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 struct list_head list
; /* list of buffer pages */
325 local_t write
; /* index for next write */
326 unsigned read
; /* index for next read */
327 local_t entries
; /* entries on this page */
328 struct buffer_data_page
*page
; /* Actual data page */
331 static void rb_init_page(struct buffer_data_page
*bpage
)
333 local_set(&bpage
->commit
, 0);
337 * ring_buffer_page_len - the size of data on the page.
338 * @page: The page to read
340 * Returns the amount of data on the page, including buffer page header.
342 size_t ring_buffer_page_len(void *page
)
344 return local_read(&((struct buffer_data_page
*)page
)->commit
)
349 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
352 static void free_buffer_page(struct buffer_page
*bpage
)
354 free_page((unsigned long)bpage
->page
);
359 * We need to fit the time_stamp delta into 27 bits.
361 static inline int test_time_stamp(u64 delta
)
363 if (delta
& TS_DELTA_TEST
)
368 #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
370 /* Max payload is BUF_PAGE_SIZE - header (8bytes) */
371 #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
373 int ring_buffer_print_page_header(struct trace_seq
*s
)
375 struct buffer_data_page field
;
378 ret
= trace_seq_printf(s
, "\tfield: u64 timestamp;\t"
379 "offset:0;\tsize:%u;\n",
380 (unsigned int)sizeof(field
.time_stamp
));
382 ret
= trace_seq_printf(s
, "\tfield: local_t commit;\t"
383 "offset:%u;\tsize:%u;\n",
384 (unsigned int)offsetof(typeof(field
), commit
),
385 (unsigned int)sizeof(field
.commit
));
387 ret
= trace_seq_printf(s
, "\tfield: char data;\t"
388 "offset:%u;\tsize:%u;\n",
389 (unsigned int)offsetof(typeof(field
), data
),
390 (unsigned int)BUF_PAGE_SIZE
);
396 * head_page == tail_page && head == tail then buffer is empty.
398 struct ring_buffer_per_cpu
{
400 struct ring_buffer
*buffer
;
401 spinlock_t reader_lock
; /* serialize readers */
403 struct lock_class_key lock_key
;
404 struct list_head pages
;
405 struct buffer_page
*head_page
; /* read from head */
406 struct buffer_page
*tail_page
; /* write to tail */
407 struct buffer_page
*commit_page
; /* committed pages */
408 struct buffer_page
*reader_page
;
409 unsigned long nmi_dropped
;
410 unsigned long commit_overrun
;
411 unsigned long overrun
;
416 atomic_t record_disabled
;
423 atomic_t record_disabled
;
424 cpumask_var_t cpumask
;
428 struct ring_buffer_per_cpu
**buffers
;
430 #ifdef CONFIG_HOTPLUG_CPU
431 struct notifier_block cpu_notify
;
436 struct ring_buffer_iter
{
437 struct ring_buffer_per_cpu
*cpu_buffer
;
439 struct buffer_page
*head_page
;
443 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
444 #define RB_WARN_ON(buffer, cond) \
446 int _____ret = unlikely(cond); \
448 atomic_inc(&buffer->record_disabled); \
454 /* Up this if you want to test the TIME_EXTENTS and normalization */
455 #define DEBUG_SHIFT 0
457 static inline u64
rb_time_stamp(struct ring_buffer
*buffer
, int cpu
)
459 /* shift to debug/test normalization and TIME_EXTENTS */
460 return buffer
->clock() << DEBUG_SHIFT
;
463 u64
ring_buffer_time_stamp(struct ring_buffer
*buffer
, int cpu
)
467 preempt_disable_notrace();
468 time
= rb_time_stamp(buffer
, cpu
);
469 preempt_enable_no_resched_notrace();
473 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp
);
475 void ring_buffer_normalize_time_stamp(struct ring_buffer
*buffer
,
478 /* Just stupid testing the normalize function and deltas */
481 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp
);
484 * check_pages - integrity check of buffer pages
485 * @cpu_buffer: CPU buffer with pages to test
487 * As a safety measure we check to make sure the data pages have not
490 static int rb_check_pages(struct ring_buffer_per_cpu
*cpu_buffer
)
492 struct list_head
*head
= &cpu_buffer
->pages
;
493 struct buffer_page
*bpage
, *tmp
;
495 if (RB_WARN_ON(cpu_buffer
, head
->next
->prev
!= head
))
497 if (RB_WARN_ON(cpu_buffer
, head
->prev
->next
!= head
))
500 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
501 if (RB_WARN_ON(cpu_buffer
,
502 bpage
->list
.next
->prev
!= &bpage
->list
))
504 if (RB_WARN_ON(cpu_buffer
,
505 bpage
->list
.prev
->next
!= &bpage
->list
))
512 static int rb_allocate_pages(struct ring_buffer_per_cpu
*cpu_buffer
,
515 struct list_head
*head
= &cpu_buffer
->pages
;
516 struct buffer_page
*bpage
, *tmp
;
521 for (i
= 0; i
< nr_pages
; i
++) {
522 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
523 GFP_KERNEL
, cpu_to_node(cpu_buffer
->cpu
));
526 list_add(&bpage
->list
, &pages
);
528 addr
= __get_free_page(GFP_KERNEL
);
531 bpage
->page
= (void *)addr
;
532 rb_init_page(bpage
->page
);
535 list_splice(&pages
, head
);
537 rb_check_pages(cpu_buffer
);
542 list_for_each_entry_safe(bpage
, tmp
, &pages
, list
) {
543 list_del_init(&bpage
->list
);
544 free_buffer_page(bpage
);
549 static struct ring_buffer_per_cpu
*
550 rb_allocate_cpu_buffer(struct ring_buffer
*buffer
, int cpu
)
552 struct ring_buffer_per_cpu
*cpu_buffer
;
553 struct buffer_page
*bpage
;
557 cpu_buffer
= kzalloc_node(ALIGN(sizeof(*cpu_buffer
), cache_line_size()),
558 GFP_KERNEL
, cpu_to_node(cpu
));
562 cpu_buffer
->cpu
= cpu
;
563 cpu_buffer
->buffer
= buffer
;
564 spin_lock_init(&cpu_buffer
->reader_lock
);
565 cpu_buffer
->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
566 INIT_LIST_HEAD(&cpu_buffer
->pages
);
568 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
569 GFP_KERNEL
, cpu_to_node(cpu
));
571 goto fail_free_buffer
;
573 cpu_buffer
->reader_page
= bpage
;
574 addr
= __get_free_page(GFP_KERNEL
);
576 goto fail_free_reader
;
577 bpage
->page
= (void *)addr
;
578 rb_init_page(bpage
->page
);
580 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
582 ret
= rb_allocate_pages(cpu_buffer
, buffer
->pages
);
584 goto fail_free_reader
;
586 cpu_buffer
->head_page
587 = list_entry(cpu_buffer
->pages
.next
, struct buffer_page
, list
);
588 cpu_buffer
->tail_page
= cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
593 free_buffer_page(cpu_buffer
->reader_page
);
600 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu
*cpu_buffer
)
602 struct list_head
*head
= &cpu_buffer
->pages
;
603 struct buffer_page
*bpage
, *tmp
;
605 free_buffer_page(cpu_buffer
->reader_page
);
607 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
608 list_del_init(&bpage
->list
);
609 free_buffer_page(bpage
);
615 * Causes compile errors if the struct buffer_page gets bigger
616 * than the struct page.
618 extern int ring_buffer_page_too_big(void);
620 #ifdef CONFIG_HOTPLUG_CPU
621 static int rb_cpu_notify(struct notifier_block
*self
,
622 unsigned long action
, void *hcpu
);
626 * ring_buffer_alloc - allocate a new ring_buffer
627 * @size: the size in bytes per cpu that is needed.
628 * @flags: attributes to set for the ring buffer.
630 * Currently the only flag that is available is the RB_FL_OVERWRITE
631 * flag. This flag means that the buffer will overwrite old data
632 * when the buffer wraps. If this flag is not set, the buffer will
633 * drop data when the tail hits the head.
635 struct ring_buffer
*ring_buffer_alloc(unsigned long size
, unsigned flags
)
637 struct ring_buffer
*buffer
;
641 /* Paranoid! Optimizes out when all is well */
642 if (sizeof(struct buffer_page
) > sizeof(struct page
))
643 ring_buffer_page_too_big();
646 /* keep it in its own cache line */
647 buffer
= kzalloc(ALIGN(sizeof(*buffer
), cache_line_size()),
652 if (!alloc_cpumask_var(&buffer
->cpumask
, GFP_KERNEL
))
653 goto fail_free_buffer
;
655 buffer
->pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
656 buffer
->flags
= flags
;
657 buffer
->clock
= trace_clock_local
;
659 /* need at least two pages */
660 if (buffer
->pages
== 1)
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
) - (PAGE_SIZE
- BUF_PAGE_SIZE
);
1009 rb_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_event(struct ring_buffer_per_cpu
*cpu_buffer
,
1024 struct ring_buffer_event
*event
)
1026 unsigned long addr
= (unsigned long)event
;
1027 unsigned long index
;
1029 index
= rb_event_index(event
);
1032 while (cpu_buffer
->commit_page
->page
!= (void *)addr
) {
1033 if (RB_WARN_ON(cpu_buffer
,
1034 cpu_buffer
->commit_page
== cpu_buffer
->tail_page
))
1036 cpu_buffer
->commit_page
->page
->commit
=
1037 cpu_buffer
->commit_page
->write
;
1038 rb_inc_page(cpu_buffer
, &cpu_buffer
->commit_page
);
1039 cpu_buffer
->write_stamp
=
1040 cpu_buffer
->commit_page
->page
->time_stamp
;
1043 /* Now set the commit to the event's index */
1044 local_set(&cpu_buffer
->commit_page
->page
->commit
, index
);
1048 rb_set_commit_to_write(struct ring_buffer_per_cpu
*cpu_buffer
)
1051 * We only race with interrupts and NMIs on this CPU.
1052 * If we own the commit event, then we can commit
1053 * all others that interrupted us, since the interruptions
1054 * are in stack format (they finish before they come
1055 * back to us). This allows us to do a simple loop to
1056 * assign the commit to the tail.
1059 while (cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
) {
1060 cpu_buffer
->commit_page
->page
->commit
=
1061 cpu_buffer
->commit_page
->write
;
1062 rb_inc_page(cpu_buffer
, &cpu_buffer
->commit_page
);
1063 cpu_buffer
->write_stamp
=
1064 cpu_buffer
->commit_page
->page
->time_stamp
;
1065 /* add barrier to keep gcc from optimizing too much */
1068 while (rb_commit_index(cpu_buffer
) !=
1069 rb_page_write(cpu_buffer
->commit_page
)) {
1070 cpu_buffer
->commit_page
->page
->commit
=
1071 cpu_buffer
->commit_page
->write
;
1075 /* again, keep gcc from optimizing */
1079 * If an interrupt came in just after the first while loop
1080 * and pushed the tail page forward, we will be left with
1081 * a dangling commit that will never go forward.
1083 if (unlikely(cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
))
1087 static void rb_reset_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
1089 cpu_buffer
->read_stamp
= cpu_buffer
->reader_page
->page
->time_stamp
;
1090 cpu_buffer
->reader_page
->read
= 0;
1093 static void rb_inc_iter(struct ring_buffer_iter
*iter
)
1095 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
1098 * The iterator could be on the reader page (it starts there).
1099 * But the head could have moved, since the reader was
1100 * found. Check for this case and assign the iterator
1101 * to the head page instead of next.
1103 if (iter
->head_page
== cpu_buffer
->reader_page
)
1104 iter
->head_page
= cpu_buffer
->head_page
;
1106 rb_inc_page(cpu_buffer
, &iter
->head_page
);
1108 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
1113 * ring_buffer_update_event - update event type and data
1114 * @event: the even to update
1115 * @type: the type of event
1116 * @length: the size of the event field in the ring buffer
1118 * Update the type and data fields of the event. The length
1119 * is the actual size that is written to the ring buffer,
1120 * and with this, we can determine what to place into the
1124 rb_update_event(struct ring_buffer_event
*event
,
1125 unsigned type
, unsigned length
)
1127 event
->type_len
= type
;
1131 case RINGBUF_TYPE_PADDING
:
1132 case RINGBUF_TYPE_TIME_EXTEND
:
1133 case RINGBUF_TYPE_TIME_STAMP
:
1137 length
-= RB_EVNT_HDR_SIZE
;
1138 if (length
> RB_MAX_SMALL_DATA
)
1139 event
->array
[0] = length
;
1141 event
->type_len
= DIV_ROUND_UP(length
, RB_ALIGNMENT
);
1148 static unsigned rb_calculate_event_length(unsigned length
)
1150 struct ring_buffer_event event
; /* Used only for sizeof array */
1152 /* zero length can cause confusions */
1156 if (length
> RB_MAX_SMALL_DATA
)
1157 length
+= sizeof(event
.array
[0]);
1159 length
+= RB_EVNT_HDR_SIZE
;
1160 length
= ALIGN(length
, RB_ALIGNMENT
);
1166 static struct ring_buffer_event
*
1167 rb_move_tail(struct ring_buffer_per_cpu
*cpu_buffer
,
1168 unsigned long length
, unsigned long tail
,
1169 struct buffer_page
*commit_page
,
1170 struct buffer_page
*tail_page
, u64
*ts
)
1172 struct buffer_page
*next_page
, *head_page
, *reader_page
;
1173 struct ring_buffer
*buffer
= cpu_buffer
->buffer
;
1174 struct ring_buffer_event
*event
;
1175 bool lock_taken
= false;
1176 unsigned long flags
;
1178 next_page
= tail_page
;
1180 local_irq_save(flags
);
1182 * Since the write to the buffer is still not
1183 * fully lockless, we must be careful with NMIs.
1184 * The locks in the writers are taken when a write
1185 * crosses to a new page. The locks protect against
1186 * races with the readers (this will soon be fixed
1187 * with a lockless solution).
1189 * Because we can not protect against NMIs, and we
1190 * want to keep traces reentrant, we need to manage
1191 * what happens when we are in an NMI.
1193 * NMIs can happen after we take the lock.
1194 * If we are in an NMI, only take the lock
1195 * if it is not already taken. Otherwise
1198 if (unlikely(in_nmi())) {
1199 if (!__raw_spin_trylock(&cpu_buffer
->lock
)) {
1200 cpu_buffer
->nmi_dropped
++;
1204 __raw_spin_lock(&cpu_buffer
->lock
);
1208 rb_inc_page(cpu_buffer
, &next_page
);
1210 head_page
= cpu_buffer
->head_page
;
1211 reader_page
= cpu_buffer
->reader_page
;
1213 /* we grabbed the lock before incrementing */
1214 if (RB_WARN_ON(cpu_buffer
, next_page
== reader_page
))
1218 * If for some reason, we had an interrupt storm that made
1219 * it all the way around the buffer, bail, and warn
1222 if (unlikely(next_page
== commit_page
)) {
1223 cpu_buffer
->commit_overrun
++;
1227 if (next_page
== head_page
) {
1228 if (!(buffer
->flags
& RB_FL_OVERWRITE
))
1231 /* tail_page has not moved yet? */
1232 if (tail_page
== cpu_buffer
->tail_page
) {
1233 /* count overflows */
1234 cpu_buffer
->overrun
+=
1235 local_read(&head_page
->entries
);
1237 rb_inc_page(cpu_buffer
, &head_page
);
1238 cpu_buffer
->head_page
= head_page
;
1239 cpu_buffer
->head_page
->read
= 0;
1244 * If the tail page is still the same as what we think
1245 * it is, then it is up to us to update the tail
1248 if (tail_page
== cpu_buffer
->tail_page
) {
1249 local_set(&next_page
->write
, 0);
1250 local_set(&next_page
->entries
, 0);
1251 local_set(&next_page
->page
->commit
, 0);
1252 cpu_buffer
->tail_page
= next_page
;
1254 /* reread the time stamp */
1255 *ts
= rb_time_stamp(buffer
, cpu_buffer
->cpu
);
1256 cpu_buffer
->tail_page
->page
->time_stamp
= *ts
;
1260 * The actual tail page has moved forward.
1262 if (tail
< BUF_PAGE_SIZE
) {
1263 /* Mark the rest of the page with padding */
1264 event
= __rb_page_index(tail_page
, tail
);
1265 rb_event_set_padding(event
);
1268 /* Set the write back to the previous setting */
1269 local_sub(length
, &tail_page
->write
);
1272 * If this was a commit entry that failed,
1273 * increment that too
1275 if (tail_page
== cpu_buffer
->commit_page
&&
1276 tail
== rb_commit_index(cpu_buffer
)) {
1277 rb_set_commit_to_write(cpu_buffer
);
1280 __raw_spin_unlock(&cpu_buffer
->lock
);
1281 local_irq_restore(flags
);
1283 /* fail and let the caller try again */
1284 return ERR_PTR(-EAGAIN
);
1288 local_sub(length
, &tail_page
->write
);
1290 if (likely(lock_taken
))
1291 __raw_spin_unlock(&cpu_buffer
->lock
);
1292 local_irq_restore(flags
);
1296 static struct ring_buffer_event
*
1297 __rb_reserve_next(struct ring_buffer_per_cpu
*cpu_buffer
,
1298 unsigned type
, unsigned long length
, u64
*ts
)
1300 struct buffer_page
*tail_page
, *commit_page
;
1301 struct ring_buffer_event
*event
;
1302 unsigned long tail
, write
;
1304 commit_page
= cpu_buffer
->commit_page
;
1305 /* we just need to protect against interrupts */
1307 tail_page
= cpu_buffer
->tail_page
;
1308 write
= local_add_return(length
, &tail_page
->write
);
1309 tail
= write
- length
;
1311 /* See if we shot pass the end of this buffer page */
1312 if (write
> BUF_PAGE_SIZE
)
1313 return rb_move_tail(cpu_buffer
, length
, tail
,
1314 commit_page
, tail_page
, ts
);
1316 /* We reserved something on the buffer */
1318 if (RB_WARN_ON(cpu_buffer
, write
> BUF_PAGE_SIZE
))
1321 event
= __rb_page_index(tail_page
, tail
);
1322 rb_update_event(event
, type
, length
);
1324 /* The passed in type is zero for DATA */
1326 local_inc(&tail_page
->entries
);
1329 * If this is a commit and the tail is zero, then update
1330 * this page's time stamp.
1332 if (!tail
&& rb_is_commit(cpu_buffer
, event
))
1333 cpu_buffer
->commit_page
->page
->time_stamp
= *ts
;
1339 rb_add_time_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
1340 u64
*ts
, u64
*delta
)
1342 struct ring_buffer_event
*event
;
1346 if (unlikely(*delta
> (1ULL << 59) && !once
++)) {
1347 printk(KERN_WARNING
"Delta way too big! %llu"
1348 " ts=%llu write stamp = %llu\n",
1349 (unsigned long long)*delta
,
1350 (unsigned long long)*ts
,
1351 (unsigned long long)cpu_buffer
->write_stamp
);
1356 * The delta is too big, we to add a
1359 event
= __rb_reserve_next(cpu_buffer
,
1360 RINGBUF_TYPE_TIME_EXTEND
,
1366 if (PTR_ERR(event
) == -EAGAIN
)
1369 /* Only a commited time event can update the write stamp */
1370 if (rb_is_commit(cpu_buffer
, event
)) {
1372 * If this is the first on the page, then we need to
1373 * update the page itself, and just put in a zero.
1375 if (rb_event_index(event
)) {
1376 event
->time_delta
= *delta
& TS_MASK
;
1377 event
->array
[0] = *delta
>> TS_SHIFT
;
1379 cpu_buffer
->commit_page
->page
->time_stamp
= *ts
;
1380 event
->time_delta
= 0;
1381 event
->array
[0] = 0;
1383 cpu_buffer
->write_stamp
= *ts
;
1384 /* let the caller know this was the commit */
1387 /* Darn, this is just wasted space */
1388 event
->time_delta
= 0;
1389 event
->array
[0] = 0;
1398 static struct ring_buffer_event
*
1399 rb_reserve_next_event(struct ring_buffer_per_cpu
*cpu_buffer
,
1400 unsigned long length
)
1402 struct ring_buffer_event
*event
;
1407 length
= rb_calculate_event_length(length
);
1410 * We allow for interrupts to reenter here and do a trace.
1411 * If one does, it will cause this original code to loop
1412 * back here. Even with heavy interrupts happening, this
1413 * should only happen a few times in a row. If this happens
1414 * 1000 times in a row, there must be either an interrupt
1415 * storm or we have something buggy.
1418 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 1000))
1421 ts
= rb_time_stamp(cpu_buffer
->buffer
, cpu_buffer
->cpu
);
1424 * Only the first commit can update the timestamp.
1425 * Yes there is a race here. If an interrupt comes in
1426 * just after the conditional and it traces too, then it
1427 * will also check the deltas. More than one timestamp may
1428 * also be made. But only the entry that did the actual
1429 * commit will be something other than zero.
1431 if (likely(cpu_buffer
->tail_page
== cpu_buffer
->commit_page
&&
1432 rb_page_write(cpu_buffer
->tail_page
) ==
1433 rb_commit_index(cpu_buffer
))) {
1435 delta
= ts
- cpu_buffer
->write_stamp
;
1437 /* make sure this delta is calculated here */
1440 /* Did the write stamp get updated already? */
1441 if (unlikely(ts
< cpu_buffer
->write_stamp
))
1444 else if (unlikely(test_time_stamp(delta
))) {
1446 commit
= rb_add_time_stamp(cpu_buffer
, &ts
, &delta
);
1448 if (commit
== -EBUSY
)
1451 if (commit
== -EAGAIN
)
1454 RB_WARN_ON(cpu_buffer
, commit
< 0);
1457 /* Non commits have zero deltas */
1460 event
= __rb_reserve_next(cpu_buffer
, 0, length
, &ts
);
1461 if (PTR_ERR(event
) == -EAGAIN
)
1465 if (unlikely(commit
))
1467 * Ouch! We needed a timestamp and it was commited. But
1468 * we didn't get our event reserved.
1470 rb_set_commit_to_write(cpu_buffer
);
1475 * If the timestamp was commited, make the commit our entry
1476 * now so that we will update it when needed.
1478 if (unlikely(commit
))
1479 rb_set_commit_event(cpu_buffer
, event
);
1480 else if (!rb_is_commit(cpu_buffer
, event
))
1483 event
->time_delta
= delta
;
1488 #define TRACE_RECURSIVE_DEPTH 16
1490 static int trace_recursive_lock(void)
1492 current
->trace_recursion
++;
1494 if (likely(current
->trace_recursion
< TRACE_RECURSIVE_DEPTH
))
1497 /* Disable all tracing before we do anything else */
1498 tracing_off_permanent();
1500 printk_once(KERN_WARNING
"Tracing recursion: depth[%ld]:"
1501 "HC[%lu]:SC[%lu]:NMI[%lu]\n",
1502 current
->trace_recursion
,
1503 hardirq_count() >> HARDIRQ_SHIFT
,
1504 softirq_count() >> SOFTIRQ_SHIFT
,
1511 static void trace_recursive_unlock(void)
1513 WARN_ON_ONCE(!current
->trace_recursion
);
1515 current
->trace_recursion
--;
1518 static DEFINE_PER_CPU(int, rb_need_resched
);
1521 * ring_buffer_lock_reserve - reserve a part of the buffer
1522 * @buffer: the ring buffer to reserve from
1523 * @length: the length of the data to reserve (excluding event header)
1525 * Returns a reseverd event on the ring buffer to copy directly to.
1526 * The user of this interface will need to get the body to write into
1527 * and can use the ring_buffer_event_data() interface.
1529 * The length is the length of the data needed, not the event length
1530 * which also includes the event header.
1532 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
1533 * If NULL is returned, then nothing has been allocated or locked.
1535 struct ring_buffer_event
*
1536 ring_buffer_lock_reserve(struct ring_buffer
*buffer
, unsigned long length
)
1538 struct ring_buffer_per_cpu
*cpu_buffer
;
1539 struct ring_buffer_event
*event
;
1542 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
1545 if (atomic_read(&buffer
->record_disabled
))
1548 /* If we are tracing schedule, we don't want to recurse */
1549 resched
= ftrace_preempt_disable();
1551 if (trace_recursive_lock())
1554 cpu
= raw_smp_processor_id();
1556 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1559 cpu_buffer
= buffer
->buffers
[cpu
];
1561 if (atomic_read(&cpu_buffer
->record_disabled
))
1564 if (length
> BUF_MAX_DATA_SIZE
)
1567 event
= rb_reserve_next_event(cpu_buffer
, length
);
1572 * Need to store resched state on this cpu.
1573 * Only the first needs to.
1576 if (preempt_count() == 1)
1577 per_cpu(rb_need_resched
, cpu
) = resched
;
1582 trace_recursive_unlock();
1585 ftrace_preempt_enable(resched
);
1588 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve
);
1590 static void rb_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
1591 struct ring_buffer_event
*event
)
1593 local_inc(&cpu_buffer
->entries
);
1595 /* Only process further if we own the commit */
1596 if (!rb_is_commit(cpu_buffer
, event
))
1599 cpu_buffer
->write_stamp
+= event
->time_delta
;
1601 rb_set_commit_to_write(cpu_buffer
);
1605 * ring_buffer_unlock_commit - commit a reserved
1606 * @buffer: The buffer to commit to
1607 * @event: The event pointer to commit.
1609 * This commits the data to the ring buffer, and releases any locks held.
1611 * Must be paired with ring_buffer_lock_reserve.
1613 int ring_buffer_unlock_commit(struct ring_buffer
*buffer
,
1614 struct ring_buffer_event
*event
)
1616 struct ring_buffer_per_cpu
*cpu_buffer
;
1617 int cpu
= raw_smp_processor_id();
1619 cpu_buffer
= buffer
->buffers
[cpu
];
1621 rb_commit(cpu_buffer
, event
);
1623 trace_recursive_unlock();
1626 * Only the last preempt count needs to restore preemption.
1628 if (preempt_count() == 1)
1629 ftrace_preempt_enable(per_cpu(rb_need_resched
, cpu
));
1631 preempt_enable_no_resched_notrace();
1635 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit
);
1637 static inline void rb_event_discard(struct ring_buffer_event
*event
)
1639 /* array[0] holds the actual length for the discarded event */
1640 event
->array
[0] = rb_event_data_length(event
) - RB_EVNT_HDR_SIZE
;
1641 event
->type_len
= RINGBUF_TYPE_PADDING
;
1642 /* time delta must be non zero */
1643 if (!event
->time_delta
)
1644 event
->time_delta
= 1;
1648 * ring_buffer_event_discard - discard any event in the ring buffer
1649 * @event: the event to discard
1651 * Sometimes a event that is in the ring buffer needs to be ignored.
1652 * This function lets the user discard an event in the ring buffer
1653 * and then that event will not be read later.
1655 * Note, it is up to the user to be careful with this, and protect
1656 * against races. If the user discards an event that has been consumed
1657 * it is possible that it could corrupt the ring buffer.
1659 void ring_buffer_event_discard(struct ring_buffer_event
*event
)
1661 rb_event_discard(event
);
1663 EXPORT_SYMBOL_GPL(ring_buffer_event_discard
);
1666 * ring_buffer_commit_discard - discard an event that has not been committed
1667 * @buffer: the ring buffer
1668 * @event: non committed event to discard
1670 * This is similar to ring_buffer_event_discard but must only be
1671 * performed on an event that has not been committed yet. The difference
1672 * is that this will also try to free the event from the ring buffer
1673 * if another event has not been added behind it.
1675 * If another event has been added behind it, it will set the event
1676 * up as discarded, and perform the commit.
1678 * If this function is called, do not call ring_buffer_unlock_commit on
1681 void ring_buffer_discard_commit(struct ring_buffer
*buffer
,
1682 struct ring_buffer_event
*event
)
1684 struct ring_buffer_per_cpu
*cpu_buffer
;
1685 unsigned long new_index
, old_index
;
1686 struct buffer_page
*bpage
;
1687 unsigned long index
;
1691 /* The event is discarded regardless */
1692 rb_event_discard(event
);
1695 * This must only be called if the event has not been
1696 * committed yet. Thus we can assume that preemption
1697 * is still disabled.
1699 RB_WARN_ON(buffer
, preemptible());
1701 cpu
= smp_processor_id();
1702 cpu_buffer
= buffer
->buffers
[cpu
];
1704 new_index
= rb_event_index(event
);
1705 old_index
= new_index
+ rb_event_length(event
);
1706 addr
= (unsigned long)event
;
1709 bpage
= cpu_buffer
->tail_page
;
1711 if (bpage
== (void *)addr
&& rb_page_write(bpage
) == old_index
) {
1713 * This is on the tail page. It is possible that
1714 * a write could come in and move the tail page
1715 * and write to the next page. That is fine
1716 * because we just shorten what is on this page.
1718 index
= local_cmpxchg(&bpage
->write
, old_index
, new_index
);
1719 if (index
== old_index
)
1724 * The commit is still visible by the reader, so we
1725 * must increment entries.
1727 local_inc(&cpu_buffer
->entries
);
1730 * If a write came in and pushed the tail page
1731 * we still need to update the commit pointer
1732 * if we were the commit.
1734 if (rb_is_commit(cpu_buffer
, event
))
1735 rb_set_commit_to_write(cpu_buffer
);
1737 trace_recursive_unlock();
1740 * Only the last preempt count needs to restore preemption.
1742 if (preempt_count() == 1)
1743 ftrace_preempt_enable(per_cpu(rb_need_resched
, cpu
));
1745 preempt_enable_no_resched_notrace();
1748 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit
);
1751 * ring_buffer_write - write data to the buffer without reserving
1752 * @buffer: The ring buffer to write to.
1753 * @length: The length of the data being written (excluding the event header)
1754 * @data: The data to write to the buffer.
1756 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
1757 * one function. If you already have the data to write to the buffer, it
1758 * may be easier to simply call this function.
1760 * Note, like ring_buffer_lock_reserve, the length is the length of the data
1761 * and not the length of the event which would hold the header.
1763 int ring_buffer_write(struct ring_buffer
*buffer
,
1764 unsigned long length
,
1767 struct ring_buffer_per_cpu
*cpu_buffer
;
1768 struct ring_buffer_event
*event
;
1773 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
1776 if (atomic_read(&buffer
->record_disabled
))
1779 resched
= ftrace_preempt_disable();
1781 cpu
= raw_smp_processor_id();
1783 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1786 cpu_buffer
= buffer
->buffers
[cpu
];
1788 if (atomic_read(&cpu_buffer
->record_disabled
))
1791 if (length
> BUF_MAX_DATA_SIZE
)
1794 event
= rb_reserve_next_event(cpu_buffer
, length
);
1798 body
= rb_event_data(event
);
1800 memcpy(body
, data
, length
);
1802 rb_commit(cpu_buffer
, event
);
1806 ftrace_preempt_enable(resched
);
1810 EXPORT_SYMBOL_GPL(ring_buffer_write
);
1812 static int rb_per_cpu_empty(struct ring_buffer_per_cpu
*cpu_buffer
)
1814 struct buffer_page
*reader
= cpu_buffer
->reader_page
;
1815 struct buffer_page
*head
= cpu_buffer
->head_page
;
1816 struct buffer_page
*commit
= cpu_buffer
->commit_page
;
1818 return reader
->read
== rb_page_commit(reader
) &&
1819 (commit
== reader
||
1821 head
->read
== rb_page_commit(commit
)));
1825 * ring_buffer_record_disable - stop all writes into the buffer
1826 * @buffer: The ring buffer to stop writes to.
1828 * This prevents all writes to the buffer. Any attempt to write
1829 * to the buffer after this will fail and return NULL.
1831 * The caller should call synchronize_sched() after this.
1833 void ring_buffer_record_disable(struct ring_buffer
*buffer
)
1835 atomic_inc(&buffer
->record_disabled
);
1837 EXPORT_SYMBOL_GPL(ring_buffer_record_disable
);
1840 * ring_buffer_record_enable - enable writes to the buffer
1841 * @buffer: The ring buffer to enable writes
1843 * Note, multiple disables will need the same number of enables
1844 * to truely enable the writing (much like preempt_disable).
1846 void ring_buffer_record_enable(struct ring_buffer
*buffer
)
1848 atomic_dec(&buffer
->record_disabled
);
1850 EXPORT_SYMBOL_GPL(ring_buffer_record_enable
);
1853 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
1854 * @buffer: The ring buffer to stop writes to.
1855 * @cpu: The CPU buffer to stop
1857 * This prevents all writes to the buffer. Any attempt to write
1858 * to the buffer after this will fail and return NULL.
1860 * The caller should call synchronize_sched() after this.
1862 void ring_buffer_record_disable_cpu(struct ring_buffer
*buffer
, int cpu
)
1864 struct ring_buffer_per_cpu
*cpu_buffer
;
1866 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1869 cpu_buffer
= buffer
->buffers
[cpu
];
1870 atomic_inc(&cpu_buffer
->record_disabled
);
1872 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu
);
1875 * ring_buffer_record_enable_cpu - enable writes to the buffer
1876 * @buffer: The ring buffer to enable writes
1877 * @cpu: The CPU to enable.
1879 * Note, multiple disables will need the same number of enables
1880 * to truely enable the writing (much like preempt_disable).
1882 void ring_buffer_record_enable_cpu(struct ring_buffer
*buffer
, int cpu
)
1884 struct ring_buffer_per_cpu
*cpu_buffer
;
1886 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1889 cpu_buffer
= buffer
->buffers
[cpu
];
1890 atomic_dec(&cpu_buffer
->record_disabled
);
1892 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu
);
1895 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
1896 * @buffer: The ring buffer
1897 * @cpu: The per CPU buffer to get the entries from.
1899 unsigned long ring_buffer_entries_cpu(struct ring_buffer
*buffer
, int cpu
)
1901 struct ring_buffer_per_cpu
*cpu_buffer
;
1904 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1907 cpu_buffer
= buffer
->buffers
[cpu
];
1908 ret
= (local_read(&cpu_buffer
->entries
) - cpu_buffer
->overrun
)
1913 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu
);
1916 * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
1917 * @buffer: The ring buffer
1918 * @cpu: The per CPU buffer to get the number of overruns from
1920 unsigned long ring_buffer_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
1922 struct ring_buffer_per_cpu
*cpu_buffer
;
1925 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1928 cpu_buffer
= buffer
->buffers
[cpu
];
1929 ret
= cpu_buffer
->overrun
;
1933 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu
);
1936 * ring_buffer_nmi_dropped_cpu - get the number of nmis that were dropped
1937 * @buffer: The ring buffer
1938 * @cpu: The per CPU buffer to get the number of overruns from
1940 unsigned long ring_buffer_nmi_dropped_cpu(struct ring_buffer
*buffer
, int cpu
)
1942 struct ring_buffer_per_cpu
*cpu_buffer
;
1945 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1948 cpu_buffer
= buffer
->buffers
[cpu
];
1949 ret
= cpu_buffer
->nmi_dropped
;
1953 EXPORT_SYMBOL_GPL(ring_buffer_nmi_dropped_cpu
);
1956 * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits
1957 * @buffer: The ring buffer
1958 * @cpu: The per CPU buffer to get the number of overruns from
1961 ring_buffer_commit_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
1963 struct ring_buffer_per_cpu
*cpu_buffer
;
1966 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1969 cpu_buffer
= buffer
->buffers
[cpu
];
1970 ret
= cpu_buffer
->commit_overrun
;
1974 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu
);
1977 * ring_buffer_entries - get the number of entries in a buffer
1978 * @buffer: The ring buffer
1980 * Returns the total number of entries in the ring buffer
1983 unsigned long ring_buffer_entries(struct ring_buffer
*buffer
)
1985 struct ring_buffer_per_cpu
*cpu_buffer
;
1986 unsigned long entries
= 0;
1989 /* if you care about this being correct, lock the buffer */
1990 for_each_buffer_cpu(buffer
, cpu
) {
1991 cpu_buffer
= buffer
->buffers
[cpu
];
1992 entries
+= (local_read(&cpu_buffer
->entries
) -
1993 cpu_buffer
->overrun
) - cpu_buffer
->read
;
1998 EXPORT_SYMBOL_GPL(ring_buffer_entries
);
2001 * ring_buffer_overrun_cpu - get the number of overruns in buffer
2002 * @buffer: The ring buffer
2004 * Returns the total number of overruns in the ring buffer
2007 unsigned long ring_buffer_overruns(struct ring_buffer
*buffer
)
2009 struct ring_buffer_per_cpu
*cpu_buffer
;
2010 unsigned long overruns
= 0;
2013 /* if you care about this being correct, lock the buffer */
2014 for_each_buffer_cpu(buffer
, cpu
) {
2015 cpu_buffer
= buffer
->buffers
[cpu
];
2016 overruns
+= cpu_buffer
->overrun
;
2021 EXPORT_SYMBOL_GPL(ring_buffer_overruns
);
2023 static void rb_iter_reset(struct ring_buffer_iter
*iter
)
2025 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2027 /* Iterator usage is expected to have record disabled */
2028 if (list_empty(&cpu_buffer
->reader_page
->list
)) {
2029 iter
->head_page
= cpu_buffer
->head_page
;
2030 iter
->head
= cpu_buffer
->head_page
->read
;
2032 iter
->head_page
= cpu_buffer
->reader_page
;
2033 iter
->head
= cpu_buffer
->reader_page
->read
;
2036 iter
->read_stamp
= cpu_buffer
->read_stamp
;
2038 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
2042 * ring_buffer_iter_reset - reset an iterator
2043 * @iter: The iterator to reset
2045 * Resets the iterator, so that it will start from the beginning
2048 void ring_buffer_iter_reset(struct ring_buffer_iter
*iter
)
2050 struct ring_buffer_per_cpu
*cpu_buffer
;
2051 unsigned long flags
;
2056 cpu_buffer
= iter
->cpu_buffer
;
2058 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2059 rb_iter_reset(iter
);
2060 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2062 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset
);
2065 * ring_buffer_iter_empty - check if an iterator has no more to read
2066 * @iter: The iterator to check
2068 int ring_buffer_iter_empty(struct ring_buffer_iter
*iter
)
2070 struct ring_buffer_per_cpu
*cpu_buffer
;
2072 cpu_buffer
= iter
->cpu_buffer
;
2074 return iter
->head_page
== cpu_buffer
->commit_page
&&
2075 iter
->head
== rb_commit_index(cpu_buffer
);
2077 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty
);
2080 rb_update_read_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
2081 struct ring_buffer_event
*event
)
2085 switch (event
->type_len
) {
2086 case RINGBUF_TYPE_PADDING
:
2089 case RINGBUF_TYPE_TIME_EXTEND
:
2090 delta
= event
->array
[0];
2092 delta
+= event
->time_delta
;
2093 cpu_buffer
->read_stamp
+= delta
;
2096 case RINGBUF_TYPE_TIME_STAMP
:
2097 /* FIXME: not implemented */
2100 case RINGBUF_TYPE_DATA
:
2101 cpu_buffer
->read_stamp
+= event
->time_delta
;
2111 rb_update_iter_read_stamp(struct ring_buffer_iter
*iter
,
2112 struct ring_buffer_event
*event
)
2116 switch (event
->type_len
) {
2117 case RINGBUF_TYPE_PADDING
:
2120 case RINGBUF_TYPE_TIME_EXTEND
:
2121 delta
= event
->array
[0];
2123 delta
+= event
->time_delta
;
2124 iter
->read_stamp
+= delta
;
2127 case RINGBUF_TYPE_TIME_STAMP
:
2128 /* FIXME: not implemented */
2131 case RINGBUF_TYPE_DATA
:
2132 iter
->read_stamp
+= event
->time_delta
;
2141 static struct buffer_page
*
2142 rb_get_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
2144 struct buffer_page
*reader
= NULL
;
2145 unsigned long flags
;
2148 local_irq_save(flags
);
2149 __raw_spin_lock(&cpu_buffer
->lock
);
2153 * This should normally only loop twice. But because the
2154 * start of the reader inserts an empty page, it causes
2155 * a case where we will loop three times. There should be no
2156 * reason to loop four times (that I know of).
2158 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 3)) {
2163 reader
= cpu_buffer
->reader_page
;
2165 /* If there's more to read, return this page */
2166 if (cpu_buffer
->reader_page
->read
< rb_page_size(reader
))
2169 /* Never should we have an index greater than the size */
2170 if (RB_WARN_ON(cpu_buffer
,
2171 cpu_buffer
->reader_page
->read
> rb_page_size(reader
)))
2174 /* check if we caught up to the tail */
2176 if (cpu_buffer
->commit_page
== cpu_buffer
->reader_page
)
2180 * Splice the empty reader page into the list around the head.
2181 * Reset the reader page to size zero.
2184 reader
= cpu_buffer
->head_page
;
2185 cpu_buffer
->reader_page
->list
.next
= reader
->list
.next
;
2186 cpu_buffer
->reader_page
->list
.prev
= reader
->list
.prev
;
2188 local_set(&cpu_buffer
->reader_page
->write
, 0);
2189 local_set(&cpu_buffer
->reader_page
->entries
, 0);
2190 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
2192 /* Make the reader page now replace the head */
2193 reader
->list
.prev
->next
= &cpu_buffer
->reader_page
->list
;
2194 reader
->list
.next
->prev
= &cpu_buffer
->reader_page
->list
;
2197 * If the tail is on the reader, then we must set the head
2198 * to the inserted page, otherwise we set it one before.
2200 cpu_buffer
->head_page
= cpu_buffer
->reader_page
;
2202 if (cpu_buffer
->commit_page
!= reader
)
2203 rb_inc_page(cpu_buffer
, &cpu_buffer
->head_page
);
2205 /* Finally update the reader page to the new head */
2206 cpu_buffer
->reader_page
= reader
;
2207 rb_reset_reader_page(cpu_buffer
);
2212 __raw_spin_unlock(&cpu_buffer
->lock
);
2213 local_irq_restore(flags
);
2218 static void rb_advance_reader(struct ring_buffer_per_cpu
*cpu_buffer
)
2220 struct ring_buffer_event
*event
;
2221 struct buffer_page
*reader
;
2224 reader
= rb_get_reader_page(cpu_buffer
);
2226 /* This function should not be called when buffer is empty */
2227 if (RB_WARN_ON(cpu_buffer
, !reader
))
2230 event
= rb_reader_event(cpu_buffer
);
2232 if (event
->type_len
<= RINGBUF_TYPE_DATA_TYPE_LEN_MAX
2233 || rb_discarded_event(event
))
2236 rb_update_read_stamp(cpu_buffer
, event
);
2238 length
= rb_event_length(event
);
2239 cpu_buffer
->reader_page
->read
+= length
;
2242 static void rb_advance_iter(struct ring_buffer_iter
*iter
)
2244 struct ring_buffer
*buffer
;
2245 struct ring_buffer_per_cpu
*cpu_buffer
;
2246 struct ring_buffer_event
*event
;
2249 cpu_buffer
= iter
->cpu_buffer
;
2250 buffer
= cpu_buffer
->buffer
;
2253 * Check if we are at the end of the buffer.
2255 if (iter
->head
>= rb_page_size(iter
->head_page
)) {
2256 if (RB_WARN_ON(buffer
,
2257 iter
->head_page
== cpu_buffer
->commit_page
))
2263 event
= rb_iter_head_event(iter
);
2265 length
= rb_event_length(event
);
2268 * This should not be called to advance the header if we are
2269 * at the tail of the buffer.
2271 if (RB_WARN_ON(cpu_buffer
,
2272 (iter
->head_page
== cpu_buffer
->commit_page
) &&
2273 (iter
->head
+ length
> rb_commit_index(cpu_buffer
))))
2276 rb_update_iter_read_stamp(iter
, event
);
2278 iter
->head
+= length
;
2280 /* check for end of page padding */
2281 if ((iter
->head
>= rb_page_size(iter
->head_page
)) &&
2282 (iter
->head_page
!= cpu_buffer
->commit_page
))
2283 rb_advance_iter(iter
);
2286 static struct ring_buffer_event
*
2287 rb_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2289 struct ring_buffer_per_cpu
*cpu_buffer
;
2290 struct ring_buffer_event
*event
;
2291 struct buffer_page
*reader
;
2294 cpu_buffer
= buffer
->buffers
[cpu
];
2298 * We repeat when a timestamp is encountered. It is possible
2299 * to get multiple timestamps from an interrupt entering just
2300 * as one timestamp is about to be written. The max times
2301 * that this can happen is the number of nested interrupts we
2302 * can have. Nesting 10 deep of interrupts is clearly
2305 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 10))
2308 reader
= rb_get_reader_page(cpu_buffer
);
2312 event
= rb_reader_event(cpu_buffer
);
2314 switch (event
->type_len
) {
2315 case RINGBUF_TYPE_PADDING
:
2316 if (rb_null_event(event
))
2317 RB_WARN_ON(cpu_buffer
, 1);
2319 * Because the writer could be discarding every
2320 * event it creates (which would probably be bad)
2321 * if we were to go back to "again" then we may never
2322 * catch up, and will trigger the warn on, or lock
2323 * the box. Return the padding, and we will release
2324 * the current locks, and try again.
2326 rb_advance_reader(cpu_buffer
);
2329 case RINGBUF_TYPE_TIME_EXTEND
:
2330 /* Internal data, OK to advance */
2331 rb_advance_reader(cpu_buffer
);
2334 case RINGBUF_TYPE_TIME_STAMP
:
2335 /* FIXME: not implemented */
2336 rb_advance_reader(cpu_buffer
);
2339 case RINGBUF_TYPE_DATA
:
2341 *ts
= cpu_buffer
->read_stamp
+ event
->time_delta
;
2342 ring_buffer_normalize_time_stamp(buffer
,
2343 cpu_buffer
->cpu
, ts
);
2353 EXPORT_SYMBOL_GPL(ring_buffer_peek
);
2355 static struct ring_buffer_event
*
2356 rb_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
2358 struct ring_buffer
*buffer
;
2359 struct ring_buffer_per_cpu
*cpu_buffer
;
2360 struct ring_buffer_event
*event
;
2363 if (ring_buffer_iter_empty(iter
))
2366 cpu_buffer
= iter
->cpu_buffer
;
2367 buffer
= cpu_buffer
->buffer
;
2371 * We repeat when a timestamp is encountered. It is possible
2372 * to get multiple timestamps from an interrupt entering just
2373 * as one timestamp is about to be written. The max times
2374 * that this can happen is the number of nested interrupts we
2375 * can have. Nesting 10 deep of interrupts is clearly
2378 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 10))
2381 if (rb_per_cpu_empty(cpu_buffer
))
2384 event
= rb_iter_head_event(iter
);
2386 switch (event
->type_len
) {
2387 case RINGBUF_TYPE_PADDING
:
2388 if (rb_null_event(event
)) {
2392 rb_advance_iter(iter
);
2395 case RINGBUF_TYPE_TIME_EXTEND
:
2396 /* Internal data, OK to advance */
2397 rb_advance_iter(iter
);
2400 case RINGBUF_TYPE_TIME_STAMP
:
2401 /* FIXME: not implemented */
2402 rb_advance_iter(iter
);
2405 case RINGBUF_TYPE_DATA
:
2407 *ts
= iter
->read_stamp
+ event
->time_delta
;
2408 ring_buffer_normalize_time_stamp(buffer
,
2409 cpu_buffer
->cpu
, ts
);
2419 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek
);
2422 * ring_buffer_peek - peek at the next event to be read
2423 * @buffer: The ring buffer to read
2424 * @cpu: The cpu to peak at
2425 * @ts: The timestamp counter of this event.
2427 * This will return the event that will be read next, but does
2428 * not consume the data.
2430 struct ring_buffer_event
*
2431 ring_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2433 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2434 struct ring_buffer_event
*event
;
2435 unsigned long flags
;
2437 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2441 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2442 event
= rb_buffer_peek(buffer
, cpu
, ts
);
2443 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2445 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2454 * ring_buffer_iter_peek - peek at the next event to be read
2455 * @iter: The ring buffer iterator
2456 * @ts: The timestamp counter of this event.
2458 * This will return the event that will be read next, but does
2459 * not increment the iterator.
2461 struct ring_buffer_event
*
2462 ring_buffer_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
2464 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2465 struct ring_buffer_event
*event
;
2466 unsigned long flags
;
2469 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2470 event
= rb_iter_peek(iter
, ts
);
2471 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2473 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2482 * ring_buffer_consume - return an event and consume it
2483 * @buffer: The ring buffer to get the next event from
2485 * Returns the next event in the ring buffer, and that event is consumed.
2486 * Meaning, that sequential reads will keep returning a different event,
2487 * and eventually empty the ring buffer if the producer is slower.
2489 struct ring_buffer_event
*
2490 ring_buffer_consume(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2492 struct ring_buffer_per_cpu
*cpu_buffer
;
2493 struct ring_buffer_event
*event
= NULL
;
2494 unsigned long flags
;
2497 /* might be called in atomic */
2500 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2503 cpu_buffer
= buffer
->buffers
[cpu
];
2504 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2506 event
= rb_buffer_peek(buffer
, cpu
, ts
);
2510 rb_advance_reader(cpu_buffer
);
2513 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2518 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2525 EXPORT_SYMBOL_GPL(ring_buffer_consume
);
2528 * ring_buffer_read_start - start a non consuming read of the buffer
2529 * @buffer: The ring buffer to read from
2530 * @cpu: The cpu buffer to iterate over
2532 * This starts up an iteration through the buffer. It also disables
2533 * the recording to the buffer until the reading is finished.
2534 * This prevents the reading from being corrupted. This is not
2535 * a consuming read, so a producer is not expected.
2537 * Must be paired with ring_buffer_finish.
2539 struct ring_buffer_iter
*
2540 ring_buffer_read_start(struct ring_buffer
*buffer
, int cpu
)
2542 struct ring_buffer_per_cpu
*cpu_buffer
;
2543 struct ring_buffer_iter
*iter
;
2544 unsigned long flags
;
2546 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2549 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
2553 cpu_buffer
= buffer
->buffers
[cpu
];
2555 iter
->cpu_buffer
= cpu_buffer
;
2557 atomic_inc(&cpu_buffer
->record_disabled
);
2558 synchronize_sched();
2560 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2561 __raw_spin_lock(&cpu_buffer
->lock
);
2562 rb_iter_reset(iter
);
2563 __raw_spin_unlock(&cpu_buffer
->lock
);
2564 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2568 EXPORT_SYMBOL_GPL(ring_buffer_read_start
);
2571 * ring_buffer_finish - finish reading the iterator of the buffer
2572 * @iter: The iterator retrieved by ring_buffer_start
2574 * This re-enables the recording to the buffer, and frees the
2578 ring_buffer_read_finish(struct ring_buffer_iter
*iter
)
2580 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2582 atomic_dec(&cpu_buffer
->record_disabled
);
2585 EXPORT_SYMBOL_GPL(ring_buffer_read_finish
);
2588 * ring_buffer_read - read the next item in the ring buffer by the iterator
2589 * @iter: The ring buffer iterator
2590 * @ts: The time stamp of the event read.
2592 * This reads the next event in the ring buffer and increments the iterator.
2594 struct ring_buffer_event
*
2595 ring_buffer_read(struct ring_buffer_iter
*iter
, u64
*ts
)
2597 struct ring_buffer_event
*event
;
2598 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2599 unsigned long flags
;
2602 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2603 event
= rb_iter_peek(iter
, ts
);
2607 rb_advance_iter(iter
);
2609 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2611 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2618 EXPORT_SYMBOL_GPL(ring_buffer_read
);
2621 * ring_buffer_size - return the size of the ring buffer (in bytes)
2622 * @buffer: The ring buffer.
2624 unsigned long ring_buffer_size(struct ring_buffer
*buffer
)
2626 return BUF_PAGE_SIZE
* buffer
->pages
;
2628 EXPORT_SYMBOL_GPL(ring_buffer_size
);
2631 rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
)
2633 cpu_buffer
->head_page
2634 = list_entry(cpu_buffer
->pages
.next
, struct buffer_page
, list
);
2635 local_set(&cpu_buffer
->head_page
->write
, 0);
2636 local_set(&cpu_buffer
->head_page
->entries
, 0);
2637 local_set(&cpu_buffer
->head_page
->page
->commit
, 0);
2639 cpu_buffer
->head_page
->read
= 0;
2641 cpu_buffer
->tail_page
= cpu_buffer
->head_page
;
2642 cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
2644 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
2645 local_set(&cpu_buffer
->reader_page
->write
, 0);
2646 local_set(&cpu_buffer
->reader_page
->entries
, 0);
2647 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
2648 cpu_buffer
->reader_page
->read
= 0;
2650 cpu_buffer
->nmi_dropped
= 0;
2651 cpu_buffer
->commit_overrun
= 0;
2652 cpu_buffer
->overrun
= 0;
2653 cpu_buffer
->read
= 0;
2654 local_set(&cpu_buffer
->entries
, 0);
2656 cpu_buffer
->write_stamp
= 0;
2657 cpu_buffer
->read_stamp
= 0;
2661 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
2662 * @buffer: The ring buffer to reset a per cpu buffer of
2663 * @cpu: The CPU buffer to be reset
2665 void ring_buffer_reset_cpu(struct ring_buffer
*buffer
, int cpu
)
2667 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2668 unsigned long flags
;
2670 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2673 atomic_inc(&cpu_buffer
->record_disabled
);
2675 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2677 __raw_spin_lock(&cpu_buffer
->lock
);
2679 rb_reset_cpu(cpu_buffer
);
2681 __raw_spin_unlock(&cpu_buffer
->lock
);
2683 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2685 atomic_dec(&cpu_buffer
->record_disabled
);
2687 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu
);
2690 * ring_buffer_reset - reset a ring buffer
2691 * @buffer: The ring buffer to reset all cpu buffers
2693 void ring_buffer_reset(struct ring_buffer
*buffer
)
2697 for_each_buffer_cpu(buffer
, cpu
)
2698 ring_buffer_reset_cpu(buffer
, cpu
);
2700 EXPORT_SYMBOL_GPL(ring_buffer_reset
);
2703 * rind_buffer_empty - is the ring buffer empty?
2704 * @buffer: The ring buffer to test
2706 int ring_buffer_empty(struct ring_buffer
*buffer
)
2708 struct ring_buffer_per_cpu
*cpu_buffer
;
2711 /* yes this is racy, but if you don't like the race, lock the buffer */
2712 for_each_buffer_cpu(buffer
, cpu
) {
2713 cpu_buffer
= buffer
->buffers
[cpu
];
2714 if (!rb_per_cpu_empty(cpu_buffer
))
2720 EXPORT_SYMBOL_GPL(ring_buffer_empty
);
2723 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
2724 * @buffer: The ring buffer
2725 * @cpu: The CPU buffer to test
2727 int ring_buffer_empty_cpu(struct ring_buffer
*buffer
, int cpu
)
2729 struct ring_buffer_per_cpu
*cpu_buffer
;
2732 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2735 cpu_buffer
= buffer
->buffers
[cpu
];
2736 ret
= rb_per_cpu_empty(cpu_buffer
);
2741 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu
);
2744 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
2745 * @buffer_a: One buffer to swap with
2746 * @buffer_b: The other buffer to swap with
2748 * This function is useful for tracers that want to take a "snapshot"
2749 * of a CPU buffer and has another back up buffer lying around.
2750 * it is expected that the tracer handles the cpu buffer not being
2751 * used at the moment.
2753 int ring_buffer_swap_cpu(struct ring_buffer
*buffer_a
,
2754 struct ring_buffer
*buffer_b
, int cpu
)
2756 struct ring_buffer_per_cpu
*cpu_buffer_a
;
2757 struct ring_buffer_per_cpu
*cpu_buffer_b
;
2760 if (!cpumask_test_cpu(cpu
, buffer_a
->cpumask
) ||
2761 !cpumask_test_cpu(cpu
, buffer_b
->cpumask
))
2764 /* At least make sure the two buffers are somewhat the same */
2765 if (buffer_a
->pages
!= buffer_b
->pages
)
2770 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
2773 if (atomic_read(&buffer_a
->record_disabled
))
2776 if (atomic_read(&buffer_b
->record_disabled
))
2779 cpu_buffer_a
= buffer_a
->buffers
[cpu
];
2780 cpu_buffer_b
= buffer_b
->buffers
[cpu
];
2782 if (atomic_read(&cpu_buffer_a
->record_disabled
))
2785 if (atomic_read(&cpu_buffer_b
->record_disabled
))
2789 * We can't do a synchronize_sched here because this
2790 * function can be called in atomic context.
2791 * Normally this will be called from the same CPU as cpu.
2792 * If not it's up to the caller to protect this.
2794 atomic_inc(&cpu_buffer_a
->record_disabled
);
2795 atomic_inc(&cpu_buffer_b
->record_disabled
);
2797 buffer_a
->buffers
[cpu
] = cpu_buffer_b
;
2798 buffer_b
->buffers
[cpu
] = cpu_buffer_a
;
2800 cpu_buffer_b
->buffer
= buffer_a
;
2801 cpu_buffer_a
->buffer
= buffer_b
;
2803 atomic_dec(&cpu_buffer_a
->record_disabled
);
2804 atomic_dec(&cpu_buffer_b
->record_disabled
);
2810 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu
);
2813 * ring_buffer_alloc_read_page - allocate a page to read from buffer
2814 * @buffer: the buffer to allocate for.
2816 * This function is used in conjunction with ring_buffer_read_page.
2817 * When reading a full page from the ring buffer, these functions
2818 * can be used to speed up the process. The calling function should
2819 * allocate a few pages first with this function. Then when it
2820 * needs to get pages from the ring buffer, it passes the result
2821 * of this function into ring_buffer_read_page, which will swap
2822 * the page that was allocated, with the read page of the buffer.
2825 * The page allocated, or NULL on error.
2827 void *ring_buffer_alloc_read_page(struct ring_buffer
*buffer
)
2829 struct buffer_data_page
*bpage
;
2832 addr
= __get_free_page(GFP_KERNEL
);
2836 bpage
= (void *)addr
;
2838 rb_init_page(bpage
);
2842 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page
);
2845 * ring_buffer_free_read_page - free an allocated read page
2846 * @buffer: the buffer the page was allocate for
2847 * @data: the page to free
2849 * Free a page allocated from ring_buffer_alloc_read_page.
2851 void ring_buffer_free_read_page(struct ring_buffer
*buffer
, void *data
)
2853 free_page((unsigned long)data
);
2855 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page
);
2858 * ring_buffer_read_page - extract a page from the ring buffer
2859 * @buffer: buffer to extract from
2860 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
2861 * @len: amount to extract
2862 * @cpu: the cpu of the buffer to extract
2863 * @full: should the extraction only happen when the page is full.
2865 * This function will pull out a page from the ring buffer and consume it.
2866 * @data_page must be the address of the variable that was returned
2867 * from ring_buffer_alloc_read_page. This is because the page might be used
2868 * to swap with a page in the ring buffer.
2871 * rpage = ring_buffer_alloc_read_page(buffer);
2874 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
2876 * process_page(rpage, ret);
2878 * When @full is set, the function will not return true unless
2879 * the writer is off the reader page.
2881 * Note: it is up to the calling functions to handle sleeps and wakeups.
2882 * The ring buffer can be used anywhere in the kernel and can not
2883 * blindly call wake_up. The layer that uses the ring buffer must be
2884 * responsible for that.
2887 * >=0 if data has been transferred, returns the offset of consumed data.
2888 * <0 if no data has been transferred.
2890 int ring_buffer_read_page(struct ring_buffer
*buffer
,
2891 void **data_page
, size_t len
, int cpu
, int full
)
2893 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2894 struct ring_buffer_event
*event
;
2895 struct buffer_data_page
*bpage
;
2896 struct buffer_page
*reader
;
2897 unsigned long flags
;
2898 unsigned int commit
;
2903 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2907 * If len is not big enough to hold the page header, then
2908 * we can not copy anything.
2910 if (len
<= BUF_PAGE_HDR_SIZE
)
2913 len
-= BUF_PAGE_HDR_SIZE
;
2922 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2924 reader
= rb_get_reader_page(cpu_buffer
);
2928 event
= rb_reader_event(cpu_buffer
);
2930 read
= reader
->read
;
2931 commit
= rb_page_commit(reader
);
2934 * If this page has been partially read or
2935 * if len is not big enough to read the rest of the page or
2936 * a writer is still on the page, then
2937 * we must copy the data from the page to the buffer.
2938 * Otherwise, we can simply swap the page with the one passed in.
2940 if (read
|| (len
< (commit
- read
)) ||
2941 cpu_buffer
->reader_page
== cpu_buffer
->commit_page
) {
2942 struct buffer_data_page
*rpage
= cpu_buffer
->reader_page
->page
;
2943 unsigned int rpos
= read
;
2944 unsigned int pos
= 0;
2950 if (len
> (commit
- read
))
2951 len
= (commit
- read
);
2953 size
= rb_event_length(event
);
2958 /* save the current timestamp, since the user will need it */
2959 save_timestamp
= cpu_buffer
->read_stamp
;
2961 /* Need to copy one event at a time */
2963 memcpy(bpage
->data
+ pos
, rpage
->data
+ rpos
, size
);
2967 rb_advance_reader(cpu_buffer
);
2968 rpos
= reader
->read
;
2971 event
= rb_reader_event(cpu_buffer
);
2972 size
= rb_event_length(event
);
2973 } while (len
> size
);
2976 local_set(&bpage
->commit
, pos
);
2977 bpage
->time_stamp
= save_timestamp
;
2979 /* we copied everything to the beginning */
2982 /* update the entry counter */
2983 cpu_buffer
->read
+= local_read(&reader
->entries
);
2985 /* swap the pages */
2986 rb_init_page(bpage
);
2987 bpage
= reader
->page
;
2988 reader
->page
= *data_page
;
2989 local_set(&reader
->write
, 0);
2990 local_set(&reader
->entries
, 0);
2997 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
3002 EXPORT_SYMBOL_GPL(ring_buffer_read_page
);
3005 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
3006 size_t cnt
, loff_t
*ppos
)
3008 unsigned long *p
= filp
->private_data
;
3012 if (test_bit(RB_BUFFERS_DISABLED_BIT
, p
))
3013 r
= sprintf(buf
, "permanently disabled\n");
3015 r
= sprintf(buf
, "%d\n", test_bit(RB_BUFFERS_ON_BIT
, p
));
3017 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3021 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
3022 size_t cnt
, loff_t
*ppos
)
3024 unsigned long *p
= filp
->private_data
;
3029 if (cnt
>= sizeof(buf
))
3032 if (copy_from_user(&buf
, ubuf
, cnt
))
3037 ret
= strict_strtoul(buf
, 10, &val
);
3042 set_bit(RB_BUFFERS_ON_BIT
, p
);
3044 clear_bit(RB_BUFFERS_ON_BIT
, p
);
3051 static const struct file_operations rb_simple_fops
= {
3052 .open
= tracing_open_generic
,
3053 .read
= rb_simple_read
,
3054 .write
= rb_simple_write
,
3058 static __init
int rb_init_debugfs(void)
3060 struct dentry
*d_tracer
;
3062 d_tracer
= tracing_init_dentry();
3064 trace_create_file("tracing_on", 0644, d_tracer
,
3065 &ring_buffer_flags
, &rb_simple_fops
);
3070 fs_initcall(rb_init_debugfs
);
3072 #ifdef CONFIG_HOTPLUG_CPU
3073 static int rb_cpu_notify(struct notifier_block
*self
,
3074 unsigned long action
, void *hcpu
)
3076 struct ring_buffer
*buffer
=
3077 container_of(self
, struct ring_buffer
, cpu_notify
);
3078 long cpu
= (long)hcpu
;
3081 case CPU_UP_PREPARE
:
3082 case CPU_UP_PREPARE_FROZEN
:
3083 if (cpu_isset(cpu
, *buffer
->cpumask
))
3086 buffer
->buffers
[cpu
] =
3087 rb_allocate_cpu_buffer(buffer
, cpu
);
3088 if (!buffer
->buffers
[cpu
]) {
3089 WARN(1, "failed to allocate ring buffer on CPU %ld\n",
3094 cpu_set(cpu
, *buffer
->cpumask
);
3096 case CPU_DOWN_PREPARE
:
3097 case CPU_DOWN_PREPARE_FROZEN
:
3100 * If we were to free the buffer, then the user would
3101 * lose any trace that was in the buffer.