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/kmemcheck.h>
14 #include <linux/module.h>
15 #include <linux/percpu.h>
16 #include <linux/mutex.h>
17 #include <linux/init.h>
18 #include <linux/hash.h>
19 #include <linux/list.h>
20 #include <linux/cpu.h>
26 * The ring buffer header is special. We must manually up keep it.
28 int ring_buffer_print_entry_header(struct trace_seq *s)
32 ret = trace_seq_printf(s, "# compressed entry header\n");
33 ret = trace_seq_printf(s, "\ttype_len : 5 bits\n");
34 ret = trace_seq_printf(s, "\ttime_delta : 27 bits\n");
35 ret = trace_seq_printf(s, "\tarray : 32 bits\n");
36 ret = trace_seq_printf(s, "\n");
37 ret = trace_seq_printf(s, "\tpadding : type == %d\n",
38 RINGBUF_TYPE_PADDING);
39 ret = trace_seq_printf(s, "\ttime_extend : type == %d\n",
40 RINGBUF_TYPE_TIME_EXTEND);
41 ret = trace_seq_printf(s, "\tdata max type_len == %d\n",
42 RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
48 * The ring buffer is made up of a list of pages. A separate list of pages is
49 * allocated for each CPU. A writer may only write to a buffer that is
50 * associated with the CPU it is currently executing on. A reader may read
51 * from any per cpu buffer.
53 * The reader is special. For each per cpu buffer, the reader has its own
54 * reader page. When a reader has read the entire reader page, this reader
55 * page is swapped with another page in the ring buffer.
57 * Now, as long as the writer is off the reader page, the reader can do what
58 * ever it wants with that page. The writer will never write to that page
59 * again (as long as it is out of the ring buffer).
61 * Here's some silly ASCII art.
64 * |reader| RING BUFFER
66 * +------+ +---+ +---+ +---+
75 * |reader| RING BUFFER
76 * |page |------------------v
77 * +------+ +---+ +---+ +---+
86 * |reader| RING BUFFER
87 * |page |------------------v
88 * +------+ +---+ +---+ +---+
93 * +------------------------------+
97 * |buffer| RING BUFFER
98 * |page |------------------v
99 * +------+ +---+ +---+ +---+
101 * | New +---+ +---+ +---+
104 * +------------------------------+
107 * After we make this swap, the reader can hand this page off to the splice
108 * code and be done with it. It can even allocate a new page if it needs to
109 * and swap that into the ring buffer.
111 * We will be using cmpxchg soon to make all this lockless.
116 * A fast way to enable or disable all ring buffers is to
117 * call tracing_on or tracing_off. Turning off the ring buffers
118 * prevents all ring buffers from being recorded to.
119 * Turning this switch on, makes it OK to write to the
120 * ring buffer, if the ring buffer is enabled itself.
122 * There's three layers that must be on in order to write
123 * to the ring buffer.
125 * 1) This global flag must be set.
126 * 2) The ring buffer must be enabled for recording.
127 * 3) The per cpu buffer must be enabled for recording.
129 * In case of an anomaly, this global flag has a bit set that
130 * will permantly disable all ring buffers.
134 * Global flag to disable all recording to ring buffers
135 * This has two bits: ON, DISABLED
139 * 0 0 : ring buffers are off
140 * 1 0 : ring buffers are on
141 * X 1 : ring buffers are permanently disabled
145 RB_BUFFERS_ON_BIT = 0,
146 RB_BUFFERS_DISABLED_BIT = 1,
150 RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT,
151 RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT,
154 static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON;
156 #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
159 * tracing_on - enable all tracing buffers
161 * This function enables all tracing buffers that may have been
162 * disabled with tracing_off.
164 void tracing_on(void)
166 set_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags);
168 EXPORT_SYMBOL_GPL(tracing_on);
171 * tracing_off - turn off all tracing buffers
173 * This function stops all tracing buffers from recording data.
174 * It does not disable any overhead the tracers themselves may
175 * be causing. This function simply causes all recording to
176 * the ring buffers to fail.
178 void tracing_off(void)
180 clear_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags);
182 EXPORT_SYMBOL_GPL(tracing_off);
185 * tracing_off_permanent - permanently disable ring buffers
187 * This function, once called, will disable all ring buffers
190 void tracing_off_permanent(void)
192 set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags);
196 * tracing_is_on - show state of ring buffers enabled
198 int tracing_is_on(void)
200 return ring_buffer_flags == RB_BUFFERS_ON;
202 EXPORT_SYMBOL_GPL(tracing_is_on);
206 #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
207 #define RB_ALIGNMENT 4U
208 #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
209 #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */
211 /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
212 #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
215 RB_LEN_TIME_EXTEND = 8,
216 RB_LEN_TIME_STAMP = 16,
219 static inline int rb_null_event(struct ring_buffer_event *event)
221 return event->type_len == RINGBUF_TYPE_PADDING
222 && event->time_delta == 0;
225 static inline int rb_discarded_event(struct ring_buffer_event *event)
227 return event->type_len == RINGBUF_TYPE_PADDING && event->time_delta;
230 static void rb_event_set_padding(struct ring_buffer_event *event)
232 event->type_len = RINGBUF_TYPE_PADDING;
233 event->time_delta = 0;
237 rb_event_data_length(struct ring_buffer_event *event)
242 length = event->type_len * RB_ALIGNMENT;
244 length = event->array[0];
245 return length + RB_EVNT_HDR_SIZE;
248 /* inline for ring buffer fast paths */
250 rb_event_length(struct ring_buffer_event *event)
252 switch (event->type_len) {
253 case RINGBUF_TYPE_PADDING:
254 if (rb_null_event(event))
257 return event->array[0] + RB_EVNT_HDR_SIZE;
259 case RINGBUF_TYPE_TIME_EXTEND:
260 return RB_LEN_TIME_EXTEND;
262 case RINGBUF_TYPE_TIME_STAMP:
263 return RB_LEN_TIME_STAMP;
265 case RINGBUF_TYPE_DATA:
266 return rb_event_data_length(event);
275 * ring_buffer_event_length - return the length of the event
276 * @event: the event to get the length of
278 unsigned ring_buffer_event_length(struct ring_buffer_event *event)
280 unsigned length = rb_event_length(event);
281 if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
283 length -= RB_EVNT_HDR_SIZE;
284 if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0]))
285 length -= sizeof(event->array[0]);
288 EXPORT_SYMBOL_GPL(ring_buffer_event_length);
290 /* inline for ring buffer fast paths */
292 rb_event_data(struct ring_buffer_event *event)
294 BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
295 /* If length is in len field, then array[0] has the data */
297 return (void *)&event->array[0];
298 /* Otherwise length is in array[0] and array[1] has the data */
299 return (void *)&event->array[1];
303 * ring_buffer_event_data - return the data of the event
304 * @event: the event to get the data from
306 void *ring_buffer_event_data(struct ring_buffer_event *event)
308 return rb_event_data(event);
310 EXPORT_SYMBOL_GPL(ring_buffer_event_data);
312 #define for_each_buffer_cpu(buffer, cpu) \
313 for_each_cpu(cpu, buffer->cpumask)
316 #define TS_MASK ((1ULL << TS_SHIFT) - 1)
317 #define TS_DELTA_TEST (~TS_MASK)
319 struct buffer_data_page {
320 u64 time_stamp; /* page time stamp */
321 local_t commit; /* write committed index */
322 unsigned char data[]; /* data of buffer page */
326 struct list_head list; /* list of buffer pages */
327 local_t write; /* index for next write */
328 unsigned read; /* index for next read */
329 local_t entries; /* entries on this page */
330 struct buffer_data_page *page; /* Actual data page */
333 static void rb_init_page(struct buffer_data_page *bpage)
335 local_set(&bpage->commit, 0);
339 * ring_buffer_page_len - the size of data on the page.
340 * @page: The page to read
342 * Returns the amount of data on the page, including buffer page header.
344 size_t ring_buffer_page_len(void *page)
346 return local_read(&((struct buffer_data_page *)page)->commit)
351 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
354 static void free_buffer_page(struct buffer_page *bpage)
356 free_page((unsigned long)bpage->page);
361 * We need to fit the time_stamp delta into 27 bits.
363 static inline int test_time_stamp(u64 delta)
365 if (delta & TS_DELTA_TEST)
370 #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
372 /* Max payload is BUF_PAGE_SIZE - header (8bytes) */
373 #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
375 /* Max number of timestamps that can fit on a page */
376 #define RB_TIMESTAMPS_PER_PAGE (BUF_PAGE_SIZE / RB_LEN_TIME_STAMP)
378 int ring_buffer_print_page_header(struct trace_seq *s)
380 struct buffer_data_page field;
383 ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t"
384 "offset:0;\tsize:%u;\n",
385 (unsigned int)sizeof(field.time_stamp));
387 ret = trace_seq_printf(s, "\tfield: local_t commit;\t"
388 "offset:%u;\tsize:%u;\n",
389 (unsigned int)offsetof(typeof(field), commit),
390 (unsigned int)sizeof(field.commit));
392 ret = trace_seq_printf(s, "\tfield: char data;\t"
393 "offset:%u;\tsize:%u;\n",
394 (unsigned int)offsetof(typeof(field), data),
395 (unsigned int)BUF_PAGE_SIZE);
401 * head_page == tail_page && head == tail then buffer is empty.
403 struct ring_buffer_per_cpu {
405 struct ring_buffer *buffer;
406 spinlock_t reader_lock; /* serialize readers */
408 struct lock_class_key lock_key;
409 struct list_head *pages;
410 struct buffer_page *head_page; /* read from head */
411 struct buffer_page *tail_page; /* write to tail */
412 struct buffer_page *commit_page; /* committed pages */
413 struct buffer_page *reader_page;
414 unsigned long nmi_dropped;
415 unsigned long commit_overrun;
416 unsigned long overrun;
423 atomic_t record_disabled;
430 atomic_t record_disabled;
431 cpumask_var_t cpumask;
433 struct lock_class_key *reader_lock_key;
437 struct ring_buffer_per_cpu **buffers;
439 #ifdef CONFIG_HOTPLUG_CPU
440 struct notifier_block cpu_notify;
445 struct ring_buffer_iter {
446 struct ring_buffer_per_cpu *cpu_buffer;
448 struct buffer_page *head_page;
452 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
453 #define RB_WARN_ON(buffer, cond) \
455 int _____ret = unlikely(cond); \
457 atomic_inc(&buffer->record_disabled); \
463 /* Up this if you want to test the TIME_EXTENTS and normalization */
464 #define DEBUG_SHIFT 0
466 static inline u64 rb_time_stamp(struct ring_buffer *buffer, int cpu)
468 /* shift to debug/test normalization and TIME_EXTENTS */
469 return buffer->clock() << DEBUG_SHIFT;
472 u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu)
476 preempt_disable_notrace();
477 time = rb_time_stamp(buffer, cpu);
478 preempt_enable_no_resched_notrace();
482 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
484 void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer,
487 /* Just stupid testing the normalize function and deltas */
490 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
493 * check_pages - integrity check of buffer pages
494 * @cpu_buffer: CPU buffer with pages to test
496 * As a safety measure we check to make sure the data pages have not
499 static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
501 struct list_head *head = cpu_buffer->pages;
502 struct buffer_page *bpage, *tmp;
504 if (RB_WARN_ON(cpu_buffer, head->next->prev != head))
506 if (RB_WARN_ON(cpu_buffer, head->prev->next != head))
509 list_for_each_entry_safe(bpage, tmp, head, list) {
510 if (RB_WARN_ON(cpu_buffer,
511 bpage->list.next->prev != &bpage->list))
513 if (RB_WARN_ON(cpu_buffer,
514 bpage->list.prev->next != &bpage->list))
521 static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
524 struct buffer_page *bpage, *tmp;
531 for (i = 0; i < nr_pages; i++) {
532 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
533 GFP_KERNEL, cpu_to_node(cpu_buffer->cpu));
536 list_add(&bpage->list, &pages);
538 addr = __get_free_page(GFP_KERNEL);
541 bpage->page = (void *)addr;
542 rb_init_page(bpage->page);
546 * The ring buffer page list is a circular list that does not
547 * start and end with a list head. All page list items point to
550 cpu_buffer->pages = pages.next;
553 rb_check_pages(cpu_buffer);
558 list_for_each_entry_safe(bpage, tmp, &pages, list) {
559 list_del_init(&bpage->list);
560 free_buffer_page(bpage);
565 static struct ring_buffer_per_cpu *
566 rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu)
568 struct ring_buffer_per_cpu *cpu_buffer;
569 struct buffer_page *bpage;
573 cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
574 GFP_KERNEL, cpu_to_node(cpu));
578 cpu_buffer->cpu = cpu;
579 cpu_buffer->buffer = buffer;
580 spin_lock_init(&cpu_buffer->reader_lock);
581 lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key);
582 cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
584 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
585 GFP_KERNEL, cpu_to_node(cpu));
587 goto fail_free_buffer;
589 cpu_buffer->reader_page = bpage;
590 addr = __get_free_page(GFP_KERNEL);
592 goto fail_free_reader;
593 bpage->page = (void *)addr;
594 rb_init_page(bpage->page);
596 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
598 ret = rb_allocate_pages(cpu_buffer, buffer->pages);
600 goto fail_free_reader;
602 cpu_buffer->head_page
603 = list_entry(cpu_buffer->pages, struct buffer_page, list);
604 cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
609 free_buffer_page(cpu_buffer->reader_page);
616 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
618 struct list_head *head = cpu_buffer->pages;
619 struct buffer_page *bpage, *tmp;
621 free_buffer_page(cpu_buffer->reader_page);
624 list_for_each_entry_safe(bpage, tmp, head, list) {
625 list_del_init(&bpage->list);
626 free_buffer_page(bpage);
628 bpage = list_entry(head, struct buffer_page, list);
629 free_buffer_page(bpage);
635 #ifdef CONFIG_HOTPLUG_CPU
636 static int rb_cpu_notify(struct notifier_block *self,
637 unsigned long action, void *hcpu);
641 * ring_buffer_alloc - allocate a new ring_buffer
642 * @size: the size in bytes per cpu that is needed.
643 * @flags: attributes to set for the ring buffer.
645 * Currently the only flag that is available is the RB_FL_OVERWRITE
646 * flag. This flag means that the buffer will overwrite old data
647 * when the buffer wraps. If this flag is not set, the buffer will
648 * drop data when the tail hits the head.
650 struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags,
651 struct lock_class_key *key)
653 struct ring_buffer *buffer;
657 /* keep it in its own cache line */
658 buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
663 if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL))
664 goto fail_free_buffer;
666 buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
667 buffer->flags = flags;
668 buffer->clock = trace_clock_local;
669 buffer->reader_lock_key = key;
671 /* need at least two pages */
672 if (buffer->pages < 2)
676 * In case of non-hotplug cpu, if the ring-buffer is allocated
677 * in early initcall, it will not be notified of secondary cpus.
678 * In that off case, we need to allocate for all possible cpus.
680 #ifdef CONFIG_HOTPLUG_CPU
682 cpumask_copy(buffer->cpumask, cpu_online_mask);
684 cpumask_copy(buffer->cpumask, cpu_possible_mask);
686 buffer->cpus = nr_cpu_ids;
688 bsize = sizeof(void *) * nr_cpu_ids;
689 buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
691 if (!buffer->buffers)
692 goto fail_free_cpumask;
694 for_each_buffer_cpu(buffer, cpu) {
695 buffer->buffers[cpu] =
696 rb_allocate_cpu_buffer(buffer, cpu);
697 if (!buffer->buffers[cpu])
698 goto fail_free_buffers;
701 #ifdef CONFIG_HOTPLUG_CPU
702 buffer->cpu_notify.notifier_call = rb_cpu_notify;
703 buffer->cpu_notify.priority = 0;
704 register_cpu_notifier(&buffer->cpu_notify);
708 mutex_init(&buffer->mutex);
713 for_each_buffer_cpu(buffer, cpu) {
714 if (buffer->buffers[cpu])
715 rb_free_cpu_buffer(buffer->buffers[cpu]);
717 kfree(buffer->buffers);
720 free_cpumask_var(buffer->cpumask);
727 EXPORT_SYMBOL_GPL(__ring_buffer_alloc);
730 * ring_buffer_free - free a ring buffer.
731 * @buffer: the buffer to free.
734 ring_buffer_free(struct ring_buffer *buffer)
740 #ifdef CONFIG_HOTPLUG_CPU
741 unregister_cpu_notifier(&buffer->cpu_notify);
744 for_each_buffer_cpu(buffer, cpu)
745 rb_free_cpu_buffer(buffer->buffers[cpu]);
749 free_cpumask_var(buffer->cpumask);
753 EXPORT_SYMBOL_GPL(ring_buffer_free);
755 void ring_buffer_set_clock(struct ring_buffer *buffer,
758 buffer->clock = clock;
761 static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
764 rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages)
766 struct buffer_page *bpage;
770 atomic_inc(&cpu_buffer->record_disabled);
773 for (i = 0; i < nr_pages; i++) {
774 if (RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)))
776 p = cpu_buffer->pages->next;
777 bpage = list_entry(p, struct buffer_page, list);
778 list_del_init(&bpage->list);
779 free_buffer_page(bpage);
781 if (RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)))
784 rb_reset_cpu(cpu_buffer);
786 rb_check_pages(cpu_buffer);
788 atomic_dec(&cpu_buffer->record_disabled);
793 rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer,
794 struct list_head *pages, unsigned nr_pages)
796 struct buffer_page *bpage;
800 atomic_inc(&cpu_buffer->record_disabled);
803 for (i = 0; i < nr_pages; i++) {
804 if (RB_WARN_ON(cpu_buffer, list_empty(pages)))
807 bpage = list_entry(p, struct buffer_page, list);
808 list_del_init(&bpage->list);
809 list_add_tail(&bpage->list, cpu_buffer->pages);
811 rb_reset_cpu(cpu_buffer);
813 rb_check_pages(cpu_buffer);
815 atomic_dec(&cpu_buffer->record_disabled);
819 * ring_buffer_resize - resize the ring buffer
820 * @buffer: the buffer to resize.
821 * @size: the new size.
823 * The tracer is responsible for making sure that the buffer is
824 * not being used while changing the size.
825 * Note: We may be able to change the above requirement by using
826 * RCU synchronizations.
828 * Minimum size is 2 * BUF_PAGE_SIZE.
830 * Returns -1 on failure.
832 int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
834 struct ring_buffer_per_cpu *cpu_buffer;
835 unsigned nr_pages, rm_pages, new_pages;
836 struct buffer_page *bpage, *tmp;
837 unsigned long buffer_size;
843 * Always succeed at resizing a non-existent buffer:
848 size = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
849 size *= BUF_PAGE_SIZE;
850 buffer_size = buffer->pages * BUF_PAGE_SIZE;
852 /* we need a minimum of two pages */
853 if (size < BUF_PAGE_SIZE * 2)
854 size = BUF_PAGE_SIZE * 2;
856 if (size == buffer_size)
859 mutex_lock(&buffer->mutex);
862 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
864 if (size < buffer_size) {
866 /* easy case, just free pages */
867 if (RB_WARN_ON(buffer, nr_pages >= buffer->pages))
870 rm_pages = buffer->pages - nr_pages;
872 for_each_buffer_cpu(buffer, cpu) {
873 cpu_buffer = buffer->buffers[cpu];
874 rb_remove_pages(cpu_buffer, rm_pages);
880 * This is a bit more difficult. We only want to add pages
881 * when we can allocate enough for all CPUs. We do this
882 * by allocating all the pages and storing them on a local
883 * link list. If we succeed in our allocation, then we
884 * add these pages to the cpu_buffers. Otherwise we just free
885 * them all and return -ENOMEM;
887 if (RB_WARN_ON(buffer, nr_pages <= buffer->pages))
890 new_pages = nr_pages - buffer->pages;
892 for_each_buffer_cpu(buffer, cpu) {
893 for (i = 0; i < new_pages; i++) {
894 bpage = kzalloc_node(ALIGN(sizeof(*bpage),
896 GFP_KERNEL, cpu_to_node(cpu));
899 list_add(&bpage->list, &pages);
900 addr = __get_free_page(GFP_KERNEL);
903 bpage->page = (void *)addr;
904 rb_init_page(bpage->page);
908 for_each_buffer_cpu(buffer, cpu) {
909 cpu_buffer = buffer->buffers[cpu];
910 rb_insert_pages(cpu_buffer, &pages, new_pages);
913 if (RB_WARN_ON(buffer, !list_empty(&pages)))
917 buffer->pages = nr_pages;
919 mutex_unlock(&buffer->mutex);
924 list_for_each_entry_safe(bpage, tmp, &pages, list) {
925 list_del_init(&bpage->list);
926 free_buffer_page(bpage);
929 mutex_unlock(&buffer->mutex);
933 * Something went totally wrong, and we are too paranoid
934 * to even clean up the mess.
938 mutex_unlock(&buffer->mutex);
941 EXPORT_SYMBOL_GPL(ring_buffer_resize);
944 __rb_data_page_index(struct buffer_data_page *bpage, unsigned index)
946 return bpage->data + index;
949 static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
951 return bpage->page->data + index;
954 static inline struct ring_buffer_event *
955 rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
957 return __rb_page_index(cpu_buffer->reader_page,
958 cpu_buffer->reader_page->read);
961 static inline struct ring_buffer_event *
962 rb_head_event(struct ring_buffer_per_cpu *cpu_buffer)
964 return __rb_page_index(cpu_buffer->head_page,
965 cpu_buffer->head_page->read);
968 static inline struct ring_buffer_event *
969 rb_iter_head_event(struct ring_buffer_iter *iter)
971 return __rb_page_index(iter->head_page, iter->head);
974 static inline unsigned rb_page_write(struct buffer_page *bpage)
976 return local_read(&bpage->write);
979 static inline unsigned rb_page_commit(struct buffer_page *bpage)
981 return local_read(&bpage->page->commit);
984 /* Size is determined by what has been commited */
985 static inline unsigned rb_page_size(struct buffer_page *bpage)
987 return rb_page_commit(bpage);
990 static inline unsigned
991 rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
993 return rb_page_commit(cpu_buffer->commit_page);
996 static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer)
998 return rb_page_commit(cpu_buffer->head_page);
1001 static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer,
1002 struct buffer_page **bpage)
1004 struct list_head *p = (*bpage)->list.next;
1006 *bpage = list_entry(p, struct buffer_page, list);
1009 static inline unsigned
1010 rb_event_index(struct ring_buffer_event *event)
1012 unsigned long addr = (unsigned long)event;
1014 return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE;
1018 rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
1019 struct ring_buffer_event *event)
1021 unsigned long addr = (unsigned long)event;
1022 unsigned long index;
1024 index = rb_event_index(event);
1027 return cpu_buffer->commit_page->page == (void *)addr &&
1028 rb_commit_index(cpu_buffer) == index;
1032 rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
1035 * We only race with interrupts and NMIs on this CPU.
1036 * If we own the commit event, then we can commit
1037 * all others that interrupted us, since the interruptions
1038 * are in stack format (they finish before they come
1039 * back to us). This allows us to do a simple loop to
1040 * assign the commit to the tail.
1043 while (cpu_buffer->commit_page != cpu_buffer->tail_page) {
1044 cpu_buffer->commit_page->page->commit =
1045 cpu_buffer->commit_page->write;
1046 rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
1047 cpu_buffer->write_stamp =
1048 cpu_buffer->commit_page->page->time_stamp;
1049 /* add barrier to keep gcc from optimizing too much */
1052 while (rb_commit_index(cpu_buffer) !=
1053 rb_page_write(cpu_buffer->commit_page)) {
1054 cpu_buffer->commit_page->page->commit =
1055 cpu_buffer->commit_page->write;
1059 /* again, keep gcc from optimizing */
1063 * If an interrupt came in just after the first while loop
1064 * and pushed the tail page forward, we will be left with
1065 * a dangling commit that will never go forward.
1067 if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page))
1071 static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
1073 cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp;
1074 cpu_buffer->reader_page->read = 0;
1077 static void rb_inc_iter(struct ring_buffer_iter *iter)
1079 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1082 * The iterator could be on the reader page (it starts there).
1083 * But the head could have moved, since the reader was
1084 * found. Check for this case and assign the iterator
1085 * to the head page instead of next.
1087 if (iter->head_page == cpu_buffer->reader_page)
1088 iter->head_page = cpu_buffer->head_page;
1090 rb_inc_page(cpu_buffer, &iter->head_page);
1092 iter->read_stamp = iter->head_page->page->time_stamp;
1097 * ring_buffer_update_event - update event type and data
1098 * @event: the even to update
1099 * @type: the type of event
1100 * @length: the size of the event field in the ring buffer
1102 * Update the type and data fields of the event. The length
1103 * is the actual size that is written to the ring buffer,
1104 * and with this, we can determine what to place into the
1108 rb_update_event(struct ring_buffer_event *event,
1109 unsigned type, unsigned length)
1111 event->type_len = type;
1115 case RINGBUF_TYPE_PADDING:
1116 case RINGBUF_TYPE_TIME_EXTEND:
1117 case RINGBUF_TYPE_TIME_STAMP:
1121 length -= RB_EVNT_HDR_SIZE;
1122 if (length > RB_MAX_SMALL_DATA)
1123 event->array[0] = length;
1125 event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT);
1132 static unsigned rb_calculate_event_length(unsigned length)
1134 struct ring_buffer_event event; /* Used only for sizeof array */
1136 /* zero length can cause confusions */
1140 if (length > RB_MAX_SMALL_DATA)
1141 length += sizeof(event.array[0]);
1143 length += RB_EVNT_HDR_SIZE;
1144 length = ALIGN(length, RB_ALIGNMENT);
1150 rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer,
1151 struct buffer_page *tail_page,
1152 unsigned long tail, unsigned long length)
1154 struct ring_buffer_event *event;
1157 * Only the event that crossed the page boundary
1158 * must fill the old tail_page with padding.
1160 if (tail >= BUF_PAGE_SIZE) {
1161 local_sub(length, &tail_page->write);
1165 event = __rb_page_index(tail_page, tail);
1166 kmemcheck_annotate_bitfield(event, bitfield);
1169 * If this event is bigger than the minimum size, then
1170 * we need to be careful that we don't subtract the
1171 * write counter enough to allow another writer to slip
1173 * We put in a discarded commit instead, to make sure
1174 * that this space is not used again.
1176 * If we are less than the minimum size, we don't need to
1179 if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) {
1180 /* No room for any events */
1182 /* Mark the rest of the page with padding */
1183 rb_event_set_padding(event);
1185 /* Set the write back to the previous setting */
1186 local_sub(length, &tail_page->write);
1190 /* Put in a discarded event */
1191 event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE;
1192 event->type_len = RINGBUF_TYPE_PADDING;
1193 /* time delta must be non zero */
1194 event->time_delta = 1;
1195 /* Account for this as an entry */
1196 local_inc(&tail_page->entries);
1197 local_inc(&cpu_buffer->entries);
1199 /* Set write to end of buffer */
1200 length = (tail + length) - BUF_PAGE_SIZE;
1201 local_sub(length, &tail_page->write);
1204 static struct ring_buffer_event *
1205 rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer,
1206 unsigned long length, unsigned long tail,
1207 struct buffer_page *commit_page,
1208 struct buffer_page *tail_page, u64 *ts)
1210 struct buffer_page *next_page, *head_page, *reader_page;
1211 struct ring_buffer *buffer = cpu_buffer->buffer;
1212 bool lock_taken = false;
1213 unsigned long flags;
1215 next_page = tail_page;
1217 local_irq_save(flags);
1219 * Since the write to the buffer is still not
1220 * fully lockless, we must be careful with NMIs.
1221 * The locks in the writers are taken when a write
1222 * crosses to a new page. The locks protect against
1223 * races with the readers (this will soon be fixed
1224 * with a lockless solution).
1226 * Because we can not protect against NMIs, and we
1227 * want to keep traces reentrant, we need to manage
1228 * what happens when we are in an NMI.
1230 * NMIs can happen after we take the lock.
1231 * If we are in an NMI, only take the lock
1232 * if it is not already taken. Otherwise
1235 if (unlikely(in_nmi())) {
1236 if (!__raw_spin_trylock(&cpu_buffer->lock)) {
1237 cpu_buffer->nmi_dropped++;
1241 __raw_spin_lock(&cpu_buffer->lock);
1245 rb_inc_page(cpu_buffer, &next_page);
1247 head_page = cpu_buffer->head_page;
1248 reader_page = cpu_buffer->reader_page;
1250 /* we grabbed the lock before incrementing */
1251 if (RB_WARN_ON(cpu_buffer, next_page == reader_page))
1255 * If for some reason, we had an interrupt storm that made
1256 * it all the way around the buffer, bail, and warn
1259 if (unlikely(next_page == commit_page)) {
1260 cpu_buffer->commit_overrun++;
1264 if (next_page == head_page) {
1265 if (!(buffer->flags & RB_FL_OVERWRITE))
1268 /* tail_page has not moved yet? */
1269 if (tail_page == cpu_buffer->tail_page) {
1270 /* count overflows */
1271 cpu_buffer->overrun +=
1272 local_read(&head_page->entries);
1274 rb_inc_page(cpu_buffer, &head_page);
1275 cpu_buffer->head_page = head_page;
1276 cpu_buffer->head_page->read = 0;
1281 * If the tail page is still the same as what we think
1282 * it is, then it is up to us to update the tail
1285 if (tail_page == cpu_buffer->tail_page) {
1286 local_set(&next_page->write, 0);
1287 local_set(&next_page->entries, 0);
1288 local_set(&next_page->page->commit, 0);
1289 cpu_buffer->tail_page = next_page;
1291 /* reread the time stamp */
1292 *ts = rb_time_stamp(buffer, cpu_buffer->cpu);
1293 cpu_buffer->tail_page->page->time_stamp = *ts;
1296 rb_reset_tail(cpu_buffer, tail_page, tail, length);
1298 __raw_spin_unlock(&cpu_buffer->lock);
1299 local_irq_restore(flags);
1301 /* fail and let the caller try again */
1302 return ERR_PTR(-EAGAIN);
1306 rb_reset_tail(cpu_buffer, tail_page, tail, length);
1308 if (likely(lock_taken))
1309 __raw_spin_unlock(&cpu_buffer->lock);
1310 local_irq_restore(flags);
1314 static struct ring_buffer_event *
1315 __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
1316 unsigned type, unsigned long length, u64 *ts)
1318 struct buffer_page *tail_page, *commit_page;
1319 struct ring_buffer_event *event;
1320 unsigned long tail, write;
1322 commit_page = cpu_buffer->commit_page;
1323 /* we just need to protect against interrupts */
1325 tail_page = cpu_buffer->tail_page;
1326 write = local_add_return(length, &tail_page->write);
1327 tail = write - length;
1329 /* See if we shot pass the end of this buffer page */
1330 if (write > BUF_PAGE_SIZE)
1331 return rb_move_tail(cpu_buffer, length, tail,
1332 commit_page, tail_page, ts);
1334 /* We reserved something on the buffer */
1336 event = __rb_page_index(tail_page, tail);
1337 kmemcheck_annotate_bitfield(event, bitfield);
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 #ifdef CONFIG_TRACING
1576 #define TRACE_RECURSIVE_DEPTH 16
1578 static int trace_recursive_lock(void)
1580 current->trace_recursion++;
1582 if (likely(current->trace_recursion < TRACE_RECURSIVE_DEPTH))
1585 /* Disable all tracing before we do anything else */
1586 tracing_off_permanent();
1588 printk_once(KERN_WARNING "Tracing recursion: depth[%ld]:"
1589 "HC[%lu]:SC[%lu]:NMI[%lu]\n",
1590 current->trace_recursion,
1591 hardirq_count() >> HARDIRQ_SHIFT,
1592 softirq_count() >> SOFTIRQ_SHIFT,
1599 static void trace_recursive_unlock(void)
1601 WARN_ON_ONCE(!current->trace_recursion);
1603 current->trace_recursion--;
1608 #define trace_recursive_lock() (0)
1609 #define trace_recursive_unlock() do { } while (0)
1613 static DEFINE_PER_CPU(int, rb_need_resched);
1616 * ring_buffer_lock_reserve - reserve a part of the buffer
1617 * @buffer: the ring buffer to reserve from
1618 * @length: the length of the data to reserve (excluding event header)
1620 * Returns a reseverd event on the ring buffer to copy directly to.
1621 * The user of this interface will need to get the body to write into
1622 * and can use the ring_buffer_event_data() interface.
1624 * The length is the length of the data needed, not the event length
1625 * which also includes the event header.
1627 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
1628 * If NULL is returned, then nothing has been allocated or locked.
1630 struct ring_buffer_event *
1631 ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length)
1633 struct ring_buffer_per_cpu *cpu_buffer;
1634 struct ring_buffer_event *event;
1637 if (ring_buffer_flags != RB_BUFFERS_ON)
1640 if (atomic_read(&buffer->record_disabled))
1643 /* If we are tracing schedule, we don't want to recurse */
1644 resched = ftrace_preempt_disable();
1646 if (trace_recursive_lock())
1649 cpu = raw_smp_processor_id();
1651 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1654 cpu_buffer = buffer->buffers[cpu];
1656 if (atomic_read(&cpu_buffer->record_disabled))
1659 if (length > BUF_MAX_DATA_SIZE)
1662 event = rb_reserve_next_event(cpu_buffer, length);
1667 * Need to store resched state on this cpu.
1668 * Only the first needs to.
1671 if (preempt_count() == 1)
1672 per_cpu(rb_need_resched, cpu) = resched;
1677 trace_recursive_unlock();
1680 ftrace_preempt_enable(resched);
1683 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve);
1685 static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
1686 struct ring_buffer_event *event)
1688 local_inc(&cpu_buffer->entries);
1691 * The event first in the commit queue updates the
1694 if (rb_event_is_commit(cpu_buffer, event))
1695 cpu_buffer->write_stamp += event->time_delta;
1697 rb_end_commit(cpu_buffer);
1701 * ring_buffer_unlock_commit - commit a reserved
1702 * @buffer: The buffer to commit to
1703 * @event: The event pointer to commit.
1705 * This commits the data to the ring buffer, and releases any locks held.
1707 * Must be paired with ring_buffer_lock_reserve.
1709 int ring_buffer_unlock_commit(struct ring_buffer *buffer,
1710 struct ring_buffer_event *event)
1712 struct ring_buffer_per_cpu *cpu_buffer;
1713 int cpu = raw_smp_processor_id();
1715 cpu_buffer = buffer->buffers[cpu];
1717 rb_commit(cpu_buffer, event);
1719 trace_recursive_unlock();
1722 * Only the last preempt count needs to restore preemption.
1724 if (preempt_count() == 1)
1725 ftrace_preempt_enable(per_cpu(rb_need_resched, cpu));
1727 preempt_enable_no_resched_notrace();
1731 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit);
1733 static inline void rb_event_discard(struct ring_buffer_event *event)
1735 /* array[0] holds the actual length for the discarded event */
1736 event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE;
1737 event->type_len = RINGBUF_TYPE_PADDING;
1738 /* time delta must be non zero */
1739 if (!event->time_delta)
1740 event->time_delta = 1;
1744 * ring_buffer_event_discard - discard any event in the ring buffer
1745 * @event: the event to discard
1747 * Sometimes a event that is in the ring buffer needs to be ignored.
1748 * This function lets the user discard an event in the ring buffer
1749 * and then that event will not be read later.
1751 * Note, it is up to the user to be careful with this, and protect
1752 * against races. If the user discards an event that has been consumed
1753 * it is possible that it could corrupt the ring buffer.
1755 void ring_buffer_event_discard(struct ring_buffer_event *event)
1757 rb_event_discard(event);
1759 EXPORT_SYMBOL_GPL(ring_buffer_event_discard);
1762 * ring_buffer_commit_discard - discard an event that has not been committed
1763 * @buffer: the ring buffer
1764 * @event: non committed event to discard
1766 * This is similar to ring_buffer_event_discard but must only be
1767 * performed on an event that has not been committed yet. The difference
1768 * is that this will also try to free the event from the ring buffer
1769 * if another event has not been added behind it.
1771 * If another event has been added behind it, it will set the event
1772 * up as discarded, and perform the commit.
1774 * If this function is called, do not call ring_buffer_unlock_commit on
1777 void ring_buffer_discard_commit(struct ring_buffer *buffer,
1778 struct ring_buffer_event *event)
1780 struct ring_buffer_per_cpu *cpu_buffer;
1783 /* The event is discarded regardless */
1784 rb_event_discard(event);
1786 cpu = smp_processor_id();
1787 cpu_buffer = buffer->buffers[cpu];
1790 * This must only be called if the event has not been
1791 * committed yet. Thus we can assume that preemption
1792 * is still disabled.
1794 RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing));
1796 if (!rb_try_to_discard(cpu_buffer, event))
1800 * The commit is still visible by the reader, so we
1801 * must increment entries.
1803 local_inc(&cpu_buffer->entries);
1805 rb_end_commit(cpu_buffer);
1807 trace_recursive_unlock();
1810 * Only the last preempt count needs to restore preemption.
1812 if (preempt_count() == 1)
1813 ftrace_preempt_enable(per_cpu(rb_need_resched, cpu));
1815 preempt_enable_no_resched_notrace();
1818 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit);
1821 * ring_buffer_write - write data to the buffer without reserving
1822 * @buffer: The ring buffer to write to.
1823 * @length: The length of the data being written (excluding the event header)
1824 * @data: The data to write to the buffer.
1826 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
1827 * one function. If you already have the data to write to the buffer, it
1828 * may be easier to simply call this function.
1830 * Note, like ring_buffer_lock_reserve, the length is the length of the data
1831 * and not the length of the event which would hold the header.
1833 int ring_buffer_write(struct ring_buffer *buffer,
1834 unsigned long length,
1837 struct ring_buffer_per_cpu *cpu_buffer;
1838 struct ring_buffer_event *event;
1843 if (ring_buffer_flags != RB_BUFFERS_ON)
1846 if (atomic_read(&buffer->record_disabled))
1849 resched = ftrace_preempt_disable();
1851 cpu = raw_smp_processor_id();
1853 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1856 cpu_buffer = buffer->buffers[cpu];
1858 if (atomic_read(&cpu_buffer->record_disabled))
1861 if (length > BUF_MAX_DATA_SIZE)
1864 event = rb_reserve_next_event(cpu_buffer, length);
1868 body = rb_event_data(event);
1870 memcpy(body, data, length);
1872 rb_commit(cpu_buffer, event);
1876 ftrace_preempt_enable(resched);
1880 EXPORT_SYMBOL_GPL(ring_buffer_write);
1882 static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
1884 struct buffer_page *reader = cpu_buffer->reader_page;
1885 struct buffer_page *head = cpu_buffer->head_page;
1886 struct buffer_page *commit = cpu_buffer->commit_page;
1888 return reader->read == rb_page_commit(reader) &&
1889 (commit == reader ||
1891 head->read == rb_page_commit(commit)));
1895 * ring_buffer_record_disable - stop all writes into the buffer
1896 * @buffer: The ring buffer to stop writes to.
1898 * This prevents all writes to the buffer. Any attempt to write
1899 * to the buffer after this will fail and return NULL.
1901 * The caller should call synchronize_sched() after this.
1903 void ring_buffer_record_disable(struct ring_buffer *buffer)
1905 atomic_inc(&buffer->record_disabled);
1907 EXPORT_SYMBOL_GPL(ring_buffer_record_disable);
1910 * ring_buffer_record_enable - enable writes to the buffer
1911 * @buffer: The ring buffer to enable writes
1913 * Note, multiple disables will need the same number of enables
1914 * to truely enable the writing (much like preempt_disable).
1916 void ring_buffer_record_enable(struct ring_buffer *buffer)
1918 atomic_dec(&buffer->record_disabled);
1920 EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
1923 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
1924 * @buffer: The ring buffer to stop writes to.
1925 * @cpu: The CPU buffer to stop
1927 * This prevents all writes to the buffer. Any attempt to write
1928 * to the buffer after this will fail and return NULL.
1930 * The caller should call synchronize_sched() after this.
1932 void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu)
1934 struct ring_buffer_per_cpu *cpu_buffer;
1936 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1939 cpu_buffer = buffer->buffers[cpu];
1940 atomic_inc(&cpu_buffer->record_disabled);
1942 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu);
1945 * ring_buffer_record_enable_cpu - enable writes to the buffer
1946 * @buffer: The ring buffer to enable writes
1947 * @cpu: The CPU to enable.
1949 * Note, multiple disables will need the same number of enables
1950 * to truely enable the writing (much like preempt_disable).
1952 void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu)
1954 struct ring_buffer_per_cpu *cpu_buffer;
1956 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1959 cpu_buffer = buffer->buffers[cpu];
1960 atomic_dec(&cpu_buffer->record_disabled);
1962 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
1965 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
1966 * @buffer: The ring buffer
1967 * @cpu: The per CPU buffer to get the entries from.
1969 unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
1971 struct ring_buffer_per_cpu *cpu_buffer;
1974 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1977 cpu_buffer = buffer->buffers[cpu];
1978 ret = (local_read(&cpu_buffer->entries) - cpu_buffer->overrun)
1983 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
1986 * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
1987 * @buffer: The ring buffer
1988 * @cpu: The per CPU buffer to get the number of overruns from
1990 unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu)
1992 struct ring_buffer_per_cpu *cpu_buffer;
1995 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1998 cpu_buffer = buffer->buffers[cpu];
1999 ret = cpu_buffer->overrun;
2003 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
2006 * ring_buffer_nmi_dropped_cpu - get the number of nmis that were dropped
2007 * @buffer: The ring buffer
2008 * @cpu: The per CPU buffer to get the number of overruns from
2010 unsigned long ring_buffer_nmi_dropped_cpu(struct ring_buffer *buffer, int cpu)
2012 struct ring_buffer_per_cpu *cpu_buffer;
2015 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2018 cpu_buffer = buffer->buffers[cpu];
2019 ret = cpu_buffer->nmi_dropped;
2023 EXPORT_SYMBOL_GPL(ring_buffer_nmi_dropped_cpu);
2026 * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits
2027 * @buffer: The ring buffer
2028 * @cpu: The per CPU buffer to get the number of overruns from
2031 ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu)
2033 struct ring_buffer_per_cpu *cpu_buffer;
2036 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2039 cpu_buffer = buffer->buffers[cpu];
2040 ret = cpu_buffer->commit_overrun;
2044 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu);
2047 * ring_buffer_entries - get the number of entries in a buffer
2048 * @buffer: The ring buffer
2050 * Returns the total number of entries in the ring buffer
2053 unsigned long ring_buffer_entries(struct ring_buffer *buffer)
2055 struct ring_buffer_per_cpu *cpu_buffer;
2056 unsigned long entries = 0;
2059 /* if you care about this being correct, lock the buffer */
2060 for_each_buffer_cpu(buffer, cpu) {
2061 cpu_buffer = buffer->buffers[cpu];
2062 entries += (local_read(&cpu_buffer->entries) -
2063 cpu_buffer->overrun) - cpu_buffer->read;
2068 EXPORT_SYMBOL_GPL(ring_buffer_entries);
2071 * ring_buffer_overrun_cpu - get the number of overruns in buffer
2072 * @buffer: The ring buffer
2074 * Returns the total number of overruns in the ring buffer
2077 unsigned long ring_buffer_overruns(struct ring_buffer *buffer)
2079 struct ring_buffer_per_cpu *cpu_buffer;
2080 unsigned long overruns = 0;
2083 /* if you care about this being correct, lock the buffer */
2084 for_each_buffer_cpu(buffer, cpu) {
2085 cpu_buffer = buffer->buffers[cpu];
2086 overruns += cpu_buffer->overrun;
2091 EXPORT_SYMBOL_GPL(ring_buffer_overruns);
2093 static void rb_iter_reset(struct ring_buffer_iter *iter)
2095 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2097 /* Iterator usage is expected to have record disabled */
2098 if (list_empty(&cpu_buffer->reader_page->list)) {
2099 iter->head_page = cpu_buffer->head_page;
2100 iter->head = cpu_buffer->head_page->read;
2102 iter->head_page = cpu_buffer->reader_page;
2103 iter->head = cpu_buffer->reader_page->read;
2106 iter->read_stamp = cpu_buffer->read_stamp;
2108 iter->read_stamp = iter->head_page->page->time_stamp;
2112 * ring_buffer_iter_reset - reset an iterator
2113 * @iter: The iterator to reset
2115 * Resets the iterator, so that it will start from the beginning
2118 void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
2120 struct ring_buffer_per_cpu *cpu_buffer;
2121 unsigned long flags;
2126 cpu_buffer = iter->cpu_buffer;
2128 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2129 rb_iter_reset(iter);
2130 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2132 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset);
2135 * ring_buffer_iter_empty - check if an iterator has no more to read
2136 * @iter: The iterator to check
2138 int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
2140 struct ring_buffer_per_cpu *cpu_buffer;
2142 cpu_buffer = iter->cpu_buffer;
2144 return iter->head_page == cpu_buffer->commit_page &&
2145 iter->head == rb_commit_index(cpu_buffer);
2147 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty);
2150 rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
2151 struct ring_buffer_event *event)
2155 switch (event->type_len) {
2156 case RINGBUF_TYPE_PADDING:
2159 case RINGBUF_TYPE_TIME_EXTEND:
2160 delta = event->array[0];
2162 delta += event->time_delta;
2163 cpu_buffer->read_stamp += delta;
2166 case RINGBUF_TYPE_TIME_STAMP:
2167 /* FIXME: not implemented */
2170 case RINGBUF_TYPE_DATA:
2171 cpu_buffer->read_stamp += event->time_delta;
2181 rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
2182 struct ring_buffer_event *event)
2186 switch (event->type_len) {
2187 case RINGBUF_TYPE_PADDING:
2190 case RINGBUF_TYPE_TIME_EXTEND:
2191 delta = event->array[0];
2193 delta += event->time_delta;
2194 iter->read_stamp += delta;
2197 case RINGBUF_TYPE_TIME_STAMP:
2198 /* FIXME: not implemented */
2201 case RINGBUF_TYPE_DATA:
2202 iter->read_stamp += event->time_delta;
2211 static struct buffer_page *
2212 rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
2214 struct buffer_page *reader = NULL;
2215 unsigned long flags;
2218 local_irq_save(flags);
2219 __raw_spin_lock(&cpu_buffer->lock);
2223 * This should normally only loop twice. But because the
2224 * start of the reader inserts an empty page, it causes
2225 * a case where we will loop three times. There should be no
2226 * reason to loop four times (that I know of).
2228 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
2233 reader = cpu_buffer->reader_page;
2235 /* If there's more to read, return this page */
2236 if (cpu_buffer->reader_page->read < rb_page_size(reader))
2239 /* Never should we have an index greater than the size */
2240 if (RB_WARN_ON(cpu_buffer,
2241 cpu_buffer->reader_page->read > rb_page_size(reader)))
2244 /* check if we caught up to the tail */
2246 if (cpu_buffer->commit_page == cpu_buffer->reader_page)
2250 * Splice the empty reader page into the list around the head.
2251 * Reset the reader page to size zero.
2254 reader = cpu_buffer->head_page;
2255 cpu_buffer->reader_page->list.next = reader->list.next;
2256 cpu_buffer->reader_page->list.prev = reader->list.prev;
2259 * cpu_buffer->pages just needs to point to the buffer, it
2260 * has no specific buffer page to point to. Lets move it out
2261 * of our way so we don't accidently swap it.
2263 cpu_buffer->pages = reader->list.prev;
2265 local_set(&cpu_buffer->reader_page->write, 0);
2266 local_set(&cpu_buffer->reader_page->entries, 0);
2267 local_set(&cpu_buffer->reader_page->page->commit, 0);
2269 /* Make the reader page now replace the head */
2270 reader->list.prev->next = &cpu_buffer->reader_page->list;
2271 reader->list.next->prev = &cpu_buffer->reader_page->list;
2274 * If the tail is on the reader, then we must set the head
2275 * to the inserted page, otherwise we set it one before.
2277 cpu_buffer->head_page = cpu_buffer->reader_page;
2279 if (cpu_buffer->commit_page != reader)
2280 rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
2282 /* Finally update the reader page to the new head */
2283 cpu_buffer->reader_page = reader;
2284 rb_reset_reader_page(cpu_buffer);
2289 __raw_spin_unlock(&cpu_buffer->lock);
2290 local_irq_restore(flags);
2295 static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
2297 struct ring_buffer_event *event;
2298 struct buffer_page *reader;
2301 reader = rb_get_reader_page(cpu_buffer);
2303 /* This function should not be called when buffer is empty */
2304 if (RB_WARN_ON(cpu_buffer, !reader))
2307 event = rb_reader_event(cpu_buffer);
2309 if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX
2310 || rb_discarded_event(event))
2313 rb_update_read_stamp(cpu_buffer, event);
2315 length = rb_event_length(event);
2316 cpu_buffer->reader_page->read += length;
2319 static void rb_advance_iter(struct ring_buffer_iter *iter)
2321 struct ring_buffer *buffer;
2322 struct ring_buffer_per_cpu *cpu_buffer;
2323 struct ring_buffer_event *event;
2326 cpu_buffer = iter->cpu_buffer;
2327 buffer = cpu_buffer->buffer;
2330 * Check if we are at the end of the buffer.
2332 if (iter->head >= rb_page_size(iter->head_page)) {
2333 /* discarded commits can make the page empty */
2334 if (iter->head_page == cpu_buffer->commit_page)
2340 event = rb_iter_head_event(iter);
2342 length = rb_event_length(event);
2345 * This should not be called to advance the header if we are
2346 * at the tail of the buffer.
2348 if (RB_WARN_ON(cpu_buffer,
2349 (iter->head_page == cpu_buffer->commit_page) &&
2350 (iter->head + length > rb_commit_index(cpu_buffer))))
2353 rb_update_iter_read_stamp(iter, event);
2355 iter->head += length;
2357 /* check for end of page padding */
2358 if ((iter->head >= rb_page_size(iter->head_page)) &&
2359 (iter->head_page != cpu_buffer->commit_page))
2360 rb_advance_iter(iter);
2363 static struct ring_buffer_event *
2364 rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
2366 struct ring_buffer_per_cpu *cpu_buffer;
2367 struct ring_buffer_event *event;
2368 struct buffer_page *reader;
2371 cpu_buffer = buffer->buffers[cpu];
2375 * We repeat when a timestamp is encountered. It is possible
2376 * to get multiple timestamps from an interrupt entering just
2377 * as one timestamp is about to be written, or from discarded
2378 * commits. The most that we can have is the number on a single page.
2380 if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE))
2383 reader = rb_get_reader_page(cpu_buffer);
2387 event = rb_reader_event(cpu_buffer);
2389 switch (event->type_len) {
2390 case RINGBUF_TYPE_PADDING:
2391 if (rb_null_event(event))
2392 RB_WARN_ON(cpu_buffer, 1);
2394 * Because the writer could be discarding every
2395 * event it creates (which would probably be bad)
2396 * if we were to go back to "again" then we may never
2397 * catch up, and will trigger the warn on, or lock
2398 * the box. Return the padding, and we will release
2399 * the current locks, and try again.
2401 rb_advance_reader(cpu_buffer);
2404 case RINGBUF_TYPE_TIME_EXTEND:
2405 /* Internal data, OK to advance */
2406 rb_advance_reader(cpu_buffer);
2409 case RINGBUF_TYPE_TIME_STAMP:
2410 /* FIXME: not implemented */
2411 rb_advance_reader(cpu_buffer);
2414 case RINGBUF_TYPE_DATA:
2416 *ts = cpu_buffer->read_stamp + event->time_delta;
2417 ring_buffer_normalize_time_stamp(buffer,
2418 cpu_buffer->cpu, ts);
2428 EXPORT_SYMBOL_GPL(ring_buffer_peek);
2430 static struct ring_buffer_event *
2431 rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
2433 struct ring_buffer *buffer;
2434 struct ring_buffer_per_cpu *cpu_buffer;
2435 struct ring_buffer_event *event;
2438 if (ring_buffer_iter_empty(iter))
2441 cpu_buffer = iter->cpu_buffer;
2442 buffer = cpu_buffer->buffer;
2446 * We repeat when a timestamp is encountered.
2447 * We can get multiple timestamps by nested interrupts or also
2448 * if filtering is on (discarding commits). Since discarding
2449 * commits can be frequent we can get a lot of timestamps.
2450 * But we limit them by not adding timestamps if they begin
2451 * at the start of a page.
2453 if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE))
2456 if (rb_per_cpu_empty(cpu_buffer))
2459 event = rb_iter_head_event(iter);
2461 switch (event->type_len) {
2462 case RINGBUF_TYPE_PADDING:
2463 if (rb_null_event(event)) {
2467 rb_advance_iter(iter);
2470 case RINGBUF_TYPE_TIME_EXTEND:
2471 /* Internal data, OK to advance */
2472 rb_advance_iter(iter);
2475 case RINGBUF_TYPE_TIME_STAMP:
2476 /* FIXME: not implemented */
2477 rb_advance_iter(iter);
2480 case RINGBUF_TYPE_DATA:
2482 *ts = iter->read_stamp + event->time_delta;
2483 ring_buffer_normalize_time_stamp(buffer,
2484 cpu_buffer->cpu, ts);
2494 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek);
2496 static inline int rb_ok_to_lock(void)
2499 * If an NMI die dumps out the content of the ring buffer
2500 * do not grab locks. We also permanently disable the ring
2501 * buffer too. A one time deal is all you get from reading
2502 * the ring buffer from an NMI.
2504 if (likely(!in_nmi() && !oops_in_progress))
2507 tracing_off_permanent();
2512 * ring_buffer_peek - peek at the next event to be read
2513 * @buffer: The ring buffer to read
2514 * @cpu: The cpu to peak at
2515 * @ts: The timestamp counter of this event.
2517 * This will return the event that will be read next, but does
2518 * not consume the data.
2520 struct ring_buffer_event *
2521 ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
2523 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2524 struct ring_buffer_event *event;
2525 unsigned long flags;
2528 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2531 dolock = rb_ok_to_lock();
2533 local_irq_save(flags);
2535 spin_lock(&cpu_buffer->reader_lock);
2536 event = rb_buffer_peek(buffer, cpu, ts);
2538 spin_unlock(&cpu_buffer->reader_lock);
2539 local_irq_restore(flags);
2541 if (event && event->type_len == RINGBUF_TYPE_PADDING) {
2550 * ring_buffer_iter_peek - peek at the next event to be read
2551 * @iter: The ring buffer iterator
2552 * @ts: The timestamp counter of this event.
2554 * This will return the event that will be read next, but does
2555 * not increment the iterator.
2557 struct ring_buffer_event *
2558 ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
2560 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2561 struct ring_buffer_event *event;
2562 unsigned long flags;
2565 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2566 event = rb_iter_peek(iter, ts);
2567 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2569 if (event && event->type_len == RINGBUF_TYPE_PADDING) {
2578 * ring_buffer_consume - return an event and consume it
2579 * @buffer: The ring buffer to get the next event from
2581 * Returns the next event in the ring buffer, and that event is consumed.
2582 * Meaning, that sequential reads will keep returning a different event,
2583 * and eventually empty the ring buffer if the producer is slower.
2585 struct ring_buffer_event *
2586 ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts)
2588 struct ring_buffer_per_cpu *cpu_buffer;
2589 struct ring_buffer_event *event = NULL;
2590 unsigned long flags;
2593 dolock = rb_ok_to_lock();
2596 /* might be called in atomic */
2599 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2602 cpu_buffer = buffer->buffers[cpu];
2603 local_irq_save(flags);
2605 spin_lock(&cpu_buffer->reader_lock);
2607 event = rb_buffer_peek(buffer, cpu, ts);
2611 rb_advance_reader(cpu_buffer);
2615 spin_unlock(&cpu_buffer->reader_lock);
2616 local_irq_restore(flags);
2621 if (event && event->type_len == RINGBUF_TYPE_PADDING) {
2628 EXPORT_SYMBOL_GPL(ring_buffer_consume);
2631 * ring_buffer_read_start - start a non consuming read of the buffer
2632 * @buffer: The ring buffer to read from
2633 * @cpu: The cpu buffer to iterate over
2635 * This starts up an iteration through the buffer. It also disables
2636 * the recording to the buffer until the reading is finished.
2637 * This prevents the reading from being corrupted. This is not
2638 * a consuming read, so a producer is not expected.
2640 * Must be paired with ring_buffer_finish.
2642 struct ring_buffer_iter *
2643 ring_buffer_read_start(struct ring_buffer *buffer, int cpu)
2645 struct ring_buffer_per_cpu *cpu_buffer;
2646 struct ring_buffer_iter *iter;
2647 unsigned long flags;
2649 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2652 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
2656 cpu_buffer = buffer->buffers[cpu];
2658 iter->cpu_buffer = cpu_buffer;
2660 atomic_inc(&cpu_buffer->record_disabled);
2661 synchronize_sched();
2663 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2664 __raw_spin_lock(&cpu_buffer->lock);
2665 rb_iter_reset(iter);
2666 __raw_spin_unlock(&cpu_buffer->lock);
2667 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2671 EXPORT_SYMBOL_GPL(ring_buffer_read_start);
2674 * ring_buffer_finish - finish reading the iterator of the buffer
2675 * @iter: The iterator retrieved by ring_buffer_start
2677 * This re-enables the recording to the buffer, and frees the
2681 ring_buffer_read_finish(struct ring_buffer_iter *iter)
2683 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2685 atomic_dec(&cpu_buffer->record_disabled);
2688 EXPORT_SYMBOL_GPL(ring_buffer_read_finish);
2691 * ring_buffer_read - read the next item in the ring buffer by the iterator
2692 * @iter: The ring buffer iterator
2693 * @ts: The time stamp of the event read.
2695 * This reads the next event in the ring buffer and increments the iterator.
2697 struct ring_buffer_event *
2698 ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
2700 struct ring_buffer_event *event;
2701 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2702 unsigned long flags;
2705 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2706 event = rb_iter_peek(iter, ts);
2710 rb_advance_iter(iter);
2712 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2714 if (event && event->type_len == RINGBUF_TYPE_PADDING) {
2721 EXPORT_SYMBOL_GPL(ring_buffer_read);
2724 * ring_buffer_size - return the size of the ring buffer (in bytes)
2725 * @buffer: The ring buffer.
2727 unsigned long ring_buffer_size(struct ring_buffer *buffer)
2729 return BUF_PAGE_SIZE * buffer->pages;
2731 EXPORT_SYMBOL_GPL(ring_buffer_size);
2734 rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
2736 cpu_buffer->head_page
2737 = list_entry(cpu_buffer->pages, struct buffer_page, list);
2738 local_set(&cpu_buffer->head_page->write, 0);
2739 local_set(&cpu_buffer->head_page->entries, 0);
2740 local_set(&cpu_buffer->head_page->page->commit, 0);
2742 cpu_buffer->head_page->read = 0;
2744 cpu_buffer->tail_page = cpu_buffer->head_page;
2745 cpu_buffer->commit_page = cpu_buffer->head_page;
2747 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
2748 local_set(&cpu_buffer->reader_page->write, 0);
2749 local_set(&cpu_buffer->reader_page->entries, 0);
2750 local_set(&cpu_buffer->reader_page->page->commit, 0);
2751 cpu_buffer->reader_page->read = 0;
2753 cpu_buffer->nmi_dropped = 0;
2754 cpu_buffer->commit_overrun = 0;
2755 cpu_buffer->overrun = 0;
2756 cpu_buffer->read = 0;
2757 local_set(&cpu_buffer->entries, 0);
2758 local_set(&cpu_buffer->committing, 0);
2759 local_set(&cpu_buffer->commits, 0);
2761 cpu_buffer->write_stamp = 0;
2762 cpu_buffer->read_stamp = 0;
2766 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
2767 * @buffer: The ring buffer to reset a per cpu buffer of
2768 * @cpu: The CPU buffer to be reset
2770 void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
2772 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2773 unsigned long flags;
2775 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2778 atomic_inc(&cpu_buffer->record_disabled);
2780 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2782 __raw_spin_lock(&cpu_buffer->lock);
2784 rb_reset_cpu(cpu_buffer);
2786 __raw_spin_unlock(&cpu_buffer->lock);
2788 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2790 atomic_dec(&cpu_buffer->record_disabled);
2792 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
2795 * ring_buffer_reset - reset a ring buffer
2796 * @buffer: The ring buffer to reset all cpu buffers
2798 void ring_buffer_reset(struct ring_buffer *buffer)
2802 for_each_buffer_cpu(buffer, cpu)
2803 ring_buffer_reset_cpu(buffer, cpu);
2805 EXPORT_SYMBOL_GPL(ring_buffer_reset);
2808 * rind_buffer_empty - is the ring buffer empty?
2809 * @buffer: The ring buffer to test
2811 int ring_buffer_empty(struct ring_buffer *buffer)
2813 struct ring_buffer_per_cpu *cpu_buffer;
2814 unsigned long flags;
2819 dolock = rb_ok_to_lock();
2821 /* yes this is racy, but if you don't like the race, lock the buffer */
2822 for_each_buffer_cpu(buffer, cpu) {
2823 cpu_buffer = buffer->buffers[cpu];
2824 local_irq_save(flags);
2826 spin_lock(&cpu_buffer->reader_lock);
2827 ret = rb_per_cpu_empty(cpu_buffer);
2829 spin_unlock(&cpu_buffer->reader_lock);
2830 local_irq_restore(flags);
2838 EXPORT_SYMBOL_GPL(ring_buffer_empty);
2841 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
2842 * @buffer: The ring buffer
2843 * @cpu: The CPU buffer to test
2845 int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu)
2847 struct ring_buffer_per_cpu *cpu_buffer;
2848 unsigned long flags;
2852 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2855 dolock = rb_ok_to_lock();
2857 cpu_buffer = buffer->buffers[cpu];
2858 local_irq_save(flags);
2860 spin_lock(&cpu_buffer->reader_lock);
2861 ret = rb_per_cpu_empty(cpu_buffer);
2863 spin_unlock(&cpu_buffer->reader_lock);
2864 local_irq_restore(flags);
2868 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
2871 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
2872 * @buffer_a: One buffer to swap with
2873 * @buffer_b: The other buffer to swap with
2875 * This function is useful for tracers that want to take a "snapshot"
2876 * of a CPU buffer and has another back up buffer lying around.
2877 * it is expected that the tracer handles the cpu buffer not being
2878 * used at the moment.
2880 int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
2881 struct ring_buffer *buffer_b, int cpu)
2883 struct ring_buffer_per_cpu *cpu_buffer_a;
2884 struct ring_buffer_per_cpu *cpu_buffer_b;
2887 if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
2888 !cpumask_test_cpu(cpu, buffer_b->cpumask))
2891 /* At least make sure the two buffers are somewhat the same */
2892 if (buffer_a->pages != buffer_b->pages)
2897 if (ring_buffer_flags != RB_BUFFERS_ON)
2900 if (atomic_read(&buffer_a->record_disabled))
2903 if (atomic_read(&buffer_b->record_disabled))
2906 cpu_buffer_a = buffer_a->buffers[cpu];
2907 cpu_buffer_b = buffer_b->buffers[cpu];
2909 if (atomic_read(&cpu_buffer_a->record_disabled))
2912 if (atomic_read(&cpu_buffer_b->record_disabled))
2916 * We can't do a synchronize_sched here because this
2917 * function can be called in atomic context.
2918 * Normally this will be called from the same CPU as cpu.
2919 * If not it's up to the caller to protect this.
2921 atomic_inc(&cpu_buffer_a->record_disabled);
2922 atomic_inc(&cpu_buffer_b->record_disabled);
2924 buffer_a->buffers[cpu] = cpu_buffer_b;
2925 buffer_b->buffers[cpu] = cpu_buffer_a;
2927 cpu_buffer_b->buffer = buffer_a;
2928 cpu_buffer_a->buffer = buffer_b;
2930 atomic_dec(&cpu_buffer_a->record_disabled);
2931 atomic_dec(&cpu_buffer_b->record_disabled);
2937 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
2940 * ring_buffer_alloc_read_page - allocate a page to read from buffer
2941 * @buffer: the buffer to allocate for.
2943 * This function is used in conjunction with ring_buffer_read_page.
2944 * When reading a full page from the ring buffer, these functions
2945 * can be used to speed up the process. The calling function should
2946 * allocate a few pages first with this function. Then when it
2947 * needs to get pages from the ring buffer, it passes the result
2948 * of this function into ring_buffer_read_page, which will swap
2949 * the page that was allocated, with the read page of the buffer.
2952 * The page allocated, or NULL on error.
2954 void *ring_buffer_alloc_read_page(struct ring_buffer *buffer)
2956 struct buffer_data_page *bpage;
2959 addr = __get_free_page(GFP_KERNEL);
2963 bpage = (void *)addr;
2965 rb_init_page(bpage);
2969 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page);
2972 * ring_buffer_free_read_page - free an allocated read page
2973 * @buffer: the buffer the page was allocate for
2974 * @data: the page to free
2976 * Free a page allocated from ring_buffer_alloc_read_page.
2978 void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data)
2980 free_page((unsigned long)data);
2982 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page);
2985 * ring_buffer_read_page - extract a page from the ring buffer
2986 * @buffer: buffer to extract from
2987 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
2988 * @len: amount to extract
2989 * @cpu: the cpu of the buffer to extract
2990 * @full: should the extraction only happen when the page is full.
2992 * This function will pull out a page from the ring buffer and consume it.
2993 * @data_page must be the address of the variable that was returned
2994 * from ring_buffer_alloc_read_page. This is because the page might be used
2995 * to swap with a page in the ring buffer.
2998 * rpage = ring_buffer_alloc_read_page(buffer);
3001 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
3003 * process_page(rpage, ret);
3005 * When @full is set, the function will not return true unless
3006 * the writer is off the reader page.
3008 * Note: it is up to the calling functions to handle sleeps and wakeups.
3009 * The ring buffer can be used anywhere in the kernel and can not
3010 * blindly call wake_up. The layer that uses the ring buffer must be
3011 * responsible for that.
3014 * >=0 if data has been transferred, returns the offset of consumed data.
3015 * <0 if no data has been transferred.
3017 int ring_buffer_read_page(struct ring_buffer *buffer,
3018 void **data_page, size_t len, int cpu, int full)
3020 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
3021 struct ring_buffer_event *event;
3022 struct buffer_data_page *bpage;
3023 struct buffer_page *reader;
3024 unsigned long flags;
3025 unsigned int commit;
3030 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3034 * If len is not big enough to hold the page header, then
3035 * we can not copy anything.
3037 if (len <= BUF_PAGE_HDR_SIZE)
3040 len -= BUF_PAGE_HDR_SIZE;
3049 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
3051 reader = rb_get_reader_page(cpu_buffer);
3055 event = rb_reader_event(cpu_buffer);
3057 read = reader->read;
3058 commit = rb_page_commit(reader);
3061 * If this page has been partially read or
3062 * if len is not big enough to read the rest of the page or
3063 * a writer is still on the page, then
3064 * we must copy the data from the page to the buffer.
3065 * Otherwise, we can simply swap the page with the one passed in.
3067 if (read || (len < (commit - read)) ||
3068 cpu_buffer->reader_page == cpu_buffer->commit_page) {
3069 struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
3070 unsigned int rpos = read;
3071 unsigned int pos = 0;
3077 if (len > (commit - read))
3078 len = (commit - read);
3080 size = rb_event_length(event);
3085 /* save the current timestamp, since the user will need it */
3086 save_timestamp = cpu_buffer->read_stamp;
3088 /* Need to copy one event at a time */
3090 memcpy(bpage->data + pos, rpage->data + rpos, size);
3094 rb_advance_reader(cpu_buffer);
3095 rpos = reader->read;
3098 event = rb_reader_event(cpu_buffer);
3099 size = rb_event_length(event);
3100 } while (len > size);
3103 local_set(&bpage->commit, pos);
3104 bpage->time_stamp = save_timestamp;
3106 /* we copied everything to the beginning */
3109 /* update the entry counter */
3110 cpu_buffer->read += local_read(&reader->entries);
3112 /* swap the pages */
3113 rb_init_page(bpage);
3114 bpage = reader->page;
3115 reader->page = *data_page;
3116 local_set(&reader->write, 0);
3117 local_set(&reader->entries, 0);
3124 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
3129 EXPORT_SYMBOL_GPL(ring_buffer_read_page);
3131 #ifdef CONFIG_TRACING
3133 rb_simple_read(struct file *filp, char __user *ubuf,
3134 size_t cnt, loff_t *ppos)
3136 unsigned long *p = filp->private_data;
3140 if (test_bit(RB_BUFFERS_DISABLED_BIT, p))
3141 r = sprintf(buf, "permanently disabled\n");
3143 r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p));
3145 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3149 rb_simple_write(struct file *filp, const char __user *ubuf,
3150 size_t cnt, loff_t *ppos)
3152 unsigned long *p = filp->private_data;
3157 if (cnt >= sizeof(buf))
3160 if (copy_from_user(&buf, ubuf, cnt))
3165 ret = strict_strtoul(buf, 10, &val);
3170 set_bit(RB_BUFFERS_ON_BIT, p);
3172 clear_bit(RB_BUFFERS_ON_BIT, p);
3179 static const struct file_operations rb_simple_fops = {
3180 .open = tracing_open_generic,
3181 .read = rb_simple_read,
3182 .write = rb_simple_write,
3186 static __init int rb_init_debugfs(void)
3188 struct dentry *d_tracer;
3190 d_tracer = tracing_init_dentry();
3192 trace_create_file("tracing_on", 0644, d_tracer,
3193 &ring_buffer_flags, &rb_simple_fops);
3198 fs_initcall(rb_init_debugfs);
3201 #ifdef CONFIG_HOTPLUG_CPU
3202 static int rb_cpu_notify(struct notifier_block *self,
3203 unsigned long action, void *hcpu)
3205 struct ring_buffer *buffer =
3206 container_of(self, struct ring_buffer, cpu_notify);
3207 long cpu = (long)hcpu;
3210 case CPU_UP_PREPARE:
3211 case CPU_UP_PREPARE_FROZEN:
3212 if (cpumask_test_cpu(cpu, buffer->cpumask))
3215 buffer->buffers[cpu] =
3216 rb_allocate_cpu_buffer(buffer, cpu);
3217 if (!buffer->buffers[cpu]) {
3218 WARN(1, "failed to allocate ring buffer on CPU %ld\n",
3223 cpumask_set_cpu(cpu, buffer->cpumask);
3225 case CPU_DOWN_PREPARE:
3226 case CPU_DOWN_PREPARE_FROZEN:
3229 * If we were to free the buffer, then the user would
3230 * lose any trace that was in the buffer.