4 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
6 #include <linux/ring_buffer.h>
7 #include <linux/trace_clock.h>
8 #include <linux/ftrace_irq.h>
9 #include <linux/spinlock.h>
10 #include <linux/debugfs.h>
11 #include <linux/uaccess.h>
12 #include <linux/hardirq.h>
13 #include <linux/module.h>
14 #include <linux/percpu.h>
15 #include <linux/mutex.h>
16 #include <linux/init.h>
17 #include <linux/hash.h>
18 #include <linux/list.h>
19 #include <linux/cpu.h>
25 * The ring buffer header is special. We must manually up keep it.
27 int ring_buffer_print_entry_header(struct trace_seq *s)
31 ret = trace_seq_printf(s, "# compressed entry header\n");
32 ret = trace_seq_printf(s, "\ttype_len : 5 bits\n");
33 ret = trace_seq_printf(s, "\ttime_delta : 27 bits\n");
34 ret = trace_seq_printf(s, "\tarray : 32 bits\n");
35 ret = trace_seq_printf(s, "\n");
36 ret = trace_seq_printf(s, "\tpadding : type == %d\n",
37 RINGBUF_TYPE_PADDING);
38 ret = trace_seq_printf(s, "\ttime_extend : type == %d\n",
39 RINGBUF_TYPE_TIME_EXTEND);
40 ret = trace_seq_printf(s, "\tdata max type_len == %d\n",
41 RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
47 * The ring buffer is made up of a list of pages. A separate list of pages is
48 * allocated for each CPU. A writer may only write to a buffer that is
49 * associated with the CPU it is currently executing on. A reader may read
50 * from any per cpu buffer.
52 * The reader is special. For each per cpu buffer, the reader has its own
53 * reader page. When a reader has read the entire reader page, this reader
54 * page is swapped with another page in the ring buffer.
56 * Now, as long as the writer is off the reader page, the reader can do what
57 * ever it wants with that page. The writer will never write to that page
58 * again (as long as it is out of the ring buffer).
60 * Here's some silly ASCII art.
63 * |reader| RING BUFFER
65 * +------+ +---+ +---+ +---+
74 * |reader| RING BUFFER
75 * |page |------------------v
76 * +------+ +---+ +---+ +---+
85 * |reader| RING BUFFER
86 * |page |------------------v
87 * +------+ +---+ +---+ +---+
92 * +------------------------------+
96 * |buffer| RING BUFFER
97 * |page |------------------v
98 * +------+ +---+ +---+ +---+
100 * | New +---+ +---+ +---+
103 * +------------------------------+
106 * After we make this swap, the reader can hand this page off to the splice
107 * code and be done with it. It can even allocate a new page if it needs to
108 * and swap that into the ring buffer.
110 * We will be using cmpxchg soon to make all this lockless.
115 * A fast way to enable or disable all ring buffers is to
116 * call tracing_on or tracing_off. Turning off the ring buffers
117 * prevents all ring buffers from being recorded to.
118 * Turning this switch on, makes it OK to write to the
119 * ring buffer, if the ring buffer is enabled itself.
121 * There's three layers that must be on in order to write
122 * to the ring buffer.
124 * 1) This global flag must be set.
125 * 2) The ring buffer must be enabled for recording.
126 * 3) The per cpu buffer must be enabled for recording.
128 * In case of an anomaly, this global flag has a bit set that
129 * will permantly disable all ring buffers.
133 * Global flag to disable all recording to ring buffers
134 * This has two bits: ON, DISABLED
138 * 0 0 : ring buffers are off
139 * 1 0 : ring buffers are on
140 * X 1 : ring buffers are permanently disabled
144 RB_BUFFERS_ON_BIT = 0,
145 RB_BUFFERS_DISABLED_BIT = 1,
149 RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT,
150 RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT,
153 static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON;
155 #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
158 * tracing_on - enable all tracing buffers
160 * This function enables all tracing buffers that may have been
161 * disabled with tracing_off.
163 void tracing_on(void)
165 set_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags);
167 EXPORT_SYMBOL_GPL(tracing_on);
170 * tracing_off - turn off all tracing buffers
172 * This function stops all tracing buffers from recording data.
173 * It does not disable any overhead the tracers themselves may
174 * be causing. This function simply causes all recording to
175 * the ring buffers to fail.
177 void tracing_off(void)
179 clear_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags);
181 EXPORT_SYMBOL_GPL(tracing_off);
184 * tracing_off_permanent - permanently disable ring buffers
186 * This function, once called, will disable all ring buffers
189 void tracing_off_permanent(void)
191 set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags);
195 * tracing_is_on - show state of ring buffers enabled
197 int tracing_is_on(void)
199 return ring_buffer_flags == RB_BUFFERS_ON;
201 EXPORT_SYMBOL_GPL(tracing_is_on);
205 #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
206 #define RB_ALIGNMENT 4U
207 #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
208 #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */
210 /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
211 #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
214 RB_LEN_TIME_EXTEND = 8,
215 RB_LEN_TIME_STAMP = 16,
218 static inline int rb_null_event(struct ring_buffer_event *event)
220 return event->type_len == RINGBUF_TYPE_PADDING
221 && event->time_delta == 0;
224 static inline int rb_discarded_event(struct ring_buffer_event *event)
226 return event->type_len == RINGBUF_TYPE_PADDING && event->time_delta;
229 static void rb_event_set_padding(struct ring_buffer_event *event)
231 event->type_len = RINGBUF_TYPE_PADDING;
232 event->time_delta = 0;
236 rb_event_data_length(struct ring_buffer_event *event)
241 length = event->type_len * RB_ALIGNMENT;
243 length = event->array[0];
244 return length + RB_EVNT_HDR_SIZE;
247 /* inline for ring buffer fast paths */
249 rb_event_length(struct ring_buffer_event *event)
251 switch (event->type_len) {
252 case RINGBUF_TYPE_PADDING:
253 if (rb_null_event(event))
256 return event->array[0] + RB_EVNT_HDR_SIZE;
258 case RINGBUF_TYPE_TIME_EXTEND:
259 return RB_LEN_TIME_EXTEND;
261 case RINGBUF_TYPE_TIME_STAMP:
262 return RB_LEN_TIME_STAMP;
264 case RINGBUF_TYPE_DATA:
265 return rb_event_data_length(event);
274 * ring_buffer_event_length - return the length of the event
275 * @event: the event to get the length of
277 unsigned ring_buffer_event_length(struct ring_buffer_event *event)
279 unsigned length = rb_event_length(event);
280 if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
282 length -= RB_EVNT_HDR_SIZE;
283 if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0]))
284 length -= sizeof(event->array[0]);
287 EXPORT_SYMBOL_GPL(ring_buffer_event_length);
289 /* inline for ring buffer fast paths */
291 rb_event_data(struct ring_buffer_event *event)
293 BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
294 /* If length is in len field, then array[0] has the data */
296 return (void *)&event->array[0];
297 /* Otherwise length is in array[0] and array[1] has the data */
298 return (void *)&event->array[1];
302 * ring_buffer_event_data - return the data of the event
303 * @event: the event to get the data from
305 void *ring_buffer_event_data(struct ring_buffer_event *event)
307 return rb_event_data(event);
309 EXPORT_SYMBOL_GPL(ring_buffer_event_data);
311 #define for_each_buffer_cpu(buffer, cpu) \
312 for_each_cpu(cpu, buffer->cpumask)
315 #define TS_MASK ((1ULL << TS_SHIFT) - 1)
316 #define TS_DELTA_TEST (~TS_MASK)
318 struct buffer_data_page {
319 u64 time_stamp; /* page time stamp */
320 local_t commit; /* write committed index */
321 unsigned char data[]; /* data of buffer page */
325 struct list_head list; /* list of buffer pages */
326 local_t write; /* index for next write */
327 unsigned read; /* index for next read */
328 local_t entries; /* entries on this page */
329 struct buffer_data_page *page; /* Actual data page */
332 static void rb_init_page(struct buffer_data_page *bpage)
334 local_set(&bpage->commit, 0);
338 * ring_buffer_page_len - the size of data on the page.
339 * @page: The page to read
341 * Returns the amount of data on the page, including buffer page header.
343 size_t ring_buffer_page_len(void *page)
345 return local_read(&((struct buffer_data_page *)page)->commit)
350 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
353 static void free_buffer_page(struct buffer_page *bpage)
355 free_page((unsigned long)bpage->page);
360 * We need to fit the time_stamp delta into 27 bits.
362 static inline int test_time_stamp(u64 delta)
364 if (delta & TS_DELTA_TEST)
369 #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
371 /* Max payload is BUF_PAGE_SIZE - header (8bytes) */
372 #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
374 /* Max number of timestamps that can fit on a page */
375 #define RB_TIMESTAMPS_PER_PAGE (BUF_PAGE_SIZE / RB_LEN_TIME_STAMP)
377 int ring_buffer_print_page_header(struct trace_seq *s)
379 struct buffer_data_page field;
382 ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t"
383 "offset:0;\tsize:%u;\n",
384 (unsigned int)sizeof(field.time_stamp));
386 ret = trace_seq_printf(s, "\tfield: local_t commit;\t"
387 "offset:%u;\tsize:%u;\n",
388 (unsigned int)offsetof(typeof(field), commit),
389 (unsigned int)sizeof(field.commit));
391 ret = trace_seq_printf(s, "\tfield: char data;\t"
392 "offset:%u;\tsize:%u;\n",
393 (unsigned int)offsetof(typeof(field), data),
394 (unsigned int)BUF_PAGE_SIZE);
400 * head_page == tail_page && head == tail then buffer is empty.
402 struct ring_buffer_per_cpu {
404 struct ring_buffer *buffer;
405 spinlock_t reader_lock; /* serialize readers */
407 struct lock_class_key lock_key;
408 struct list_head pages;
409 struct buffer_page *head_page; /* read from head */
410 struct buffer_page *tail_page; /* write to tail */
411 struct buffer_page *commit_page; /* committed pages */
412 struct buffer_page *reader_page;
413 unsigned long nmi_dropped;
414 unsigned long commit_overrun;
415 unsigned long overrun;
422 atomic_t record_disabled;
429 atomic_t record_disabled;
430 cpumask_var_t cpumask;
432 struct lock_class_key *reader_lock_key;
436 struct ring_buffer_per_cpu **buffers;
438 #ifdef CONFIG_HOTPLUG_CPU
439 struct notifier_block cpu_notify;
444 struct ring_buffer_iter {
445 struct ring_buffer_per_cpu *cpu_buffer;
447 struct buffer_page *head_page;
451 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
452 #define RB_WARN_ON(buffer, cond) \
454 int _____ret = unlikely(cond); \
456 atomic_inc(&buffer->record_disabled); \
462 /* Up this if you want to test the TIME_EXTENTS and normalization */
463 #define DEBUG_SHIFT 0
465 static inline u64 rb_time_stamp(struct ring_buffer *buffer, int cpu)
467 /* shift to debug/test normalization and TIME_EXTENTS */
468 return buffer->clock() << DEBUG_SHIFT;
471 u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu)
475 preempt_disable_notrace();
476 time = rb_time_stamp(buffer, cpu);
477 preempt_enable_no_resched_notrace();
481 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
483 void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer,
486 /* Just stupid testing the normalize function and deltas */
489 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
492 * check_pages - integrity check of buffer pages
493 * @cpu_buffer: CPU buffer with pages to test
495 * As a safety measure we check to make sure the data pages have not
498 static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
500 struct list_head *head = &cpu_buffer->pages;
501 struct buffer_page *bpage, *tmp;
503 if (RB_WARN_ON(cpu_buffer, head->next->prev != head))
505 if (RB_WARN_ON(cpu_buffer, head->prev->next != head))
508 list_for_each_entry_safe(bpage, tmp, head, list) {
509 if (RB_WARN_ON(cpu_buffer,
510 bpage->list.next->prev != &bpage->list))
512 if (RB_WARN_ON(cpu_buffer,
513 bpage->list.prev->next != &bpage->list))
520 static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
523 struct list_head *head = &cpu_buffer->pages;
524 struct buffer_page *bpage, *tmp;
529 for (i = 0; i < nr_pages; i++) {
530 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
531 GFP_KERNEL, cpu_to_node(cpu_buffer->cpu));
534 list_add(&bpage->list, &pages);
536 addr = __get_free_page(GFP_KERNEL);
539 bpage->page = (void *)addr;
540 rb_init_page(bpage->page);
543 list_splice(&pages, head);
545 rb_check_pages(cpu_buffer);
550 list_for_each_entry_safe(bpage, tmp, &pages, list) {
551 list_del_init(&bpage->list);
552 free_buffer_page(bpage);
557 static struct ring_buffer_per_cpu *
558 rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu)
560 struct ring_buffer_per_cpu *cpu_buffer;
561 struct buffer_page *bpage;
565 cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
566 GFP_KERNEL, cpu_to_node(cpu));
570 cpu_buffer->cpu = cpu;
571 cpu_buffer->buffer = buffer;
572 spin_lock_init(&cpu_buffer->reader_lock);
573 lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key);
574 cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
575 INIT_LIST_HEAD(&cpu_buffer->pages);
577 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
578 GFP_KERNEL, cpu_to_node(cpu));
580 goto fail_free_buffer;
582 cpu_buffer->reader_page = bpage;
583 addr = __get_free_page(GFP_KERNEL);
585 goto fail_free_reader;
586 bpage->page = (void *)addr;
587 rb_init_page(bpage->page);
589 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
591 ret = rb_allocate_pages(cpu_buffer, buffer->pages);
593 goto fail_free_reader;
595 cpu_buffer->head_page
596 = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
597 cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
602 free_buffer_page(cpu_buffer->reader_page);
609 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
611 struct list_head *head = &cpu_buffer->pages;
612 struct buffer_page *bpage, *tmp;
614 free_buffer_page(cpu_buffer->reader_page);
616 list_for_each_entry_safe(bpage, tmp, head, list) {
617 list_del_init(&bpage->list);
618 free_buffer_page(bpage);
624 * Causes compile errors if the struct buffer_page gets bigger
625 * than the struct page.
627 extern int ring_buffer_page_too_big(void);
629 #ifdef CONFIG_HOTPLUG_CPU
630 static int rb_cpu_notify(struct notifier_block *self,
631 unsigned long action, void *hcpu);
635 * ring_buffer_alloc - allocate a new ring_buffer
636 * @size: the size in bytes per cpu that is needed.
637 * @flags: attributes to set for the ring buffer.
639 * Currently the only flag that is available is the RB_FL_OVERWRITE
640 * flag. This flag means that the buffer will overwrite old data
641 * when the buffer wraps. If this flag is not set, the buffer will
642 * drop data when the tail hits the head.
644 struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags,
645 struct lock_class_key *key)
647 struct ring_buffer *buffer;
651 /* Paranoid! Optimizes out when all is well */
652 if (sizeof(struct buffer_page) > sizeof(struct page))
653 ring_buffer_page_too_big();
656 /* keep it in its own cache line */
657 buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
662 if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL))
663 goto fail_free_buffer;
665 buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
666 buffer->flags = flags;
667 buffer->clock = trace_clock_local;
668 buffer->reader_lock_key = key;
670 /* need at least two pages */
671 if (buffer->pages == 1)
675 * In case of non-hotplug cpu, if the ring-buffer is allocated
676 * in early initcall, it will not be notified of secondary cpus.
677 * In that off case, we need to allocate for all possible cpus.
679 #ifdef CONFIG_HOTPLUG_CPU
681 cpumask_copy(buffer->cpumask, cpu_online_mask);
683 cpumask_copy(buffer->cpumask, cpu_possible_mask);
685 buffer->cpus = nr_cpu_ids;
687 bsize = sizeof(void *) * nr_cpu_ids;
688 buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
690 if (!buffer->buffers)
691 goto fail_free_cpumask;
693 for_each_buffer_cpu(buffer, cpu) {
694 buffer->buffers[cpu] =
695 rb_allocate_cpu_buffer(buffer, cpu);
696 if (!buffer->buffers[cpu])
697 goto fail_free_buffers;
700 #ifdef CONFIG_HOTPLUG_CPU
701 buffer->cpu_notify.notifier_call = rb_cpu_notify;
702 buffer->cpu_notify.priority = 0;
703 register_cpu_notifier(&buffer->cpu_notify);
707 mutex_init(&buffer->mutex);
712 for_each_buffer_cpu(buffer, cpu) {
713 if (buffer->buffers[cpu])
714 rb_free_cpu_buffer(buffer->buffers[cpu]);
716 kfree(buffer->buffers);
719 free_cpumask_var(buffer->cpumask);
726 EXPORT_SYMBOL_GPL(__ring_buffer_alloc);
729 * ring_buffer_free - free a ring buffer.
730 * @buffer: the buffer to free.
733 ring_buffer_free(struct ring_buffer *buffer)
739 #ifdef CONFIG_HOTPLUG_CPU
740 unregister_cpu_notifier(&buffer->cpu_notify);
743 for_each_buffer_cpu(buffer, cpu)
744 rb_free_cpu_buffer(buffer->buffers[cpu]);
748 free_cpumask_var(buffer->cpumask);
752 EXPORT_SYMBOL_GPL(ring_buffer_free);
754 void ring_buffer_set_clock(struct ring_buffer *buffer,
757 buffer->clock = clock;
760 static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
763 rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages)
765 struct buffer_page *bpage;
769 atomic_inc(&cpu_buffer->record_disabled);
772 for (i = 0; i < nr_pages; i++) {
773 if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages)))
775 p = cpu_buffer->pages.next;
776 bpage = list_entry(p, struct buffer_page, list);
777 list_del_init(&bpage->list);
778 free_buffer_page(bpage);
780 if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages)))
783 rb_reset_cpu(cpu_buffer);
785 rb_check_pages(cpu_buffer);
787 atomic_dec(&cpu_buffer->record_disabled);
792 rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer,
793 struct list_head *pages, unsigned nr_pages)
795 struct buffer_page *bpage;
799 atomic_inc(&cpu_buffer->record_disabled);
802 for (i = 0; i < nr_pages; i++) {
803 if (RB_WARN_ON(cpu_buffer, list_empty(pages)))
806 bpage = list_entry(p, struct buffer_page, list);
807 list_del_init(&bpage->list);
808 list_add_tail(&bpage->list, &cpu_buffer->pages);
810 rb_reset_cpu(cpu_buffer);
812 rb_check_pages(cpu_buffer);
814 atomic_dec(&cpu_buffer->record_disabled);
818 * ring_buffer_resize - resize the ring buffer
819 * @buffer: the buffer to resize.
820 * @size: the new size.
822 * The tracer is responsible for making sure that the buffer is
823 * not being used while changing the size.
824 * Note: We may be able to change the above requirement by using
825 * RCU synchronizations.
827 * Minimum size is 2 * BUF_PAGE_SIZE.
829 * Returns -1 on failure.
831 int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
833 struct ring_buffer_per_cpu *cpu_buffer;
834 unsigned nr_pages, rm_pages, new_pages;
835 struct buffer_page *bpage, *tmp;
836 unsigned long buffer_size;
842 * Always succeed at resizing a non-existent buffer:
847 size = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
848 size *= BUF_PAGE_SIZE;
849 buffer_size = buffer->pages * BUF_PAGE_SIZE;
851 /* we need a minimum of two pages */
852 if (size < BUF_PAGE_SIZE * 2)
853 size = BUF_PAGE_SIZE * 2;
855 if (size == buffer_size)
858 mutex_lock(&buffer->mutex);
861 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
863 if (size < buffer_size) {
865 /* easy case, just free pages */
866 if (RB_WARN_ON(buffer, nr_pages >= buffer->pages))
869 rm_pages = buffer->pages - nr_pages;
871 for_each_buffer_cpu(buffer, cpu) {
872 cpu_buffer = buffer->buffers[cpu];
873 rb_remove_pages(cpu_buffer, rm_pages);
879 * This is a bit more difficult. We only want to add pages
880 * when we can allocate enough for all CPUs. We do this
881 * by allocating all the pages and storing them on a local
882 * link list. If we succeed in our allocation, then we
883 * add these pages to the cpu_buffers. Otherwise we just free
884 * them all and return -ENOMEM;
886 if (RB_WARN_ON(buffer, nr_pages <= buffer->pages))
889 new_pages = nr_pages - buffer->pages;
891 for_each_buffer_cpu(buffer, cpu) {
892 for (i = 0; i < new_pages; i++) {
893 bpage = kzalloc_node(ALIGN(sizeof(*bpage),
895 GFP_KERNEL, cpu_to_node(cpu));
898 list_add(&bpage->list, &pages);
899 addr = __get_free_page(GFP_KERNEL);
902 bpage->page = (void *)addr;
903 rb_init_page(bpage->page);
907 for_each_buffer_cpu(buffer, cpu) {
908 cpu_buffer = buffer->buffers[cpu];
909 rb_insert_pages(cpu_buffer, &pages, new_pages);
912 if (RB_WARN_ON(buffer, !list_empty(&pages)))
916 buffer->pages = nr_pages;
918 mutex_unlock(&buffer->mutex);
923 list_for_each_entry_safe(bpage, tmp, &pages, list) {
924 list_del_init(&bpage->list);
925 free_buffer_page(bpage);
928 mutex_unlock(&buffer->mutex);
932 * Something went totally wrong, and we are too paranoid
933 * to even clean up the mess.
937 mutex_unlock(&buffer->mutex);
940 EXPORT_SYMBOL_GPL(ring_buffer_resize);
943 __rb_data_page_index(struct buffer_data_page *bpage, unsigned index)
945 return bpage->data + index;
948 static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
950 return bpage->page->data + index;
953 static inline struct ring_buffer_event *
954 rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
956 return __rb_page_index(cpu_buffer->reader_page,
957 cpu_buffer->reader_page->read);
960 static inline struct ring_buffer_event *
961 rb_head_event(struct ring_buffer_per_cpu *cpu_buffer)
963 return __rb_page_index(cpu_buffer->head_page,
964 cpu_buffer->head_page->read);
967 static inline struct ring_buffer_event *
968 rb_iter_head_event(struct ring_buffer_iter *iter)
970 return __rb_page_index(iter->head_page, iter->head);
973 static inline unsigned rb_page_write(struct buffer_page *bpage)
975 return local_read(&bpage->write);
978 static inline unsigned rb_page_commit(struct buffer_page *bpage)
980 return local_read(&bpage->page->commit);
983 /* Size is determined by what has been commited */
984 static inline unsigned rb_page_size(struct buffer_page *bpage)
986 return rb_page_commit(bpage);
989 static inline unsigned
990 rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
992 return rb_page_commit(cpu_buffer->commit_page);
995 static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer)
997 return rb_page_commit(cpu_buffer->head_page);
1000 static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer,
1001 struct buffer_page **bpage)
1003 struct list_head *p = (*bpage)->list.next;
1005 if (p == &cpu_buffer->pages)
1008 *bpage = list_entry(p, struct buffer_page, list);
1011 static inline unsigned
1012 rb_event_index(struct ring_buffer_event *event)
1014 unsigned long addr = (unsigned long)event;
1016 return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE;
1020 rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
1021 struct ring_buffer_event *event)
1023 unsigned long addr = (unsigned long)event;
1024 unsigned long index;
1026 index = rb_event_index(event);
1029 return cpu_buffer->commit_page->page == (void *)addr &&
1030 rb_commit_index(cpu_buffer) == index;
1034 rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
1037 * We only race with interrupts and NMIs on this CPU.
1038 * If we own the commit event, then we can commit
1039 * all others that interrupted us, since the interruptions
1040 * are in stack format (they finish before they come
1041 * back to us). This allows us to do a simple loop to
1042 * assign the commit to the tail.
1045 while (cpu_buffer->commit_page != cpu_buffer->tail_page) {
1046 cpu_buffer->commit_page->page->commit =
1047 cpu_buffer->commit_page->write;
1048 rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
1049 cpu_buffer->write_stamp =
1050 cpu_buffer->commit_page->page->time_stamp;
1051 /* add barrier to keep gcc from optimizing too much */
1054 while (rb_commit_index(cpu_buffer) !=
1055 rb_page_write(cpu_buffer->commit_page)) {
1056 cpu_buffer->commit_page->page->commit =
1057 cpu_buffer->commit_page->write;
1061 /* again, keep gcc from optimizing */
1065 * If an interrupt came in just after the first while loop
1066 * and pushed the tail page forward, we will be left with
1067 * a dangling commit that will never go forward.
1069 if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page))
1073 static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
1075 cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp;
1076 cpu_buffer->reader_page->read = 0;
1079 static void rb_inc_iter(struct ring_buffer_iter *iter)
1081 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1084 * The iterator could be on the reader page (it starts there).
1085 * But the head could have moved, since the reader was
1086 * found. Check for this case and assign the iterator
1087 * to the head page instead of next.
1089 if (iter->head_page == cpu_buffer->reader_page)
1090 iter->head_page = cpu_buffer->head_page;
1092 rb_inc_page(cpu_buffer, &iter->head_page);
1094 iter->read_stamp = iter->head_page->page->time_stamp;
1099 * ring_buffer_update_event - update event type and data
1100 * @event: the even to update
1101 * @type: the type of event
1102 * @length: the size of the event field in the ring buffer
1104 * Update the type and data fields of the event. The length
1105 * is the actual size that is written to the ring buffer,
1106 * and with this, we can determine what to place into the
1110 rb_update_event(struct ring_buffer_event *event,
1111 unsigned type, unsigned length)
1113 event->type_len = type;
1117 case RINGBUF_TYPE_PADDING:
1118 case RINGBUF_TYPE_TIME_EXTEND:
1119 case RINGBUF_TYPE_TIME_STAMP:
1123 length -= RB_EVNT_HDR_SIZE;
1124 if (length > RB_MAX_SMALL_DATA)
1125 event->array[0] = length;
1127 event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT);
1134 static unsigned rb_calculate_event_length(unsigned length)
1136 struct ring_buffer_event event; /* Used only for sizeof array */
1138 /* zero length can cause confusions */
1142 if (length > RB_MAX_SMALL_DATA)
1143 length += sizeof(event.array[0]);
1145 length += RB_EVNT_HDR_SIZE;
1146 length = ALIGN(length, RB_ALIGNMENT);
1152 rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer,
1153 struct buffer_page *tail_page,
1154 unsigned long tail, unsigned long length)
1156 struct ring_buffer_event *event;
1159 * Only the event that crossed the page boundary
1160 * must fill the old tail_page with padding.
1162 if (tail >= BUF_PAGE_SIZE) {
1163 local_sub(length, &tail_page->write);
1167 event = __rb_page_index(tail_page, tail);
1170 * If this event is bigger than the minimum size, then
1171 * we need to be careful that we don't subtract the
1172 * write counter enough to allow another writer to slip
1174 * We put in a discarded commit instead, to make sure
1175 * that this space is not used again.
1177 * If we are less than the minimum size, we don't need to
1180 if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) {
1181 /* No room for any events */
1183 /* Mark the rest of the page with padding */
1184 rb_event_set_padding(event);
1186 /* Set the write back to the previous setting */
1187 local_sub(length, &tail_page->write);
1191 /* Put in a discarded event */
1192 event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE;
1193 event->type_len = RINGBUF_TYPE_PADDING;
1194 /* time delta must be non zero */
1195 event->time_delta = 1;
1196 /* Account for this as an entry */
1197 local_inc(&tail_page->entries);
1198 local_inc(&cpu_buffer->entries);
1200 /* Set write to end of buffer */
1201 length = (tail + length) - BUF_PAGE_SIZE;
1202 local_sub(length, &tail_page->write);
1205 static struct ring_buffer_event *
1206 rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer,
1207 unsigned long length, unsigned long tail,
1208 struct buffer_page *commit_page,
1209 struct buffer_page *tail_page, u64 *ts)
1211 struct buffer_page *next_page, *head_page, *reader_page;
1212 struct ring_buffer *buffer = cpu_buffer->buffer;
1213 bool lock_taken = false;
1214 unsigned long flags;
1216 next_page = tail_page;
1218 local_irq_save(flags);
1220 * Since the write to the buffer is still not
1221 * fully lockless, we must be careful with NMIs.
1222 * The locks in the writers are taken when a write
1223 * crosses to a new page. The locks protect against
1224 * races with the readers (this will soon be fixed
1225 * with a lockless solution).
1227 * Because we can not protect against NMIs, and we
1228 * want to keep traces reentrant, we need to manage
1229 * what happens when we are in an NMI.
1231 * NMIs can happen after we take the lock.
1232 * If we are in an NMI, only take the lock
1233 * if it is not already taken. Otherwise
1236 if (unlikely(in_nmi())) {
1237 if (!__raw_spin_trylock(&cpu_buffer->lock)) {
1238 cpu_buffer->nmi_dropped++;
1242 __raw_spin_lock(&cpu_buffer->lock);
1246 rb_inc_page(cpu_buffer, &next_page);
1248 head_page = cpu_buffer->head_page;
1249 reader_page = cpu_buffer->reader_page;
1251 /* we grabbed the lock before incrementing */
1252 if (RB_WARN_ON(cpu_buffer, next_page == reader_page))
1256 * If for some reason, we had an interrupt storm that made
1257 * it all the way around the buffer, bail, and warn
1260 if (unlikely(next_page == commit_page)) {
1261 cpu_buffer->commit_overrun++;
1265 if (next_page == head_page) {
1266 if (!(buffer->flags & RB_FL_OVERWRITE))
1269 /* tail_page has not moved yet? */
1270 if (tail_page == cpu_buffer->tail_page) {
1271 /* count overflows */
1272 cpu_buffer->overrun +=
1273 local_read(&head_page->entries);
1275 rb_inc_page(cpu_buffer, &head_page);
1276 cpu_buffer->head_page = head_page;
1277 cpu_buffer->head_page->read = 0;
1282 * If the tail page is still the same as what we think
1283 * it is, then it is up to us to update the tail
1286 if (tail_page == cpu_buffer->tail_page) {
1287 local_set(&next_page->write, 0);
1288 local_set(&next_page->entries, 0);
1289 local_set(&next_page->page->commit, 0);
1290 cpu_buffer->tail_page = next_page;
1292 /* reread the time stamp */
1293 *ts = rb_time_stamp(buffer, cpu_buffer->cpu);
1294 cpu_buffer->tail_page->page->time_stamp = *ts;
1297 rb_reset_tail(cpu_buffer, tail_page, tail, length);
1299 __raw_spin_unlock(&cpu_buffer->lock);
1300 local_irq_restore(flags);
1302 /* fail and let the caller try again */
1303 return ERR_PTR(-EAGAIN);
1307 rb_reset_tail(cpu_buffer, tail_page, tail, length);
1309 if (likely(lock_taken))
1310 __raw_spin_unlock(&cpu_buffer->lock);
1311 local_irq_restore(flags);
1315 static struct ring_buffer_event *
1316 __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
1317 unsigned type, unsigned long length, u64 *ts)
1319 struct buffer_page *tail_page, *commit_page;
1320 struct ring_buffer_event *event;
1321 unsigned long tail, write;
1323 commit_page = cpu_buffer->commit_page;
1324 /* we just need to protect against interrupts */
1326 tail_page = cpu_buffer->tail_page;
1327 write = local_add_return(length, &tail_page->write);
1328 tail = write - length;
1330 /* See if we shot pass the end of this buffer page */
1331 if (write > BUF_PAGE_SIZE)
1332 return rb_move_tail(cpu_buffer, length, tail,
1333 commit_page, tail_page, ts);
1335 /* We reserved something on the buffer */
1337 event = __rb_page_index(tail_page, tail);
1338 rb_update_event(event, type, length);
1340 /* The passed in type is zero for DATA */
1342 local_inc(&tail_page->entries);
1345 * If this is the first commit on the page, then update
1349 tail_page->page->time_stamp = *ts;
1355 rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer,
1356 struct ring_buffer_event *event)
1358 unsigned long new_index, old_index;
1359 struct buffer_page *bpage;
1360 unsigned long index;
1363 new_index = rb_event_index(event);
1364 old_index = new_index + rb_event_length(event);
1365 addr = (unsigned long)event;
1368 bpage = cpu_buffer->tail_page;
1370 if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) {
1372 * This is on the tail page. It is possible that
1373 * a write could come in and move the tail page
1374 * and write to the next page. That is fine
1375 * because we just shorten what is on this page.
1377 index = local_cmpxchg(&bpage->write, old_index, new_index);
1378 if (index == old_index)
1382 /* could not discard */
1387 rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer,
1388 u64 *ts, u64 *delta)
1390 struct ring_buffer_event *event;
1394 if (unlikely(*delta > (1ULL << 59) && !once++)) {
1395 printk(KERN_WARNING "Delta way too big! %llu"
1396 " ts=%llu write stamp = %llu\n",
1397 (unsigned long long)*delta,
1398 (unsigned long long)*ts,
1399 (unsigned long long)cpu_buffer->write_stamp);
1404 * The delta is too big, we to add a
1407 event = __rb_reserve_next(cpu_buffer,
1408 RINGBUF_TYPE_TIME_EXTEND,
1414 if (PTR_ERR(event) == -EAGAIN)
1417 /* Only a commited time event can update the write stamp */
1418 if (rb_event_is_commit(cpu_buffer, event)) {
1420 * If this is the first on the page, then it was
1421 * updated with the page itself. Try to discard it
1422 * and if we can't just make it zero.
1424 if (rb_event_index(event)) {
1425 event->time_delta = *delta & TS_MASK;
1426 event->array[0] = *delta >> TS_SHIFT;
1428 /* try to discard, since we do not need this */
1429 if (!rb_try_to_discard(cpu_buffer, event)) {
1430 /* nope, just zero it */
1431 event->time_delta = 0;
1432 event->array[0] = 0;
1435 cpu_buffer->write_stamp = *ts;
1436 /* let the caller know this was the commit */
1439 /* Try to discard the event */
1440 if (!rb_try_to_discard(cpu_buffer, event)) {
1441 /* Darn, this is just wasted space */
1442 event->time_delta = 0;
1443 event->array[0] = 0;
1453 static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer)
1455 local_inc(&cpu_buffer->committing);
1456 local_inc(&cpu_buffer->commits);
1459 static void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer)
1461 unsigned long commits;
1463 if (RB_WARN_ON(cpu_buffer,
1464 !local_read(&cpu_buffer->committing)))
1468 commits = local_read(&cpu_buffer->commits);
1469 /* synchronize with interrupts */
1471 if (local_read(&cpu_buffer->committing) == 1)
1472 rb_set_commit_to_write(cpu_buffer);
1474 local_dec(&cpu_buffer->committing);
1476 /* synchronize with interrupts */
1480 * Need to account for interrupts coming in between the
1481 * updating of the commit page and the clearing of the
1482 * committing counter.
1484 if (unlikely(local_read(&cpu_buffer->commits) != commits) &&
1485 !local_read(&cpu_buffer->committing)) {
1486 local_inc(&cpu_buffer->committing);
1491 static struct ring_buffer_event *
1492 rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer,
1493 unsigned long length)
1495 struct ring_buffer_event *event;
1500 rb_start_commit(cpu_buffer);
1502 length = rb_calculate_event_length(length);
1505 * We allow for interrupts to reenter here and do a trace.
1506 * If one does, it will cause this original code to loop
1507 * back here. Even with heavy interrupts happening, this
1508 * should only happen a few times in a row. If this happens
1509 * 1000 times in a row, there must be either an interrupt
1510 * storm or we have something buggy.
1513 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
1516 ts = rb_time_stamp(cpu_buffer->buffer, cpu_buffer->cpu);
1519 * Only the first commit can update the timestamp.
1520 * Yes there is a race here. If an interrupt comes in
1521 * just after the conditional and it traces too, then it
1522 * will also check the deltas. More than one timestamp may
1523 * also be made. But only the entry that did the actual
1524 * commit will be something other than zero.
1526 if (likely(cpu_buffer->tail_page == cpu_buffer->commit_page &&
1527 rb_page_write(cpu_buffer->tail_page) ==
1528 rb_commit_index(cpu_buffer))) {
1531 diff = ts - cpu_buffer->write_stamp;
1533 /* make sure this diff is calculated here */
1536 /* Did the write stamp get updated already? */
1537 if (unlikely(ts < cpu_buffer->write_stamp))
1541 if (unlikely(test_time_stamp(delta))) {
1543 commit = rb_add_time_stamp(cpu_buffer, &ts, &delta);
1544 if (commit == -EBUSY)
1547 if (commit == -EAGAIN)
1550 RB_WARN_ON(cpu_buffer, commit < 0);
1555 event = __rb_reserve_next(cpu_buffer, 0, length, &ts);
1556 if (unlikely(PTR_ERR(event) == -EAGAIN))
1562 if (!rb_event_is_commit(cpu_buffer, event))
1565 event->time_delta = delta;
1570 rb_end_commit(cpu_buffer);
1574 #define TRACE_RECURSIVE_DEPTH 16
1576 static int trace_recursive_lock(void)
1578 current->trace_recursion++;
1580 if (likely(current->trace_recursion < TRACE_RECURSIVE_DEPTH))
1583 /* Disable all tracing before we do anything else */
1584 tracing_off_permanent();
1586 printk_once(KERN_WARNING "Tracing recursion: depth[%ld]:"
1587 "HC[%lu]:SC[%lu]:NMI[%lu]\n",
1588 current->trace_recursion,
1589 hardirq_count() >> HARDIRQ_SHIFT,
1590 softirq_count() >> SOFTIRQ_SHIFT,
1597 static void trace_recursive_unlock(void)
1599 WARN_ON_ONCE(!current->trace_recursion);
1601 current->trace_recursion--;
1604 static DEFINE_PER_CPU(int, rb_need_resched);
1607 * ring_buffer_lock_reserve - reserve a part of the buffer
1608 * @buffer: the ring buffer to reserve from
1609 * @length: the length of the data to reserve (excluding event header)
1611 * Returns a reseverd event on the ring buffer to copy directly to.
1612 * The user of this interface will need to get the body to write into
1613 * and can use the ring_buffer_event_data() interface.
1615 * The length is the length of the data needed, not the event length
1616 * which also includes the event header.
1618 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
1619 * If NULL is returned, then nothing has been allocated or locked.
1621 struct ring_buffer_event *
1622 ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length)
1624 struct ring_buffer_per_cpu *cpu_buffer;
1625 struct ring_buffer_event *event;
1628 if (ring_buffer_flags != RB_BUFFERS_ON)
1631 if (atomic_read(&buffer->record_disabled))
1634 /* If we are tracing schedule, we don't want to recurse */
1635 resched = ftrace_preempt_disable();
1637 if (trace_recursive_lock())
1640 cpu = raw_smp_processor_id();
1642 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1645 cpu_buffer = buffer->buffers[cpu];
1647 if (atomic_read(&cpu_buffer->record_disabled))
1650 if (length > BUF_MAX_DATA_SIZE)
1653 event = rb_reserve_next_event(cpu_buffer, length);
1658 * Need to store resched state on this cpu.
1659 * Only the first needs to.
1662 if (preempt_count() == 1)
1663 per_cpu(rb_need_resched, cpu) = resched;
1668 trace_recursive_unlock();
1671 ftrace_preempt_enable(resched);
1674 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve);
1676 static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
1677 struct ring_buffer_event *event)
1679 local_inc(&cpu_buffer->entries);
1682 * The event first in the commit queue updates the
1685 if (rb_event_is_commit(cpu_buffer, event))
1686 cpu_buffer->write_stamp += event->time_delta;
1688 rb_end_commit(cpu_buffer);
1692 * ring_buffer_unlock_commit - commit a reserved
1693 * @buffer: The buffer to commit to
1694 * @event: The event pointer to commit.
1696 * This commits the data to the ring buffer, and releases any locks held.
1698 * Must be paired with ring_buffer_lock_reserve.
1700 int ring_buffer_unlock_commit(struct ring_buffer *buffer,
1701 struct ring_buffer_event *event)
1703 struct ring_buffer_per_cpu *cpu_buffer;
1704 int cpu = raw_smp_processor_id();
1706 cpu_buffer = buffer->buffers[cpu];
1708 rb_commit(cpu_buffer, event);
1710 trace_recursive_unlock();
1713 * Only the last preempt count needs to restore preemption.
1715 if (preempt_count() == 1)
1716 ftrace_preempt_enable(per_cpu(rb_need_resched, cpu));
1718 preempt_enable_no_resched_notrace();
1722 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit);
1724 static inline void rb_event_discard(struct ring_buffer_event *event)
1726 /* array[0] holds the actual length for the discarded event */
1727 event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE;
1728 event->type_len = RINGBUF_TYPE_PADDING;
1729 /* time delta must be non zero */
1730 if (!event->time_delta)
1731 event->time_delta = 1;
1735 * ring_buffer_event_discard - discard any event in the ring buffer
1736 * @event: the event to discard
1738 * Sometimes a event that is in the ring buffer needs to be ignored.
1739 * This function lets the user discard an event in the ring buffer
1740 * and then that event will not be read later.
1742 * Note, it is up to the user to be careful with this, and protect
1743 * against races. If the user discards an event that has been consumed
1744 * it is possible that it could corrupt the ring buffer.
1746 void ring_buffer_event_discard(struct ring_buffer_event *event)
1748 rb_event_discard(event);
1750 EXPORT_SYMBOL_GPL(ring_buffer_event_discard);
1753 * ring_buffer_commit_discard - discard an event that has not been committed
1754 * @buffer: the ring buffer
1755 * @event: non committed event to discard
1757 * This is similar to ring_buffer_event_discard but must only be
1758 * performed on an event that has not been committed yet. The difference
1759 * is that this will also try to free the event from the ring buffer
1760 * if another event has not been added behind it.
1762 * If another event has been added behind it, it will set the event
1763 * up as discarded, and perform the commit.
1765 * If this function is called, do not call ring_buffer_unlock_commit on
1768 void ring_buffer_discard_commit(struct ring_buffer *buffer,
1769 struct ring_buffer_event *event)
1771 struct ring_buffer_per_cpu *cpu_buffer;
1774 /* The event is discarded regardless */
1775 rb_event_discard(event);
1777 cpu = smp_processor_id();
1778 cpu_buffer = buffer->buffers[cpu];
1781 * This must only be called if the event has not been
1782 * committed yet. Thus we can assume that preemption
1783 * is still disabled.
1785 RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing));
1787 if (!rb_try_to_discard(cpu_buffer, event))
1791 * The commit is still visible by the reader, so we
1792 * must increment entries.
1794 local_inc(&cpu_buffer->entries);
1796 rb_end_commit(cpu_buffer);
1798 trace_recursive_unlock();
1801 * Only the last preempt count needs to restore preemption.
1803 if (preempt_count() == 1)
1804 ftrace_preempt_enable(per_cpu(rb_need_resched, cpu));
1806 preempt_enable_no_resched_notrace();
1809 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit);
1812 * ring_buffer_write - write data to the buffer without reserving
1813 * @buffer: The ring buffer to write to.
1814 * @length: The length of the data being written (excluding the event header)
1815 * @data: The data to write to the buffer.
1817 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
1818 * one function. If you already have the data to write to the buffer, it
1819 * may be easier to simply call this function.
1821 * Note, like ring_buffer_lock_reserve, the length is the length of the data
1822 * and not the length of the event which would hold the header.
1824 int ring_buffer_write(struct ring_buffer *buffer,
1825 unsigned long length,
1828 struct ring_buffer_per_cpu *cpu_buffer;
1829 struct ring_buffer_event *event;
1834 if (ring_buffer_flags != RB_BUFFERS_ON)
1837 if (atomic_read(&buffer->record_disabled))
1840 resched = ftrace_preempt_disable();
1842 cpu = raw_smp_processor_id();
1844 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1847 cpu_buffer = buffer->buffers[cpu];
1849 if (atomic_read(&cpu_buffer->record_disabled))
1852 if (length > BUF_MAX_DATA_SIZE)
1855 event = rb_reserve_next_event(cpu_buffer, length);
1859 body = rb_event_data(event);
1861 memcpy(body, data, length);
1863 rb_commit(cpu_buffer, event);
1867 ftrace_preempt_enable(resched);
1871 EXPORT_SYMBOL_GPL(ring_buffer_write);
1873 static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
1875 struct buffer_page *reader = cpu_buffer->reader_page;
1876 struct buffer_page *head = cpu_buffer->head_page;
1877 struct buffer_page *commit = cpu_buffer->commit_page;
1879 return reader->read == rb_page_commit(reader) &&
1880 (commit == reader ||
1882 head->read == rb_page_commit(commit)));
1886 * ring_buffer_record_disable - stop all writes into the buffer
1887 * @buffer: The ring buffer to stop writes to.
1889 * This prevents all writes to the buffer. Any attempt to write
1890 * to the buffer after this will fail and return NULL.
1892 * The caller should call synchronize_sched() after this.
1894 void ring_buffer_record_disable(struct ring_buffer *buffer)
1896 atomic_inc(&buffer->record_disabled);
1898 EXPORT_SYMBOL_GPL(ring_buffer_record_disable);
1901 * ring_buffer_record_enable - enable writes to the buffer
1902 * @buffer: The ring buffer to enable writes
1904 * Note, multiple disables will need the same number of enables
1905 * to truely enable the writing (much like preempt_disable).
1907 void ring_buffer_record_enable(struct ring_buffer *buffer)
1909 atomic_dec(&buffer->record_disabled);
1911 EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
1914 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
1915 * @buffer: The ring buffer to stop writes to.
1916 * @cpu: The CPU buffer to stop
1918 * This prevents all writes to the buffer. Any attempt to write
1919 * to the buffer after this will fail and return NULL.
1921 * The caller should call synchronize_sched() after this.
1923 void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu)
1925 struct ring_buffer_per_cpu *cpu_buffer;
1927 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1930 cpu_buffer = buffer->buffers[cpu];
1931 atomic_inc(&cpu_buffer->record_disabled);
1933 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu);
1936 * ring_buffer_record_enable_cpu - enable writes to the buffer
1937 * @buffer: The ring buffer to enable writes
1938 * @cpu: The CPU to enable.
1940 * Note, multiple disables will need the same number of enables
1941 * to truely enable the writing (much like preempt_disable).
1943 void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu)
1945 struct ring_buffer_per_cpu *cpu_buffer;
1947 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1950 cpu_buffer = buffer->buffers[cpu];
1951 atomic_dec(&cpu_buffer->record_disabled);
1953 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
1956 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
1957 * @buffer: The ring buffer
1958 * @cpu: The per CPU buffer to get the entries from.
1960 unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
1962 struct ring_buffer_per_cpu *cpu_buffer;
1965 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1968 cpu_buffer = buffer->buffers[cpu];
1969 ret = (local_read(&cpu_buffer->entries) - cpu_buffer->overrun)
1974 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
1977 * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
1978 * @buffer: The ring buffer
1979 * @cpu: The per CPU buffer to get the number of overruns from
1981 unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu)
1983 struct ring_buffer_per_cpu *cpu_buffer;
1986 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1989 cpu_buffer = buffer->buffers[cpu];
1990 ret = cpu_buffer->overrun;
1994 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
1997 * ring_buffer_nmi_dropped_cpu - get the number of nmis that were dropped
1998 * @buffer: The ring buffer
1999 * @cpu: The per CPU buffer to get the number of overruns from
2001 unsigned long ring_buffer_nmi_dropped_cpu(struct ring_buffer *buffer, int cpu)
2003 struct ring_buffer_per_cpu *cpu_buffer;
2006 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2009 cpu_buffer = buffer->buffers[cpu];
2010 ret = cpu_buffer->nmi_dropped;
2014 EXPORT_SYMBOL_GPL(ring_buffer_nmi_dropped_cpu);
2017 * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits
2018 * @buffer: The ring buffer
2019 * @cpu: The per CPU buffer to get the number of overruns from
2022 ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu)
2024 struct ring_buffer_per_cpu *cpu_buffer;
2027 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2030 cpu_buffer = buffer->buffers[cpu];
2031 ret = cpu_buffer->commit_overrun;
2035 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu);
2038 * ring_buffer_entries - get the number of entries in a buffer
2039 * @buffer: The ring buffer
2041 * Returns the total number of entries in the ring buffer
2044 unsigned long ring_buffer_entries(struct ring_buffer *buffer)
2046 struct ring_buffer_per_cpu *cpu_buffer;
2047 unsigned long entries = 0;
2050 /* if you care about this being correct, lock the buffer */
2051 for_each_buffer_cpu(buffer, cpu) {
2052 cpu_buffer = buffer->buffers[cpu];
2053 entries += (local_read(&cpu_buffer->entries) -
2054 cpu_buffer->overrun) - cpu_buffer->read;
2059 EXPORT_SYMBOL_GPL(ring_buffer_entries);
2062 * ring_buffer_overrun_cpu - get the number of overruns in buffer
2063 * @buffer: The ring buffer
2065 * Returns the total number of overruns in the ring buffer
2068 unsigned long ring_buffer_overruns(struct ring_buffer *buffer)
2070 struct ring_buffer_per_cpu *cpu_buffer;
2071 unsigned long overruns = 0;
2074 /* if you care about this being correct, lock the buffer */
2075 for_each_buffer_cpu(buffer, cpu) {
2076 cpu_buffer = buffer->buffers[cpu];
2077 overruns += cpu_buffer->overrun;
2082 EXPORT_SYMBOL_GPL(ring_buffer_overruns);
2084 static void rb_iter_reset(struct ring_buffer_iter *iter)
2086 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2088 /* Iterator usage is expected to have record disabled */
2089 if (list_empty(&cpu_buffer->reader_page->list)) {
2090 iter->head_page = cpu_buffer->head_page;
2091 iter->head = cpu_buffer->head_page->read;
2093 iter->head_page = cpu_buffer->reader_page;
2094 iter->head = cpu_buffer->reader_page->read;
2097 iter->read_stamp = cpu_buffer->read_stamp;
2099 iter->read_stamp = iter->head_page->page->time_stamp;
2103 * ring_buffer_iter_reset - reset an iterator
2104 * @iter: The iterator to reset
2106 * Resets the iterator, so that it will start from the beginning
2109 void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
2111 struct ring_buffer_per_cpu *cpu_buffer;
2112 unsigned long flags;
2117 cpu_buffer = iter->cpu_buffer;
2119 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2120 rb_iter_reset(iter);
2121 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2123 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset);
2126 * ring_buffer_iter_empty - check if an iterator has no more to read
2127 * @iter: The iterator to check
2129 int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
2131 struct ring_buffer_per_cpu *cpu_buffer;
2133 cpu_buffer = iter->cpu_buffer;
2135 return iter->head_page == cpu_buffer->commit_page &&
2136 iter->head == rb_commit_index(cpu_buffer);
2138 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty);
2141 rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
2142 struct ring_buffer_event *event)
2146 switch (event->type_len) {
2147 case RINGBUF_TYPE_PADDING:
2150 case RINGBUF_TYPE_TIME_EXTEND:
2151 delta = event->array[0];
2153 delta += event->time_delta;
2154 cpu_buffer->read_stamp += delta;
2157 case RINGBUF_TYPE_TIME_STAMP:
2158 /* FIXME: not implemented */
2161 case RINGBUF_TYPE_DATA:
2162 cpu_buffer->read_stamp += event->time_delta;
2172 rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
2173 struct ring_buffer_event *event)
2177 switch (event->type_len) {
2178 case RINGBUF_TYPE_PADDING:
2181 case RINGBUF_TYPE_TIME_EXTEND:
2182 delta = event->array[0];
2184 delta += event->time_delta;
2185 iter->read_stamp += delta;
2188 case RINGBUF_TYPE_TIME_STAMP:
2189 /* FIXME: not implemented */
2192 case RINGBUF_TYPE_DATA:
2193 iter->read_stamp += event->time_delta;
2202 static struct buffer_page *
2203 rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
2205 struct buffer_page *reader = NULL;
2206 unsigned long flags;
2209 local_irq_save(flags);
2210 __raw_spin_lock(&cpu_buffer->lock);
2214 * This should normally only loop twice. But because the
2215 * start of the reader inserts an empty page, it causes
2216 * a case where we will loop three times. There should be no
2217 * reason to loop four times (that I know of).
2219 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
2224 reader = cpu_buffer->reader_page;
2226 /* If there's more to read, return this page */
2227 if (cpu_buffer->reader_page->read < rb_page_size(reader))
2230 /* Never should we have an index greater than the size */
2231 if (RB_WARN_ON(cpu_buffer,
2232 cpu_buffer->reader_page->read > rb_page_size(reader)))
2235 /* check if we caught up to the tail */
2237 if (cpu_buffer->commit_page == cpu_buffer->reader_page)
2241 * Splice the empty reader page into the list around the head.
2242 * Reset the reader page to size zero.
2245 reader = cpu_buffer->head_page;
2246 cpu_buffer->reader_page->list.next = reader->list.next;
2247 cpu_buffer->reader_page->list.prev = reader->list.prev;
2249 local_set(&cpu_buffer->reader_page->write, 0);
2250 local_set(&cpu_buffer->reader_page->entries, 0);
2251 local_set(&cpu_buffer->reader_page->page->commit, 0);
2253 /* Make the reader page now replace the head */
2254 reader->list.prev->next = &cpu_buffer->reader_page->list;
2255 reader->list.next->prev = &cpu_buffer->reader_page->list;
2258 * If the tail is on the reader, then we must set the head
2259 * to the inserted page, otherwise we set it one before.
2261 cpu_buffer->head_page = cpu_buffer->reader_page;
2263 if (cpu_buffer->commit_page != reader)
2264 rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
2266 /* Finally update the reader page to the new head */
2267 cpu_buffer->reader_page = reader;
2268 rb_reset_reader_page(cpu_buffer);
2273 __raw_spin_unlock(&cpu_buffer->lock);
2274 local_irq_restore(flags);
2279 static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
2281 struct ring_buffer_event *event;
2282 struct buffer_page *reader;
2285 reader = rb_get_reader_page(cpu_buffer);
2287 /* This function should not be called when buffer is empty */
2288 if (RB_WARN_ON(cpu_buffer, !reader))
2291 event = rb_reader_event(cpu_buffer);
2293 if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX
2294 || rb_discarded_event(event))
2297 rb_update_read_stamp(cpu_buffer, event);
2299 length = rb_event_length(event);
2300 cpu_buffer->reader_page->read += length;
2303 static void rb_advance_iter(struct ring_buffer_iter *iter)
2305 struct ring_buffer *buffer;
2306 struct ring_buffer_per_cpu *cpu_buffer;
2307 struct ring_buffer_event *event;
2310 cpu_buffer = iter->cpu_buffer;
2311 buffer = cpu_buffer->buffer;
2314 * Check if we are at the end of the buffer.
2316 if (iter->head >= rb_page_size(iter->head_page)) {
2317 /* discarded commits can make the page empty */
2318 if (iter->head_page == cpu_buffer->commit_page)
2324 event = rb_iter_head_event(iter);
2326 length = rb_event_length(event);
2329 * This should not be called to advance the header if we are
2330 * at the tail of the buffer.
2332 if (RB_WARN_ON(cpu_buffer,
2333 (iter->head_page == cpu_buffer->commit_page) &&
2334 (iter->head + length > rb_commit_index(cpu_buffer))))
2337 rb_update_iter_read_stamp(iter, event);
2339 iter->head += length;
2341 /* check for end of page padding */
2342 if ((iter->head >= rb_page_size(iter->head_page)) &&
2343 (iter->head_page != cpu_buffer->commit_page))
2344 rb_advance_iter(iter);
2347 static struct ring_buffer_event *
2348 rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
2350 struct ring_buffer_per_cpu *cpu_buffer;
2351 struct ring_buffer_event *event;
2352 struct buffer_page *reader;
2355 cpu_buffer = buffer->buffers[cpu];
2359 * We repeat when a timestamp is encountered. It is possible
2360 * to get multiple timestamps from an interrupt entering just
2361 * as one timestamp is about to be written, or from discarded
2362 * commits. The most that we can have is the number on a single page.
2364 if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE))
2367 reader = rb_get_reader_page(cpu_buffer);
2371 event = rb_reader_event(cpu_buffer);
2373 switch (event->type_len) {
2374 case RINGBUF_TYPE_PADDING:
2375 if (rb_null_event(event))
2376 RB_WARN_ON(cpu_buffer, 1);
2378 * Because the writer could be discarding every
2379 * event it creates (which would probably be bad)
2380 * if we were to go back to "again" then we may never
2381 * catch up, and will trigger the warn on, or lock
2382 * the box. Return the padding, and we will release
2383 * the current locks, and try again.
2385 rb_advance_reader(cpu_buffer);
2388 case RINGBUF_TYPE_TIME_EXTEND:
2389 /* Internal data, OK to advance */
2390 rb_advance_reader(cpu_buffer);
2393 case RINGBUF_TYPE_TIME_STAMP:
2394 /* FIXME: not implemented */
2395 rb_advance_reader(cpu_buffer);
2398 case RINGBUF_TYPE_DATA:
2400 *ts = cpu_buffer->read_stamp + event->time_delta;
2401 ring_buffer_normalize_time_stamp(buffer,
2402 cpu_buffer->cpu, ts);
2412 EXPORT_SYMBOL_GPL(ring_buffer_peek);
2414 static struct ring_buffer_event *
2415 rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
2417 struct ring_buffer *buffer;
2418 struct ring_buffer_per_cpu *cpu_buffer;
2419 struct ring_buffer_event *event;
2422 if (ring_buffer_iter_empty(iter))
2425 cpu_buffer = iter->cpu_buffer;
2426 buffer = cpu_buffer->buffer;
2430 * We repeat when a timestamp is encountered.
2431 * We can get multiple timestamps by nested interrupts or also
2432 * if filtering is on (discarding commits). Since discarding
2433 * commits can be frequent we can get a lot of timestamps.
2434 * But we limit them by not adding timestamps if they begin
2435 * at the start of a page.
2437 if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE))
2440 if (rb_per_cpu_empty(cpu_buffer))
2443 event = rb_iter_head_event(iter);
2445 switch (event->type_len) {
2446 case RINGBUF_TYPE_PADDING:
2447 if (rb_null_event(event)) {
2451 rb_advance_iter(iter);
2454 case RINGBUF_TYPE_TIME_EXTEND:
2455 /* Internal data, OK to advance */
2456 rb_advance_iter(iter);
2459 case RINGBUF_TYPE_TIME_STAMP:
2460 /* FIXME: not implemented */
2461 rb_advance_iter(iter);
2464 case RINGBUF_TYPE_DATA:
2466 *ts = iter->read_stamp + event->time_delta;
2467 ring_buffer_normalize_time_stamp(buffer,
2468 cpu_buffer->cpu, ts);
2478 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek);
2481 * ring_buffer_peek - peek at the next event to be read
2482 * @buffer: The ring buffer to read
2483 * @cpu: The cpu to peak at
2484 * @ts: The timestamp counter of this event.
2486 * This will return the event that will be read next, but does
2487 * not consume the data.
2489 struct ring_buffer_event *
2490 ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
2492 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2493 struct ring_buffer_event *event;
2494 unsigned long flags;
2496 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2500 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2501 event = rb_buffer_peek(buffer, cpu, ts);
2502 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2504 if (event && event->type_len == RINGBUF_TYPE_PADDING) {
2513 * ring_buffer_iter_peek - peek at the next event to be read
2514 * @iter: The ring buffer iterator
2515 * @ts: The timestamp counter of this event.
2517 * This will return the event that will be read next, but does
2518 * not increment the iterator.
2520 struct ring_buffer_event *
2521 ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
2523 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2524 struct ring_buffer_event *event;
2525 unsigned long flags;
2528 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2529 event = rb_iter_peek(iter, ts);
2530 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2532 if (event && event->type_len == RINGBUF_TYPE_PADDING) {
2541 * ring_buffer_consume - return an event and consume it
2542 * @buffer: The ring buffer to get the next event from
2544 * Returns the next event in the ring buffer, and that event is consumed.
2545 * Meaning, that sequential reads will keep returning a different event,
2546 * and eventually empty the ring buffer if the producer is slower.
2548 struct ring_buffer_event *
2549 ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts)
2551 struct ring_buffer_per_cpu *cpu_buffer;
2552 struct ring_buffer_event *event = NULL;
2553 unsigned long flags;
2556 /* might be called in atomic */
2559 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2562 cpu_buffer = buffer->buffers[cpu];
2563 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2565 event = rb_buffer_peek(buffer, cpu, ts);
2569 rb_advance_reader(cpu_buffer);
2572 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2577 if (event && event->type_len == RINGBUF_TYPE_PADDING) {
2584 EXPORT_SYMBOL_GPL(ring_buffer_consume);
2587 * ring_buffer_read_start - start a non consuming read of the buffer
2588 * @buffer: The ring buffer to read from
2589 * @cpu: The cpu buffer to iterate over
2591 * This starts up an iteration through the buffer. It also disables
2592 * the recording to the buffer until the reading is finished.
2593 * This prevents the reading from being corrupted. This is not
2594 * a consuming read, so a producer is not expected.
2596 * Must be paired with ring_buffer_finish.
2598 struct ring_buffer_iter *
2599 ring_buffer_read_start(struct ring_buffer *buffer, int cpu)
2601 struct ring_buffer_per_cpu *cpu_buffer;
2602 struct ring_buffer_iter *iter;
2603 unsigned long flags;
2605 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2608 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
2612 cpu_buffer = buffer->buffers[cpu];
2614 iter->cpu_buffer = cpu_buffer;
2616 atomic_inc(&cpu_buffer->record_disabled);
2617 synchronize_sched();
2619 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2620 __raw_spin_lock(&cpu_buffer->lock);
2621 rb_iter_reset(iter);
2622 __raw_spin_unlock(&cpu_buffer->lock);
2623 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2627 EXPORT_SYMBOL_GPL(ring_buffer_read_start);
2630 * ring_buffer_finish - finish reading the iterator of the buffer
2631 * @iter: The iterator retrieved by ring_buffer_start
2633 * This re-enables the recording to the buffer, and frees the
2637 ring_buffer_read_finish(struct ring_buffer_iter *iter)
2639 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2641 atomic_dec(&cpu_buffer->record_disabled);
2644 EXPORT_SYMBOL_GPL(ring_buffer_read_finish);
2647 * ring_buffer_read - read the next item in the ring buffer by the iterator
2648 * @iter: The ring buffer iterator
2649 * @ts: The time stamp of the event read.
2651 * This reads the next event in the ring buffer and increments the iterator.
2653 struct ring_buffer_event *
2654 ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
2656 struct ring_buffer_event *event;
2657 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2658 unsigned long flags;
2661 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2662 event = rb_iter_peek(iter, ts);
2666 rb_advance_iter(iter);
2668 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2670 if (event && event->type_len == RINGBUF_TYPE_PADDING) {
2677 EXPORT_SYMBOL_GPL(ring_buffer_read);
2680 * ring_buffer_size - return the size of the ring buffer (in bytes)
2681 * @buffer: The ring buffer.
2683 unsigned long ring_buffer_size(struct ring_buffer *buffer)
2685 return BUF_PAGE_SIZE * buffer->pages;
2687 EXPORT_SYMBOL_GPL(ring_buffer_size);
2690 rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
2692 cpu_buffer->head_page
2693 = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
2694 local_set(&cpu_buffer->head_page->write, 0);
2695 local_set(&cpu_buffer->head_page->entries, 0);
2696 local_set(&cpu_buffer->head_page->page->commit, 0);
2698 cpu_buffer->head_page->read = 0;
2700 cpu_buffer->tail_page = cpu_buffer->head_page;
2701 cpu_buffer->commit_page = cpu_buffer->head_page;
2703 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
2704 local_set(&cpu_buffer->reader_page->write, 0);
2705 local_set(&cpu_buffer->reader_page->entries, 0);
2706 local_set(&cpu_buffer->reader_page->page->commit, 0);
2707 cpu_buffer->reader_page->read = 0;
2709 cpu_buffer->nmi_dropped = 0;
2710 cpu_buffer->commit_overrun = 0;
2711 cpu_buffer->overrun = 0;
2712 cpu_buffer->read = 0;
2713 local_set(&cpu_buffer->entries, 0);
2714 local_set(&cpu_buffer->committing, 0);
2715 local_set(&cpu_buffer->commits, 0);
2717 cpu_buffer->write_stamp = 0;
2718 cpu_buffer->read_stamp = 0;
2722 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
2723 * @buffer: The ring buffer to reset a per cpu buffer of
2724 * @cpu: The CPU buffer to be reset
2726 void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
2728 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2729 unsigned long flags;
2731 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2734 atomic_inc(&cpu_buffer->record_disabled);
2736 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2738 __raw_spin_lock(&cpu_buffer->lock);
2740 rb_reset_cpu(cpu_buffer);
2742 __raw_spin_unlock(&cpu_buffer->lock);
2744 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2746 atomic_dec(&cpu_buffer->record_disabled);
2748 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
2751 * ring_buffer_reset - reset a ring buffer
2752 * @buffer: The ring buffer to reset all cpu buffers
2754 void ring_buffer_reset(struct ring_buffer *buffer)
2758 for_each_buffer_cpu(buffer, cpu)
2759 ring_buffer_reset_cpu(buffer, cpu);
2761 EXPORT_SYMBOL_GPL(ring_buffer_reset);
2764 * rind_buffer_empty - is the ring buffer empty?
2765 * @buffer: The ring buffer to test
2767 int ring_buffer_empty(struct ring_buffer *buffer)
2769 struct ring_buffer_per_cpu *cpu_buffer;
2772 /* yes this is racy, but if you don't like the race, lock the buffer */
2773 for_each_buffer_cpu(buffer, cpu) {
2774 cpu_buffer = buffer->buffers[cpu];
2775 if (!rb_per_cpu_empty(cpu_buffer))
2781 EXPORT_SYMBOL_GPL(ring_buffer_empty);
2784 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
2785 * @buffer: The ring buffer
2786 * @cpu: The CPU buffer to test
2788 int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu)
2790 struct ring_buffer_per_cpu *cpu_buffer;
2793 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2796 cpu_buffer = buffer->buffers[cpu];
2797 ret = rb_per_cpu_empty(cpu_buffer);
2802 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
2805 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
2806 * @buffer_a: One buffer to swap with
2807 * @buffer_b: The other buffer to swap with
2809 * This function is useful for tracers that want to take a "snapshot"
2810 * of a CPU buffer and has another back up buffer lying around.
2811 * it is expected that the tracer handles the cpu buffer not being
2812 * used at the moment.
2814 int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
2815 struct ring_buffer *buffer_b, int cpu)
2817 struct ring_buffer_per_cpu *cpu_buffer_a;
2818 struct ring_buffer_per_cpu *cpu_buffer_b;
2821 if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
2822 !cpumask_test_cpu(cpu, buffer_b->cpumask))
2825 /* At least make sure the two buffers are somewhat the same */
2826 if (buffer_a->pages != buffer_b->pages)
2831 if (ring_buffer_flags != RB_BUFFERS_ON)
2834 if (atomic_read(&buffer_a->record_disabled))
2837 if (atomic_read(&buffer_b->record_disabled))
2840 cpu_buffer_a = buffer_a->buffers[cpu];
2841 cpu_buffer_b = buffer_b->buffers[cpu];
2843 if (atomic_read(&cpu_buffer_a->record_disabled))
2846 if (atomic_read(&cpu_buffer_b->record_disabled))
2850 * We can't do a synchronize_sched here because this
2851 * function can be called in atomic context.
2852 * Normally this will be called from the same CPU as cpu.
2853 * If not it's up to the caller to protect this.
2855 atomic_inc(&cpu_buffer_a->record_disabled);
2856 atomic_inc(&cpu_buffer_b->record_disabled);
2858 buffer_a->buffers[cpu] = cpu_buffer_b;
2859 buffer_b->buffers[cpu] = cpu_buffer_a;
2861 cpu_buffer_b->buffer = buffer_a;
2862 cpu_buffer_a->buffer = buffer_b;
2864 atomic_dec(&cpu_buffer_a->record_disabled);
2865 atomic_dec(&cpu_buffer_b->record_disabled);
2871 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
2874 * ring_buffer_alloc_read_page - allocate a page to read from buffer
2875 * @buffer: the buffer to allocate for.
2877 * This function is used in conjunction with ring_buffer_read_page.
2878 * When reading a full page from the ring buffer, these functions
2879 * can be used to speed up the process. The calling function should
2880 * allocate a few pages first with this function. Then when it
2881 * needs to get pages from the ring buffer, it passes the result
2882 * of this function into ring_buffer_read_page, which will swap
2883 * the page that was allocated, with the read page of the buffer.
2886 * The page allocated, or NULL on error.
2888 void *ring_buffer_alloc_read_page(struct ring_buffer *buffer)
2890 struct buffer_data_page *bpage;
2893 addr = __get_free_page(GFP_KERNEL);
2897 bpage = (void *)addr;
2899 rb_init_page(bpage);
2903 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page);
2906 * ring_buffer_free_read_page - free an allocated read page
2907 * @buffer: the buffer the page was allocate for
2908 * @data: the page to free
2910 * Free a page allocated from ring_buffer_alloc_read_page.
2912 void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data)
2914 free_page((unsigned long)data);
2916 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page);
2919 * ring_buffer_read_page - extract a page from the ring buffer
2920 * @buffer: buffer to extract from
2921 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
2922 * @len: amount to extract
2923 * @cpu: the cpu of the buffer to extract
2924 * @full: should the extraction only happen when the page is full.
2926 * This function will pull out a page from the ring buffer and consume it.
2927 * @data_page must be the address of the variable that was returned
2928 * from ring_buffer_alloc_read_page. This is because the page might be used
2929 * to swap with a page in the ring buffer.
2932 * rpage = ring_buffer_alloc_read_page(buffer);
2935 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
2937 * process_page(rpage, ret);
2939 * When @full is set, the function will not return true unless
2940 * the writer is off the reader page.
2942 * Note: it is up to the calling functions to handle sleeps and wakeups.
2943 * The ring buffer can be used anywhere in the kernel and can not
2944 * blindly call wake_up. The layer that uses the ring buffer must be
2945 * responsible for that.
2948 * >=0 if data has been transferred, returns the offset of consumed data.
2949 * <0 if no data has been transferred.
2951 int ring_buffer_read_page(struct ring_buffer *buffer,
2952 void **data_page, size_t len, int cpu, int full)
2954 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2955 struct ring_buffer_event *event;
2956 struct buffer_data_page *bpage;
2957 struct buffer_page *reader;
2958 unsigned long flags;
2959 unsigned int commit;
2964 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2968 * If len is not big enough to hold the page header, then
2969 * we can not copy anything.
2971 if (len <= BUF_PAGE_HDR_SIZE)
2974 len -= BUF_PAGE_HDR_SIZE;
2983 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2985 reader = rb_get_reader_page(cpu_buffer);
2989 event = rb_reader_event(cpu_buffer);
2991 read = reader->read;
2992 commit = rb_page_commit(reader);
2995 * If this page has been partially read or
2996 * if len is not big enough to read the rest of the page or
2997 * a writer is still on the page, then
2998 * we must copy the data from the page to the buffer.
2999 * Otherwise, we can simply swap the page with the one passed in.
3001 if (read || (len < (commit - read)) ||
3002 cpu_buffer->reader_page == cpu_buffer->commit_page) {
3003 struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
3004 unsigned int rpos = read;
3005 unsigned int pos = 0;
3011 if (len > (commit - read))
3012 len = (commit - read);
3014 size = rb_event_length(event);
3019 /* save the current timestamp, since the user will need it */
3020 save_timestamp = cpu_buffer->read_stamp;
3022 /* Need to copy one event at a time */
3024 memcpy(bpage->data + pos, rpage->data + rpos, size);
3028 rb_advance_reader(cpu_buffer);
3029 rpos = reader->read;
3032 event = rb_reader_event(cpu_buffer);
3033 size = rb_event_length(event);
3034 } while (len > size);
3037 local_set(&bpage->commit, pos);
3038 bpage->time_stamp = save_timestamp;
3040 /* we copied everything to the beginning */
3043 /* update the entry counter */
3044 cpu_buffer->read += local_read(&reader->entries);
3046 /* swap the pages */
3047 rb_init_page(bpage);
3048 bpage = reader->page;
3049 reader->page = *data_page;
3050 local_set(&reader->write, 0);
3051 local_set(&reader->entries, 0);
3058 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
3063 EXPORT_SYMBOL_GPL(ring_buffer_read_page);
3066 rb_simple_read(struct file *filp, char __user *ubuf,
3067 size_t cnt, loff_t *ppos)
3069 unsigned long *p = filp->private_data;
3073 if (test_bit(RB_BUFFERS_DISABLED_BIT, p))
3074 r = sprintf(buf, "permanently disabled\n");
3076 r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p));
3078 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3082 rb_simple_write(struct file *filp, const char __user *ubuf,
3083 size_t cnt, loff_t *ppos)
3085 unsigned long *p = filp->private_data;
3090 if (cnt >= sizeof(buf))
3093 if (copy_from_user(&buf, ubuf, cnt))
3098 ret = strict_strtoul(buf, 10, &val);
3103 set_bit(RB_BUFFERS_ON_BIT, p);
3105 clear_bit(RB_BUFFERS_ON_BIT, p);
3112 static const struct file_operations rb_simple_fops = {
3113 .open = tracing_open_generic,
3114 .read = rb_simple_read,
3115 .write = rb_simple_write,
3119 static __init int rb_init_debugfs(void)
3121 struct dentry *d_tracer;
3123 d_tracer = tracing_init_dentry();
3125 trace_create_file("tracing_on", 0644, d_tracer,
3126 &ring_buffer_flags, &rb_simple_fops);
3131 fs_initcall(rb_init_debugfs);
3133 #ifdef CONFIG_HOTPLUG_CPU
3134 static int rb_cpu_notify(struct notifier_block *self,
3135 unsigned long action, void *hcpu)
3137 struct ring_buffer *buffer =
3138 container_of(self, struct ring_buffer, cpu_notify);
3139 long cpu = (long)hcpu;
3142 case CPU_UP_PREPARE:
3143 case CPU_UP_PREPARE_FROZEN:
3144 if (cpumask_test_cpu(cpu, buffer->cpumask))
3147 buffer->buffers[cpu] =
3148 rb_allocate_cpu_buffer(buffer, cpu);
3149 if (!buffer->buffers[cpu]) {
3150 WARN(1, "failed to allocate ring buffer on CPU %ld\n",
3155 cpumask_set_cpu(cpu, buffer->cpumask);
3157 case CPU_DOWN_PREPARE:
3158 case CPU_DOWN_PREPARE_FROZEN:
3161 * If we were to free the buffer, then the user would
3162 * lose any trace that was in the buffer.