
->pages
)
1010 *bpage
= list_entry(p
, struct buffer_page
, list
);
1013 static inline unsigned
1014 rb_event_index(struct ring_buffer_event
*event
)
1016 unsigned long addr
= (unsigned long)event
;
1018 return (addr
& ~PAGE_MASK
) - (PAGE_SIZE
- BUF_PAGE_SIZE
);
1022 rb_is_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
1023 struct ring_buffer_event
*event
)
1025 unsigned long addr
= (unsigned long)event
;
1026 unsigned long index
;
1028 index
= rb_event_index(event
);
1031 return cpu_buffer
->commit_page
->page
== (void *)addr
&&
1032 rb_commit_index(cpu_buffer
) == index
;
1036 rb_set_commit_event(struct ring_buffer_per_cpu
*cpu_buffer
,
1037 struct ring_buffer_event
*event
)
1039 unsigned long addr
= (unsigned long)event
;
1040 unsigned long index
;
1042 index
= rb_event_index(event
);
1045 while (cpu_buffer
->commit_page
->page
!= (void *)addr
) {
1046 if (RB_WARN_ON(cpu_buffer
,
1047 cpu_buffer
->commit_page
== cpu_buffer
->tail_page
))
1049 cpu_buffer
->commit_page
->page
->commit
=
1050 cpu_buffer
->commit_page
->write
;
1051 rb_inc_page(cpu_buffer
, &cpu_buffer
->commit_page
);
1052 cpu_buffer
->write_stamp
=
1053 cpu_buffer
->commit_page
->page
->time_stamp
;
1056 /* Now set the commit to the event's index */
1057 local_set(&cpu_buffer
->commit_page
->page
->commit
, index
);
1061 rb_set_commit_to_write(struct ring_buffer_per_cpu
*cpu_buffer
)
1064 * We only race with interrupts and NMIs on this CPU.
1065 * If we own the commit event, then we can commit
1066 * all others that interrupted us, since the interruptions
1067 * are in stack format (they finish before they come
1068 * back to us). This allows us to do a simple loop to
1069 * assign the commit to the tail.
1072 while (cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
) {
1073 cpu_buffer
->commit_page
->page
->commit
=
1074 cpu_buffer
->commit_page
->write
;
1075 rb_inc_page(cpu_buffer
, &cpu_buffer
->commit_page
);
1076 cpu_buffer
->write_stamp
=
1077 cpu_buffer
->commit_page
->page
->time_stamp
;
1078 /* add barrier to keep gcc from optimizing too much */
1081 while (rb_commit_index(cpu_buffer
) !=
1082 rb_page_write(cpu_buffer
->commit_page
)) {
1083 cpu_buffer
->commit_page
->page
->commit
=
1084 cpu_buffer
->commit_page
->write
;
1088 /* again, keep gcc from optimizing */
1092 * If an interrupt came in just after the first while loop
1093 * and pushed the tail page forward, we will be left with
1094 * a dangling commit that will never go forward.
1096 if (unlikely(cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
))
1100 static void rb_reset_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
1102 cpu_buffer
->read_stamp
= cpu_buffer
->reader_page
->page
->time_stamp
;
1103 cpu_buffer
->reader_page
->read
= 0;
1106 static void rb_inc_iter(struct ring_buffer_iter
*iter
)
1108 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
1111 * The iterator could be on the reader page (it starts there).
1112 * But the head could have moved, since the reader was
1113 * found. Check for this case and assign the iterator
1114 * to the head page instead of next.
1116 if (iter
->head_page
== cpu_buffer
->reader_page
)
1117 iter
->head_page
= cpu_buffer
->head_page
;
1119 rb_inc_page(cpu_buffer
, &iter
->head_page
);
1121 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
1126 * ring_buffer_update_event - update event type and data
1127 * @event: the even to update
1128 * @type: the type of event
1129 * @length: the size of the event field in the ring buffer
1131 * Update the type and data fields of the event. The length
1132 * is the actual size that is written to the ring buffer,
1133 * and with this, we can determine what to place into the
1137 rb_update_event(struct ring_buffer_event
*event
,
1138 unsigned type
, unsigned length
)
1140 event
->type_len
= type
;
1144 case RINGBUF_TYPE_PADDING
:
1145 case RINGBUF_TYPE_TIME_EXTEND
:
1146 case RINGBUF_TYPE_TIME_STAMP
:
1150 length
-= RB_EVNT_HDR_SIZE
;
1151 if (length
> RB_MAX_SMALL_DATA
)
1152 event
->array
[0] = length
;
1154 event
->type_len
= DIV_ROUND_UP(length
, RB_ALIGNMENT
);
1161 static unsigned rb_calculate_event_length(unsigned length
)
1163 struct ring_buffer_event event
; /* Used only for sizeof array */
1165 /* zero length can cause confusions */
1169 if (length
> RB_MAX_SMALL_DATA
)
1170 length
+= sizeof(event
.array
[0]);
1172 length
+= RB_EVNT_HDR_SIZE
;
1173 length
= ALIGN(length
, RB_ALIGNMENT
);
1178 static struct ring_buffer_event
*
1179 __rb_reserve_next(struct ring_buffer_per_cpu
*cpu_buffer
,
1180 unsigned type
, unsigned long length
, u64
*ts
)
1182 struct buffer_page
*tail_page
, *head_page
, *reader_page
, *commit_page
;
1183 unsigned long tail
, write
;
1184 struct ring_buffer
*buffer
= cpu_buffer
->buffer
;
1185 struct ring_buffer_event
*event
;
1186 unsigned long flags
;
1187 bool lock_taken
= false;
1189 commit_page
= cpu_buffer
->commit_page
;
1190 /* we just need to protect against interrupts */
1192 tail_page
= cpu_buffer
->tail_page
;
1193 write
= local_add_return(length
, &tail_page
->write
);
1194 tail
= write
- length
;
1196 /* See if we shot pass the end of this buffer page */
1197 if (write
> BUF_PAGE_SIZE
) {
1198 struct buffer_page
*next_page
= tail_page
;
1200 local_irq_save(flags
);
1202 * Since the write to the buffer is still not
1203 * fully lockless, we must be careful with NMIs.
1204 * The locks in the writers are taken when a write
1205 * crosses to a new page. The locks protect against
1206 * races with the readers (this will soon be fixed
1207 * with a lockless solution).
1209 * Because we can not protect against NMIs, and we
1210 * want to keep traces reentrant, we need to manage
1211 * what happens when we are in an NMI.
1213 * NMIs can happen after we take the lock.
1214 * If we are in an NMI, only take the lock
1215 * if it is not already taken. Otherwise
1218 if (unlikely(in_nmi())) {
1219 if (!__raw_spin_trylock(&cpu_buffer
->lock
))
1222 __raw_spin_lock(&cpu_buffer
->lock
);
1226 rb_inc_page(cpu_buffer
, &next_page
);
1228 head_page
= cpu_buffer
->head_page
;
1229 reader_page
= cpu_buffer
->reader_page
;
1231 /* we grabbed the lock before incrementing */
1232 if (RB_WARN_ON(cpu_buffer
, next_page
== reader_page
))
1236 * If for some reason, we had an interrupt storm that made
1237 * it all the way around the buffer, bail, and warn
1240 if (unlikely(next_page
== commit_page
)) {
1241 /* This can easily happen on small ring buffers */
1242 WARN_ON_ONCE(buffer
->pages
> 2);
1246 if (next_page
== head_page
) {
1247 if (!(buffer
->flags
& RB_FL_OVERWRITE
))
1250 /* tail_page has not moved yet? */
1251 if (tail_page
== cpu_buffer
->tail_page
) {
1252 /* count overflows */
1253 rb_update_overflow(cpu_buffer
);
1255 rb_inc_page(cpu_buffer
, &head_page
);
1256 cpu_buffer
->head_page
= head_page
;
1257 cpu_buffer
->head_page
->read
= 0;
1262 * If the tail page is still the same as what we think
1263 * it is, then it is up to us to update the tail
1266 if (tail_page
== cpu_buffer
->tail_page
) {
1267 local_set(&next_page
->write
, 0);
1268 local_set(&next_page
->page
->commit
, 0);
1269 cpu_buffer
->tail_page
= next_page
;
1271 /* reread the time stamp */
1272 *ts
= ring_buffer_time_stamp(buffer
, cpu_buffer
->cpu
);
1273 cpu_buffer
->tail_page
->page
->time_stamp
= *ts
;
1277 * The actual tail page has moved forward.
1279 if (tail
< BUF_PAGE_SIZE
) {
1280 /* Mark the rest of the page with padding */
1281 event
= __rb_page_index(tail_page
, tail
);
1282 rb_event_set_padding(event
);
1285 if (tail
<= BUF_PAGE_SIZE
)
1286 /* Set the write back to the previous setting */
1287 local_set(&tail_page
->write
, tail
);
1290 * If this was a commit entry that failed,
1291 * increment that too
1293 if (tail_page
== cpu_buffer
->commit_page
&&
1294 tail
== rb_commit_index(cpu_buffer
)) {
1295 rb_set_commit_to_write(cpu_buffer
);
1298 __raw_spin_unlock(&cpu_buffer
->lock
);
1299 local_irq_restore(flags
);
1301 /* fail and let the caller try again */
1302 return ERR_PTR(-EAGAIN
);
1305 /* We reserved something on the buffer */
1307 if (RB_WARN_ON(cpu_buffer
, write
> BUF_PAGE_SIZE
))
1310 event
= __rb_page_index(tail_page
, tail
);
1311 rb_update_event(event
, type
, length
);
1314 * If this is a commit and the tail is zero, then update
1315 * this page's time stamp.
1317 if (!tail
&& rb_is_commit(cpu_buffer
, event
))
1318 cpu_buffer
->commit_page
->page
->time_stamp
= *ts
;
1324 if (tail
<= BUF_PAGE_SIZE
)
1325 local_set(&tail_page
->write
, tail
);
1327 if (likely(lock_taken
))
1328 __raw_spin_unlock(&cpu_buffer
->lock
);
1329 local_irq_restore(flags
);
1334 rb_add_time_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
1335 u64
*ts
, u64
*delta
)
1337 struct ring_buffer_event
*event
;
1341 if (unlikely(*delta
> (1ULL << 59) && !once
++)) {
1342 printk(KERN_WARNING
"Delta way too big! %llu"
1343 " ts=%llu write stamp = %llu\n",
1344 (unsigned long long)*delta
,
1345 (unsigned long long)*ts
,
1346 (unsigned long long)cpu_buffer
->write_stamp
);
1351 * The delta is too big, we to add a
1354 event
= __rb_reserve_next(cpu_buffer
,
1355 RINGBUF_TYPE_TIME_EXTEND
,
1361 if (PTR_ERR(event
) == -EAGAIN
)
1364 /* Only a commited time event can update the write stamp */
1365 if (rb_is_commit(cpu_buffer
, event
)) {
1367 * If this is the first on the page, then we need to
1368 * update the page itself, and just put in a zero.
1370 if (rb_event_index(event
)) {
1371 event
->time_delta
= *delta
& TS_MASK
;
1372 event
->array
[0] = *delta
>> TS_SHIFT
;
1374 cpu_buffer
->commit_page
->page
->time_stamp
= *ts
;
1375 event
->time_delta
= 0;
1376 event
->array
[0] = 0;
1378 cpu_buffer
->write_stamp
= *ts
;
1379 /* let the caller know this was the commit */
1382 /* Darn, this is just wasted space */
1383 event
->time_delta
= 0;
1384 event
->array
[0] = 0;
1393 static struct ring_buffer_event
*
1394 rb_reserve_next_event(struct ring_buffer_per_cpu
*cpu_buffer
,
1395 unsigned type
, unsigned long length
)
1397 struct ring_buffer_event
*event
;
1404 * We allow for interrupts to reenter here and do a trace.
1405 * If one does, it will cause this original code to loop
1406 * back here. Even with heavy interrupts happening, this
1407 * should only happen a few times in a row. If this happens
1408 * 1000 times in a row, there must be either an interrupt
1409 * storm or we have something buggy.
1412 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 1000))
1415 ts
= ring_buffer_time_stamp(cpu_buffer
->buffer
, cpu_buffer
->cpu
);
1418 * Only the first commit can update the timestamp.
1419 * Yes there is a race here. If an interrupt comes in
1420 * just after the conditional and it traces too, then it
1421 * will also check the deltas. More than one timestamp may
1422 * also be made. But only the entry that did the actual
1423 * commit will be something other than zero.
1425 if (cpu_buffer
->tail_page
== cpu_buffer
->commit_page
&&
1426 rb_page_write(cpu_buffer
->tail_page
) ==
1427 rb_commit_index(cpu_buffer
)) {
1429 delta
= ts
- cpu_buffer
->write_stamp
;
1431 /* make sure this delta is calculated here */
1434 /* Did the write stamp get updated already? */
1435 if (unlikely(ts
< cpu_buffer
->write_stamp
))
1438 if (test_time_stamp(delta
)) {
1440 commit
= rb_add_time_stamp(cpu_buffer
, &ts
, &delta
);
1442 if (commit
== -EBUSY
)
1445 if (commit
== -EAGAIN
)
1448 RB_WARN_ON(cpu_buffer
, commit
< 0);
1451 /* Non commits have zero deltas */
1454 event
= __rb_reserve_next(cpu_buffer
, type
, length
, &ts
);
1455 if (PTR_ERR(event
) == -EAGAIN
)
1459 if (unlikely(commit
))
1461 * Ouch! We needed a timestamp and it was commited. But
1462 * we didn't get our event reserved.
1464 rb_set_commit_to_write(cpu_buffer
);
1469 * If the timestamp was commited, make the commit our entry
1470 * now so that we will update it when needed.
1473 rb_set_commit_event(cpu_buffer
, event
);
1474 else if (!rb_is_commit(cpu_buffer
, event
))
1477 event
->time_delta
= delta
;
1482 #define TRACE_RECURSIVE_DEPTH 16
1484 static int trace_recursive_lock(void)
1486 current
->trace_recursion
++;
1488 if (likely(current
->trace_recursion
< TRACE_RECURSIVE_DEPTH
))
1491 /* Disable all tracing before we do anything else */
1492 tracing_off_permanent();
1494 printk_once(KERN_WARNING
"Tracing recursion: depth[%ld]:"
1495 "HC[%lu]:SC[%lu]:NMI[%lu]\n",
1496 current
->trace_recursion
,
1497 hardirq_count() >> HARDIRQ_SHIFT
,
1498 softirq_count() >> SOFTIRQ_SHIFT
,
1505 static void trace_recursive_unlock(void)
1507 WARN_ON_ONCE(!current
->trace_recursion
);
1509 current
->trace_recursion
--;
1512 static DEFINE_PER_CPU(int, rb_need_resched
);
1515 * ring_buffer_lock_reserve - reserve a part of the buffer
1516 * @buffer: the ring buffer to reserve from
1517 * @length: the length of the data to reserve (excluding event header)
1519 * Returns a reseverd event on the ring buffer to copy directly to.
1520 * The user of this interface will need to get the body to write into
1521 * and can use the ring_buffer_event_data() interface.
1523 * The length is the length of the data needed, not the event length
1524 * which also includes the event header.
1526 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
1527 * If NULL is returned, then nothing has been allocated or locked.
1529 struct ring_buffer_event
*
1530 ring_buffer_lock_reserve(struct ring_buffer
*buffer
, unsigned long length
)
1532 struct ring_buffer_per_cpu
*cpu_buffer
;
1533 struct ring_buffer_event
*event
;
1536 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
1539 if (atomic_read(&buffer
->record_disabled
))
1542 /* If we are tracing schedule, we don't want to recurse */
1543 resched
= ftrace_preempt_disable();
1545 if (trace_recursive_lock())
1548 cpu
= raw_smp_processor_id();
1550 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1553 cpu_buffer
= buffer
->buffers
[cpu
];
1555 if (atomic_read(&cpu_buffer
->record_disabled
))
1558 length
= rb_calculate_event_length(length
);
1559 if (length
> BUF_PAGE_SIZE
)
1562 event
= rb_reserve_next_event(cpu_buffer
, 0, length
);
1567 * Need to store resched state on this cpu.
1568 * Only the first needs to.
1571 if (preempt_count() == 1)
1572 per_cpu(rb_need_resched
, cpu
) = resched
;
1577 trace_recursive_unlock();
1580 ftrace_preempt_enable(resched
);
1583 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve
);
1585 static void rb_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
1586 struct ring_buffer_event
*event
)
1588 cpu_buffer
->entries
++;
1590 /* Only process further if we own the commit */
1591 if (!rb_is_commit(cpu_buffer
, event
))
1594 cpu_buffer
->write_stamp
+= event
->time_delta
;
1596 rb_set_commit_to_write(cpu_buffer
);
1600 * ring_buffer_unlock_commit - commit a reserved
1601 * @buffer: The buffer to commit to
1602 * @event: The event pointer to commit.
1604 * This commits the data to the ring buffer, and releases any locks held.
1606 * Must be paired with ring_buffer_lock_reserve.
1608 int ring_buffer_unlock_commit(struct ring_buffer
*buffer
,
1609 struct ring_buffer_event
*event
)
1611 struct ring_buffer_per_cpu
*cpu_buffer
;
1612 int cpu
= raw_smp_processor_id();
1614 cpu_buffer
= buffer
->buffers
[cpu
];
1616 rb_commit(cpu_buffer
, event
);
1618 trace_recursive_unlock();
1621 * Only the last preempt count needs to restore preemption.
1623 if (preempt_count() == 1)
1624 ftrace_preempt_enable(per_cpu(rb_need_resched
, cpu
));
1626 preempt_enable_no_resched_notrace();
1630 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit
);
1632 static inline void rb_event_discard(struct ring_buffer_event
*event
)
1634 /* array[0] holds the actual length for the discarded event */
1635 event
->array
[0] = rb_event_data_length(event
) - RB_EVNT_HDR_SIZE
;
1636 event
->type_len
= RINGBUF_TYPE_PADDING
;
1637 /* time delta must be non zero */
1638 if (!event
->time_delta
)
1639 event
->time_delta
= 1;
1643 * ring_buffer_event_discard - discard any event in the ring buffer
1644 * @event: the event to discard
1646 * Sometimes a event that is in the ring buffer needs to be ignored.
1647 * This function lets the user discard an event in the ring buffer
1648 * and then that event will not be read later.
1650 * Note, it is up to the user to be careful with this, and protect
1651 * against races. If the user discards an event that has been consumed
1652 * it is possible that it could corrupt the ring buffer.
1654 void ring_buffer_event_discard(struct ring_buffer_event
*event
)
1656 rb_event_discard(event
);
1658 EXPORT_SYMBOL_GPL(ring_buffer_event_discard
);
1661 * ring_buffer_commit_discard - discard an event that has not been committed
1662 * @buffer: the ring buffer
1663 * @event: non committed event to discard
1665 * This is similar to ring_buffer_event_discard but must only be
1666 * performed on an event that has not been committed yet. The difference
1667 * is that this will also try to free the event from the ring buffer
1668 * if another event has not been added behind it.
1670 * If another event has been added behind it, it will set the event
1671 * up as discarded, and perform the commit.
1673 * If this function is called, do not call ring_buffer_unlock_commit on
1676 void ring_buffer_discard_commit(struct ring_buffer
*buffer
,
1677 struct ring_buffer_event
*event
)
1679 struct ring_buffer_per_cpu
*cpu_buffer
;
1680 unsigned long new_index
, old_index
;
1681 struct buffer_page
*bpage
;
1682 unsigned long index
;
1686 /* The event is discarded regardless */
1687 rb_event_discard(event
);
1690 * This must only be called if the event has not been
1691 * committed yet. Thus we can assume that preemption
1692 * is still disabled.
1694 RB_WARN_ON(buffer
, !preempt_count());
1696 cpu
= smp_processor_id();
1697 cpu_buffer
= buffer
->buffers
[cpu
];
1699 new_index
= rb_event_index(event
);
1700 old_index
= new_index
+ rb_event_length(event
);
1701 addr
= (unsigned long)event
;
1704 bpage
= cpu_buffer
->tail_page
;
1706 if (bpage
== (void *)addr
&& rb_page_write(bpage
) == old_index
) {
1708 * This is on the tail page. It is possible that
1709 * a write could come in and move the tail page
1710 * and write to the next page. That is fine
1711 * because we just shorten what is on this page.
1713 index
= local_cmpxchg(&bpage
->write
, old_index
, new_index
);
1714 if (index
== old_index
)
1719 * The commit is still visible by the reader, so we
1720 * must increment entries.
1722 cpu_buffer
->entries
++;
1725 * If a write came in and pushed the tail page
1726 * we still need to update the commit pointer
1727 * if we were the commit.
1729 if (rb_is_commit(cpu_buffer
, event
))
1730 rb_set_commit_to_write(cpu_buffer
);
1732 trace_recursive_unlock();
1735 * Only the last preempt count needs to restore preemption.
1737 if (preempt_count() == 1)
1738 ftrace_preempt_enable(per_cpu(rb_need_resched
, cpu
));
1740 preempt_enable_no_resched_notrace();
1743 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit
);
1746 * ring_buffer_write - write data to the buffer without reserving
1747 * @buffer: The ring buffer to write to.
1748 * @length: The length of the data being written (excluding the event header)
1749 * @data: The data to write to the buffer.
1751 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
1752 * one function. If you already have the data to write to the buffer, it
1753 * may be easier to simply call this function.
1755 * Note, like ring_buffer_lock_reserve, the length is the length of the data
1756 * and not the length of the event which would hold the header.
1758 int ring_buffer_write(struct ring_buffer
*buffer
,
1759 unsigned long length
,
1762 struct ring_buffer_per_cpu
*cpu_buffer
;
1763 struct ring_buffer_event
*event
;
1764 unsigned long event_length
;
1769 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
1772 if (atomic_read(&buffer
->record_disabled
))
1775 resched
= ftrace_preempt_disable();
1777 cpu
= raw_smp_processor_id();
1779 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1782 cpu_buffer
= buffer
->buffers
[cpu
];
1784 if (atomic_read(&cpu_buffer
->record_disabled
))
1787 event_length
= rb_calculate_event_length(length
);
1788 event
= rb_reserve_next_event(cpu_buffer
, 0, event_length
);
1792 body
= rb_event_data(event
);
1794 memcpy(body
, data
, length
);
1796 rb_commit(cpu_buffer
, event
);
1800 ftrace_preempt_enable(resched
);
1804 EXPORT_SYMBOL_GPL(ring_buffer_write
);
1806 static int rb_per_cpu_empty(struct ring_buffer_per_cpu
*cpu_buffer
)
1808 struct buffer_page
*reader
= cpu_buffer
->reader_page
;
1809 struct buffer_page
*head
= cpu_buffer
->head_page
;
1810 struct buffer_page
*commit
= cpu_buffer
->commit_page
;
1812 return reader
->read
== rb_page_commit(reader
) &&
1813 (commit
== reader
||
1815 head
->read
== rb_page_commit(commit
)));
1819 * ring_buffer_record_disable - stop all writes into the buffer
1820 * @buffer: The ring buffer to stop writes to.
1822 * This prevents all writes to the buffer. Any attempt to write
1823 * to the buffer after this will fail and return NULL.
1825 * The caller should call synchronize_sched() after this.
1827 void ring_buffer_record_disable(struct ring_buffer
*buffer
)
1829 atomic_inc(&buffer
->record_disabled
);
1831 EXPORT_SYMBOL_GPL(ring_buffer_record_disable
);
1834 * ring_buffer_record_enable - enable writes to the buffer
1835 * @buffer: The ring buffer to enable writes
1837 * Note, multiple disables will need the same number of enables
1838 * to truely enable the writing (much like preempt_disable).
1840 void ring_buffer_record_enable(struct ring_buffer
*buffer
)
1842 atomic_dec(&buffer
->record_disabled
);
1844 EXPORT_SYMBOL_GPL(ring_buffer_record_enable
);
1847 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
1848 * @buffer: The ring buffer to stop writes to.
1849 * @cpu: The CPU buffer to stop
1851 * This prevents all writes to the buffer. Any attempt to write
1852 * to the buffer after this will fail and return NULL.
1854 * The caller should call synchronize_sched() after this.
1856 void ring_buffer_record_disable_cpu(struct ring_buffer
*buffer
, int cpu
)
1858 struct ring_buffer_per_cpu
*cpu_buffer
;
1860 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1863 cpu_buffer
= buffer
->buffers
[cpu
];
1864 atomic_inc(&cpu_buffer
->record_disabled
);
1866 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu
);
1869 * ring_buffer_record_enable_cpu - enable writes to the buffer
1870 * @buffer: The ring buffer to enable writes
1871 * @cpu: The CPU to enable.
1873 * Note, multiple disables will need the same number of enables
1874 * to truely enable the writing (much like preempt_disable).
1876 void ring_buffer_record_enable_cpu(struct ring_buffer
*buffer
, int cpu
)
1878 struct ring_buffer_per_cpu
*cpu_buffer
;
1880 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1883 cpu_buffer
= buffer
->buffers
[cpu
];
1884 atomic_dec(&cpu_buffer
->record_disabled
);
1886 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu
);
1889 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
1890 * @buffer: The ring buffer
1891 * @cpu: The per CPU buffer to get the entries from.
1893 unsigned long ring_buffer_entries_cpu(struct ring_buffer
*buffer
, int cpu
)
1895 struct ring_buffer_per_cpu
*cpu_buffer
;
1898 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1901 cpu_buffer
= buffer
->buffers
[cpu
];
1902 ret
= cpu_buffer
->entries
;
1906 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu
);
1909 * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
1910 * @buffer: The ring buffer
1911 * @cpu: The per CPU buffer to get the number of overruns from
1913 unsigned long ring_buffer_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
1915 struct ring_buffer_per_cpu
*cpu_buffer
;
1918 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1921 cpu_buffer
= buffer
->buffers
[cpu
];
1922 ret
= cpu_buffer
->overrun
;
1926 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu
);
1929 * ring_buffer_entries - get the number of entries in a buffer
1930 * @buffer: The ring buffer
1932 * Returns the total number of entries in the ring buffer
1935 unsigned long ring_buffer_entries(struct ring_buffer
*buffer
)
1937 struct ring_buffer_per_cpu
*cpu_buffer
;
1938 unsigned long entries
= 0;
1941 /* if you care about this being correct, lock the buffer */
1942 for_each_buffer_cpu(buffer
, cpu
) {
1943 cpu_buffer
= buffer
->buffers
[cpu
];
1944 entries
+= cpu_buffer
->entries
;
1949 EXPORT_SYMBOL_GPL(ring_buffer_entries
);
1952 * ring_buffer_overrun_cpu - get the number of overruns in buffer
1953 * @buffer: The ring buffer
1955 * Returns the total number of overruns in the ring buffer
1958 unsigned long ring_buffer_overruns(struct ring_buffer
*buffer
)
1960 struct ring_buffer_per_cpu
*cpu_buffer
;
1961 unsigned long overruns
= 0;
1964 /* if you care about this being correct, lock the buffer */
1965 for_each_buffer_cpu(buffer
, cpu
) {
1966 cpu_buffer
= buffer
->buffers
[cpu
];
1967 overruns
+= cpu_buffer
->overrun
;
1972 EXPORT_SYMBOL_GPL(ring_buffer_overruns
);
1974 static void rb_iter_reset(struct ring_buffer_iter
*iter
)
1976 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
1978 /* Iterator usage is expected to have record disabled */
1979 if (list_empty(&cpu_buffer
->reader_page
->list
)) {
1980 iter
->head_page
= cpu_buffer
->head_page
;
1981 iter
->head
= cpu_buffer
->head_page
->read
;
1983 iter
->head_page
= cpu_buffer
->reader_page
;
1984 iter
->head
= cpu_buffer
->reader_page
->read
;
1987 iter
->read_stamp
= cpu_buffer
->read_stamp
;
1989 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
1993 * ring_buffer_iter_reset - reset an iterator
1994 * @iter: The iterator to reset
1996 * Resets the iterator, so that it will start from the beginning
1999 void ring_buffer_iter_reset(struct ring_buffer_iter
*iter
)
2001 struct ring_buffer_per_cpu
*cpu_buffer
;
2002 unsigned long flags
;
2007 cpu_buffer
= iter
->cpu_buffer
;
2009 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2010 rb_iter_reset(iter
);
2011 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2013 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset
);
2016 * ring_buffer_iter_empty - check if an iterator has no more to read
2017 * @iter: The iterator to check
2019 int ring_buffer_iter_empty(struct ring_buffer_iter
*iter
)
2021 struct ring_buffer_per_cpu
*cpu_buffer
;
2023 cpu_buffer
= iter
->cpu_buffer
;
2025 return iter
->head_page
== cpu_buffer
->commit_page
&&
2026 iter
->head
== rb_commit_index(cpu_buffer
);
2028 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty
);
2031 rb_update_read_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
2032 struct ring_buffer_event
*event
)
2036 switch (event
->type_len
) {
2037 case RINGBUF_TYPE_PADDING
:
2040 case RINGBUF_TYPE_TIME_EXTEND
:
2041 delta
= event
->array
[0];
2043 delta
+= event
->time_delta
;
2044 cpu_buffer
->read_stamp
+= delta
;
2047 case RINGBUF_TYPE_TIME_STAMP
:
2048 /* FIXME: not implemented */
2051 case RINGBUF_TYPE_DATA
:
2052 cpu_buffer
->read_stamp
+= event
->time_delta
;
2062 rb_update_iter_read_stamp(struct ring_buffer_iter
*iter
,
2063 struct ring_buffer_event
*event
)
2067 switch (event
->type_len
) {
2068 case RINGBUF_TYPE_PADDING
:
2071 case RINGBUF_TYPE_TIME_EXTEND
:
2072 delta
= event
->array
[0];
2074 delta
+= event
->time_delta
;
2075 iter
->read_stamp
+= delta
;
2078 case RINGBUF_TYPE_TIME_STAMP
:
2079 /* FIXME: not implemented */
2082 case RINGBUF_TYPE_DATA
:
2083 iter
->read_stamp
+= event
->time_delta
;
2092 static struct buffer_page
*
2093 rb_get_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
2095 struct buffer_page
*reader
= NULL
;
2096 unsigned long flags
;
2099 local_irq_save(flags
);
2100 __raw_spin_lock(&cpu_buffer
->lock
);
2104 * This should normally only loop twice. But because the
2105 * start of the reader inserts an empty page, it causes
2106 * a case where we will loop three times. There should be no
2107 * reason to loop four times (that I know of).
2109 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 3)) {
2114 reader
= cpu_buffer
->reader_page
;
2116 /* If there's more to read, return this page */
2117 if (cpu_buffer
->reader_page
->read
< rb_page_size(reader
))
2120 /* Never should we have an index greater than the size */
2121 if (RB_WARN_ON(cpu_buffer
,
2122 cpu_buffer
->reader_page
->read
> rb_page_size(reader
)))
2125 /* check if we caught up to the tail */
2127 if (cpu_buffer
->commit_page
== cpu_buffer
->reader_page
)
2131 * Splice the empty reader page into the list around the head.
2132 * Reset the reader page to size zero.
2135 reader
= cpu_buffer
->head_page
;
2136 cpu_buffer
->reader_page
->list
.next
= reader
->list
.next
;
2137 cpu_buffer
->reader_page
->list
.prev
= reader
->list
.prev
;
2139 local_set(&cpu_buffer
->reader_page
->write
, 0);
2140 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
2142 /* Make the reader page now replace the head */
2143 reader
->list
.prev
->next
= &cpu_buffer
->reader_page
->list
;
2144 reader
->list
.next
->prev
= &cpu_buffer
->reader_page
->list
;
2147 * If the tail is on the reader, then we must set the head
2148 * to the inserted page, otherwise we set it one before.
2150 cpu_buffer
->head_page
= cpu_buffer
->reader_page
;
2152 if (cpu_buffer
->commit_page
!= reader
)
2153 rb_inc_page(cpu_buffer
, &cpu_buffer
->head_page
);
2155 /* Finally update the reader page to the new head */
2156 cpu_buffer
->reader_page
= reader
;
2157 rb_reset_reader_page(cpu_buffer
);
2162 __raw_spin_unlock(&cpu_buffer
->lock
);
2163 local_irq_restore(flags
);
2168 static void rb_advance_reader(struct ring_buffer_per_cpu
*cpu_buffer
)
2170 struct ring_buffer_event
*event
;
2171 struct buffer_page
*reader
;
2174 reader
= rb_get_reader_page(cpu_buffer
);
2176 /* This function should not be called when buffer is empty */
2177 if (RB_WARN_ON(cpu_buffer
, !reader
))
2180 event
= rb_reader_event(cpu_buffer
);
2182 if (event
->type_len
<= RINGBUF_TYPE_DATA_TYPE_LEN_MAX
2183 || rb_discarded_event(event
))
2184 cpu_buffer
->entries
--;
2186 rb_update_read_stamp(cpu_buffer
, event
);
2188 length
= rb_event_length(event
);
2189 cpu_buffer
->reader_page
->read
+= length
;
2192 static void rb_advance_iter(struct ring_buffer_iter
*iter
)
2194 struct ring_buffer
*buffer
;
2195 struct ring_buffer_per_cpu
*cpu_buffer
;
2196 struct ring_buffer_event
*event
;
2199 cpu_buffer
= iter
->cpu_buffer
;
2200 buffer
= cpu_buffer
->buffer
;
2203 * Check if we are at the end of the buffer.
2205 if (iter
->head
>= rb_page_size(iter
->head_page
)) {
2206 if (RB_WARN_ON(buffer
,
2207 iter
->head_page
== cpu_buffer
->commit_page
))
2213 event
= rb_iter_head_event(iter
);
2215 length
= rb_event_length(event
);
2218 * This should not be called to advance the header if we are
2219 * at the tail of the buffer.
2221 if (RB_WARN_ON(cpu_buffer
,
2222 (iter
->head_page
== cpu_buffer
->commit_page
) &&
2223 (iter
->head
+ length
> rb_commit_index(cpu_buffer
))))
2226 rb_update_iter_read_stamp(iter
, event
);
2228 iter
->head
+= length
;
2230 /* check for end of page padding */
2231 if ((iter
->head
>= rb_page_size(iter
->head_page
)) &&
2232 (iter
->head_page
!= cpu_buffer
->commit_page
))
2233 rb_advance_iter(iter
);
2236 static struct ring_buffer_event
*
2237 rb_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2239 struct ring_buffer_per_cpu
*cpu_buffer
;
2240 struct ring_buffer_event
*event
;
2241 struct buffer_page
*reader
;
2244 cpu_buffer
= buffer
->buffers
[cpu
];
2248 * We repeat when a timestamp is encountered. It is possible
2249 * to get multiple timestamps from an interrupt entering just
2250 * as one timestamp is about to be written. The max times
2251 * that this can happen is the number of nested interrupts we
2252 * can have. Nesting 10 deep of interrupts is clearly
2255 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 10))
2258 reader
= rb_get_reader_page(cpu_buffer
);
2262 event
= rb_reader_event(cpu_buffer
);
2264 switch (event
->type_len
) {
2265 case RINGBUF_TYPE_PADDING
:
2266 if (rb_null_event(event
))
2267 RB_WARN_ON(cpu_buffer
, 1);
2269 * Because the writer could be discarding every
2270 * event it creates (which would probably be bad)
2271 * if we were to go back to "again" then we may never
2272 * catch up, and will trigger the warn on, or lock
2273 * the box. Return the padding, and we will release
2274 * the current locks, and try again.
2276 rb_advance_reader(cpu_buffer
);
2279 case RINGBUF_TYPE_TIME_EXTEND
:
2280 /* Internal data, OK to advance */
2281 rb_advance_reader(cpu_buffer
);
2284 case RINGBUF_TYPE_TIME_STAMP
:
2285 /* FIXME: not implemented */
2286 rb_advance_reader(cpu_buffer
);
2289 case RINGBUF_TYPE_DATA
:
2291 *ts
= cpu_buffer
->read_stamp
+ event
->time_delta
;
2292 ring_buffer_normalize_time_stamp(buffer
,
2293 cpu_buffer
->cpu
, ts
);
2303 EXPORT_SYMBOL_GPL(ring_buffer_peek
);
2305 static struct ring_buffer_event
*
2306 rb_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
2308 struct ring_buffer
*buffer
;
2309 struct ring_buffer_per_cpu
*cpu_buffer
;
2310 struct ring_buffer_event
*event
;
2313 if (ring_buffer_iter_empty(iter
))
2316 cpu_buffer
= iter
->cpu_buffer
;
2317 buffer
= cpu_buffer
->buffer
;
2321 * We repeat when a timestamp is encountered. It is possible
2322 * to get multiple timestamps from an interrupt entering just
2323 * as one timestamp is about to be written. The max times
2324 * that this can happen is the number of nested interrupts we
2325 * can have. Nesting 10 deep of interrupts is clearly
2328 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 10))
2331 if (rb_per_cpu_empty(cpu_buffer
))
2334 event
= rb_iter_head_event(iter
);
2336 switch (event
->type_len
) {
2337 case RINGBUF_TYPE_PADDING
:
2338 if (rb_null_event(event
)) {
2342 rb_advance_iter(iter
);
2345 case RINGBUF_TYPE_TIME_EXTEND
:
2346 /* Internal data, OK to advance */
2347 rb_advance_iter(iter
);
2350 case RINGBUF_TYPE_TIME_STAMP
:
2351 /* FIXME: not implemented */
2352 rb_advance_iter(iter
);
2355 case RINGBUF_TYPE_DATA
:
2357 *ts
= iter
->read_stamp
+ event
->time_delta
;
2358 ring_buffer_normalize_time_stamp(buffer
,
2359 cpu_buffer
->cpu
, ts
);
2369 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek
);
2372 * ring_buffer_peek - peek at the next event to be read
2373 * @buffer: The ring buffer to read
2374 * @cpu: The cpu to peak at
2375 * @ts: The timestamp counter of this event.
2377 * This will return the event that will be read next, but does
2378 * not consume the data.
2380 struct ring_buffer_event
*
2381 ring_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2383 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2384 struct ring_buffer_event
*event
;
2385 unsigned long flags
;
2387 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2391 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2392 event
= rb_buffer_peek(buffer
, cpu
, ts
);
2393 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2395 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2404 * ring_buffer_iter_peek - peek at the next event to be read
2405 * @iter: The ring buffer iterator
2406 * @ts: The timestamp counter of this event.
2408 * This will return the event that will be read next, but does
2409 * not increment the iterator.
2411 struct ring_buffer_event
*
2412 ring_buffer_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
2414 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2415 struct ring_buffer_event
*event
;
2416 unsigned long flags
;
2419 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2420 event
= rb_iter_peek(iter
, ts
);
2421 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2423 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2432 * ring_buffer_consume - return an event and consume it
2433 * @buffer: The ring buffer to get the next event from
2435 * Returns the next event in the ring buffer, and that event is consumed.
2436 * Meaning, that sequential reads will keep returning a different event,
2437 * and eventually empty the ring buffer if the producer is slower.
2439 struct ring_buffer_event
*
2440 ring_buffer_consume(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2442 struct ring_buffer_per_cpu
*cpu_buffer
;
2443 struct ring_buffer_event
*event
= NULL
;
2444 unsigned long flags
;
2447 /* might be called in atomic */
2450 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2453 cpu_buffer
= buffer
->buffers
[cpu
];
2454 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2456 event
= rb_buffer_peek(buffer
, cpu
, ts
);
2460 rb_advance_reader(cpu_buffer
);
2463 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2468 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2475 EXPORT_SYMBOL_GPL(ring_buffer_consume
);
2478 * ring_buffer_read_start - start a non consuming read of the buffer
2479 * @buffer: The ring buffer to read from
2480 * @cpu: The cpu buffer to iterate over
2482 * This starts up an iteration through the buffer. It also disables
2483 * the recording to the buffer until the reading is finished.
2484 * This prevents the reading from being corrupted. This is not
2485 * a consuming read, so a producer is not expected.
2487 * Must be paired with ring_buffer_finish.
2489 struct ring_buffer_iter
*
2490 ring_buffer_read_start(struct ring_buffer
*buffer
, int cpu
)
2492 struct ring_buffer_per_cpu
*cpu_buffer
;
2493 struct ring_buffer_iter
*iter
;
2494 unsigned long flags
;
2496 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2499 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
2503 cpu_buffer
= buffer
->buffers
[cpu
];
2505 iter
->cpu_buffer
= cpu_buffer
;
2507 atomic_inc(&cpu_buffer
->record_disabled
);
2508 synchronize_sched();
2510 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2511 __raw_spin_lock(&cpu_buffer
->lock
);
2512 rb_iter_reset(iter
);
2513 __raw_spin_unlock(&cpu_buffer
->lock
);
2514 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2518 EXPORT_SYMBOL_GPL(ring_buffer_read_start
);
2521 * ring_buffer_finish - finish reading the iterator of the buffer
2522 * @iter: The iterator retrieved by ring_buffer_start
2524 * This re-enables the recording to the buffer, and frees the
2528 ring_buffer_read_finish(struct ring_buffer_iter
*iter
)
2530 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2532 atomic_dec(&cpu_buffer
->record_disabled
);
2535 EXPORT_SYMBOL_GPL(ring_buffer_read_finish
);
2538 * ring_buffer_read - read the next item in the ring buffer by the iterator
2539 * @iter: The ring buffer iterator
2540 * @ts: The time stamp of the event read.
2542 * This reads the next event in the ring buffer and increments the iterator.
2544 struct ring_buffer_event
*
2545 ring_buffer_read(struct ring_buffer_iter
*iter
, u64
*ts
)
2547 struct ring_buffer_event
*event
;
2548 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2549 unsigned long flags
;
2552 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2553 event
= rb_iter_peek(iter
, ts
);
2557 rb_advance_iter(iter
);
2559 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2561 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2568 EXPORT_SYMBOL_GPL(ring_buffer_read
);
2571 * ring_buffer_size - return the size of the ring buffer (in bytes)
2572 * @buffer: The ring buffer.
2574 unsigned long ring_buffer_size(struct ring_buffer
*buffer
)
2576 return BUF_PAGE_SIZE
* buffer
->pages
;
2578 EXPORT_SYMBOL_GPL(ring_buffer_size
);
2581 rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
)
2583 cpu_buffer
->head_page
2584 = list_entry(cpu_buffer
->pages
.next
, struct buffer_page
, list
);
2585 local_set(&cpu_buffer
->head_page
->write
, 0);
2586 local_set(&cpu_buffer
->head_page
->page
->commit
, 0);
2588 cpu_buffer
->head_page
->read
= 0;
2590 cpu_buffer
->tail_page
= cpu_buffer
->head_page
;
2591 cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
2593 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
2594 local_set(&cpu_buffer
->reader_page
->write
, 0);
2595 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
2596 cpu_buffer
->reader_page
->read
= 0;
2598 cpu_buffer
->overrun
= 0;
2599 cpu_buffer
->entries
= 0;
2601 cpu_buffer
->write_stamp
= 0;
2602 cpu_buffer
->read_stamp
= 0;
2606 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
2607 * @buffer: The ring buffer to reset a per cpu buffer of
2608 * @cpu: The CPU buffer to be reset
2610 void ring_buffer_reset_cpu(struct ring_buffer
*buffer
, int cpu
)
2612 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2613 unsigned long flags
;
2615 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2618 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2620 __raw_spin_lock(&cpu_buffer
->lock
);
2622 rb_reset_cpu(cpu_buffer
);
2624 __raw_spin_unlock(&cpu_buffer
->lock
);
2626 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2628 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu
);
2631 * ring_buffer_reset - reset a ring buffer
2632 * @buffer: The ring buffer to reset all cpu buffers
2634 void ring_buffer_reset(struct ring_buffer
*buffer
)
2638 for_each_buffer_cpu(buffer
, cpu
)
2639 ring_buffer_reset_cpu(buffer
, cpu
);
2641 EXPORT_SYMBOL_GPL(ring_buffer_reset
);
2644 * rind_buffer_empty - is the ring buffer empty?
2645 * @buffer: The ring buffer to test
2647 int ring_buffer_empty(struct ring_buffer
*buffer
)
2649 struct ring_buffer_per_cpu
*cpu_buffer
;
2652 /* yes this is racy, but if you don't like the race, lock the buffer */
2653 for_each_buffer_cpu(buffer
, cpu
) {
2654 cpu_buffer
= buffer
->buffers
[cpu
];
2655 if (!rb_per_cpu_empty(cpu_buffer
))
2661 EXPORT_SYMBOL_GPL(ring_buffer_empty
);
2664 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
2665 * @buffer: The ring buffer
2666 * @cpu: The CPU buffer to test
2668 int ring_buffer_empty_cpu(struct ring_buffer
*buffer
, int cpu
)
2670 struct ring_buffer_per_cpu
*cpu_buffer
;
2673 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2676 cpu_buffer
= buffer
->buffers
[cpu
];
2677 ret
= rb_per_cpu_empty(cpu_buffer
);
2682 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu
);
2685 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
2686 * @buffer_a: One buffer to swap with
2687 * @buffer_b: The other buffer to swap with
2689 * This function is useful for tracers that want to take a "snapshot"
2690 * of a CPU buffer and has another back up buffer lying around.
2691 * it is expected that the tracer handles the cpu buffer not being
2692 * used at the moment.
2694 int ring_buffer_swap_cpu(struct ring_buffer
*buffer_a
,
2695 struct ring_buffer
*buffer_b
, int cpu
)
2697 struct ring_buffer_per_cpu
*cpu_buffer_a
;
2698 struct ring_buffer_per_cpu
*cpu_buffer_b
;
2701 if (!cpumask_test_cpu(cpu
, buffer_a
->cpumask
) ||
2702 !cpumask_test_cpu(cpu
, buffer_b
->cpumask
))
2705 /* At least make sure the two buffers are somewhat the same */
2706 if (buffer_a
->pages
!= buffer_b
->pages
)
2711 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
2714 if (atomic_read(&buffer_a
->record_disabled
))
2717 if (atomic_read(&buffer_b
->record_disabled
))
2720 cpu_buffer_a
= buffer_a
->buffers
[cpu
];
2721 cpu_buffer_b
= buffer_b
->buffers
[cpu
];
2723 if (atomic_read(&cpu_buffer_a
->record_disabled
))
2726 if (atomic_read(&cpu_buffer_b
->record_disabled
))
2730 * We can't do a synchronize_sched here because this
2731 * function can be called in atomic context.
2732 * Normally this will be called from the same CPU as cpu.
2733 * If not it's up to the caller to protect this.
2735 atomic_inc(&cpu_buffer_a
->record_disabled
);
2736 atomic_inc(&cpu_buffer_b
->record_disabled
);
2738 buffer_a
->buffers
[cpu
] = cpu_buffer_b
;
2739 buffer_b
->buffers
[cpu
] = cpu_buffer_a
;
2741 cpu_buffer_b
->buffer
= buffer_a
;
2742 cpu_buffer_a
->buffer
= buffer_b
;
2744 atomic_dec(&cpu_buffer_a
->record_disabled
);
2745 atomic_dec(&cpu_buffer_b
->record_disabled
);
2751 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu
);
2753 static void rb_remove_entries(struct ring_buffer_per_cpu
*cpu_buffer
,
2754 struct buffer_data_page
*bpage
,
2755 unsigned int offset
)
2757 struct ring_buffer_event
*event
;
2760 __raw_spin_lock(&cpu_buffer
->lock
);
2761 for (head
= offset
; head
< local_read(&bpage
->commit
);
2762 head
+= rb_event_length(event
)) {
2764 event
= __rb_data_page_index(bpage
, head
);
2765 if (RB_WARN_ON(cpu_buffer
, rb_null_event(event
)))
2767 /* Only count data entries */
2768 if (event
->type_len
> RINGBUF_TYPE_DATA_TYPE_LEN_MAX
)
2770 cpu_buffer
->entries
--;
2772 __raw_spin_unlock(&cpu_buffer
->lock
);
2776 * ring_buffer_alloc_read_page - allocate a page to read from buffer
2777 * @buffer: the buffer to allocate for.
2779 * This function is used in conjunction with ring_buffer_read_page.
2780 * When reading a full page from the ring buffer, these functions
2781 * can be used to speed up the process. The calling function should
2782 * allocate a few pages first with this function. Then when it
2783 * needs to get pages from the ring buffer, it passes the result
2784 * of this function into ring_buffer_read_page, which will swap
2785 * the page that was allocated, with the read page of the buffer.
2788 * The page allocated, or NULL on error.
2790 void *ring_buffer_alloc_read_page(struct ring_buffer
*buffer
)
2792 struct buffer_data_page
*bpage
;
2795 addr
= __get_free_page(GFP_KERNEL
);
2799 bpage
= (void *)addr
;
2801 rb_init_page(bpage
);
2807 * ring_buffer_free_read_page - free an allocated read page
2808 * @buffer: the buffer the page was allocate for
2809 * @data: the page to free
2811 * Free a page allocated from ring_buffer_alloc_read_page.
2813 void ring_buffer_free_read_page(struct ring_buffer
*buffer
, void *data
)
2815 free_page((unsigned long)data
);
2819 * ring_buffer_read_page - extract a page from the ring buffer
2820 * @buffer: buffer to extract from
2821 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
2822 * @len: amount to extract
2823 * @cpu: the cpu of the buffer to extract
2824 * @full: should the extraction only happen when the page is full.
2826 * This function will pull out a page from the ring buffer and consume it.
2827 * @data_page must be the address of the variable that was returned
2828 * from ring_buffer_alloc_read_page. This is because the page might be used
2829 * to swap with a page in the ring buffer.
2832 * rpage = ring_buffer_alloc_read_page(buffer);
2835 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
2837 * process_page(rpage, ret);
2839 * When @full is set, the function will not return true unless
2840 * the writer is off the reader page.
2842 * Note: it is up to the calling functions to handle sleeps and wakeups.
2843 * The ring buffer can be used anywhere in the kernel and can not
2844 * blindly call wake_up. The layer that uses the ring buffer must be
2845 * responsible for that.
2848 * >=0 if data has been transferred, returns the offset of consumed data.
2849 * <0 if no data has been transferred.
2851 int ring_buffer_read_page(struct ring_buffer
*buffer
,
2852 void **data_page
, size_t len
, int cpu
, int full
)
2854 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2855 struct ring_buffer_event
*event
;
2856 struct buffer_data_page
*bpage
;
2857 struct buffer_page
*reader
;
2858 unsigned long flags
;
2859 unsigned int commit
;
2864 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2868 * If len is not big enough to hold the page header, then
2869 * we can not copy anything.
2871 if (len
<= BUF_PAGE_HDR_SIZE
)
2874 len
-= BUF_PAGE_HDR_SIZE
;
2883 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2885 reader
= rb_get_reader_page(cpu_buffer
);
2889 event
= rb_reader_event(cpu_buffer
);
2891 read
= reader
->read
;
2892 commit
= rb_page_commit(reader
);
2895 * If this page has been partially read or
2896 * if len is not big enough to read the rest of the page or
2897 * a writer is still on the page, then
2898 * we must copy the data from the page to the buffer.
2899 * Otherwise, we can simply swap the page with the one passed in.
2901 if (read
|| (len
< (commit
- read
)) ||
2902 cpu_buffer
->reader_page
== cpu_buffer
->commit_page
) {
2903 struct buffer_data_page
*rpage
= cpu_buffer
->reader_page
->page
;
2904 unsigned int rpos
= read
;
2905 unsigned int pos
= 0;
2911 if (len
> (commit
- read
))
2912 len
= (commit
- read
);
2914 size
= rb_event_length(event
);
2919 /* save the current timestamp, since the user will need it */
2920 save_timestamp
= cpu_buffer
->read_stamp
;
2922 /* Need to copy one event at a time */
2924 memcpy(bpage
->data
+ pos
, rpage
->data
+ rpos
, size
);
2928 rb_advance_reader(cpu_buffer
);
2929 rpos
= reader
->read
;
2932 event
= rb_reader_event(cpu_buffer
);
2933 size
= rb_event_length(event
);
2934 } while (len
> size
);
2937 local_set(&bpage
->commit
, pos
);
2938 bpage
->time_stamp
= save_timestamp
;
2940 /* we copied everything to the beginning */
2943 /* swap the pages */
2944 rb_init_page(bpage
);
2945 bpage
= reader
->page
;
2946 reader
->page
= *data_page
;
2947 local_set(&reader
->write
, 0);
2951 /* update the entry counter */
2952 rb_remove_entries(cpu_buffer
, bpage
, read
);
2957 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2964 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
2965 size_t cnt
, loff_t
*ppos
)
2967 unsigned long *p
= filp
->private_data
;
2971 if (test_bit(RB_BUFFERS_DISABLED_BIT
, p
))
2972 r
= sprintf(buf
, "permanently disabled\n");
2974 r
= sprintf(buf
, "%d\n", test_bit(RB_BUFFERS_ON_BIT
, p
));
2976 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
2980 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
2981 size_t cnt
, loff_t
*ppos
)
2983 unsigned long *p
= filp
->private_data
;
2988 if (cnt
>= sizeof(buf
))
2991 if (copy_from_user(&buf
, ubuf
, cnt
))
2996 ret
= strict_strtoul(buf
, 10, &val
);
3001 set_bit(RB_BUFFERS_ON_BIT
, p
);
3003 clear_bit(RB_BUFFERS_ON_BIT
, p
);
3010 static const struct file_operations rb_simple_fops
= {
3011 .open
= tracing_open_generic
,
3012 .read
= rb_simple_read
,
3013 .write
= rb_simple_write
,
3017 static __init
int rb_init_debugfs(void)
3019 struct dentry
*d_tracer
;
3021 d_tracer
= tracing_init_dentry();
3023 trace_create_file("tracing_on", 0644, d_tracer
,
3024 &ring_buffer_flags
, &rb_simple_fops
);
3029 fs_initcall(rb_init_debugfs
);
3031 #ifdef CONFIG_HOTPLUG_CPU
3032 static int rb_cpu_notify(struct notifier_block
*self
,
3033 unsigned long action
, void *hcpu
)
3035 struct ring_buffer
*buffer
=
3036 container_of(self
, struct ring_buffer
, cpu_notify
);
3037 long cpu
= (long)hcpu
;
3040 case CPU_UP_PREPARE
:
3041 case CPU_UP_PREPARE_FROZEN
:
3042 if (cpu_isset(cpu
, *buffer
->cpumask
))
3045 buffer
->buffers
[cpu
] =
3046 rb_allocate_cpu_buffer(buffer
, cpu
);
3047 if (!buffer
->buffers
[cpu
]) {
3048 WARN(1, "failed to allocate ring buffer on CPU %ld\n",
3053 cpu_set(cpu
, *buffer
->cpumask
);
3055 case CPU_DOWN_PREPARE
:
3056 case CPU_DOWN_PREPARE_FROZEN
:
3059 * If we were to free the buffer, then the user would
3060 * lose any trace that was in the buffer.