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)
208 #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */
210 /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
211 #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
214 RB_LEN_TIME_EXTEND
= 8,
215 RB_LEN_TIME_STAMP
= 16,
218 static inline int rb_null_event(struct ring_buffer_event
*event
)
220 return event
->type_len
== RINGBUF_TYPE_PADDING
221 && event
->time_delta
== 0;
224 static inline int rb_discarded_event(struct ring_buffer_event
*event
)
226 return event
->type_len
== RINGBUF_TYPE_PADDING
&& event
->time_delta
;
229 static void rb_event_set_padding(struct ring_buffer_event
*event
)
231 event
->type_len
= RINGBUF_TYPE_PADDING
;
232 event
->time_delta
= 0;
236 rb_event_data_length(struct ring_buffer_event
*event
)
241 length
= event
->type_len
* RB_ALIGNMENT
;
243 length
= event
->array
[0];
244 return length
+ RB_EVNT_HDR_SIZE
;
247 /* inline for ring buffer fast paths */
249 rb_event_length(struct ring_buffer_event
*event
)
251 switch (event
->type_len
) {
252 case RINGBUF_TYPE_PADDING
:
253 if (rb_null_event(event
))
256 return event
->array
[0] + RB_EVNT_HDR_SIZE
;
258 case RINGBUF_TYPE_TIME_EXTEND
:
259 return RB_LEN_TIME_EXTEND
;
261 case RINGBUF_TYPE_TIME_STAMP
:
262 return RB_LEN_TIME_STAMP
;
264 case RINGBUF_TYPE_DATA
:
265 return rb_event_data_length(event
);
274 * ring_buffer_event_length - return the length of the event
275 * @event: the event to get the length of
277 unsigned ring_buffer_event_length(struct ring_buffer_event
*event
)
279 unsigned length
= rb_event_length(event
);
280 if (event
->type_len
> RINGBUF_TYPE_DATA_TYPE_LEN_MAX
)
282 length
-= RB_EVNT_HDR_SIZE
;
283 if (length
> RB_MAX_SMALL_DATA
+ sizeof(event
->array
[0]))
284 length
-= sizeof(event
->array
[0]);
287 EXPORT_SYMBOL_GPL(ring_buffer_event_length
);
289 /* inline for ring buffer fast paths */
291 rb_event_data(struct ring_buffer_event
*event
)
293 BUG_ON(event
->type_len
> RINGBUF_TYPE_DATA_TYPE_LEN_MAX
);
294 /* If length is in len field, then array[0] has the data */
296 return (void *)&event
->array
[0];
297 /* Otherwise length is in array[0] and array[1] has the data */
298 return (void *)&event
->array
[1];
302 * ring_buffer_event_data - return the data of the event
303 * @event: the event to get the data from
305 void *ring_buffer_event_data(struct ring_buffer_event
*event
)
307 return rb_event_data(event
);
309 EXPORT_SYMBOL_GPL(ring_buffer_event_data
);
311 #define for_each_buffer_cpu(buffer, cpu) \
312 for_each_cpu(cpu, buffer->cpumask)
315 #define TS_MASK ((1ULL << TS_SHIFT) - 1)
316 #define TS_DELTA_TEST (~TS_MASK)
318 struct buffer_data_page
{
319 u64 time_stamp
; /* page time stamp */
320 local_t commit
; /* write committed index */
321 unsigned char data
[]; /* data of buffer page */
325 struct list_head list
; /* list of buffer pages */
326 local_t write
; /* index for next write */
327 unsigned read
; /* index for next read */
328 local_t entries
; /* entries on this page */
329 struct buffer_data_page
*page
; /* Actual data page */
332 static void rb_init_page(struct buffer_data_page
*bpage
)
334 local_set(&bpage
->commit
, 0);
338 * ring_buffer_page_len - the size of data on the page.
339 * @page: The page to read
341 * Returns the amount of data on the page, including buffer page header.
343 size_t ring_buffer_page_len(void *page
)
345 return local_read(&((struct buffer_data_page
*)page
)->commit
)
350 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
353 static void free_buffer_page(struct buffer_page
*bpage
)
355 free_page((unsigned long)bpage
->page
);
360 * We need to fit the time_stamp delta into 27 bits.
362 static inline int test_time_stamp(u64 delta
)
364 if (delta
& TS_DELTA_TEST
)
369 #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
371 /* Max payload is BUF_PAGE_SIZE - header (8bytes) */
372 #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
374 /* Max number of timestamps that can fit on a page */
375 #define RB_TIMESTAMPS_PER_PAGE (BUF_PAGE_SIZE / RB_LEN_TIME_STAMP)
377 int ring_buffer_print_page_header(struct trace_seq
*s
)
379 struct buffer_data_page field
;
382 ret
= trace_seq_printf(s
, "\tfield: u64 timestamp;\t"
383 "offset:0;\tsize:%u;\n",
384 (unsigned int)sizeof(field
.time_stamp
));
386 ret
= trace_seq_printf(s
, "\tfield: local_t commit;\t"
387 "offset:%u;\tsize:%u;\n",
388 (unsigned int)offsetof(typeof(field
), commit
),
389 (unsigned int)sizeof(field
.commit
));
391 ret
= trace_seq_printf(s
, "\tfield: char data;\t"
392 "offset:%u;\tsize:%u;\n",
393 (unsigned int)offsetof(typeof(field
), data
),
394 (unsigned int)BUF_PAGE_SIZE
);
400 * head_page == tail_page && head == tail then buffer is empty.
402 struct ring_buffer_per_cpu
{
404 struct ring_buffer
*buffer
;
405 spinlock_t reader_lock
; /* serialize readers */
407 struct lock_class_key lock_key
;
408 struct list_head pages
;
409 struct buffer_page
*head_page
; /* read from head */
410 struct buffer_page
*tail_page
; /* write to tail */
411 struct buffer_page
*commit_page
; /* committed pages */
412 struct buffer_page
*reader_page
;
413 unsigned long nmi_dropped
;
414 unsigned long commit_overrun
;
415 unsigned long overrun
;
422 atomic_t record_disabled
;
429 atomic_t record_disabled
;
430 cpumask_var_t cpumask
;
432 struct lock_class_key
*reader_lock_key
;
436 struct ring_buffer_per_cpu
**buffers
;
438 #ifdef CONFIG_HOTPLUG_CPU
439 struct notifier_block cpu_notify
;
444 struct ring_buffer_iter
{
445 struct ring_buffer_per_cpu
*cpu_buffer
;
447 struct buffer_page
*head_page
;
451 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
452 #define RB_WARN_ON(buffer, cond) \
454 int _____ret = unlikely(cond); \
456 atomic_inc(&buffer->record_disabled); \
462 /* Up this if you want to test the TIME_EXTENTS and normalization */
463 #define DEBUG_SHIFT 0
465 static inline u64
rb_time_stamp(struct ring_buffer
*buffer
, int cpu
)
467 /* shift to debug/test normalization and TIME_EXTENTS */
468 return buffer
->clock() << DEBUG_SHIFT
;
471 u64
ring_buffer_time_stamp(struct ring_buffer
*buffer
, int cpu
)
475 preempt_disable_notrace();
476 time
= rb_time_stamp(buffer
, cpu
);
477 preempt_enable_no_resched_notrace();
481 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp
);
483 void ring_buffer_normalize_time_stamp(struct ring_buffer
*buffer
,
486 /* Just stupid testing the normalize function and deltas */
489 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp
);
492 * check_pages - integrity check of buffer pages
493 * @cpu_buffer: CPU buffer with pages to test
495 * As a safety measure we check to make sure the data pages have not
498 static int rb_check_pages(struct ring_buffer_per_cpu
*cpu_buffer
)
500 struct list_head
*head
= &cpu_buffer
->pages
;
501 struct buffer_page
*bpage
, *tmp
;
503 if (RB_WARN_ON(cpu_buffer
, head
->next
->prev
!= head
))
505 if (RB_WARN_ON(cpu_buffer
, head
->prev
->next
!= head
))
508 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
509 if (RB_WARN_ON(cpu_buffer
,
510 bpage
->list
.next
->prev
!= &bpage
->list
))
512 if (RB_WARN_ON(cpu_buffer
,
513 bpage
->list
.prev
->next
!= &bpage
->list
))
520 static int rb_allocate_pages(struct ring_buffer_per_cpu
*cpu_buffer
,
523 struct list_head
*head
= &cpu_buffer
->pages
;
524 struct buffer_page
*bpage
, *tmp
;
529 for (i
= 0; i
< nr_pages
; i
++) {
530 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
531 GFP_KERNEL
, cpu_to_node(cpu_buffer
->cpu
));
534 list_add(&bpage
->list
, &pages
);
536 addr
= __get_free_page(GFP_KERNEL
);
539 bpage
->page
= (void *)addr
;
540 rb_init_page(bpage
->page
);
543 list_splice(&pages
, head
);
545 rb_check_pages(cpu_buffer
);
550 list_for_each_entry_safe(bpage
, tmp
, &pages
, list
) {
551 list_del_init(&bpage
->list
);
552 free_buffer_page(bpage
);
557 static struct ring_buffer_per_cpu
*
558 rb_allocate_cpu_buffer(struct ring_buffer
*buffer
, int cpu
)
560 struct ring_buffer_per_cpu
*cpu_buffer
;
561 struct buffer_page
*bpage
;
565 cpu_buffer
= kzalloc_node(ALIGN(sizeof(*cpu_buffer
), cache_line_size()),
566 GFP_KERNEL
, cpu_to_node(cpu
));
570 cpu_buffer
->cpu
= cpu
;
571 cpu_buffer
->buffer
= buffer
;
572 spin_lock_init(&cpu_buffer
->reader_lock
);
573 lockdep_set_class(&cpu_buffer
->reader_lock
, buffer
->reader_lock_key
);
574 cpu_buffer
->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
575 INIT_LIST_HEAD(&cpu_buffer
->pages
);
577 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
578 GFP_KERNEL
, cpu_to_node(cpu
));
580 goto fail_free_buffer
;
582 cpu_buffer
->reader_page
= bpage
;
583 addr
= __get_free_page(GFP_KERNEL
);
585 goto fail_free_reader
;
586 bpage
->page
= (void *)addr
;
587 rb_init_page(bpage
->page
);
589 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
591 ret
= rb_allocate_pages(cpu_buffer
, buffer
->pages
);
593 goto fail_free_reader
;
595 cpu_buffer
->head_page
596 = list_entry(cpu_buffer
->pages
.next
, struct buffer_page
, list
);
597 cpu_buffer
->tail_page
= cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
602 free_buffer_page(cpu_buffer
->reader_page
);
609 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu
*cpu_buffer
)
611 struct list_head
*head
= &cpu_buffer
->pages
;
612 struct buffer_page
*bpage
, *tmp
;
614 free_buffer_page(cpu_buffer
->reader_page
);
616 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
617 list_del_init(&bpage
->list
);
618 free_buffer_page(bpage
);
623 #ifdef CONFIG_HOTPLUG_CPU
624 static int rb_cpu_notify(struct notifier_block
*self
,
625 unsigned long action
, void *hcpu
);
629 * ring_buffer_alloc - allocate a new ring_buffer
630 * @size: the size in bytes per cpu that is needed.
631 * @flags: attributes to set for the ring buffer.
633 * Currently the only flag that is available is the RB_FL_OVERWRITE
634 * flag. This flag means that the buffer will overwrite old data
635 * when the buffer wraps. If this flag is not set, the buffer will
636 * drop data when the tail hits the head.
638 struct ring_buffer
*__ring_buffer_alloc(unsigned long size
, unsigned flags
,
639 struct lock_class_key
*key
)
641 struct ring_buffer
*buffer
;
645 /* keep it in its own cache line */
646 buffer
= kzalloc(ALIGN(sizeof(*buffer
), cache_line_size()),
651 if (!alloc_cpumask_var(&buffer
->cpumask
, GFP_KERNEL
))
652 goto fail_free_buffer
;
654 buffer
->pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
655 buffer
->flags
= flags
;
656 buffer
->clock
= trace_clock_local
;
657 buffer
->reader_lock_key
= key
;
659 /* need at least two pages */
660 if (buffer
->pages
< 2)
664 * In case of non-hotplug cpu, if the ring-buffer is allocated
665 * in early initcall, it will not be notified of secondary cpus.
666 * In that off case, we need to allocate for all possible cpus.
668 #ifdef CONFIG_HOTPLUG_CPU
670 cpumask_copy(buffer
->cpumask
, cpu_online_mask
);
672 cpumask_copy(buffer
->cpumask
, cpu_possible_mask
);
674 buffer
->cpus
= nr_cpu_ids
;
676 bsize
= sizeof(void *) * nr_cpu_ids
;
677 buffer
->buffers
= kzalloc(ALIGN(bsize
, cache_line_size()),
679 if (!buffer
->buffers
)
680 goto fail_free_cpumask
;
682 for_each_buffer_cpu(buffer
, cpu
) {
683 buffer
->buffers
[cpu
] =
684 rb_allocate_cpu_buffer(buffer
, cpu
);
685 if (!buffer
->buffers
[cpu
])
686 goto fail_free_buffers
;
689 #ifdef CONFIG_HOTPLUG_CPU
690 buffer
->cpu_notify
.notifier_call
= rb_cpu_notify
;
691 buffer
->cpu_notify
.priority
= 0;
692 register_cpu_notifier(&buffer
->cpu_notify
);
696 mutex_init(&buffer
->mutex
);
701 for_each_buffer_cpu(buffer
, cpu
) {
702 if (buffer
->buffers
[cpu
])
703 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
705 kfree(buffer
->buffers
);
708 free_cpumask_var(buffer
->cpumask
);
715 EXPORT_SYMBOL_GPL(__ring_buffer_alloc
);
718 * ring_buffer_free - free a ring buffer.
719 * @buffer: the buffer to free.
722 ring_buffer_free(struct ring_buffer
*buffer
)
728 #ifdef CONFIG_HOTPLUG_CPU
729 unregister_cpu_notifier(&buffer
->cpu_notify
);
732 for_each_buffer_cpu(buffer
, cpu
)
733 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
737 free_cpumask_var(buffer
->cpumask
);
741 EXPORT_SYMBOL_GPL(ring_buffer_free
);
743 void ring_buffer_set_clock(struct ring_buffer
*buffer
,
746 buffer
->clock
= clock
;
749 static void rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
);
752 rb_remove_pages(struct ring_buffer_per_cpu
*cpu_buffer
, unsigned nr_pages
)
754 struct buffer_page
*bpage
;
758 atomic_inc(&cpu_buffer
->record_disabled
);
761 for (i
= 0; i
< nr_pages
; i
++) {
762 if (RB_WARN_ON(cpu_buffer
, list_empty(&cpu_buffer
->pages
)))
764 p
= cpu_buffer
->pages
.next
;
765 bpage
= list_entry(p
, struct buffer_page
, list
);
766 list_del_init(&bpage
->list
);
767 free_buffer_page(bpage
);
769 if (RB_WARN_ON(cpu_buffer
, list_empty(&cpu_buffer
->pages
)))
772 rb_reset_cpu(cpu_buffer
);
774 rb_check_pages(cpu_buffer
);
776 atomic_dec(&cpu_buffer
->record_disabled
);
781 rb_insert_pages(struct ring_buffer_per_cpu
*cpu_buffer
,
782 struct list_head
*pages
, unsigned nr_pages
)
784 struct buffer_page
*bpage
;
788 atomic_inc(&cpu_buffer
->record_disabled
);
791 for (i
= 0; i
< nr_pages
; i
++) {
792 if (RB_WARN_ON(cpu_buffer
, list_empty(pages
)))
795 bpage
= list_entry(p
, struct buffer_page
, list
);
796 list_del_init(&bpage
->list
);
797 list_add_tail(&bpage
->list
, &cpu_buffer
->pages
);
799 rb_reset_cpu(cpu_buffer
);
801 rb_check_pages(cpu_buffer
);
803 atomic_dec(&cpu_buffer
->record_disabled
);
807 * ring_buffer_resize - resize the ring buffer
808 * @buffer: the buffer to resize.
809 * @size: the new size.
811 * The tracer is responsible for making sure that the buffer is
812 * not being used while changing the size.
813 * Note: We may be able to change the above requirement by using
814 * RCU synchronizations.
816 * Minimum size is 2 * BUF_PAGE_SIZE.
818 * Returns -1 on failure.
820 int ring_buffer_resize(struct ring_buffer
*buffer
, unsigned long size
)
822 struct ring_buffer_per_cpu
*cpu_buffer
;
823 unsigned nr_pages
, rm_pages
, new_pages
;
824 struct buffer_page
*bpage
, *tmp
;
825 unsigned long buffer_size
;
831 * Always succeed at resizing a non-existent buffer:
836 size
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
837 size
*= BUF_PAGE_SIZE
;
838 buffer_size
= buffer
->pages
* BUF_PAGE_SIZE
;
840 /* we need a minimum of two pages */
841 if (size
< BUF_PAGE_SIZE
* 2)
842 size
= BUF_PAGE_SIZE
* 2;
844 if (size
== buffer_size
)
847 mutex_lock(&buffer
->mutex
);
850 nr_pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
852 if (size
< buffer_size
) {
854 /* easy case, just free pages */
855 if (RB_WARN_ON(buffer
, nr_pages
>= buffer
->pages
))
858 rm_pages
= buffer
->pages
- nr_pages
;
860 for_each_buffer_cpu(buffer
, cpu
) {
861 cpu_buffer
= buffer
->buffers
[cpu
];
862 rb_remove_pages(cpu_buffer
, rm_pages
);
868 * This is a bit more difficult. We only want to add pages
869 * when we can allocate enough for all CPUs. We do this
870 * by allocating all the pages and storing them on a local
871 * link list. If we succeed in our allocation, then we
872 * add these pages to the cpu_buffers. Otherwise we just free
873 * them all and return -ENOMEM;
875 if (RB_WARN_ON(buffer
, nr_pages
<= buffer
->pages
))
878 new_pages
= nr_pages
- buffer
->pages
;
880 for_each_buffer_cpu(buffer
, cpu
) {
881 for (i
= 0; i
< new_pages
; i
++) {
882 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
),
884 GFP_KERNEL
, cpu_to_node(cpu
));
887 list_add(&bpage
->list
, &pages
);
888 addr
= __get_free_page(GFP_KERNEL
);
891 bpage
->page
= (void *)addr
;
892 rb_init_page(bpage
->page
);
896 for_each_buffer_cpu(buffer
, cpu
) {
897 cpu_buffer
= buffer
->buffers
[cpu
];
898 rb_insert_pages(cpu_buffer
, &pages
, new_pages
);
901 if (RB_WARN_ON(buffer
, !list_empty(&pages
)))
905 buffer
->pages
= nr_pages
;
907 mutex_unlock(&buffer
->mutex
);
912 list_for_each_entry_safe(bpage
, tmp
, &pages
, list
) {
913 list_del_init(&bpage
->list
);
914 free_buffer_page(bpage
);
917 mutex_unlock(&buffer
->mutex
);
921 * Something went totally wrong, and we are too paranoid
922 * to even clean up the mess.
926 mutex_unlock(&buffer
->mutex
);
929 EXPORT_SYMBOL_GPL(ring_buffer_resize
);
932 __rb_data_page_index(struct buffer_data_page
*bpage
, unsigned index
)
934 return bpage
->data
+ index
;
937 static inline void *__rb_page_index(struct buffer_page
*bpage
, unsigned index
)
939 return bpage
->page
->data
+ index
;
942 static inline struct ring_buffer_event
*
943 rb_reader_event(struct ring_buffer_per_cpu
*cpu_buffer
)
945 return __rb_page_index(cpu_buffer
->reader_page
,
946 cpu_buffer
->reader_page
->read
);
949 static inline struct ring_buffer_event
*
950 rb_head_event(struct ring_buffer_per_cpu
*cpu_buffer
)
952 return __rb_page_index(cpu_buffer
->head_page
,
953 cpu_buffer
->head_page
->read
);
956 static inline struct ring_buffer_event
*
957 rb_iter_head_event(struct ring_buffer_iter
*iter
)
959 return __rb_page_index(iter
->head_page
, iter
->head
);
962 static inline unsigned rb_page_write(struct buffer_page
*bpage
)
964 return local_read(&bpage
->write
);
967 static inline unsigned rb_page_commit(struct buffer_page
*bpage
)
969 return local_read(&bpage
->page
->commit
);
972 /* Size is determined by what has been commited */
973 static inline unsigned rb_page_size(struct buffer_page
*bpage
)
975 return rb_page_commit(bpage
);
978 static inline unsigned
979 rb_commit_index(struct ring_buffer_per_cpu
*cpu_buffer
)
981 return rb_page_commit(cpu_buffer
->commit_page
);
984 static inline unsigned rb_head_size(struct ring_buffer_per_cpu
*cpu_buffer
)
986 return rb_page_commit(cpu_buffer
->head_page
);
989 static inline void rb_inc_page(struct ring_buffer_per_cpu
*cpu_buffer
,
990 struct buffer_page
**bpage
)
992 struct list_head
*p
= (*bpage
)->list
.next
;
994 if (p
== &cpu_buffer
->pages
)
997 *bpage
= list_entry(p
, struct buffer_page
, list
);
1000 static inline unsigned
1001 rb_event_index(struct ring_buffer_event
*event
)
1003 unsigned long addr
= (unsigned long)event
;
1005 return (addr
& ~PAGE_MASK
) - BUF_PAGE_HDR_SIZE
;
1009 rb_event_is_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
1010 struct ring_buffer_event
*event
)
1012 unsigned long addr
= (unsigned long)event
;
1013 unsigned long index
;
1015 index
= rb_event_index(event
);
1018 return cpu_buffer
->commit_page
->page
== (void *)addr
&&
1019 rb_commit_index(cpu_buffer
) == index
;
1023 rb_set_commit_to_write(struct ring_buffer_per_cpu
*cpu_buffer
)
1026 * We only race with interrupts and NMIs on this CPU.
1027 * If we own the commit event, then we can commit
1028 * all others that interrupted us, since the interruptions
1029 * are in stack format (they finish before they come
1030 * back to us). This allows us to do a simple loop to
1031 * assign the commit to the tail.
1034 while (cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
) {
1035 cpu_buffer
->commit_page
->page
->commit
=
1036 cpu_buffer
->commit_page
->write
;
1037 rb_inc_page(cpu_buffer
, &cpu_buffer
->commit_page
);
1038 cpu_buffer
->write_stamp
=
1039 cpu_buffer
->commit_page
->page
->time_stamp
;
1040 /* add barrier to keep gcc from optimizing too much */
1043 while (rb_commit_index(cpu_buffer
) !=
1044 rb_page_write(cpu_buffer
->commit_page
)) {
1045 cpu_buffer
->commit_page
->page
->commit
=
1046 cpu_buffer
->commit_page
->write
;
1050 /* again, keep gcc from optimizing */
1054 * If an interrupt came in just after the first while loop
1055 * and pushed the tail page forward, we will be left with
1056 * a dangling commit that will never go forward.
1058 if (unlikely(cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
))
1062 static void rb_reset_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
1064 cpu_buffer
->read_stamp
= cpu_buffer
->reader_page
->page
->time_stamp
;
1065 cpu_buffer
->reader_page
->read
= 0;
1068 static void rb_inc_iter(struct ring_buffer_iter
*iter
)
1070 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
1073 * The iterator could be on the reader page (it starts there).
1074 * But the head could have moved, since the reader was
1075 * found. Check for this case and assign the iterator
1076 * to the head page instead of next.
1078 if (iter
->head_page
== cpu_buffer
->reader_page
)
1079 iter
->head_page
= cpu_buffer
->head_page
;
1081 rb_inc_page(cpu_buffer
, &iter
->head_page
);
1083 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
1088 * ring_buffer_update_event - update event type and data
1089 * @event: the even to update
1090 * @type: the type of event
1091 * @length: the size of the event field in the ring buffer
1093 * Update the type and data fields of the event. The length
1094 * is the actual size that is written to the ring buffer,
1095 * and with this, we can determine what to place into the
1099 rb_update_event(struct ring_buffer_event
*event
,
1100 unsigned type
, unsigned length
)
1102 event
->type_len
= type
;
1106 case RINGBUF_TYPE_PADDING
:
1107 case RINGBUF_TYPE_TIME_EXTEND
:
1108 case RINGBUF_TYPE_TIME_STAMP
:
1112 length
-= RB_EVNT_HDR_SIZE
;
1113 if (length
> RB_MAX_SMALL_DATA
)
1114 event
->array
[0] = length
;
1116 event
->type_len
= DIV_ROUND_UP(length
, RB_ALIGNMENT
);
1123 static unsigned rb_calculate_event_length(unsigned length
)
1125 struct ring_buffer_event event
; /* Used only for sizeof array */
1127 /* zero length can cause confusions */
1131 if (length
> RB_MAX_SMALL_DATA
)
1132 length
+= sizeof(event
.array
[0]);
1134 length
+= RB_EVNT_HDR_SIZE
;
1135 length
= ALIGN(length
, RB_ALIGNMENT
);
1141 rb_reset_tail(struct ring_buffer_per_cpu
*cpu_buffer
,
1142 struct buffer_page
*tail_page
,
1143 unsigned long tail
, unsigned long length
)
1145 struct ring_buffer_event
*event
;
1148 * Only the event that crossed the page boundary
1149 * must fill the old tail_page with padding.
1151 if (tail
>= BUF_PAGE_SIZE
) {
1152 local_sub(length
, &tail_page
->write
);
1156 event
= __rb_page_index(tail_page
, tail
);
1159 * If this event is bigger than the minimum size, then
1160 * we need to be careful that we don't subtract the
1161 * write counter enough to allow another writer to slip
1163 * We put in a discarded commit instead, to make sure
1164 * that this space is not used again.
1166 * If we are less than the minimum size, we don't need to
1169 if (tail
> (BUF_PAGE_SIZE
- RB_EVNT_MIN_SIZE
)) {
1170 /* No room for any events */
1172 /* Mark the rest of the page with padding */
1173 rb_event_set_padding(event
);
1175 /* Set the write back to the previous setting */
1176 local_sub(length
, &tail_page
->write
);
1180 /* Put in a discarded event */
1181 event
->array
[0] = (BUF_PAGE_SIZE
- tail
) - RB_EVNT_HDR_SIZE
;
1182 event
->type_len
= RINGBUF_TYPE_PADDING
;
1183 /* time delta must be non zero */
1184 event
->time_delta
= 1;
1185 /* Account for this as an entry */
1186 local_inc(&tail_page
->entries
);
1187 local_inc(&cpu_buffer
->entries
);
1189 /* Set write to end of buffer */
1190 length
= (tail
+ length
) - BUF_PAGE_SIZE
;
1191 local_sub(length
, &tail_page
->write
);
1194 static struct ring_buffer_event
*
1195 rb_move_tail(struct ring_buffer_per_cpu
*cpu_buffer
,
1196 unsigned long length
, unsigned long tail
,
1197 struct buffer_page
*commit_page
,
1198 struct buffer_page
*tail_page
, u64
*ts
)
1200 struct buffer_page
*next_page
, *head_page
, *reader_page
;
1201 struct ring_buffer
*buffer
= cpu_buffer
->buffer
;
1202 bool lock_taken
= false;
1203 unsigned long flags
;
1205 next_page
= tail_page
;
1207 local_irq_save(flags
);
1209 * Since the write to the buffer is still not
1210 * fully lockless, we must be careful with NMIs.
1211 * The locks in the writers are taken when a write
1212 * crosses to a new page. The locks protect against
1213 * races with the readers (this will soon be fixed
1214 * with a lockless solution).
1216 * Because we can not protect against NMIs, and we
1217 * want to keep traces reentrant, we need to manage
1218 * what happens when we are in an NMI.
1220 * NMIs can happen after we take the lock.
1221 * If we are in an NMI, only take the lock
1222 * if it is not already taken. Otherwise
1225 if (unlikely(in_nmi())) {
1226 if (!__raw_spin_trylock(&cpu_buffer
->lock
)) {
1227 cpu_buffer
->nmi_dropped
++;
1231 __raw_spin_lock(&cpu_buffer
->lock
);
1235 rb_inc_page(cpu_buffer
, &next_page
);
1237 head_page
= cpu_buffer
->head_page
;
1238 reader_page
= cpu_buffer
->reader_page
;
1240 /* we grabbed the lock before incrementing */
1241 if (RB_WARN_ON(cpu_buffer
, next_page
== reader_page
))
1245 * If for some reason, we had an interrupt storm that made
1246 * it all the way around the buffer, bail, and warn
1249 if (unlikely(next_page
== commit_page
)) {
1250 cpu_buffer
->commit_overrun
++;
1254 if (next_page
== head_page
) {
1255 if (!(buffer
->flags
& RB_FL_OVERWRITE
))
1258 /* tail_page has not moved yet? */
1259 if (tail_page
== cpu_buffer
->tail_page
) {
1260 /* count overflows */
1261 cpu_buffer
->overrun
+=
1262 local_read(&head_page
->entries
);
1264 rb_inc_page(cpu_buffer
, &head_page
);
1265 cpu_buffer
->head_page
= head_page
;
1266 cpu_buffer
->head_page
->read
= 0;
1271 * If the tail page is still the same as what we think
1272 * it is, then it is up to us to update the tail
1275 if (tail_page
== cpu_buffer
->tail_page
) {
1276 local_set(&next_page
->write
, 0);
1277 local_set(&next_page
->entries
, 0);
1278 local_set(&next_page
->page
->commit
, 0);
1279 cpu_buffer
->tail_page
= next_page
;
1281 /* reread the time stamp */
1282 *ts
= rb_time_stamp(buffer
, cpu_buffer
->cpu
);
1283 cpu_buffer
->tail_page
->page
->time_stamp
= *ts
;
1286 rb_reset_tail(cpu_buffer
, tail_page
, tail
, length
);
1288 __raw_spin_unlock(&cpu_buffer
->lock
);
1289 local_irq_restore(flags
);
1291 /* fail and let the caller try again */
1292 return ERR_PTR(-EAGAIN
);
1296 rb_reset_tail(cpu_buffer
, tail_page
, tail
, length
);
1298 if (likely(lock_taken
))
1299 __raw_spin_unlock(&cpu_buffer
->lock
);
1300 local_irq_restore(flags
);
1304 static struct ring_buffer_event
*
1305 __rb_reserve_next(struct ring_buffer_per_cpu
*cpu_buffer
,
1306 unsigned type
, unsigned long length
, u64
*ts
)
1308 struct buffer_page
*tail_page
, *commit_page
;
1309 struct ring_buffer_event
*event
;
1310 unsigned long tail
, write
;
1312 commit_page
= cpu_buffer
->commit_page
;
1313 /* we just need to protect against interrupts */
1315 tail_page
= cpu_buffer
->tail_page
;
1316 write
= local_add_return(length
, &tail_page
->write
);
1317 tail
= write
- length
;
1319 /* See if we shot pass the end of this buffer page */
1320 if (write
> BUF_PAGE_SIZE
)
1321 return rb_move_tail(cpu_buffer
, length
, tail
,
1322 commit_page
, tail_page
, ts
);
1324 /* We reserved something on the buffer */
1326 event
= __rb_page_index(tail_page
, tail
);
1327 rb_update_event(event
, type
, length
);
1329 /* The passed in type is zero for DATA */
1331 local_inc(&tail_page
->entries
);
1334 * If this is the first commit on the page, then update
1338 tail_page
->page
->time_stamp
= *ts
;
1344 rb_try_to_discard(struct ring_buffer_per_cpu
*cpu_buffer
,
1345 struct ring_buffer_event
*event
)
1347 unsigned long new_index
, old_index
;
1348 struct buffer_page
*bpage
;
1349 unsigned long index
;
1352 new_index
= rb_event_index(event
);
1353 old_index
= new_index
+ rb_event_length(event
);
1354 addr
= (unsigned long)event
;
1357 bpage
= cpu_buffer
->tail_page
;
1359 if (bpage
->page
== (void *)addr
&& rb_page_write(bpage
) == old_index
) {
1361 * This is on the tail page. It is possible that
1362 * a write could come in and move the tail page
1363 * and write to the next page. That is fine
1364 * because we just shorten what is on this page.
1366 index
= local_cmpxchg(&bpage
->write
, old_index
, new_index
);
1367 if (index
== old_index
)
1371 /* could not discard */
1376 rb_add_time_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
1377 u64
*ts
, u64
*delta
)
1379 struct ring_buffer_event
*event
;
1383 if (unlikely(*delta
> (1ULL << 59) && !once
++)) {
1384 printk(KERN_WARNING
"Delta way too big! %llu"
1385 " ts=%llu write stamp = %llu\n",
1386 (unsigned long long)*delta
,
1387 (unsigned long long)*ts
,
1388 (unsigned long long)cpu_buffer
->write_stamp
);
1393 * The delta is too big, we to add a
1396 event
= __rb_reserve_next(cpu_buffer
,
1397 RINGBUF_TYPE_TIME_EXTEND
,
1403 if (PTR_ERR(event
) == -EAGAIN
)
1406 /* Only a commited time event can update the write stamp */
1407 if (rb_event_is_commit(cpu_buffer
, event
)) {
1409 * If this is the first on the page, then it was
1410 * updated with the page itself. Try to discard it
1411 * and if we can't just make it zero.
1413 if (rb_event_index(event
)) {
1414 event
->time_delta
= *delta
& TS_MASK
;
1415 event
->array
[0] = *delta
>> TS_SHIFT
;
1417 /* try to discard, since we do not need this */
1418 if (!rb_try_to_discard(cpu_buffer
, event
)) {
1419 /* nope, just zero it */
1420 event
->time_delta
= 0;
1421 event
->array
[0] = 0;
1424 cpu_buffer
->write_stamp
= *ts
;
1425 /* let the caller know this was the commit */
1428 /* Try to discard the event */
1429 if (!rb_try_to_discard(cpu_buffer
, event
)) {
1430 /* Darn, this is just wasted space */
1431 event
->time_delta
= 0;
1432 event
->array
[0] = 0;
1442 static void rb_start_commit(struct ring_buffer_per_cpu
*cpu_buffer
)
1444 local_inc(&cpu_buffer
->committing
);
1445 local_inc(&cpu_buffer
->commits
);
1448 static void rb_end_commit(struct ring_buffer_per_cpu
*cpu_buffer
)
1450 unsigned long commits
;
1452 if (RB_WARN_ON(cpu_buffer
,
1453 !local_read(&cpu_buffer
->committing
)))
1457 commits
= local_read(&cpu_buffer
->commits
);
1458 /* synchronize with interrupts */
1460 if (local_read(&cpu_buffer
->committing
) == 1)
1461 rb_set_commit_to_write(cpu_buffer
);
1463 local_dec(&cpu_buffer
->committing
);
1465 /* synchronize with interrupts */
1469 * Need to account for interrupts coming in between the
1470 * updating of the commit page and the clearing of the
1471 * committing counter.
1473 if (unlikely(local_read(&cpu_buffer
->commits
) != commits
) &&
1474 !local_read(&cpu_buffer
->committing
)) {
1475 local_inc(&cpu_buffer
->committing
);
1480 static struct ring_buffer_event
*
1481 rb_reserve_next_event(struct ring_buffer_per_cpu
*cpu_buffer
,
1482 unsigned long length
)
1484 struct ring_buffer_event
*event
;
1489 rb_start_commit(cpu_buffer
);
1491 length
= rb_calculate_event_length(length
);
1494 * We allow for interrupts to reenter here and do a trace.
1495 * If one does, it will cause this original code to loop
1496 * back here. Even with heavy interrupts happening, this
1497 * should only happen a few times in a row. If this happens
1498 * 1000 times in a row, there must be either an interrupt
1499 * storm or we have something buggy.
1502 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 1000))
1505 ts
= rb_time_stamp(cpu_buffer
->buffer
, cpu_buffer
->cpu
);
1508 * Only the first commit can update the timestamp.
1509 * Yes there is a race here. If an interrupt comes in
1510 * just after the conditional and it traces too, then it
1511 * will also check the deltas. More than one timestamp may
1512 * also be made. But only the entry that did the actual
1513 * commit will be something other than zero.
1515 if (likely(cpu_buffer
->tail_page
== cpu_buffer
->commit_page
&&
1516 rb_page_write(cpu_buffer
->tail_page
) ==
1517 rb_commit_index(cpu_buffer
))) {
1520 diff
= ts
- cpu_buffer
->write_stamp
;
1522 /* make sure this diff is calculated here */
1525 /* Did the write stamp get updated already? */
1526 if (unlikely(ts
< cpu_buffer
->write_stamp
))
1530 if (unlikely(test_time_stamp(delta
))) {
1532 commit
= rb_add_time_stamp(cpu_buffer
, &ts
, &delta
);
1533 if (commit
== -EBUSY
)
1536 if (commit
== -EAGAIN
)
1539 RB_WARN_ON(cpu_buffer
, commit
< 0);
1544 event
= __rb_reserve_next(cpu_buffer
, 0, length
, &ts
);
1545 if (unlikely(PTR_ERR(event
) == -EAGAIN
))
1551 if (!rb_event_is_commit(cpu_buffer
, event
))
1554 event
->time_delta
= delta
;
1559 rb_end_commit(cpu_buffer
);
1563 #define TRACE_RECURSIVE_DEPTH 16
1565 static int trace_recursive_lock(void)
1567 current
->trace_recursion
++;
1569 if (likely(current
->trace_recursion
< TRACE_RECURSIVE_DEPTH
))
1572 /* Disable all tracing before we do anything else */
1573 tracing_off_permanent();
1575 printk_once(KERN_WARNING
"Tracing recursion: depth[%ld]:"
1576 "HC[%lu]:SC[%lu]:NMI[%lu]\n",
1577 current
->trace_recursion
,
1578 hardirq_count() >> HARDIRQ_SHIFT
,
1579 softirq_count() >> SOFTIRQ_SHIFT
,
1586 static void trace_recursive_unlock(void)
1588 WARN_ON_ONCE(!current
->trace_recursion
);
1590 current
->trace_recursion
--;
1593 static DEFINE_PER_CPU(int, rb_need_resched
);
1596 * ring_buffer_lock_reserve - reserve a part of the buffer
1597 * @buffer: the ring buffer to reserve from
1598 * @length: the length of the data to reserve (excluding event header)
1600 * Returns a reseverd event on the ring buffer to copy directly to.
1601 * The user of this interface will need to get the body to write into
1602 * and can use the ring_buffer_event_data() interface.
1604 * The length is the length of the data needed, not the event length
1605 * which also includes the event header.
1607 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
1608 * If NULL is returned, then nothing has been allocated or locked.
1610 struct ring_buffer_event
*
1611 ring_buffer_lock_reserve(struct ring_buffer
*buffer
, unsigned long length
)
1613 struct ring_buffer_per_cpu
*cpu_buffer
;
1614 struct ring_buffer_event
*event
;
1617 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
1620 if (atomic_read(&buffer
->record_disabled
))
1623 /* If we are tracing schedule, we don't want to recurse */
1624 resched
= ftrace_preempt_disable();
1626 if (trace_recursive_lock())
1629 cpu
= raw_smp_processor_id();
1631 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1634 cpu_buffer
= buffer
->buffers
[cpu
];
1636 if (atomic_read(&cpu_buffer
->record_disabled
))
1639 if (length
> BUF_MAX_DATA_SIZE
)
1642 event
= rb_reserve_next_event(cpu_buffer
, length
);
1647 * Need to store resched state on this cpu.
1648 * Only the first needs to.
1651 if (preempt_count() == 1)
1652 per_cpu(rb_need_resched
, cpu
) = resched
;
1657 trace_recursive_unlock();
1660 ftrace_preempt_enable(resched
);
1663 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve
);
1665 static void rb_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
1666 struct ring_buffer_event
*event
)
1668 local_inc(&cpu_buffer
->entries
);
1671 * The event first in the commit queue updates the
1674 if (rb_event_is_commit(cpu_buffer
, event
))
1675 cpu_buffer
->write_stamp
+= event
->time_delta
;
1677 rb_end_commit(cpu_buffer
);
1681 * ring_buffer_unlock_commit - commit a reserved
1682 * @buffer: The buffer to commit to
1683 * @event: The event pointer to commit.
1685 * This commits the data to the ring buffer, and releases any locks held.
1687 * Must be paired with ring_buffer_lock_reserve.
1689 int ring_buffer_unlock_commit(struct ring_buffer
*buffer
,
1690 struct ring_buffer_event
*event
)
1692 struct ring_buffer_per_cpu
*cpu_buffer
;
1693 int cpu
= raw_smp_processor_id();
1695 cpu_buffer
= buffer
->buffers
[cpu
];
1697 rb_commit(cpu_buffer
, event
);
1699 trace_recursive_unlock();
1702 * Only the last preempt count needs to restore preemption.
1704 if (preempt_count() == 1)
1705 ftrace_preempt_enable(per_cpu(rb_need_resched
, cpu
));
1707 preempt_enable_no_resched_notrace();
1711 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit
);
1713 static inline void rb_event_discard(struct ring_buffer_event
*event
)
1715 /* array[0] holds the actual length for the discarded event */
1716 event
->array
[0] = rb_event_data_length(event
) - RB_EVNT_HDR_SIZE
;
1717 event
->type_len
= RINGBUF_TYPE_PADDING
;
1718 /* time delta must be non zero */
1719 if (!event
->time_delta
)
1720 event
->time_delta
= 1;
1724 * ring_buffer_event_discard - discard any event in the ring buffer
1725 * @event: the event to discard
1727 * Sometimes a event that is in the ring buffer needs to be ignored.
1728 * This function lets the user discard an event in the ring buffer
1729 * and then that event will not be read later.
1731 * Note, it is up to the user to be careful with this, and protect
1732 * against races. If the user discards an event that has been consumed
1733 * it is possible that it could corrupt the ring buffer.
1735 void ring_buffer_event_discard(struct ring_buffer_event
*event
)
1737 rb_event_discard(event
);
1739 EXPORT_SYMBOL_GPL(ring_buffer_event_discard
);
1742 * ring_buffer_commit_discard - discard an event that has not been committed
1743 * @buffer: the ring buffer
1744 * @event: non committed event to discard
1746 * This is similar to ring_buffer_event_discard but must only be
1747 * performed on an event that has not been committed yet. The difference
1748 * is that this will also try to free the event from the ring buffer
1749 * if another event has not been added behind it.
1751 * If another event has been added behind it, it will set the event
1752 * up as discarded, and perform the commit.
1754 * If this function is called, do not call ring_buffer_unlock_commit on
1757 void ring_buffer_discard_commit(struct ring_buffer
*buffer
,
1758 struct ring_buffer_event
*event
)
1760 struct ring_buffer_per_cpu
*cpu_buffer
;
1763 /* The event is discarded regardless */
1764 rb_event_discard(event
);
1766 cpu
= smp_processor_id();
1767 cpu_buffer
= buffer
->buffers
[cpu
];
1770 * This must only be called if the event has not been
1771 * committed yet. Thus we can assume that preemption
1772 * is still disabled.
1774 RB_WARN_ON(buffer
, !local_read(&cpu_buffer
->committing
));
1776 if (!rb_try_to_discard(cpu_buffer
, event
))
1780 * The commit is still visible by the reader, so we
1781 * must increment entries.
1783 local_inc(&cpu_buffer
->entries
);
1785 rb_end_commit(cpu_buffer
);
1787 trace_recursive_unlock();
1790 * Only the last preempt count needs to restore preemption.
1792 if (preempt_count() == 1)
1793 ftrace_preempt_enable(per_cpu(rb_need_resched
, cpu
));
1795 preempt_enable_no_resched_notrace();
1798 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit
);
1801 * ring_buffer_write - write data to the buffer without reserving
1802 * @buffer: The ring buffer to write to.
1803 * @length: The length of the data being written (excluding the event header)
1804 * @data: The data to write to the buffer.
1806 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
1807 * one function. If you already have the data to write to the buffer, it
1808 * may be easier to simply call this function.
1810 * Note, like ring_buffer_lock_reserve, the length is the length of the data
1811 * and not the length of the event which would hold the header.
1813 int ring_buffer_write(struct ring_buffer
*buffer
,
1814 unsigned long length
,
1817 struct ring_buffer_per_cpu
*cpu_buffer
;
1818 struct ring_buffer_event
*event
;
1823 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
1826 if (atomic_read(&buffer
->record_disabled
))
1829 resched
= ftrace_preempt_disable();
1831 cpu
= raw_smp_processor_id();
1833 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1836 cpu_buffer
= buffer
->buffers
[cpu
];
1838 if (atomic_read(&cpu_buffer
->record_disabled
))
1841 if (length
> BUF_MAX_DATA_SIZE
)
1844 event
= rb_reserve_next_event(cpu_buffer
, length
);
1848 body
= rb_event_data(event
);
1850 memcpy(body
, data
, length
);
1852 rb_commit(cpu_buffer
, event
);
1856 ftrace_preempt_enable(resched
);
1860 EXPORT_SYMBOL_GPL(ring_buffer_write
);
1862 static int rb_per_cpu_empty(struct ring_buffer_per_cpu
*cpu_buffer
)
1864 struct buffer_page
*reader
= cpu_buffer
->reader_page
;
1865 struct buffer_page
*head
= cpu_buffer
->head_page
;
1866 struct buffer_page
*commit
= cpu_buffer
->commit_page
;
1868 return reader
->read
== rb_page_commit(reader
) &&
1869 (commit
== reader
||
1871 head
->read
== rb_page_commit(commit
)));
1875 * ring_buffer_record_disable - stop all writes into the buffer
1876 * @buffer: The ring buffer to stop writes to.
1878 * This prevents all writes to the buffer. Any attempt to write
1879 * to the buffer after this will fail and return NULL.
1881 * The caller should call synchronize_sched() after this.
1883 void ring_buffer_record_disable(struct ring_buffer
*buffer
)
1885 atomic_inc(&buffer
->record_disabled
);
1887 EXPORT_SYMBOL_GPL(ring_buffer_record_disable
);
1890 * ring_buffer_record_enable - enable writes to the buffer
1891 * @buffer: The ring buffer to enable writes
1893 * Note, multiple disables will need the same number of enables
1894 * to truely enable the writing (much like preempt_disable).
1896 void ring_buffer_record_enable(struct ring_buffer
*buffer
)
1898 atomic_dec(&buffer
->record_disabled
);
1900 EXPORT_SYMBOL_GPL(ring_buffer_record_enable
);
1903 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
1904 * @buffer: The ring buffer to stop writes to.
1905 * @cpu: The CPU buffer to stop
1907 * This prevents all writes to the buffer. Any attempt to write
1908 * to the buffer after this will fail and return NULL.
1910 * The caller should call synchronize_sched() after this.
1912 void ring_buffer_record_disable_cpu(struct ring_buffer
*buffer
, int cpu
)
1914 struct ring_buffer_per_cpu
*cpu_buffer
;
1916 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1919 cpu_buffer
= buffer
->buffers
[cpu
];
1920 atomic_inc(&cpu_buffer
->record_disabled
);
1922 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu
);
1925 * ring_buffer_record_enable_cpu - enable writes to the buffer
1926 * @buffer: The ring buffer to enable writes
1927 * @cpu: The CPU to enable.
1929 * Note, multiple disables will need the same number of enables
1930 * to truely enable the writing (much like preempt_disable).
1932 void ring_buffer_record_enable_cpu(struct ring_buffer
*buffer
, int cpu
)
1934 struct ring_buffer_per_cpu
*cpu_buffer
;
1936 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1939 cpu_buffer
= buffer
->buffers
[cpu
];
1940 atomic_dec(&cpu_buffer
->record_disabled
);
1942 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu
);
1945 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
1946 * @buffer: The ring buffer
1947 * @cpu: The per CPU buffer to get the entries from.
1949 unsigned long ring_buffer_entries_cpu(struct ring_buffer
*buffer
, int cpu
)
1951 struct ring_buffer_per_cpu
*cpu_buffer
;
1954 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1957 cpu_buffer
= buffer
->buffers
[cpu
];
1958 ret
= (local_read(&cpu_buffer
->entries
) - cpu_buffer
->overrun
)
1963 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu
);
1966 * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
1967 * @buffer: The ring buffer
1968 * @cpu: The per CPU buffer to get the number of overruns from
1970 unsigned long ring_buffer_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
1972 struct ring_buffer_per_cpu
*cpu_buffer
;
1975 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1978 cpu_buffer
= buffer
->buffers
[cpu
];
1979 ret
= cpu_buffer
->overrun
;
1983 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu
);
1986 * ring_buffer_nmi_dropped_cpu - get the number of nmis that were dropped
1987 * @buffer: The ring buffer
1988 * @cpu: The per CPU buffer to get the number of overruns from
1990 unsigned long ring_buffer_nmi_dropped_cpu(struct ring_buffer
*buffer
, int cpu
)
1992 struct ring_buffer_per_cpu
*cpu_buffer
;
1995 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1998 cpu_buffer
= buffer
->buffers
[cpu
];
1999 ret
= cpu_buffer
->nmi_dropped
;
2003 EXPORT_SYMBOL_GPL(ring_buffer_nmi_dropped_cpu
);
2006 * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits
2007 * @buffer: The ring buffer
2008 * @cpu: The per CPU buffer to get the number of overruns from
2011 ring_buffer_commit_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
2013 struct ring_buffer_per_cpu
*cpu_buffer
;
2016 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2019 cpu_buffer
= buffer
->buffers
[cpu
];
2020 ret
= cpu_buffer
->commit_overrun
;
2024 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu
);
2027 * ring_buffer_entries - get the number of entries in a buffer
2028 * @buffer: The ring buffer
2030 * Returns the total number of entries in the ring buffer
2033 unsigned long ring_buffer_entries(struct ring_buffer
*buffer
)
2035 struct ring_buffer_per_cpu
*cpu_buffer
;
2036 unsigned long entries
= 0;
2039 /* if you care about this being correct, lock the buffer */
2040 for_each_buffer_cpu(buffer
, cpu
) {
2041 cpu_buffer
= buffer
->buffers
[cpu
];
2042 entries
+= (local_read(&cpu_buffer
->entries
) -
2043 cpu_buffer
->overrun
) - cpu_buffer
->read
;
2048 EXPORT_SYMBOL_GPL(ring_buffer_entries
);
2051 * ring_buffer_overrun_cpu - get the number of overruns in buffer
2052 * @buffer: The ring buffer
2054 * Returns the total number of overruns in the ring buffer
2057 unsigned long ring_buffer_overruns(struct ring_buffer
*buffer
)
2059 struct ring_buffer_per_cpu
*cpu_buffer
;
2060 unsigned long overruns
= 0;
2063 /* if you care about this being correct, lock the buffer */
2064 for_each_buffer_cpu(buffer
, cpu
) {
2065 cpu_buffer
= buffer
->buffers
[cpu
];
2066 overruns
+= cpu_buffer
->overrun
;
2071 EXPORT_SYMBOL_GPL(ring_buffer_overruns
);
2073 static void rb_iter_reset(struct ring_buffer_iter
*iter
)
2075 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2077 /* Iterator usage is expected to have record disabled */
2078 if (list_empty(&cpu_buffer
->reader_page
->list
)) {
2079 iter
->head_page
= cpu_buffer
->head_page
;
2080 iter
->head
= cpu_buffer
->head_page
->read
;
2082 iter
->head_page
= cpu_buffer
->reader_page
;
2083 iter
->head
= cpu_buffer
->reader_page
->read
;
2086 iter
->read_stamp
= cpu_buffer
->read_stamp
;
2088 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
2092 * ring_buffer_iter_reset - reset an iterator
2093 * @iter: The iterator to reset
2095 * Resets the iterator, so that it will start from the beginning
2098 void ring_buffer_iter_reset(struct ring_buffer_iter
*iter
)
2100 struct ring_buffer_per_cpu
*cpu_buffer
;
2101 unsigned long flags
;
2106 cpu_buffer
= iter
->cpu_buffer
;
2108 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2109 rb_iter_reset(iter
);
2110 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2112 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset
);
2115 * ring_buffer_iter_empty - check if an iterator has no more to read
2116 * @iter: The iterator to check
2118 int ring_buffer_iter_empty(struct ring_buffer_iter
*iter
)
2120 struct ring_buffer_per_cpu
*cpu_buffer
;
2122 cpu_buffer
= iter
->cpu_buffer
;
2124 return iter
->head_page
== cpu_buffer
->commit_page
&&
2125 iter
->head
== rb_commit_index(cpu_buffer
);
2127 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty
);
2130 rb_update_read_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
2131 struct ring_buffer_event
*event
)
2135 switch (event
->type_len
) {
2136 case RINGBUF_TYPE_PADDING
:
2139 case RINGBUF_TYPE_TIME_EXTEND
:
2140 delta
= event
->array
[0];
2142 delta
+= event
->time_delta
;
2143 cpu_buffer
->read_stamp
+= delta
;
2146 case RINGBUF_TYPE_TIME_STAMP
:
2147 /* FIXME: not implemented */
2150 case RINGBUF_TYPE_DATA
:
2151 cpu_buffer
->read_stamp
+= event
->time_delta
;
2161 rb_update_iter_read_stamp(struct ring_buffer_iter
*iter
,
2162 struct ring_buffer_event
*event
)
2166 switch (event
->type_len
) {
2167 case RINGBUF_TYPE_PADDING
:
2170 case RINGBUF_TYPE_TIME_EXTEND
:
2171 delta
= event
->array
[0];
2173 delta
+= event
->time_delta
;
2174 iter
->read_stamp
+= delta
;
2177 case RINGBUF_TYPE_TIME_STAMP
:
2178 /* FIXME: not implemented */
2181 case RINGBUF_TYPE_DATA
:
2182 iter
->read_stamp
+= event
->time_delta
;
2191 static struct buffer_page
*
2192 rb_get_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
2194 struct buffer_page
*reader
= NULL
;
2195 unsigned long flags
;
2198 local_irq_save(flags
);
2199 __raw_spin_lock(&cpu_buffer
->lock
);
2203 * This should normally only loop twice. But because the
2204 * start of the reader inserts an empty page, it causes
2205 * a case where we will loop three times. There should be no
2206 * reason to loop four times (that I know of).
2208 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 3)) {
2213 reader
= cpu_buffer
->reader_page
;
2215 /* If there's more to read, return this page */
2216 if (cpu_buffer
->reader_page
->read
< rb_page_size(reader
))
2219 /* Never should we have an index greater than the size */
2220 if (RB_WARN_ON(cpu_buffer
,
2221 cpu_buffer
->reader_page
->read
> rb_page_size(reader
)))
2224 /* check if we caught up to the tail */
2226 if (cpu_buffer
->commit_page
== cpu_buffer
->reader_page
)
2230 * Splice the empty reader page into the list around the head.
2231 * Reset the reader page to size zero.
2234 reader
= cpu_buffer
->head_page
;
2235 cpu_buffer
->reader_page
->list
.next
= reader
->list
.next
;
2236 cpu_buffer
->reader_page
->list
.prev
= reader
->list
.prev
;
2238 local_set(&cpu_buffer
->reader_page
->write
, 0);
2239 local_set(&cpu_buffer
->reader_page
->entries
, 0);
2240 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
2242 /* Make the reader page now replace the head */
2243 reader
->list
.prev
->next
= &cpu_buffer
->reader_page
->list
;
2244 reader
->list
.next
->prev
= &cpu_buffer
->reader_page
->list
;
2247 * If the tail is on the reader, then we must set the head
2248 * to the inserted page, otherwise we set it one before.
2250 cpu_buffer
->head_page
= cpu_buffer
->reader_page
;
2252 if (cpu_buffer
->commit_page
!= reader
)
2253 rb_inc_page(cpu_buffer
, &cpu_buffer
->head_page
);
2255 /* Finally update the reader page to the new head */
2256 cpu_buffer
->reader_page
= reader
;
2257 rb_reset_reader_page(cpu_buffer
);
2262 __raw_spin_unlock(&cpu_buffer
->lock
);
2263 local_irq_restore(flags
);
2268 static void rb_advance_reader(struct ring_buffer_per_cpu
*cpu_buffer
)
2270 struct ring_buffer_event
*event
;
2271 struct buffer_page
*reader
;
2274 reader
= rb_get_reader_page(cpu_buffer
);
2276 /* This function should not be called when buffer is empty */
2277 if (RB_WARN_ON(cpu_buffer
, !reader
))
2280 event
= rb_reader_event(cpu_buffer
);
2282 if (event
->type_len
<= RINGBUF_TYPE_DATA_TYPE_LEN_MAX
2283 || rb_discarded_event(event
))
2286 rb_update_read_stamp(cpu_buffer
, event
);
2288 length
= rb_event_length(event
);
2289 cpu_buffer
->reader_page
->read
+= length
;
2292 static void rb_advance_iter(struct ring_buffer_iter
*iter
)
2294 struct ring_buffer
*buffer
;
2295 struct ring_buffer_per_cpu
*cpu_buffer
;
2296 struct ring_buffer_event
*event
;
2299 cpu_buffer
= iter
->cpu_buffer
;
2300 buffer
= cpu_buffer
->buffer
;
2303 * Check if we are at the end of the buffer.
2305 if (iter
->head
>= rb_page_size(iter
->head_page
)) {
2306 /* discarded commits can make the page empty */
2307 if (iter
->head_page
== cpu_buffer
->commit_page
)
2313 event
= rb_iter_head_event(iter
);
2315 length
= rb_event_length(event
);
2318 * This should not be called to advance the header if we are
2319 * at the tail of the buffer.
2321 if (RB_WARN_ON(cpu_buffer
,
2322 (iter
->head_page
== cpu_buffer
->commit_page
) &&
2323 (iter
->head
+ length
> rb_commit_index(cpu_buffer
))))
2326 rb_update_iter_read_stamp(iter
, event
);
2328 iter
->head
+= length
;
2330 /* check for end of page padding */
2331 if ((iter
->head
>= rb_page_size(iter
->head_page
)) &&
2332 (iter
->head_page
!= cpu_buffer
->commit_page
))
2333 rb_advance_iter(iter
);
2336 static struct ring_buffer_event
*
2337 rb_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2339 struct ring_buffer_per_cpu
*cpu_buffer
;
2340 struct ring_buffer_event
*event
;
2341 struct buffer_page
*reader
;
2344 cpu_buffer
= buffer
->buffers
[cpu
];
2348 * We repeat when a timestamp is encountered. It is possible
2349 * to get multiple timestamps from an interrupt entering just
2350 * as one timestamp is about to be written, or from discarded
2351 * commits. The most that we can have is the number on a single page.
2353 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> RB_TIMESTAMPS_PER_PAGE
))
2356 reader
= rb_get_reader_page(cpu_buffer
);
2360 event
= rb_reader_event(cpu_buffer
);
2362 switch (event
->type_len
) {
2363 case RINGBUF_TYPE_PADDING
:
2364 if (rb_null_event(event
))
2365 RB_WARN_ON(cpu_buffer
, 1);
2367 * Because the writer could be discarding every
2368 * event it creates (which would probably be bad)
2369 * if we were to go back to "again" then we may never
2370 * catch up, and will trigger the warn on, or lock
2371 * the box. Return the padding, and we will release
2372 * the current locks, and try again.
2374 rb_advance_reader(cpu_buffer
);
2377 case RINGBUF_TYPE_TIME_EXTEND
:
2378 /* Internal data, OK to advance */
2379 rb_advance_reader(cpu_buffer
);
2382 case RINGBUF_TYPE_TIME_STAMP
:
2383 /* FIXME: not implemented */
2384 rb_advance_reader(cpu_buffer
);
2387 case RINGBUF_TYPE_DATA
:
2389 *ts
= cpu_buffer
->read_stamp
+ event
->time_delta
;
2390 ring_buffer_normalize_time_stamp(buffer
,
2391 cpu_buffer
->cpu
, ts
);
2401 EXPORT_SYMBOL_GPL(ring_buffer_peek
);
2403 static struct ring_buffer_event
*
2404 rb_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
2406 struct ring_buffer
*buffer
;
2407 struct ring_buffer_per_cpu
*cpu_buffer
;
2408 struct ring_buffer_event
*event
;
2411 if (ring_buffer_iter_empty(iter
))
2414 cpu_buffer
= iter
->cpu_buffer
;
2415 buffer
= cpu_buffer
->buffer
;
2419 * We repeat when a timestamp is encountered.
2420 * We can get multiple timestamps by nested interrupts or also
2421 * if filtering is on (discarding commits). Since discarding
2422 * commits can be frequent we can get a lot of timestamps.
2423 * But we limit them by not adding timestamps if they begin
2424 * at the start of a page.
2426 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> RB_TIMESTAMPS_PER_PAGE
))
2429 if (rb_per_cpu_empty(cpu_buffer
))
2432 event
= rb_iter_head_event(iter
);
2434 switch (event
->type_len
) {
2435 case RINGBUF_TYPE_PADDING
:
2436 if (rb_null_event(event
)) {
2440 rb_advance_iter(iter
);
2443 case RINGBUF_TYPE_TIME_EXTEND
:
2444 /* Internal data, OK to advance */
2445 rb_advance_iter(iter
);
2448 case RINGBUF_TYPE_TIME_STAMP
:
2449 /* FIXME: not implemented */
2450 rb_advance_iter(iter
);
2453 case RINGBUF_TYPE_DATA
:
2455 *ts
= iter
->read_stamp
+ event
->time_delta
;
2456 ring_buffer_normalize_time_stamp(buffer
,
2457 cpu_buffer
->cpu
, ts
);
2467 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek
);
2470 * ring_buffer_peek - peek at the next event to be read
2471 * @buffer: The ring buffer to read
2472 * @cpu: The cpu to peak at
2473 * @ts: The timestamp counter of this event.
2475 * This will return the event that will be read next, but does
2476 * not consume the data.
2478 struct ring_buffer_event
*
2479 ring_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2481 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2482 struct ring_buffer_event
*event
;
2483 unsigned long flags
;
2485 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2489 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2490 event
= rb_buffer_peek(buffer
, cpu
, ts
);
2491 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2493 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2502 * ring_buffer_iter_peek - peek at the next event to be read
2503 * @iter: The ring buffer iterator
2504 * @ts: The timestamp counter of this event.
2506 * This will return the event that will be read next, but does
2507 * not increment the iterator.
2509 struct ring_buffer_event
*
2510 ring_buffer_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
2512 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2513 struct ring_buffer_event
*event
;
2514 unsigned long flags
;
2517 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2518 event
= rb_iter_peek(iter
, ts
);
2519 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2521 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2530 * ring_buffer_consume - return an event and consume it
2531 * @buffer: The ring buffer to get the next event from
2533 * Returns the next event in the ring buffer, and that event is consumed.
2534 * Meaning, that sequential reads will keep returning a different event,
2535 * and eventually empty the ring buffer if the producer is slower.
2537 struct ring_buffer_event
*
2538 ring_buffer_consume(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2540 struct ring_buffer_per_cpu
*cpu_buffer
;
2541 struct ring_buffer_event
*event
= NULL
;
2542 unsigned long flags
;
2545 /* might be called in atomic */
2548 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2551 cpu_buffer
= buffer
->buffers
[cpu
];
2552 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2554 event
= rb_buffer_peek(buffer
, cpu
, ts
);
2558 rb_advance_reader(cpu_buffer
);
2561 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2566 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2573 EXPORT_SYMBOL_GPL(ring_buffer_consume
);
2576 * ring_buffer_read_start - start a non consuming read of the buffer
2577 * @buffer: The ring buffer to read from
2578 * @cpu: The cpu buffer to iterate over
2580 * This starts up an iteration through the buffer. It also disables
2581 * the recording to the buffer until the reading is finished.
2582 * This prevents the reading from being corrupted. This is not
2583 * a consuming read, so a producer is not expected.
2585 * Must be paired with ring_buffer_finish.
2587 struct ring_buffer_iter
*
2588 ring_buffer_read_start(struct ring_buffer
*buffer
, int cpu
)
2590 struct ring_buffer_per_cpu
*cpu_buffer
;
2591 struct ring_buffer_iter
*iter
;
2592 unsigned long flags
;
2594 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2597 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
2601 cpu_buffer
= buffer
->buffers
[cpu
];
2603 iter
->cpu_buffer
= cpu_buffer
;
2605 atomic_inc(&cpu_buffer
->record_disabled
);
2606 synchronize_sched();
2608 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2609 __raw_spin_lock(&cpu_buffer
->lock
);
2610 rb_iter_reset(iter
);
2611 __raw_spin_unlock(&cpu_buffer
->lock
);
2612 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2616 EXPORT_SYMBOL_GPL(ring_buffer_read_start
);
2619 * ring_buffer_finish - finish reading the iterator of the buffer
2620 * @iter: The iterator retrieved by ring_buffer_start
2622 * This re-enables the recording to the buffer, and frees the
2626 ring_buffer_read_finish(struct ring_buffer_iter
*iter
)
2628 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2630 atomic_dec(&cpu_buffer
->record_disabled
);
2633 EXPORT_SYMBOL_GPL(ring_buffer_read_finish
);
2636 * ring_buffer_read - read the next item in the ring buffer by the iterator
2637 * @iter: The ring buffer iterator
2638 * @ts: The time stamp of the event read.
2640 * This reads the next event in the ring buffer and increments the iterator.
2642 struct ring_buffer_event
*
2643 ring_buffer_read(struct ring_buffer_iter
*iter
, u64
*ts
)
2645 struct ring_buffer_event
*event
;
2646 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2647 unsigned long flags
;
2650 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2651 event
= rb_iter_peek(iter
, ts
);
2655 rb_advance_iter(iter
);
2657 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2659 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2666 EXPORT_SYMBOL_GPL(ring_buffer_read
);
2669 * ring_buffer_size - return the size of the ring buffer (in bytes)
2670 * @buffer: The ring buffer.
2672 unsigned long ring_buffer_size(struct ring_buffer
*buffer
)
2674 return BUF_PAGE_SIZE
* buffer
->pages
;
2676 EXPORT_SYMBOL_GPL(ring_buffer_size
);
2679 rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
)
2681 cpu_buffer
->head_page
2682 = list_entry(cpu_buffer
->pages
.next
, struct buffer_page
, list
);
2683 local_set(&cpu_buffer
->head_page
->write
, 0);
2684 local_set(&cpu_buffer
->head_page
->entries
, 0);
2685 local_set(&cpu_buffer
->head_page
->page
->commit
, 0);
2687 cpu_buffer
->head_page
->read
= 0;
2689 cpu_buffer
->tail_page
= cpu_buffer
->head_page
;
2690 cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
2692 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
2693 local_set(&cpu_buffer
->reader_page
->write
, 0);
2694 local_set(&cpu_buffer
->reader_page
->entries
, 0);
2695 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
2696 cpu_buffer
->reader_page
->read
= 0;
2698 cpu_buffer
->nmi_dropped
= 0;
2699 cpu_buffer
->commit_overrun
= 0;
2700 cpu_buffer
->overrun
= 0;
2701 cpu_buffer
->read
= 0;
2702 local_set(&cpu_buffer
->entries
, 0);
2703 local_set(&cpu_buffer
->committing
, 0);
2704 local_set(&cpu_buffer
->commits
, 0);
2706 cpu_buffer
->write_stamp
= 0;
2707 cpu_buffer
->read_stamp
= 0;
2711 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
2712 * @buffer: The ring buffer to reset a per cpu buffer of
2713 * @cpu: The CPU buffer to be reset
2715 void ring_buffer_reset_cpu(struct ring_buffer
*buffer
, int cpu
)
2717 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2718 unsigned long flags
;
2720 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2723 atomic_inc(&cpu_buffer
->record_disabled
);
2725 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2727 __raw_spin_lock(&cpu_buffer
->lock
);
2729 rb_reset_cpu(cpu_buffer
);
2731 __raw_spin_unlock(&cpu_buffer
->lock
);
2733 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2735 atomic_dec(&cpu_buffer
->record_disabled
);
2737 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu
);
2740 * ring_buffer_reset - reset a ring buffer
2741 * @buffer: The ring buffer to reset all cpu buffers
2743 void ring_buffer_reset(struct ring_buffer
*buffer
)
2747 for_each_buffer_cpu(buffer
, cpu
)
2748 ring_buffer_reset_cpu(buffer
, cpu
);
2750 EXPORT_SYMBOL_GPL(ring_buffer_reset
);
2753 * rind_buffer_empty - is the ring buffer empty?
2754 * @buffer: The ring buffer to test
2756 int ring_buffer_empty(struct ring_buffer
*buffer
)
2758 struct ring_buffer_per_cpu
*cpu_buffer
;
2759 unsigned long flags
;
2763 /* yes this is racy, but if you don't like the race, lock the buffer */
2764 for_each_buffer_cpu(buffer
, cpu
) {
2765 cpu_buffer
= buffer
->buffers
[cpu
];
2766 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2767 ret
= rb_per_cpu_empty(cpu_buffer
);
2768 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2775 EXPORT_SYMBOL_GPL(ring_buffer_empty
);
2778 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
2779 * @buffer: The ring buffer
2780 * @cpu: The CPU buffer to test
2782 int ring_buffer_empty_cpu(struct ring_buffer
*buffer
, int cpu
)
2784 struct ring_buffer_per_cpu
*cpu_buffer
;
2785 unsigned long flags
;
2788 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2791 cpu_buffer
= buffer
->buffers
[cpu
];
2792 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2793 ret
= rb_per_cpu_empty(cpu_buffer
);
2794 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2798 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu
);
2801 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
2802 * @buffer_a: One buffer to swap with
2803 * @buffer_b: The other buffer to swap with
2805 * This function is useful for tracers that want to take a "snapshot"
2806 * of a CPU buffer and has another back up buffer lying around.
2807 * it is expected that the tracer handles the cpu buffer not being
2808 * used at the moment.
2810 int ring_buffer_swap_cpu(struct ring_buffer
*buffer_a
,
2811 struct ring_buffer
*buffer_b
, int cpu
)
2813 struct ring_buffer_per_cpu
*cpu_buffer_a
;
2814 struct ring_buffer_per_cpu
*cpu_buffer_b
;
2817 if (!cpumask_test_cpu(cpu
, buffer_a
->cpumask
) ||
2818 !cpumask_test_cpu(cpu
, buffer_b
->cpumask
))
2821 /* At least make sure the two buffers are somewhat the same */
2822 if (buffer_a
->pages
!= buffer_b
->pages
)
2827 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
2830 if (atomic_read(&buffer_a
->record_disabled
))
2833 if (atomic_read(&buffer_b
->record_disabled
))
2836 cpu_buffer_a
= buffer_a
->buffers
[cpu
];
2837 cpu_buffer_b
= buffer_b
->buffers
[cpu
];
2839 if (atomic_read(&cpu_buffer_a
->record_disabled
))
2842 if (atomic_read(&cpu_buffer_b
->record_disabled
))
2846 * We can't do a synchronize_sched here because this
2847 * function can be called in atomic context.
2848 * Normally this will be called from the same CPU as cpu.
2849 * If not it's up to the caller to protect this.
2851 atomic_inc(&cpu_buffer_a
->record_disabled
);
2852 atomic_inc(&cpu_buffer_b
->record_disabled
);
2854 buffer_a
->buffers
[cpu
] = cpu_buffer_b
;
2855 buffer_b
->buffers
[cpu
] = cpu_buffer_a
;
2857 cpu_buffer_b
->buffer
= buffer_a
;
2858 cpu_buffer_a
->buffer
= buffer_b
;
2860 atomic_dec(&cpu_buffer_a
->record_disabled
);
2861 atomic_dec(&cpu_buffer_b
->record_disabled
);
2867 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu
);
2870 * ring_buffer_alloc_read_page - allocate a page to read from buffer
2871 * @buffer: the buffer to allocate for.
2873 * This function is used in conjunction with ring_buffer_read_page.
2874 * When reading a full page from the ring buffer, these functions
2875 * can be used to speed up the process. The calling function should
2876 * allocate a few pages first with this function. Then when it
2877 * needs to get pages from the ring buffer, it passes the result
2878 * of this function into ring_buffer_read_page, which will swap
2879 * the page that was allocated, with the read page of the buffer.
2882 * The page allocated, or NULL on error.
2884 void *ring_buffer_alloc_read_page(struct ring_buffer
*buffer
)
2886 struct buffer_data_page
*bpage
;
2889 addr
= __get_free_page(GFP_KERNEL
);
2893 bpage
= (void *)addr
;
2895 rb_init_page(bpage
);
2899 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page
);
2902 * ring_buffer_free_read_page - free an allocated read page
2903 * @buffer: the buffer the page was allocate for
2904 * @data: the page to free
2906 * Free a page allocated from ring_buffer_alloc_read_page.
2908 void ring_buffer_free_read_page(struct ring_buffer
*buffer
, void *data
)
2910 free_page((unsigned long)data
);
2912 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page
);
2915 * ring_buffer_read_page - extract a page from the ring buffer
2916 * @buffer: buffer to extract from
2917 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
2918 * @len: amount to extract
2919 * @cpu: the cpu of the buffer to extract
2920 * @full: should the extraction only happen when the page is full.
2922 * This function will pull out a page from the ring buffer and consume it.
2923 * @data_page must be the address of the variable that was returned
2924 * from ring_buffer_alloc_read_page. This is because the page might be used
2925 * to swap with a page in the ring buffer.
2928 * rpage = ring_buffer_alloc_read_page(buffer);
2931 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
2933 * process_page(rpage, ret);
2935 * When @full is set, the function will not return true unless
2936 * the writer is off the reader page.
2938 * Note: it is up to the calling functions to handle sleeps and wakeups.
2939 * The ring buffer can be used anywhere in the kernel and can not
2940 * blindly call wake_up. The layer that uses the ring buffer must be
2941 * responsible for that.
2944 * >=0 if data has been transferred, returns the offset of consumed data.
2945 * <0 if no data has been transferred.
2947 int ring_buffer_read_page(struct ring_buffer
*buffer
,
2948 void **data_page
, size_t len
, int cpu
, int full
)
2950 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2951 struct ring_buffer_event
*event
;
2952 struct buffer_data_page
*bpage
;
2953 struct buffer_page
*reader
;
2954 unsigned long flags
;
2955 unsigned int commit
;
2960 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2964 * If len is not big enough to hold the page header, then
2965 * we can not copy anything.
2967 if (len
<= BUF_PAGE_HDR_SIZE
)
2970 len
-= BUF_PAGE_HDR_SIZE
;
2979 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2981 reader
= rb_get_reader_page(cpu_buffer
);
2985 event
= rb_reader_event(cpu_buffer
);
2987 read
= reader
->read
;
2988 commit
= rb_page_commit(reader
);
2991 * If this page has been partially read or
2992 * if len is not big enough to read the rest of the page or
2993 * a writer is still on the page, then
2994 * we must copy the data from the page to the buffer.
2995 * Otherwise, we can simply swap the page with the one passed in.
2997 if (read
|| (len
< (commit
- read
)) ||
2998 cpu_buffer
->reader_page
== cpu_buffer
->commit_page
) {
2999 struct buffer_data_page
*rpage
= cpu_buffer
->reader_page
->page
;
3000 unsigned int rpos
= read
;
3001 unsigned int pos
= 0;
3007 if (len
> (commit
- read
))
3008 len
= (commit
- read
);
3010 size
= rb_event_length(event
);
3015 /* save the current timestamp, since the user will need it */
3016 save_timestamp
= cpu_buffer
->read_stamp
;
3018 /* Need to copy one event at a time */
3020 memcpy(bpage
->data
+ pos
, rpage
->data
+ rpos
, size
);
3024 rb_advance_reader(cpu_buffer
);
3025 rpos
= reader
->read
;
3028 event
= rb_reader_event(cpu_buffer
);
3029 size
= rb_event_length(event
);
3030 } while (len
> size
);
3033 local_set(&bpage
->commit
, pos
);
3034 bpage
->time_stamp
= save_timestamp
;
3036 /* we copied everything to the beginning */
3039 /* update the entry counter */
3040 cpu_buffer
->read
+= local_read(&reader
->entries
);
3042 /* swap the pages */
3043 rb_init_page(bpage
);
3044 bpage
= reader
->page
;
3045 reader
->page
= *data_page
;
3046 local_set(&reader
->write
, 0);
3047 local_set(&reader
->entries
, 0);
3054 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
3059 EXPORT_SYMBOL_GPL(ring_buffer_read_page
);
3062 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
3063 size_t cnt
, loff_t
*ppos
)
3065 unsigned long *p
= filp
->private_data
;
3069 if (test_bit(RB_BUFFERS_DISABLED_BIT
, p
))
3070 r
= sprintf(buf
, "permanently disabled\n");
3072 r
= sprintf(buf
, "%d\n", test_bit(RB_BUFFERS_ON_BIT
, p
));
3074 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3078 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
3079 size_t cnt
, loff_t
*ppos
)
3081 unsigned long *p
= filp
->private_data
;
3086 if (cnt
>= sizeof(buf
))
3089 if (copy_from_user(&buf
, ubuf
, cnt
))
3094 ret
= strict_strtoul(buf
, 10, &val
);
3099 set_bit(RB_BUFFERS_ON_BIT
, p
);
3101 clear_bit(RB_BUFFERS_ON_BIT
, p
);
3108 static const struct file_operations rb_simple_fops
= {
3109 .open
= tracing_open_generic
,
3110 .read
= rb_simple_read
,
3111 .write
= rb_simple_write
,
3115 static __init
int rb_init_debugfs(void)
3117 struct dentry
*d_tracer
;
3119 d_tracer
= tracing_init_dentry();
3121 trace_create_file("tracing_on", 0644, d_tracer
,
3122 &ring_buffer_flags
, &rb_simple_fops
);
3127 fs_initcall(rb_init_debugfs
);
3129 #ifdef CONFIG_HOTPLUG_CPU
3130 static int rb_cpu_notify(struct notifier_block
*self
,
3131 unsigned long action
, void *hcpu
)
3133 struct ring_buffer
*buffer
=
3134 container_of(self
, struct ring_buffer
, cpu_notify
);
3135 long cpu
= (long)hcpu
;
3138 case CPU_UP_PREPARE
:
3139 case CPU_UP_PREPARE_FROZEN
:
3140 if (cpumask_test_cpu(cpu
, buffer
->cpumask
))
3143 buffer
->buffers
[cpu
] =
3144 rb_allocate_cpu_buffer(buffer
, cpu
);
3145 if (!buffer
->buffers
[cpu
]) {
3146 WARN(1, "failed to allocate ring buffer on CPU %ld\n",
3151 cpumask_set_cpu(cpu
, buffer
->cpumask
);
3153 case CPU_DOWN_PREPARE
:
3154 case CPU_DOWN_PREPARE_FROZEN
:
3157 * If we were to free the buffer, then the user would
3158 * lose any trace that was in the buffer.