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 int ring_buffer_print_page_header(struct trace_seq
*s
)
372 struct buffer_data_page field
;
375 ret
= trace_seq_printf(s
, "\tfield: u64 timestamp;\t"
376 "offset:0;\tsize:%u;\n",
377 (unsigned int)sizeof(field
.time_stamp
));
379 ret
= trace_seq_printf(s
, "\tfield: local_t commit;\t"
380 "offset:%u;\tsize:%u;\n",
381 (unsigned int)offsetof(typeof(field
), commit
),
382 (unsigned int)sizeof(field
.commit
));
384 ret
= trace_seq_printf(s
, "\tfield: char data;\t"
385 "offset:%u;\tsize:%u;\n",
386 (unsigned int)offsetof(typeof(field
), data
),
387 (unsigned int)BUF_PAGE_SIZE
);
393 * head_page == tail_page && head == tail then buffer is empty.
395 struct ring_buffer_per_cpu
{
397 struct ring_buffer
*buffer
;
398 spinlock_t reader_lock
; /* serialize readers */
400 struct lock_class_key lock_key
;
401 struct list_head pages
;
402 struct buffer_page
*head_page
; /* read from head */
403 struct buffer_page
*tail_page
; /* write to tail */
404 struct buffer_page
*commit_page
; /* committed pages */
405 struct buffer_page
*reader_page
;
406 unsigned long nmi_dropped
;
407 unsigned long commit_overrun
;
408 unsigned long overrun
;
413 atomic_t record_disabled
;
420 atomic_t record_disabled
;
421 cpumask_var_t cpumask
;
425 struct ring_buffer_per_cpu
**buffers
;
427 #ifdef CONFIG_HOTPLUG_CPU
428 struct notifier_block cpu_notify
;
433 struct ring_buffer_iter
{
434 struct ring_buffer_per_cpu
*cpu_buffer
;
436 struct buffer_page
*head_page
;
440 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
441 #define RB_WARN_ON(buffer, cond) \
443 int _____ret = unlikely(cond); \
445 atomic_inc(&buffer->record_disabled); \
451 /* Up this if you want to test the TIME_EXTENTS and normalization */
452 #define DEBUG_SHIFT 0
454 u64
ring_buffer_time_stamp(struct ring_buffer
*buffer
, int cpu
)
458 preempt_disable_notrace();
459 /* shift to debug/test normalization and TIME_EXTENTS */
460 time
= buffer
->clock() << DEBUG_SHIFT
;
461 preempt_enable_no_resched_notrace();
465 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp
);
467 void ring_buffer_normalize_time_stamp(struct ring_buffer
*buffer
,
470 /* Just stupid testing the normalize function and deltas */
473 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp
);
476 * check_pages - integrity check of buffer pages
477 * @cpu_buffer: CPU buffer with pages to test
479 * As a safety measure we check to make sure the data pages have not
482 static int rb_check_pages(struct ring_buffer_per_cpu
*cpu_buffer
)
484 struct list_head
*head
= &cpu_buffer
->pages
;
485 struct buffer_page
*bpage
, *tmp
;
487 if (RB_WARN_ON(cpu_buffer
, head
->next
->prev
!= head
))
489 if (RB_WARN_ON(cpu_buffer
, head
->prev
->next
!= head
))
492 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
493 if (RB_WARN_ON(cpu_buffer
,
494 bpage
->list
.next
->prev
!= &bpage
->list
))
496 if (RB_WARN_ON(cpu_buffer
,
497 bpage
->list
.prev
->next
!= &bpage
->list
))
504 static int rb_allocate_pages(struct ring_buffer_per_cpu
*cpu_buffer
,
507 struct list_head
*head
= &cpu_buffer
->pages
;
508 struct buffer_page
*bpage
, *tmp
;
513 for (i
= 0; i
< nr_pages
; i
++) {
514 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
515 GFP_KERNEL
, cpu_to_node(cpu_buffer
->cpu
));
518 list_add(&bpage
->list
, &pages
);
520 addr
= __get_free_page(GFP_KERNEL
);
523 bpage
->page
= (void *)addr
;
524 rb_init_page(bpage
->page
);
527 list_splice(&pages
, head
);
529 rb_check_pages(cpu_buffer
);
534 list_for_each_entry_safe(bpage
, tmp
, &pages
, list
) {
535 list_del_init(&bpage
->list
);
536 free_buffer_page(bpage
);
541 static struct ring_buffer_per_cpu
*
542 rb_allocate_cpu_buffer(struct ring_buffer
*buffer
, int cpu
)
544 struct ring_buffer_per_cpu
*cpu_buffer
;
545 struct buffer_page
*bpage
;
549 cpu_buffer
= kzalloc_node(ALIGN(sizeof(*cpu_buffer
), cache_line_size()),
550 GFP_KERNEL
, cpu_to_node(cpu
));
554 cpu_buffer
->cpu
= cpu
;
555 cpu_buffer
->buffer
= buffer
;
556 spin_lock_init(&cpu_buffer
->reader_lock
);
557 cpu_buffer
->lock
= (raw_spinlock_t
)__RAW_SPIN_LOCK_UNLOCKED
;
558 INIT_LIST_HEAD(&cpu_buffer
->pages
);
560 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
561 GFP_KERNEL
, cpu_to_node(cpu
));
563 goto fail_free_buffer
;
565 cpu_buffer
->reader_page
= bpage
;
566 addr
= __get_free_page(GFP_KERNEL
);
568 goto fail_free_reader
;
569 bpage
->page
= (void *)addr
;
570 rb_init_page(bpage
->page
);
572 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
574 ret
= rb_allocate_pages(cpu_buffer
, buffer
->pages
);
576 goto fail_free_reader
;
578 cpu_buffer
->head_page
579 = list_entry(cpu_buffer
->pages
.next
, struct buffer_page
, list
);
580 cpu_buffer
->tail_page
= cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
585 free_buffer_page(cpu_buffer
->reader_page
);
592 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu
*cpu_buffer
)
594 struct list_head
*head
= &cpu_buffer
->pages
;
595 struct buffer_page
*bpage
, *tmp
;
597 free_buffer_page(cpu_buffer
->reader_page
);
599 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
600 list_del_init(&bpage
->list
);
601 free_buffer_page(bpage
);
607 * Causes compile errors if the struct buffer_page gets bigger
608 * than the struct page.
610 extern int ring_buffer_page_too_big(void);
612 #ifdef CONFIG_HOTPLUG_CPU
613 static int rb_cpu_notify(struct notifier_block
*self
,
614 unsigned long action
, void *hcpu
);
618 * ring_buffer_alloc - allocate a new ring_buffer
619 * @size: the size in bytes per cpu that is needed.
620 * @flags: attributes to set for the ring buffer.
622 * Currently the only flag that is available is the RB_FL_OVERWRITE
623 * flag. This flag means that the buffer will overwrite old data
624 * when the buffer wraps. If this flag is not set, the buffer will
625 * drop data when the tail hits the head.
627 struct ring_buffer
*ring_buffer_alloc(unsigned long size
, unsigned flags
)
629 struct ring_buffer
*buffer
;
633 /* Paranoid! Optimizes out when all is well */
634 if (sizeof(struct buffer_page
) > sizeof(struct page
))
635 ring_buffer_page_too_big();
638 /* keep it in its own cache line */
639 buffer
= kzalloc(ALIGN(sizeof(*buffer
), cache_line_size()),
644 if (!alloc_cpumask_var(&buffer
->cpumask
, GFP_KERNEL
))
645 goto fail_free_buffer
;
647 buffer
->pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
648 buffer
->flags
= flags
;
649 buffer
->clock
= trace_clock_local
;
651 /* need at least two pages */
652 if (buffer
->pages
== 1)
656 * In case of non-hotplug cpu, if the ring-buffer is allocated
657 * in early initcall, it will not be notified of secondary cpus.
658 * In that off case, we need to allocate for all possible cpus.
660 #ifdef CONFIG_HOTPLUG_CPU
662 cpumask_copy(buffer
->cpumask
, cpu_online_mask
);
664 cpumask_copy(buffer
->cpumask
, cpu_possible_mask
);
666 buffer
->cpus
= nr_cpu_ids
;
668 bsize
= sizeof(void *) * nr_cpu_ids
;
669 buffer
->buffers
= kzalloc(ALIGN(bsize
, cache_line_size()),
671 if (!buffer
->buffers
)
672 goto fail_free_cpumask
;
674 for_each_buffer_cpu(buffer
, cpu
) {
675 buffer
->buffers
[cpu
] =
676 rb_allocate_cpu_buffer(buffer
, cpu
);
677 if (!buffer
->buffers
[cpu
])
678 goto fail_free_buffers
;
681 #ifdef CONFIG_HOTPLUG_CPU
682 buffer
->cpu_notify
.notifier_call
= rb_cpu_notify
;
683 buffer
->cpu_notify
.priority
= 0;
684 register_cpu_notifier(&buffer
->cpu_notify
);
688 mutex_init(&buffer
->mutex
);
693 for_each_buffer_cpu(buffer
, cpu
) {
694 if (buffer
->buffers
[cpu
])
695 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
697 kfree(buffer
->buffers
);
700 free_cpumask_var(buffer
->cpumask
);
707 EXPORT_SYMBOL_GPL(ring_buffer_alloc
);
710 * ring_buffer_free - free a ring buffer.
711 * @buffer: the buffer to free.
714 ring_buffer_free(struct ring_buffer
*buffer
)
720 #ifdef CONFIG_HOTPLUG_CPU
721 unregister_cpu_notifier(&buffer
->cpu_notify
);
724 for_each_buffer_cpu(buffer
, cpu
)
725 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
729 free_cpumask_var(buffer
->cpumask
);
733 EXPORT_SYMBOL_GPL(ring_buffer_free
);
735 void ring_buffer_set_clock(struct ring_buffer
*buffer
,
738 buffer
->clock
= clock
;
741 static void rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
);
744 rb_remove_pages(struct ring_buffer_per_cpu
*cpu_buffer
, unsigned nr_pages
)
746 struct buffer_page
*bpage
;
750 atomic_inc(&cpu_buffer
->record_disabled
);
753 for (i
= 0; i
< nr_pages
; i
++) {
754 if (RB_WARN_ON(cpu_buffer
, list_empty(&cpu_buffer
->pages
)))
756 p
= cpu_buffer
->pages
.next
;
757 bpage
= list_entry(p
, struct buffer_page
, list
);
758 list_del_init(&bpage
->list
);
759 free_buffer_page(bpage
);
761 if (RB_WARN_ON(cpu_buffer
, list_empty(&cpu_buffer
->pages
)))
764 rb_reset_cpu(cpu_buffer
);
766 rb_check_pages(cpu_buffer
);
768 atomic_dec(&cpu_buffer
->record_disabled
);
773 rb_insert_pages(struct ring_buffer_per_cpu
*cpu_buffer
,
774 struct list_head
*pages
, unsigned nr_pages
)
776 struct buffer_page
*bpage
;
780 atomic_inc(&cpu_buffer
->record_disabled
);
783 for (i
= 0; i
< nr_pages
; i
++) {
784 if (RB_WARN_ON(cpu_buffer
, list_empty(pages
)))
787 bpage
= list_entry(p
, struct buffer_page
, list
);
788 list_del_init(&bpage
->list
);
789 list_add_tail(&bpage
->list
, &cpu_buffer
->pages
);
791 rb_reset_cpu(cpu_buffer
);
793 rb_check_pages(cpu_buffer
);
795 atomic_dec(&cpu_buffer
->record_disabled
);
799 * ring_buffer_resize - resize the ring buffer
800 * @buffer: the buffer to resize.
801 * @size: the new size.
803 * The tracer is responsible for making sure that the buffer is
804 * not being used while changing the size.
805 * Note: We may be able to change the above requirement by using
806 * RCU synchronizations.
808 * Minimum size is 2 * BUF_PAGE_SIZE.
810 * Returns -1 on failure.
812 int ring_buffer_resize(struct ring_buffer
*buffer
, unsigned long size
)
814 struct ring_buffer_per_cpu
*cpu_buffer
;
815 unsigned nr_pages
, rm_pages
, new_pages
;
816 struct buffer_page
*bpage
, *tmp
;
817 unsigned long buffer_size
;
823 * Always succeed at resizing a non-existent buffer:
828 size
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
829 size
*= BUF_PAGE_SIZE
;
830 buffer_size
= buffer
->pages
* BUF_PAGE_SIZE
;
832 /* we need a minimum of two pages */
833 if (size
< BUF_PAGE_SIZE
* 2)
834 size
= BUF_PAGE_SIZE
* 2;
836 if (size
== buffer_size
)
839 mutex_lock(&buffer
->mutex
);
842 nr_pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
844 if (size
< buffer_size
) {
846 /* easy case, just free pages */
847 if (RB_WARN_ON(buffer
, nr_pages
>= buffer
->pages
))
850 rm_pages
= buffer
->pages
- nr_pages
;
852 for_each_buffer_cpu(buffer
, cpu
) {
853 cpu_buffer
= buffer
->buffers
[cpu
];
854 rb_remove_pages(cpu_buffer
, rm_pages
);
860 * This is a bit more difficult. We only want to add pages
861 * when we can allocate enough for all CPUs. We do this
862 * by allocating all the pages and storing them on a local
863 * link list. If we succeed in our allocation, then we
864 * add these pages to the cpu_buffers. Otherwise we just free
865 * them all and return -ENOMEM;
867 if (RB_WARN_ON(buffer
, nr_pages
<= buffer
->pages
))
870 new_pages
= nr_pages
- buffer
->pages
;
872 for_each_buffer_cpu(buffer
, cpu
) {
873 for (i
= 0; i
< new_pages
; i
++) {
874 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
),
876 GFP_KERNEL
, cpu_to_node(cpu
));
879 list_add(&bpage
->list
, &pages
);
880 addr
= __get_free_page(GFP_KERNEL
);
883 bpage
->page
= (void *)addr
;
884 rb_init_page(bpage
->page
);
888 for_each_buffer_cpu(buffer
, cpu
) {
889 cpu_buffer
= buffer
->buffers
[cpu
];
890 rb_insert_pages(cpu_buffer
, &pages
, new_pages
);
893 if (RB_WARN_ON(buffer
, !list_empty(&pages
)))
897 buffer
->pages
= nr_pages
;
899 mutex_unlock(&buffer
->mutex
);
904 list_for_each_entry_safe(bpage
, tmp
, &pages
, list
) {
905 list_del_init(&bpage
->list
);
906 free_buffer_page(bpage
);
909 mutex_unlock(&buffer
->mutex
);
913 * Something went totally wrong, and we are too paranoid
914 * to even clean up the mess.
918 mutex_unlock(&buffer
->mutex
);
921 EXPORT_SYMBOL_GPL(ring_buffer_resize
);
924 __rb_data_page_index(struct buffer_data_page
*bpage
, unsigned index
)
926 return bpage
->data
+ index
;
929 static inline void *__rb_page_index(struct buffer_page
*bpage
, unsigned index
)
931 return bpage
->page
->data
+ index
;
934 static inline struct ring_buffer_event
*
935 rb_reader_event(struct ring_buffer_per_cpu
*cpu_buffer
)
937 return __rb_page_index(cpu_buffer
->reader_page
,
938 cpu_buffer
->reader_page
->read
);
941 static inline struct ring_buffer_event
*
942 rb_head_event(struct ring_buffer_per_cpu
*cpu_buffer
)
944 return __rb_page_index(cpu_buffer
->head_page
,
945 cpu_buffer
->head_page
->read
);
948 static inline struct ring_buffer_event
*
949 rb_iter_head_event(struct ring_buffer_iter
*iter
)
951 return __rb_page_index(iter
->head_page
, iter
->head
);
954 static inline unsigned rb_page_write(struct buffer_page
*bpage
)
956 return local_read(&bpage
->write
);
959 static inline unsigned rb_page_commit(struct buffer_page
*bpage
)
961 return local_read(&bpage
->page
->commit
);
964 /* Size is determined by what has been commited */
965 static inline unsigned rb_page_size(struct buffer_page
*bpage
)
967 return rb_page_commit(bpage
);
970 static inline unsigned
971 rb_commit_index(struct ring_buffer_per_cpu
*cpu_buffer
)
973 return rb_page_commit(cpu_buffer
->commit_page
);
976 static inline unsigned rb_head_size(struct ring_buffer_per_cpu
*cpu_buffer
)
978 return rb_page_commit(cpu_buffer
->head_page
);
981 static inline void rb_inc_page(struct ring_buffer_per_cpu
*cpu_buffer
,
982 struct buffer_page
**bpage
)
984 struct list_head
*p
= (*bpage
)->list
.next
;
986 if (p
== &cpu_buffer
->pages
)
989 *bpage
= list_entry(p
, struct buffer_page
, list
);
992 static inline unsigned
993 rb_event_index(struct ring_buffer_event
*event
)
995 unsigned long addr
= (unsigned long)event
;
997 return (addr
& ~PAGE_MASK
) - (PAGE_SIZE
- BUF_PAGE_SIZE
);
1001 rb_is_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
1002 struct ring_buffer_event
*event
)
1004 unsigned long addr
= (unsigned long)event
;
1005 unsigned long index
;
1007 index
= rb_event_index(event
);
1010 return cpu_buffer
->commit_page
->page
== (void *)addr
&&
1011 rb_commit_index(cpu_buffer
) == index
;
1015 rb_set_commit_event(struct ring_buffer_per_cpu
*cpu_buffer
,
1016 struct ring_buffer_event
*event
)
1018 unsigned long addr
= (unsigned long)event
;
1019 unsigned long index
;
1021 index
= rb_event_index(event
);
1024 while (cpu_buffer
->commit_page
->page
!= (void *)addr
) {
1025 if (RB_WARN_ON(cpu_buffer
,
1026 cpu_buffer
->commit_page
== cpu_buffer
->tail_page
))
1028 cpu_buffer
->commit_page
->page
->commit
=
1029 cpu_buffer
->commit_page
->write
;
1030 rb_inc_page(cpu_buffer
, &cpu_buffer
->commit_page
);
1031 cpu_buffer
->write_stamp
=
1032 cpu_buffer
->commit_page
->page
->time_stamp
;
1035 /* Now set the commit to the event's index */
1036 local_set(&cpu_buffer
->commit_page
->page
->commit
, index
);
1040 rb_set_commit_to_write(struct ring_buffer_per_cpu
*cpu_buffer
)
1043 * We only race with interrupts and NMIs on this CPU.
1044 * If we own the commit event, then we can commit
1045 * all others that interrupted us, since the interruptions
1046 * are in stack format (they finish before they come
1047 * back to us). This allows us to do a simple loop to
1048 * assign the commit to the tail.
1051 while (cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
) {
1052 cpu_buffer
->commit_page
->page
->commit
=
1053 cpu_buffer
->commit_page
->write
;
1054 rb_inc_page(cpu_buffer
, &cpu_buffer
->commit_page
);
1055 cpu_buffer
->write_stamp
=
1056 cpu_buffer
->commit_page
->page
->time_stamp
;
1057 /* add barrier to keep gcc from optimizing too much */
1060 while (rb_commit_index(cpu_buffer
) !=
1061 rb_page_write(cpu_buffer
->commit_page
)) {
1062 cpu_buffer
->commit_page
->page
->commit
=
1063 cpu_buffer
->commit_page
->write
;
1067 /* again, keep gcc from optimizing */
1071 * If an interrupt came in just after the first while loop
1072 * and pushed the tail page forward, we will be left with
1073 * a dangling commit that will never go forward.
1075 if (unlikely(cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
))
1079 static void rb_reset_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
1081 cpu_buffer
->read_stamp
= cpu_buffer
->reader_page
->page
->time_stamp
;
1082 cpu_buffer
->reader_page
->read
= 0;
1085 static void rb_inc_iter(struct ring_buffer_iter
*iter
)
1087 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
1090 * The iterator could be on the reader page (it starts there).
1091 * But the head could have moved, since the reader was
1092 * found. Check for this case and assign the iterator
1093 * to the head page instead of next.
1095 if (iter
->head_page
== cpu_buffer
->reader_page
)
1096 iter
->head_page
= cpu_buffer
->head_page
;
1098 rb_inc_page(cpu_buffer
, &iter
->head_page
);
1100 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
1105 * ring_buffer_update_event - update event type and data
1106 * @event: the even to update
1107 * @type: the type of event
1108 * @length: the size of the event field in the ring buffer
1110 * Update the type and data fields of the event. The length
1111 * is the actual size that is written to the ring buffer,
1112 * and with this, we can determine what to place into the
1116 rb_update_event(struct ring_buffer_event
*event
,
1117 unsigned type
, unsigned length
)
1119 event
->type_len
= type
;
1123 case RINGBUF_TYPE_PADDING
:
1124 case RINGBUF_TYPE_TIME_EXTEND
:
1125 case RINGBUF_TYPE_TIME_STAMP
:
1129 length
-= RB_EVNT_HDR_SIZE
;
1130 if (length
> RB_MAX_SMALL_DATA
)
1131 event
->array
[0] = length
;
1133 event
->type_len
= DIV_ROUND_UP(length
, RB_ALIGNMENT
);
1140 static unsigned rb_calculate_event_length(unsigned length
)
1142 struct ring_buffer_event event
; /* Used only for sizeof array */
1144 /* zero length can cause confusions */
1148 if (length
> RB_MAX_SMALL_DATA
)
1149 length
+= sizeof(event
.array
[0]);
1151 length
+= RB_EVNT_HDR_SIZE
;
1152 length
= ALIGN(length
, RB_ALIGNMENT
);
1157 static struct ring_buffer_event
*
1158 __rb_reserve_next(struct ring_buffer_per_cpu
*cpu_buffer
,
1159 unsigned type
, unsigned long length
, u64
*ts
)
1161 struct buffer_page
*tail_page
, *head_page
, *reader_page
, *commit_page
;
1162 unsigned long tail
, write
;
1163 struct ring_buffer
*buffer
= cpu_buffer
->buffer
;
1164 struct ring_buffer_event
*event
;
1165 unsigned long flags
;
1166 bool lock_taken
= false;
1168 commit_page
= cpu_buffer
->commit_page
;
1169 /* we just need to protect against interrupts */
1171 tail_page
= cpu_buffer
->tail_page
;
1172 write
= local_add_return(length
, &tail_page
->write
);
1173 tail
= write
- length
;
1175 /* See if we shot pass the end of this buffer page */
1176 if (write
> BUF_PAGE_SIZE
) {
1177 struct buffer_page
*next_page
= tail_page
;
1179 local_irq_save(flags
);
1181 * Since the write to the buffer is still not
1182 * fully lockless, we must be careful with NMIs.
1183 * The locks in the writers are taken when a write
1184 * crosses to a new page. The locks protect against
1185 * races with the readers (this will soon be fixed
1186 * with a lockless solution).
1188 * Because we can not protect against NMIs, and we
1189 * want to keep traces reentrant, we need to manage
1190 * what happens when we are in an NMI.
1192 * NMIs can happen after we take the lock.
1193 * If we are in an NMI, only take the lock
1194 * if it is not already taken. Otherwise
1197 if (unlikely(in_nmi())) {
1198 if (!__raw_spin_trylock(&cpu_buffer
->lock
)) {
1199 cpu_buffer
->nmi_dropped
++;
1203 __raw_spin_lock(&cpu_buffer
->lock
);
1207 rb_inc_page(cpu_buffer
, &next_page
);
1209 head_page
= cpu_buffer
->head_page
;
1210 reader_page
= cpu_buffer
->reader_page
;
1212 /* we grabbed the lock before incrementing */
1213 if (RB_WARN_ON(cpu_buffer
, next_page
== reader_page
))
1217 * If for some reason, we had an interrupt storm that made
1218 * it all the way around the buffer, bail, and warn
1221 if (unlikely(next_page
== commit_page
)) {
1222 cpu_buffer
->commit_overrun
++;
1226 if (next_page
== head_page
) {
1227 if (!(buffer
->flags
& RB_FL_OVERWRITE
))
1230 /* tail_page has not moved yet? */
1231 if (tail_page
== cpu_buffer
->tail_page
) {
1232 /* count overflows */
1233 cpu_buffer
->overrun
+=
1234 local_read(&head_page
->entries
);
1236 rb_inc_page(cpu_buffer
, &head_page
);
1237 cpu_buffer
->head_page
= head_page
;
1238 cpu_buffer
->head_page
->read
= 0;
1243 * If the tail page is still the same as what we think
1244 * it is, then it is up to us to update the tail
1247 if (tail_page
== cpu_buffer
->tail_page
) {
1248 local_set(&next_page
->write
, 0);
1249 local_set(&next_page
->entries
, 0);
1250 local_set(&next_page
->page
->commit
, 0);
1251 cpu_buffer
->tail_page
= next_page
;
1253 /* reread the time stamp */
1254 *ts
= ring_buffer_time_stamp(buffer
, cpu_buffer
->cpu
);
1255 cpu_buffer
->tail_page
->page
->time_stamp
= *ts
;
1259 * The actual tail page has moved forward.
1261 if (tail
< BUF_PAGE_SIZE
) {
1262 /* Mark the rest of the page with padding */
1263 event
= __rb_page_index(tail_page
, tail
);
1264 rb_event_set_padding(event
);
1267 if (tail
<= BUF_PAGE_SIZE
)
1268 /* Set the write back to the previous setting */
1269 local_set(&tail_page
->write
, tail
);
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
);
1287 /* We reserved something on the buffer */
1289 if (RB_WARN_ON(cpu_buffer
, write
> BUF_PAGE_SIZE
))
1292 event
= __rb_page_index(tail_page
, tail
);
1293 rb_update_event(event
, type
, length
);
1295 /* The passed in type is zero for DATA */
1297 local_inc(&tail_page
->entries
);
1300 * If this is a commit and the tail is zero, then update
1301 * this page's time stamp.
1303 if (!tail
&& rb_is_commit(cpu_buffer
, event
))
1304 cpu_buffer
->commit_page
->page
->time_stamp
= *ts
;
1310 if (tail
<= BUF_PAGE_SIZE
)
1311 local_set(&tail_page
->write
, tail
);
1313 if (likely(lock_taken
))
1314 __raw_spin_unlock(&cpu_buffer
->lock
);
1315 local_irq_restore(flags
);
1320 rb_add_time_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
1321 u64
*ts
, u64
*delta
)
1323 struct ring_buffer_event
*event
;
1327 if (unlikely(*delta
> (1ULL << 59) && !once
++)) {
1328 printk(KERN_WARNING
"Delta way too big! %llu"
1329 " ts=%llu write stamp = %llu\n",
1330 (unsigned long long)*delta
,
1331 (unsigned long long)*ts
,
1332 (unsigned long long)cpu_buffer
->write_stamp
);
1337 * The delta is too big, we to add a
1340 event
= __rb_reserve_next(cpu_buffer
,
1341 RINGBUF_TYPE_TIME_EXTEND
,
1347 if (PTR_ERR(event
) == -EAGAIN
)
1350 /* Only a commited time event can update the write stamp */
1351 if (rb_is_commit(cpu_buffer
, event
)) {
1353 * If this is the first on the page, then we need to
1354 * update the page itself, and just put in a zero.
1356 if (rb_event_index(event
)) {
1357 event
->time_delta
= *delta
& TS_MASK
;
1358 event
->array
[0] = *delta
>> TS_SHIFT
;
1360 cpu_buffer
->commit_page
->page
->time_stamp
= *ts
;
1361 event
->time_delta
= 0;
1362 event
->array
[0] = 0;
1364 cpu_buffer
->write_stamp
= *ts
;
1365 /* let the caller know this was the commit */
1368 /* Darn, this is just wasted space */
1369 event
->time_delta
= 0;
1370 event
->array
[0] = 0;
1379 static struct ring_buffer_event
*
1380 rb_reserve_next_event(struct ring_buffer_per_cpu
*cpu_buffer
,
1381 unsigned type
, unsigned long length
)
1383 struct ring_buffer_event
*event
;
1390 * We allow for interrupts to reenter here and do a trace.
1391 * If one does, it will cause this original code to loop
1392 * back here. Even with heavy interrupts happening, this
1393 * should only happen a few times in a row. If this happens
1394 * 1000 times in a row, there must be either an interrupt
1395 * storm or we have something buggy.
1398 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 1000))
1401 ts
= ring_buffer_time_stamp(cpu_buffer
->buffer
, cpu_buffer
->cpu
);
1404 * Only the first commit can update the timestamp.
1405 * Yes there is a race here. If an interrupt comes in
1406 * just after the conditional and it traces too, then it
1407 * will also check the deltas. More than one timestamp may
1408 * also be made. But only the entry that did the actual
1409 * commit will be something other than zero.
1411 if (cpu_buffer
->tail_page
== cpu_buffer
->commit_page
&&
1412 rb_page_write(cpu_buffer
->tail_page
) ==
1413 rb_commit_index(cpu_buffer
)) {
1415 delta
= ts
- cpu_buffer
->write_stamp
;
1417 /* make sure this delta is calculated here */
1420 /* Did the write stamp get updated already? */
1421 if (unlikely(ts
< cpu_buffer
->write_stamp
))
1424 if (test_time_stamp(delta
)) {
1426 commit
= rb_add_time_stamp(cpu_buffer
, &ts
, &delta
);
1428 if (commit
== -EBUSY
)
1431 if (commit
== -EAGAIN
)
1434 RB_WARN_ON(cpu_buffer
, commit
< 0);
1437 /* Non commits have zero deltas */
1440 event
= __rb_reserve_next(cpu_buffer
, type
, length
, &ts
);
1441 if (PTR_ERR(event
) == -EAGAIN
)
1445 if (unlikely(commit
))
1447 * Ouch! We needed a timestamp and it was commited. But
1448 * we didn't get our event reserved.
1450 rb_set_commit_to_write(cpu_buffer
);
1455 * If the timestamp was commited, make the commit our entry
1456 * now so that we will update it when needed.
1459 rb_set_commit_event(cpu_buffer
, event
);
1460 else if (!rb_is_commit(cpu_buffer
, event
))
1463 event
->time_delta
= delta
;
1468 #define TRACE_RECURSIVE_DEPTH 16
1470 static int trace_recursive_lock(void)
1472 current
->trace_recursion
++;
1474 if (likely(current
->trace_recursion
< TRACE_RECURSIVE_DEPTH
))
1477 /* Disable all tracing before we do anything else */
1478 tracing_off_permanent();
1480 printk_once(KERN_WARNING
"Tracing recursion: depth[%ld]:"
1481 "HC[%lu]:SC[%lu]:NMI[%lu]\n",
1482 current
->trace_recursion
,
1483 hardirq_count() >> HARDIRQ_SHIFT
,
1484 softirq_count() >> SOFTIRQ_SHIFT
,
1491 static void trace_recursive_unlock(void)
1493 WARN_ON_ONCE(!current
->trace_recursion
);
1495 current
->trace_recursion
--;
1498 static DEFINE_PER_CPU(int, rb_need_resched
);
1501 * ring_buffer_lock_reserve - reserve a part of the buffer
1502 * @buffer: the ring buffer to reserve from
1503 * @length: the length of the data to reserve (excluding event header)
1505 * Returns a reseverd event on the ring buffer to copy directly to.
1506 * The user of this interface will need to get the body to write into
1507 * and can use the ring_buffer_event_data() interface.
1509 * The length is the length of the data needed, not the event length
1510 * which also includes the event header.
1512 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
1513 * If NULL is returned, then nothing has been allocated or locked.
1515 struct ring_buffer_event
*
1516 ring_buffer_lock_reserve(struct ring_buffer
*buffer
, unsigned long length
)
1518 struct ring_buffer_per_cpu
*cpu_buffer
;
1519 struct ring_buffer_event
*event
;
1522 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
1525 if (atomic_read(&buffer
->record_disabled
))
1528 /* If we are tracing schedule, we don't want to recurse */
1529 resched
= ftrace_preempt_disable();
1531 if (trace_recursive_lock())
1534 cpu
= raw_smp_processor_id();
1536 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1539 cpu_buffer
= buffer
->buffers
[cpu
];
1541 if (atomic_read(&cpu_buffer
->record_disabled
))
1544 length
= rb_calculate_event_length(length
);
1545 if (length
> BUF_PAGE_SIZE
)
1548 event
= rb_reserve_next_event(cpu_buffer
, 0, length
);
1553 * Need to store resched state on this cpu.
1554 * Only the first needs to.
1557 if (preempt_count() == 1)
1558 per_cpu(rb_need_resched
, cpu
) = resched
;
1563 trace_recursive_unlock();
1566 ftrace_preempt_enable(resched
);
1569 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve
);
1571 static void rb_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
1572 struct ring_buffer_event
*event
)
1574 local_inc(&cpu_buffer
->entries
);
1576 /* Only process further if we own the commit */
1577 if (!rb_is_commit(cpu_buffer
, event
))
1580 cpu_buffer
->write_stamp
+= event
->time_delta
;
1582 rb_set_commit_to_write(cpu_buffer
);
1586 * ring_buffer_unlock_commit - commit a reserved
1587 * @buffer: The buffer to commit to
1588 * @event: The event pointer to commit.
1590 * This commits the data to the ring buffer, and releases any locks held.
1592 * Must be paired with ring_buffer_lock_reserve.
1594 int ring_buffer_unlock_commit(struct ring_buffer
*buffer
,
1595 struct ring_buffer_event
*event
)
1597 struct ring_buffer_per_cpu
*cpu_buffer
;
1598 int cpu
= raw_smp_processor_id();
1600 cpu_buffer
= buffer
->buffers
[cpu
];
1602 rb_commit(cpu_buffer
, event
);
1604 trace_recursive_unlock();
1607 * Only the last preempt count needs to restore preemption.
1609 if (preempt_count() == 1)
1610 ftrace_preempt_enable(per_cpu(rb_need_resched
, cpu
));
1612 preempt_enable_no_resched_notrace();
1616 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit
);
1618 static inline void rb_event_discard(struct ring_buffer_event
*event
)
1620 /* array[0] holds the actual length for the discarded event */
1621 event
->array
[0] = rb_event_data_length(event
) - RB_EVNT_HDR_SIZE
;
1622 event
->type_len
= RINGBUF_TYPE_PADDING
;
1623 /* time delta must be non zero */
1624 if (!event
->time_delta
)
1625 event
->time_delta
= 1;
1629 * ring_buffer_event_discard - discard any event in the ring buffer
1630 * @event: the event to discard
1632 * Sometimes a event that is in the ring buffer needs to be ignored.
1633 * This function lets the user discard an event in the ring buffer
1634 * and then that event will not be read later.
1636 * Note, it is up to the user to be careful with this, and protect
1637 * against races. If the user discards an event that has been consumed
1638 * it is possible that it could corrupt the ring buffer.
1640 void ring_buffer_event_discard(struct ring_buffer_event
*event
)
1642 rb_event_discard(event
);
1644 EXPORT_SYMBOL_GPL(ring_buffer_event_discard
);
1647 * ring_buffer_commit_discard - discard an event that has not been committed
1648 * @buffer: the ring buffer
1649 * @event: non committed event to discard
1651 * This is similar to ring_buffer_event_discard but must only be
1652 * performed on an event that has not been committed yet. The difference
1653 * is that this will also try to free the event from the ring buffer
1654 * if another event has not been added behind it.
1656 * If another event has been added behind it, it will set the event
1657 * up as discarded, and perform the commit.
1659 * If this function is called, do not call ring_buffer_unlock_commit on
1662 void ring_buffer_discard_commit(struct ring_buffer
*buffer
,
1663 struct ring_buffer_event
*event
)
1665 struct ring_buffer_per_cpu
*cpu_buffer
;
1666 unsigned long new_index
, old_index
;
1667 struct buffer_page
*bpage
;
1668 unsigned long index
;
1672 /* The event is discarded regardless */
1673 rb_event_discard(event
);
1676 * This must only be called if the event has not been
1677 * committed yet. Thus we can assume that preemption
1678 * is still disabled.
1680 RB_WARN_ON(buffer
, !preempt_count());
1682 cpu
= smp_processor_id();
1683 cpu_buffer
= buffer
->buffers
[cpu
];
1685 new_index
= rb_event_index(event
);
1686 old_index
= new_index
+ rb_event_length(event
);
1687 addr
= (unsigned long)event
;
1690 bpage
= cpu_buffer
->tail_page
;
1692 if (bpage
== (void *)addr
&& rb_page_write(bpage
) == old_index
) {
1694 * This is on the tail page. It is possible that
1695 * a write could come in and move the tail page
1696 * and write to the next page. That is fine
1697 * because we just shorten what is on this page.
1699 index
= local_cmpxchg(&bpage
->write
, old_index
, new_index
);
1700 if (index
== old_index
)
1705 * The commit is still visible by the reader, so we
1706 * must increment entries.
1708 local_inc(&cpu_buffer
->entries
);
1711 * If a write came in and pushed the tail page
1712 * we still need to update the commit pointer
1713 * if we were the commit.
1715 if (rb_is_commit(cpu_buffer
, event
))
1716 rb_set_commit_to_write(cpu_buffer
);
1718 trace_recursive_unlock();
1721 * Only the last preempt count needs to restore preemption.
1723 if (preempt_count() == 1)
1724 ftrace_preempt_enable(per_cpu(rb_need_resched
, cpu
));
1726 preempt_enable_no_resched_notrace();
1729 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit
);
1732 * ring_buffer_write - write data to the buffer without reserving
1733 * @buffer: The ring buffer to write to.
1734 * @length: The length of the data being written (excluding the event header)
1735 * @data: The data to write to the buffer.
1737 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
1738 * one function. If you already have the data to write to the buffer, it
1739 * may be easier to simply call this function.
1741 * Note, like ring_buffer_lock_reserve, the length is the length of the data
1742 * and not the length of the event which would hold the header.
1744 int ring_buffer_write(struct ring_buffer
*buffer
,
1745 unsigned long length
,
1748 struct ring_buffer_per_cpu
*cpu_buffer
;
1749 struct ring_buffer_event
*event
;
1750 unsigned long event_length
;
1755 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
1758 if (atomic_read(&buffer
->record_disabled
))
1761 resched
= ftrace_preempt_disable();
1763 cpu
= raw_smp_processor_id();
1765 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1768 cpu_buffer
= buffer
->buffers
[cpu
];
1770 if (atomic_read(&cpu_buffer
->record_disabled
))
1773 event_length
= rb_calculate_event_length(length
);
1774 event
= rb_reserve_next_event(cpu_buffer
, 0, event_length
);
1778 body
= rb_event_data(event
);
1780 memcpy(body
, data
, length
);
1782 rb_commit(cpu_buffer
, event
);
1786 ftrace_preempt_enable(resched
);
1790 EXPORT_SYMBOL_GPL(ring_buffer_write
);
1792 static int rb_per_cpu_empty(struct ring_buffer_per_cpu
*cpu_buffer
)
1794 struct buffer_page
*reader
= cpu_buffer
->reader_page
;
1795 struct buffer_page
*head
= cpu_buffer
->head_page
;
1796 struct buffer_page
*commit
= cpu_buffer
->commit_page
;
1798 return reader
->read
== rb_page_commit(reader
) &&
1799 (commit
== reader
||
1801 head
->read
== rb_page_commit(commit
)));
1805 * ring_buffer_record_disable - stop all writes into the buffer
1806 * @buffer: The ring buffer to stop writes to.
1808 * This prevents all writes to the buffer. Any attempt to write
1809 * to the buffer after this will fail and return NULL.
1811 * The caller should call synchronize_sched() after this.
1813 void ring_buffer_record_disable(struct ring_buffer
*buffer
)
1815 atomic_inc(&buffer
->record_disabled
);
1817 EXPORT_SYMBOL_GPL(ring_buffer_record_disable
);
1820 * ring_buffer_record_enable - enable writes to the buffer
1821 * @buffer: The ring buffer to enable writes
1823 * Note, multiple disables will need the same number of enables
1824 * to truely enable the writing (much like preempt_disable).
1826 void ring_buffer_record_enable(struct ring_buffer
*buffer
)
1828 atomic_dec(&buffer
->record_disabled
);
1830 EXPORT_SYMBOL_GPL(ring_buffer_record_enable
);
1833 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
1834 * @buffer: The ring buffer to stop writes to.
1835 * @cpu: The CPU buffer to stop
1837 * This prevents all writes to the buffer. Any attempt to write
1838 * to the buffer after this will fail and return NULL.
1840 * The caller should call synchronize_sched() after this.
1842 void ring_buffer_record_disable_cpu(struct ring_buffer
*buffer
, int cpu
)
1844 struct ring_buffer_per_cpu
*cpu_buffer
;
1846 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1849 cpu_buffer
= buffer
->buffers
[cpu
];
1850 atomic_inc(&cpu_buffer
->record_disabled
);
1852 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu
);
1855 * ring_buffer_record_enable_cpu - enable writes to the buffer
1856 * @buffer: The ring buffer to enable writes
1857 * @cpu: The CPU to enable.
1859 * Note, multiple disables will need the same number of enables
1860 * to truely enable the writing (much like preempt_disable).
1862 void ring_buffer_record_enable_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_dec(&cpu_buffer
->record_disabled
);
1872 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu
);
1875 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
1876 * @buffer: The ring buffer
1877 * @cpu: The per CPU buffer to get the entries from.
1879 unsigned long ring_buffer_entries_cpu(struct ring_buffer
*buffer
, int cpu
)
1881 struct ring_buffer_per_cpu
*cpu_buffer
;
1884 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1887 cpu_buffer
= buffer
->buffers
[cpu
];
1888 ret
= (local_read(&cpu_buffer
->entries
) - cpu_buffer
->overrun
)
1893 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu
);
1896 * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
1897 * @buffer: The ring buffer
1898 * @cpu: The per CPU buffer to get the number of overruns from
1900 unsigned long ring_buffer_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
1902 struct ring_buffer_per_cpu
*cpu_buffer
;
1905 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1908 cpu_buffer
= buffer
->buffers
[cpu
];
1909 ret
= cpu_buffer
->overrun
;
1913 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu
);
1916 * ring_buffer_nmi_dropped_cpu - get the number of nmis that were dropped
1917 * @buffer: The ring buffer
1918 * @cpu: The per CPU buffer to get the number of overruns from
1920 unsigned long ring_buffer_nmi_dropped_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
->nmi_dropped
;
1933 EXPORT_SYMBOL_GPL(ring_buffer_nmi_dropped_cpu
);
1936 * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits
1937 * @buffer: The ring buffer
1938 * @cpu: The per CPU buffer to get the number of overruns from
1941 ring_buffer_commit_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
1943 struct ring_buffer_per_cpu
*cpu_buffer
;
1946 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
1949 cpu_buffer
= buffer
->buffers
[cpu
];
1950 ret
= cpu_buffer
->commit_overrun
;
1954 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu
);
1957 * ring_buffer_entries - get the number of entries in a buffer
1958 * @buffer: The ring buffer
1960 * Returns the total number of entries in the ring buffer
1963 unsigned long ring_buffer_entries(struct ring_buffer
*buffer
)
1965 struct ring_buffer_per_cpu
*cpu_buffer
;
1966 unsigned long entries
= 0;
1969 /* if you care about this being correct, lock the buffer */
1970 for_each_buffer_cpu(buffer
, cpu
) {
1971 cpu_buffer
= buffer
->buffers
[cpu
];
1972 entries
+= (local_read(&cpu_buffer
->entries
) -
1973 cpu_buffer
->overrun
) - cpu_buffer
->read
;
1978 EXPORT_SYMBOL_GPL(ring_buffer_entries
);
1981 * ring_buffer_overrun_cpu - get the number of overruns in buffer
1982 * @buffer: The ring buffer
1984 * Returns the total number of overruns in the ring buffer
1987 unsigned long ring_buffer_overruns(struct ring_buffer
*buffer
)
1989 struct ring_buffer_per_cpu
*cpu_buffer
;
1990 unsigned long overruns
= 0;
1993 /* if you care about this being correct, lock the buffer */
1994 for_each_buffer_cpu(buffer
, cpu
) {
1995 cpu_buffer
= buffer
->buffers
[cpu
];
1996 overruns
+= cpu_buffer
->overrun
;
2001 EXPORT_SYMBOL_GPL(ring_buffer_overruns
);
2003 static void rb_iter_reset(struct ring_buffer_iter
*iter
)
2005 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2007 /* Iterator usage is expected to have record disabled */
2008 if (list_empty(&cpu_buffer
->reader_page
->list
)) {
2009 iter
->head_page
= cpu_buffer
->head_page
;
2010 iter
->head
= cpu_buffer
->head_page
->read
;
2012 iter
->head_page
= cpu_buffer
->reader_page
;
2013 iter
->head
= cpu_buffer
->reader_page
->read
;
2016 iter
->read_stamp
= cpu_buffer
->read_stamp
;
2018 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
2022 * ring_buffer_iter_reset - reset an iterator
2023 * @iter: The iterator to reset
2025 * Resets the iterator, so that it will start from the beginning
2028 void ring_buffer_iter_reset(struct ring_buffer_iter
*iter
)
2030 struct ring_buffer_per_cpu
*cpu_buffer
;
2031 unsigned long flags
;
2036 cpu_buffer
= iter
->cpu_buffer
;
2038 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2039 rb_iter_reset(iter
);
2040 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2042 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset
);
2045 * ring_buffer_iter_empty - check if an iterator has no more to read
2046 * @iter: The iterator to check
2048 int ring_buffer_iter_empty(struct ring_buffer_iter
*iter
)
2050 struct ring_buffer_per_cpu
*cpu_buffer
;
2052 cpu_buffer
= iter
->cpu_buffer
;
2054 return iter
->head_page
== cpu_buffer
->commit_page
&&
2055 iter
->head
== rb_commit_index(cpu_buffer
);
2057 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty
);
2060 rb_update_read_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
2061 struct ring_buffer_event
*event
)
2065 switch (event
->type_len
) {
2066 case RINGBUF_TYPE_PADDING
:
2069 case RINGBUF_TYPE_TIME_EXTEND
:
2070 delta
= event
->array
[0];
2072 delta
+= event
->time_delta
;
2073 cpu_buffer
->read_stamp
+= delta
;
2076 case RINGBUF_TYPE_TIME_STAMP
:
2077 /* FIXME: not implemented */
2080 case RINGBUF_TYPE_DATA
:
2081 cpu_buffer
->read_stamp
+= event
->time_delta
;
2091 rb_update_iter_read_stamp(struct ring_buffer_iter
*iter
,
2092 struct ring_buffer_event
*event
)
2096 switch (event
->type_len
) {
2097 case RINGBUF_TYPE_PADDING
:
2100 case RINGBUF_TYPE_TIME_EXTEND
:
2101 delta
= event
->array
[0];
2103 delta
+= event
->time_delta
;
2104 iter
->read_stamp
+= delta
;
2107 case RINGBUF_TYPE_TIME_STAMP
:
2108 /* FIXME: not implemented */
2111 case RINGBUF_TYPE_DATA
:
2112 iter
->read_stamp
+= event
->time_delta
;
2121 static struct buffer_page
*
2122 rb_get_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
2124 struct buffer_page
*reader
= NULL
;
2125 unsigned long flags
;
2128 local_irq_save(flags
);
2129 __raw_spin_lock(&cpu_buffer
->lock
);
2133 * This should normally only loop twice. But because the
2134 * start of the reader inserts an empty page, it causes
2135 * a case where we will loop three times. There should be no
2136 * reason to loop four times (that I know of).
2138 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 3)) {
2143 reader
= cpu_buffer
->reader_page
;
2145 /* If there's more to read, return this page */
2146 if (cpu_buffer
->reader_page
->read
< rb_page_size(reader
))
2149 /* Never should we have an index greater than the size */
2150 if (RB_WARN_ON(cpu_buffer
,
2151 cpu_buffer
->reader_page
->read
> rb_page_size(reader
)))
2154 /* check if we caught up to the tail */
2156 if (cpu_buffer
->commit_page
== cpu_buffer
->reader_page
)
2160 * Splice the empty reader page into the list around the head.
2161 * Reset the reader page to size zero.
2164 reader
= cpu_buffer
->head_page
;
2165 cpu_buffer
->reader_page
->list
.next
= reader
->list
.next
;
2166 cpu_buffer
->reader_page
->list
.prev
= reader
->list
.prev
;
2168 local_set(&cpu_buffer
->reader_page
->write
, 0);
2169 local_set(&cpu_buffer
->reader_page
->entries
, 0);
2170 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
2172 /* Make the reader page now replace the head */
2173 reader
->list
.prev
->next
= &cpu_buffer
->reader_page
->list
;
2174 reader
->list
.next
->prev
= &cpu_buffer
->reader_page
->list
;
2177 * If the tail is on the reader, then we must set the head
2178 * to the inserted page, otherwise we set it one before.
2180 cpu_buffer
->head_page
= cpu_buffer
->reader_page
;
2182 if (cpu_buffer
->commit_page
!= reader
)
2183 rb_inc_page(cpu_buffer
, &cpu_buffer
->head_page
);
2185 /* Finally update the reader page to the new head */
2186 cpu_buffer
->reader_page
= reader
;
2187 rb_reset_reader_page(cpu_buffer
);
2192 __raw_spin_unlock(&cpu_buffer
->lock
);
2193 local_irq_restore(flags
);
2198 static void rb_advance_reader(struct ring_buffer_per_cpu
*cpu_buffer
)
2200 struct ring_buffer_event
*event
;
2201 struct buffer_page
*reader
;
2204 reader
= rb_get_reader_page(cpu_buffer
);
2206 /* This function should not be called when buffer is empty */
2207 if (RB_WARN_ON(cpu_buffer
, !reader
))
2210 event
= rb_reader_event(cpu_buffer
);
2212 if (event
->type_len
<= RINGBUF_TYPE_DATA_TYPE_LEN_MAX
2213 || rb_discarded_event(event
))
2216 rb_update_read_stamp(cpu_buffer
, event
);
2218 length
= rb_event_length(event
);
2219 cpu_buffer
->reader_page
->read
+= length
;
2222 static void rb_advance_iter(struct ring_buffer_iter
*iter
)
2224 struct ring_buffer
*buffer
;
2225 struct ring_buffer_per_cpu
*cpu_buffer
;
2226 struct ring_buffer_event
*event
;
2229 cpu_buffer
= iter
->cpu_buffer
;
2230 buffer
= cpu_buffer
->buffer
;
2233 * Check if we are at the end of the buffer.
2235 if (iter
->head
>= rb_page_size(iter
->head_page
)) {
2236 if (RB_WARN_ON(buffer
,
2237 iter
->head_page
== cpu_buffer
->commit_page
))
2243 event
= rb_iter_head_event(iter
);
2245 length
= rb_event_length(event
);
2248 * This should not be called to advance the header if we are
2249 * at the tail of the buffer.
2251 if (RB_WARN_ON(cpu_buffer
,
2252 (iter
->head_page
== cpu_buffer
->commit_page
) &&
2253 (iter
->head
+ length
> rb_commit_index(cpu_buffer
))))
2256 rb_update_iter_read_stamp(iter
, event
);
2258 iter
->head
+= length
;
2260 /* check for end of page padding */
2261 if ((iter
->head
>= rb_page_size(iter
->head_page
)) &&
2262 (iter
->head_page
!= cpu_buffer
->commit_page
))
2263 rb_advance_iter(iter
);
2266 static struct ring_buffer_event
*
2267 rb_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2269 struct ring_buffer_per_cpu
*cpu_buffer
;
2270 struct ring_buffer_event
*event
;
2271 struct buffer_page
*reader
;
2274 cpu_buffer
= buffer
->buffers
[cpu
];
2278 * We repeat when a timestamp is encountered. It is possible
2279 * to get multiple timestamps from an interrupt entering just
2280 * as one timestamp is about to be written. The max times
2281 * that this can happen is the number of nested interrupts we
2282 * can have. Nesting 10 deep of interrupts is clearly
2285 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 10))
2288 reader
= rb_get_reader_page(cpu_buffer
);
2292 event
= rb_reader_event(cpu_buffer
);
2294 switch (event
->type_len
) {
2295 case RINGBUF_TYPE_PADDING
:
2296 if (rb_null_event(event
))
2297 RB_WARN_ON(cpu_buffer
, 1);
2299 * Because the writer could be discarding every
2300 * event it creates (which would probably be bad)
2301 * if we were to go back to "again" then we may never
2302 * catch up, and will trigger the warn on, or lock
2303 * the box. Return the padding, and we will release
2304 * the current locks, and try again.
2306 rb_advance_reader(cpu_buffer
);
2309 case RINGBUF_TYPE_TIME_EXTEND
:
2310 /* Internal data, OK to advance */
2311 rb_advance_reader(cpu_buffer
);
2314 case RINGBUF_TYPE_TIME_STAMP
:
2315 /* FIXME: not implemented */
2316 rb_advance_reader(cpu_buffer
);
2319 case RINGBUF_TYPE_DATA
:
2321 *ts
= cpu_buffer
->read_stamp
+ event
->time_delta
;
2322 ring_buffer_normalize_time_stamp(buffer
,
2323 cpu_buffer
->cpu
, ts
);
2333 EXPORT_SYMBOL_GPL(ring_buffer_peek
);
2335 static struct ring_buffer_event
*
2336 rb_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
2338 struct ring_buffer
*buffer
;
2339 struct ring_buffer_per_cpu
*cpu_buffer
;
2340 struct ring_buffer_event
*event
;
2343 if (ring_buffer_iter_empty(iter
))
2346 cpu_buffer
= iter
->cpu_buffer
;
2347 buffer
= cpu_buffer
->buffer
;
2351 * We repeat when a timestamp is encountered. It is possible
2352 * to get multiple timestamps from an interrupt entering just
2353 * as one timestamp is about to be written. The max times
2354 * that this can happen is the number of nested interrupts we
2355 * can have. Nesting 10 deep of interrupts is clearly
2358 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 10))
2361 if (rb_per_cpu_empty(cpu_buffer
))
2364 event
= rb_iter_head_event(iter
);
2366 switch (event
->type_len
) {
2367 case RINGBUF_TYPE_PADDING
:
2368 if (rb_null_event(event
)) {
2372 rb_advance_iter(iter
);
2375 case RINGBUF_TYPE_TIME_EXTEND
:
2376 /* Internal data, OK to advance */
2377 rb_advance_iter(iter
);
2380 case RINGBUF_TYPE_TIME_STAMP
:
2381 /* FIXME: not implemented */
2382 rb_advance_iter(iter
);
2385 case RINGBUF_TYPE_DATA
:
2387 *ts
= iter
->read_stamp
+ event
->time_delta
;
2388 ring_buffer_normalize_time_stamp(buffer
,
2389 cpu_buffer
->cpu
, ts
);
2399 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek
);
2402 * ring_buffer_peek - peek at the next event to be read
2403 * @buffer: The ring buffer to read
2404 * @cpu: The cpu to peak at
2405 * @ts: The timestamp counter of this event.
2407 * This will return the event that will be read next, but does
2408 * not consume the data.
2410 struct ring_buffer_event
*
2411 ring_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2413 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2414 struct ring_buffer_event
*event
;
2415 unsigned long flags
;
2417 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2421 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2422 event
= rb_buffer_peek(buffer
, cpu
, ts
);
2423 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2425 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2434 * ring_buffer_iter_peek - peek at the next event to be read
2435 * @iter: The ring buffer iterator
2436 * @ts: The timestamp counter of this event.
2438 * This will return the event that will be read next, but does
2439 * not increment the iterator.
2441 struct ring_buffer_event
*
2442 ring_buffer_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
2444 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2445 struct ring_buffer_event
*event
;
2446 unsigned long flags
;
2449 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2450 event
= rb_iter_peek(iter
, ts
);
2451 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2453 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2462 * ring_buffer_consume - return an event and consume it
2463 * @buffer: The ring buffer to get the next event from
2465 * Returns the next event in the ring buffer, and that event is consumed.
2466 * Meaning, that sequential reads will keep returning a different event,
2467 * and eventually empty the ring buffer if the producer is slower.
2469 struct ring_buffer_event
*
2470 ring_buffer_consume(struct ring_buffer
*buffer
, int cpu
, u64
*ts
)
2472 struct ring_buffer_per_cpu
*cpu_buffer
;
2473 struct ring_buffer_event
*event
= NULL
;
2474 unsigned long flags
;
2477 /* might be called in atomic */
2480 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2483 cpu_buffer
= buffer
->buffers
[cpu
];
2484 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2486 event
= rb_buffer_peek(buffer
, cpu
, ts
);
2490 rb_advance_reader(cpu_buffer
);
2493 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2498 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2505 EXPORT_SYMBOL_GPL(ring_buffer_consume
);
2508 * ring_buffer_read_start - start a non consuming read of the buffer
2509 * @buffer: The ring buffer to read from
2510 * @cpu: The cpu buffer to iterate over
2512 * This starts up an iteration through the buffer. It also disables
2513 * the recording to the buffer until the reading is finished.
2514 * This prevents the reading from being corrupted. This is not
2515 * a consuming read, so a producer is not expected.
2517 * Must be paired with ring_buffer_finish.
2519 struct ring_buffer_iter
*
2520 ring_buffer_read_start(struct ring_buffer
*buffer
, int cpu
)
2522 struct ring_buffer_per_cpu
*cpu_buffer
;
2523 struct ring_buffer_iter
*iter
;
2524 unsigned long flags
;
2526 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2529 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
2533 cpu_buffer
= buffer
->buffers
[cpu
];
2535 iter
->cpu_buffer
= cpu_buffer
;
2537 atomic_inc(&cpu_buffer
->record_disabled
);
2538 synchronize_sched();
2540 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2541 __raw_spin_lock(&cpu_buffer
->lock
);
2542 rb_iter_reset(iter
);
2543 __raw_spin_unlock(&cpu_buffer
->lock
);
2544 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2548 EXPORT_SYMBOL_GPL(ring_buffer_read_start
);
2551 * ring_buffer_finish - finish reading the iterator of the buffer
2552 * @iter: The iterator retrieved by ring_buffer_start
2554 * This re-enables the recording to the buffer, and frees the
2558 ring_buffer_read_finish(struct ring_buffer_iter
*iter
)
2560 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2562 atomic_dec(&cpu_buffer
->record_disabled
);
2565 EXPORT_SYMBOL_GPL(ring_buffer_read_finish
);
2568 * ring_buffer_read - read the next item in the ring buffer by the iterator
2569 * @iter: The ring buffer iterator
2570 * @ts: The time stamp of the event read.
2572 * This reads the next event in the ring buffer and increments the iterator.
2574 struct ring_buffer_event
*
2575 ring_buffer_read(struct ring_buffer_iter
*iter
, u64
*ts
)
2577 struct ring_buffer_event
*event
;
2578 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
2579 unsigned long flags
;
2582 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2583 event
= rb_iter_peek(iter
, ts
);
2587 rb_advance_iter(iter
);
2589 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2591 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
) {
2598 EXPORT_SYMBOL_GPL(ring_buffer_read
);
2601 * ring_buffer_size - return the size of the ring buffer (in bytes)
2602 * @buffer: The ring buffer.
2604 unsigned long ring_buffer_size(struct ring_buffer
*buffer
)
2606 return BUF_PAGE_SIZE
* buffer
->pages
;
2608 EXPORT_SYMBOL_GPL(ring_buffer_size
);
2611 rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
)
2613 cpu_buffer
->head_page
2614 = list_entry(cpu_buffer
->pages
.next
, struct buffer_page
, list
);
2615 local_set(&cpu_buffer
->head_page
->write
, 0);
2616 local_set(&cpu_buffer
->head_page
->entries
, 0);
2617 local_set(&cpu_buffer
->head_page
->page
->commit
, 0);
2619 cpu_buffer
->head_page
->read
= 0;
2621 cpu_buffer
->tail_page
= cpu_buffer
->head_page
;
2622 cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
2624 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
2625 local_set(&cpu_buffer
->reader_page
->write
, 0);
2626 local_set(&cpu_buffer
->reader_page
->entries
, 0);
2627 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
2628 cpu_buffer
->reader_page
->read
= 0;
2630 cpu_buffer
->nmi_dropped
= 0;
2631 cpu_buffer
->commit_overrun
= 0;
2632 cpu_buffer
->overrun
= 0;
2633 cpu_buffer
->read
= 0;
2634 local_set(&cpu_buffer
->entries
, 0);
2636 cpu_buffer
->write_stamp
= 0;
2637 cpu_buffer
->read_stamp
= 0;
2641 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
2642 * @buffer: The ring buffer to reset a per cpu buffer of
2643 * @cpu: The CPU buffer to be reset
2645 void ring_buffer_reset_cpu(struct ring_buffer
*buffer
, int cpu
)
2647 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2648 unsigned long flags
;
2650 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2653 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2655 __raw_spin_lock(&cpu_buffer
->lock
);
2657 rb_reset_cpu(cpu_buffer
);
2659 __raw_spin_unlock(&cpu_buffer
->lock
);
2661 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
2663 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu
);
2666 * ring_buffer_reset - reset a ring buffer
2667 * @buffer: The ring buffer to reset all cpu buffers
2669 void ring_buffer_reset(struct ring_buffer
*buffer
)
2673 for_each_buffer_cpu(buffer
, cpu
)
2674 ring_buffer_reset_cpu(buffer
, cpu
);
2676 EXPORT_SYMBOL_GPL(ring_buffer_reset
);
2679 * rind_buffer_empty - is the ring buffer empty?
2680 * @buffer: The ring buffer to test
2682 int ring_buffer_empty(struct ring_buffer
*buffer
)
2684 struct ring_buffer_per_cpu
*cpu_buffer
;
2687 /* yes this is racy, but if you don't like the race, lock the buffer */
2688 for_each_buffer_cpu(buffer
, cpu
) {
2689 cpu_buffer
= buffer
->buffers
[cpu
];
2690 if (!rb_per_cpu_empty(cpu_buffer
))
2696 EXPORT_SYMBOL_GPL(ring_buffer_empty
);
2699 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
2700 * @buffer: The ring buffer
2701 * @cpu: The CPU buffer to test
2703 int ring_buffer_empty_cpu(struct ring_buffer
*buffer
, int cpu
)
2705 struct ring_buffer_per_cpu
*cpu_buffer
;
2708 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2711 cpu_buffer
= buffer
->buffers
[cpu
];
2712 ret
= rb_per_cpu_empty(cpu_buffer
);
2717 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu
);
2720 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
2721 * @buffer_a: One buffer to swap with
2722 * @buffer_b: The other buffer to swap with
2724 * This function is useful for tracers that want to take a "snapshot"
2725 * of a CPU buffer and has another back up buffer lying around.
2726 * it is expected that the tracer handles the cpu buffer not being
2727 * used at the moment.
2729 int ring_buffer_swap_cpu(struct ring_buffer
*buffer_a
,
2730 struct ring_buffer
*buffer_b
, int cpu
)
2732 struct ring_buffer_per_cpu
*cpu_buffer_a
;
2733 struct ring_buffer_per_cpu
*cpu_buffer_b
;
2736 if (!cpumask_test_cpu(cpu
, buffer_a
->cpumask
) ||
2737 !cpumask_test_cpu(cpu
, buffer_b
->cpumask
))
2740 /* At least make sure the two buffers are somewhat the same */
2741 if (buffer_a
->pages
!= buffer_b
->pages
)
2746 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
2749 if (atomic_read(&buffer_a
->record_disabled
))
2752 if (atomic_read(&buffer_b
->record_disabled
))
2755 cpu_buffer_a
= buffer_a
->buffers
[cpu
];
2756 cpu_buffer_b
= buffer_b
->buffers
[cpu
];
2758 if (atomic_read(&cpu_buffer_a
->record_disabled
))
2761 if (atomic_read(&cpu_buffer_b
->record_disabled
))
2765 * We can't do a synchronize_sched here because this
2766 * function can be called in atomic context.
2767 * Normally this will be called from the same CPU as cpu.
2768 * If not it's up to the caller to protect this.
2770 atomic_inc(&cpu_buffer_a
->record_disabled
);
2771 atomic_inc(&cpu_buffer_b
->record_disabled
);
2773 buffer_a
->buffers
[cpu
] = cpu_buffer_b
;
2774 buffer_b
->buffers
[cpu
] = cpu_buffer_a
;
2776 cpu_buffer_b
->buffer
= buffer_a
;
2777 cpu_buffer_a
->buffer
= buffer_b
;
2779 atomic_dec(&cpu_buffer_a
->record_disabled
);
2780 atomic_dec(&cpu_buffer_b
->record_disabled
);
2786 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu
);
2788 static void rb_remove_entries(struct ring_buffer_per_cpu
*cpu_buffer
,
2789 struct buffer_data_page
*bpage
,
2790 unsigned int offset
)
2792 struct ring_buffer_event
*event
;
2795 __raw_spin_lock(&cpu_buffer
->lock
);
2796 for (head
= offset
; head
< local_read(&bpage
->commit
);
2797 head
+= rb_event_length(event
)) {
2799 event
= __rb_data_page_index(bpage
, head
);
2800 if (RB_WARN_ON(cpu_buffer
, rb_null_event(event
)))
2802 /* Only count data entries */
2803 if (event
->type_len
> RINGBUF_TYPE_DATA_TYPE_LEN_MAX
)
2807 __raw_spin_unlock(&cpu_buffer
->lock
);
2811 * ring_buffer_alloc_read_page - allocate a page to read from buffer
2812 * @buffer: the buffer to allocate for.
2814 * This function is used in conjunction with ring_buffer_read_page.
2815 * When reading a full page from the ring buffer, these functions
2816 * can be used to speed up the process. The calling function should
2817 * allocate a few pages first with this function. Then when it
2818 * needs to get pages from the ring buffer, it passes the result
2819 * of this function into ring_buffer_read_page, which will swap
2820 * the page that was allocated, with the read page of the buffer.
2823 * The page allocated, or NULL on error.
2825 void *ring_buffer_alloc_read_page(struct ring_buffer
*buffer
)
2827 struct buffer_data_page
*bpage
;
2830 addr
= __get_free_page(GFP_KERNEL
);
2834 bpage
= (void *)addr
;
2836 rb_init_page(bpage
);
2840 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page
);
2843 * ring_buffer_free_read_page - free an allocated read page
2844 * @buffer: the buffer the page was allocate for
2845 * @data: the page to free
2847 * Free a page allocated from ring_buffer_alloc_read_page.
2849 void ring_buffer_free_read_page(struct ring_buffer
*buffer
, void *data
)
2851 free_page((unsigned long)data
);
2853 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page
);
2856 * ring_buffer_read_page - extract a page from the ring buffer
2857 * @buffer: buffer to extract from
2858 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
2859 * @len: amount to extract
2860 * @cpu: the cpu of the buffer to extract
2861 * @full: should the extraction only happen when the page is full.
2863 * This function will pull out a page from the ring buffer and consume it.
2864 * @data_page must be the address of the variable that was returned
2865 * from ring_buffer_alloc_read_page. This is because the page might be used
2866 * to swap with a page in the ring buffer.
2869 * rpage = ring_buffer_alloc_read_page(buffer);
2872 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
2874 * process_page(rpage, ret);
2876 * When @full is set, the function will not return true unless
2877 * the writer is off the reader page.
2879 * Note: it is up to the calling functions to handle sleeps and wakeups.
2880 * The ring buffer can be used anywhere in the kernel and can not
2881 * blindly call wake_up. The layer that uses the ring buffer must be
2882 * responsible for that.
2885 * >=0 if data has been transferred, returns the offset of consumed data.
2886 * <0 if no data has been transferred.
2888 int ring_buffer_read_page(struct ring_buffer
*buffer
,
2889 void **data_page
, size_t len
, int cpu
, int full
)
2891 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
2892 struct ring_buffer_event
*event
;
2893 struct buffer_data_page
*bpage
;
2894 struct buffer_page
*reader
;
2895 unsigned long flags
;
2896 unsigned int commit
;
2901 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2905 * If len is not big enough to hold the page header, then
2906 * we can not copy anything.
2908 if (len
<= BUF_PAGE_HDR_SIZE
)
2911 len
-= BUF_PAGE_HDR_SIZE
;
2920 spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
2922 reader
= rb_get_reader_page(cpu_buffer
);
2926 event
= rb_reader_event(cpu_buffer
);
2928 read
= reader
->read
;
2929 commit
= rb_page_commit(reader
);
2932 * If this page has been partially read or
2933 * if len is not big enough to read the rest of the page or
2934 * a writer is still on the page, then
2935 * we must copy the data from the page to the buffer.
2936 * Otherwise, we can simply swap the page with the one passed in.
2938 if (read
|| (len
< (commit
- read
)) ||
2939 cpu_buffer
->reader_page
== cpu_buffer
->commit_page
) {
2940 struct buffer_data_page
*rpage
= cpu_buffer
->reader_page
->page
;
2941 unsigned int rpos
= read
;
2942 unsigned int pos
= 0;
2948 if (len
> (commit
- read
))
2949 len
= (commit
- read
);
2951 size
= rb_event_length(event
);
2956 /* save the current timestamp, since the user will need it */
2957 save_timestamp
= cpu_buffer
->read_stamp
;
2959 /* Need to copy one event at a time */
2961 memcpy(bpage
->data
+ pos
, rpage
->data
+ rpos
, size
);
2965 rb_advance_reader(cpu_buffer
);
2966 rpos
= reader
->read
;
2969 event
= rb_reader_event(cpu_buffer
);
2970 size
= rb_event_length(event
);
2971 } while (len
> size
);
2974 local_set(&bpage
->commit
, pos
);
2975 bpage
->time_stamp
= save_timestamp
;
2977 /* we copied everything to the beginning */
2980 /* swap the pages */
2981 rb_init_page(bpage
);
2982 bpage
= reader
->page
;
2983 reader
->page
= *data_page
;
2984 local_set(&reader
->write
, 0);
2985 local_set(&reader
->entries
, 0);
2989 /* update the entry counter */
2990 rb_remove_entries(cpu_buffer
, bpage
, read
);
2995 spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
3000 EXPORT_SYMBOL_GPL(ring_buffer_read_page
);
3003 rb_simple_read(struct file
*filp
, char __user
*ubuf
,
3004 size_t cnt
, loff_t
*ppos
)
3006 unsigned long *p
= filp
->private_data
;
3010 if (test_bit(RB_BUFFERS_DISABLED_BIT
, p
))
3011 r
= sprintf(buf
, "permanently disabled\n");
3013 r
= sprintf(buf
, "%d\n", test_bit(RB_BUFFERS_ON_BIT
, p
));
3015 return simple_read_from_buffer(ubuf
, cnt
, ppos
, buf
, r
);
3019 rb_simple_write(struct file
*filp
, const char __user
*ubuf
,
3020 size_t cnt
, loff_t
*ppos
)
3022 unsigned long *p
= filp
->private_data
;
3027 if (cnt
>= sizeof(buf
))
3030 if (copy_from_user(&buf
, ubuf
, cnt
))
3035 ret
= strict_strtoul(buf
, 10, &val
);
3040 set_bit(RB_BUFFERS_ON_BIT
, p
);
3042 clear_bit(RB_BUFFERS_ON_BIT
, p
);
3049 static const struct file_operations rb_simple_fops
= {
3050 .open
= tracing_open_generic
,
3051 .read
= rb_simple_read
,
3052 .write
= rb_simple_write
,
3056 static __init
int rb_init_debugfs(void)
3058 struct dentry
*d_tracer
;
3060 d_tracer
= tracing_init_dentry();
3062 trace_create_file("tracing_on", 0644, d_tracer
,
3063 &ring_buffer_flags
, &rb_simple_fops
);
3068 fs_initcall(rb_init_debugfs
);
3070 #ifdef CONFIG_HOTPLUG_CPU
3071 static int rb_cpu_notify(struct notifier_block
*self
,
3072 unsigned long action
, void *hcpu
)
3074 struct ring_buffer
*buffer
=
3075 container_of(self
, struct ring_buffer
, cpu_notify
);
3076 long cpu
= (long)hcpu
;
3079 case CPU_UP_PREPARE
:
3080 case CPU_UP_PREPARE_FROZEN
:
3081 if (cpu_isset(cpu
, *buffer
->cpumask
))
3084 buffer
->buffers
[cpu
] =
3085 rb_allocate_cpu_buffer(buffer
, cpu
);
3086 if (!buffer
->buffers
[cpu
]) {
3087 WARN(1, "failed to allocate ring buffer on CPU %ld\n",
3092 cpu_set(cpu
, *buffer
->cpumask
);
3094 case CPU_DOWN_PREPARE
:
3095 case CPU_DOWN_PREPARE_FROZEN
:
3098 * If we were to free the buffer, then the user would
3099 * lose any trace that was in the buffer.