aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/trace/ring_buffer.c
diff options
context:
space:
mode:
Diffstat (limited to 'kernel/trace/ring_buffer.c')
-rw-r--r--kernel/trace/ring_buffer.c2072
1 files changed, 2072 insertions, 0 deletions
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
new file mode 100644
index 000000000000..3f3380638646
--- /dev/null
+++ b/kernel/trace/ring_buffer.c
@@ -0,0 +1,2072 @@
1/*
2 * Generic ring buffer
3 *
4 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
5 */
6#include <linux/ring_buffer.h>
7#include <linux/spinlock.h>
8#include <linux/debugfs.h>
9#include <linux/uaccess.h>
10#include <linux/module.h>
11#include <linux/percpu.h>
12#include <linux/mutex.h>
13#include <linux/sched.h> /* used for sched_clock() (for now) */
14#include <linux/init.h>
15#include <linux/hash.h>
16#include <linux/list.h>
17#include <linux/fs.h>
18
19/* Up this if you want to test the TIME_EXTENTS and normalization */
20#define DEBUG_SHIFT 0
21
22/* FIXME!!! */
23u64 ring_buffer_time_stamp(int cpu)
24{
25 /* shift to debug/test normalization and TIME_EXTENTS */
26 return sched_clock() << DEBUG_SHIFT;
27}
28
29void ring_buffer_normalize_time_stamp(int cpu, u64 *ts)
30{
31 /* Just stupid testing the normalize function and deltas */
32 *ts >>= DEBUG_SHIFT;
33}
34
35#define RB_EVNT_HDR_SIZE (sizeof(struct ring_buffer_event))
36#define RB_ALIGNMENT_SHIFT 2
37#define RB_ALIGNMENT (1 << RB_ALIGNMENT_SHIFT)
38#define RB_MAX_SMALL_DATA 28
39
40enum {
41 RB_LEN_TIME_EXTEND = 8,
42 RB_LEN_TIME_STAMP = 16,
43};
44
45/* inline for ring buffer fast paths */
46static inline unsigned
47rb_event_length(struct ring_buffer_event *event)
48{
49 unsigned length;
50
51 switch (event->type) {
52 case RINGBUF_TYPE_PADDING:
53 /* undefined */
54 return -1;
55
56 case RINGBUF_TYPE_TIME_EXTEND:
57 return RB_LEN_TIME_EXTEND;
58
59 case RINGBUF_TYPE_TIME_STAMP:
60 return RB_LEN_TIME_STAMP;
61
62 case RINGBUF_TYPE_DATA:
63 if (event->len)
64 length = event->len << RB_ALIGNMENT_SHIFT;
65 else
66 length = event->array[0];
67 return length + RB_EVNT_HDR_SIZE;
68 default:
69 BUG();
70 }
71 /* not hit */
72 return 0;
73}
74
75/**
76 * ring_buffer_event_length - return the length of the event
77 * @event: the event to get the length of
78 */
79unsigned ring_buffer_event_length(struct ring_buffer_event *event)
80{
81 return rb_event_length(event);
82}
83
84/* inline for ring buffer fast paths */
85static inline void *
86rb_event_data(struct ring_buffer_event *event)
87{
88 BUG_ON(event->type != RINGBUF_TYPE_DATA);
89 /* If length is in len field, then array[0] has the data */
90 if (event->len)
91 return (void *)&event->array[0];
92 /* Otherwise length is in array[0] and array[1] has the data */
93 return (void *)&event->array[1];
94}
95
96/**
97 * ring_buffer_event_data - return the data of the event
98 * @event: the event to get the data from
99 */
100void *ring_buffer_event_data(struct ring_buffer_event *event)
101{
102 return rb_event_data(event);
103}
104
105#define for_each_buffer_cpu(buffer, cpu) \
106 for_each_cpu_mask(cpu, buffer->cpumask)
107
108#define TS_SHIFT 27
109#define TS_MASK ((1ULL << TS_SHIFT) - 1)
110#define TS_DELTA_TEST (~TS_MASK)
111
112/*
113 * This hack stolen from mm/slob.c.
114 * We can store per page timing information in the page frame of the page.
115 * Thanks to Peter Zijlstra for suggesting this idea.
116 */
117struct buffer_page {
118 u64 time_stamp; /* page time stamp */
119 local_t write; /* index for next write */
120 local_t commit; /* write commited index */
121 unsigned read; /* index for next read */
122 struct list_head list; /* list of free pages */
123 void *page; /* Actual data page */
124};
125
126/*
127 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
128 * this issue out.
129 */
130static inline void free_buffer_page(struct buffer_page *bpage)
131{
132 if (bpage->page)
133 free_page((unsigned long)bpage->page);
134 kfree(bpage);
135}
136
137/*
138 * We need to fit the time_stamp delta into 27 bits.
139 */
140static inline int test_time_stamp(u64 delta)
141{
142 if (delta & TS_DELTA_TEST)
143 return 1;
144 return 0;
145}
146
147#define BUF_PAGE_SIZE PAGE_SIZE
148
149/*
150 * head_page == tail_page && head == tail then buffer is empty.
151 */
152struct ring_buffer_per_cpu {
153 int cpu;
154 struct ring_buffer *buffer;
155 spinlock_t lock;
156 struct lock_class_key lock_key;
157 struct list_head pages;
158 struct buffer_page *head_page; /* read from head */
159 struct buffer_page *tail_page; /* write to tail */
160 struct buffer_page *commit_page; /* commited pages */
161 struct buffer_page *reader_page;
162 unsigned long overrun;
163 unsigned long entries;
164 u64 write_stamp;
165 u64 read_stamp;
166 atomic_t record_disabled;
167};
168
169struct ring_buffer {
170 unsigned long size;
171 unsigned pages;
172 unsigned flags;
173 int cpus;
174 cpumask_t cpumask;
175 atomic_t record_disabled;
176
177 struct mutex mutex;
178
179 struct ring_buffer_per_cpu **buffers;
180};
181
182struct ring_buffer_iter {
183 struct ring_buffer_per_cpu *cpu_buffer;
184 unsigned long head;
185 struct buffer_page *head_page;
186 u64 read_stamp;
187};
188
189#define RB_WARN_ON(buffer, cond) \
190 do { \
191 if (unlikely(cond)) { \
192 atomic_inc(&buffer->record_disabled); \
193 WARN_ON(1); \
194 } \
195 } while (0)
196
197#define RB_WARN_ON_RET(buffer, cond) \
198 do { \
199 if (unlikely(cond)) { \
200 atomic_inc(&buffer->record_disabled); \
201 WARN_ON(1); \
202 return -1; \
203 } \
204 } while (0)
205
206#define RB_WARN_ON_ONCE(buffer, cond) \
207 do { \
208 static int once; \
209 if (unlikely(cond) && !once) { \
210 once++; \
211 atomic_inc(&buffer->record_disabled); \
212 WARN_ON(1); \
213 } \
214 } while (0)
215
216/**
217 * check_pages - integrity check of buffer pages
218 * @cpu_buffer: CPU buffer with pages to test
219 *
220 * As a safty measure we check to make sure the data pages have not
221 * been corrupted.
222 */
223static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
224{
225 struct list_head *head = &cpu_buffer->pages;
226 struct buffer_page *page, *tmp;
227
228 RB_WARN_ON_RET(cpu_buffer, head->next->prev != head);
229 RB_WARN_ON_RET(cpu_buffer, head->prev->next != head);
230
231 list_for_each_entry_safe(page, tmp, head, list) {
232 RB_WARN_ON_RET(cpu_buffer,
233 page->list.next->prev != &page->list);
234 RB_WARN_ON_RET(cpu_buffer,
235 page->list.prev->next != &page->list);
236 }
237
238 return 0;
239}
240
241static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
242 unsigned nr_pages)
243{
244 struct list_head *head = &cpu_buffer->pages;
245 struct buffer_page *page, *tmp;
246 unsigned long addr;
247 LIST_HEAD(pages);
248 unsigned i;
249
250 for (i = 0; i < nr_pages; i++) {
251 page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()),
252 GFP_KERNEL, cpu_to_node(cpu_buffer->cpu));
253 if (!page)
254 goto free_pages;
255 list_add(&page->list, &pages);
256
257 addr = __get_free_page(GFP_KERNEL);
258 if (!addr)
259 goto free_pages;
260 page->page = (void *)addr;
261 }
262
263 list_splice(&pages, head);
264
265 rb_check_pages(cpu_buffer);
266
267 return 0;
268
269 free_pages:
270 list_for_each_entry_safe(page, tmp, &pages, list) {
271 list_del_init(&page->list);
272 free_buffer_page(page);
273 }
274 return -ENOMEM;
275}
276
277static struct ring_buffer_per_cpu *
278rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu)
279{
280 struct ring_buffer_per_cpu *cpu_buffer;
281 struct buffer_page *page;
282 unsigned long addr;
283 int ret;
284
285 cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
286 GFP_KERNEL, cpu_to_node(cpu));
287 if (!cpu_buffer)
288 return NULL;
289
290 cpu_buffer->cpu = cpu;
291 cpu_buffer->buffer = buffer;
292 spin_lock_init(&cpu_buffer->lock);
293 INIT_LIST_HEAD(&cpu_buffer->pages);
294
295 page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()),
296 GFP_KERNEL, cpu_to_node(cpu));
297 if (!page)
298 goto fail_free_buffer;
299
300 cpu_buffer->reader_page = page;
301 addr = __get_free_page(GFP_KERNEL);
302 if (!addr)
303 goto fail_free_reader;
304 page->page = (void *)addr;
305
306 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
307
308 ret = rb_allocate_pages(cpu_buffer, buffer->pages);
309 if (ret < 0)
310 goto fail_free_reader;
311
312 cpu_buffer->head_page
313 = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
314 cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
315
316 return cpu_buffer;
317
318 fail_free_reader:
319 free_buffer_page(cpu_buffer->reader_page);
320
321 fail_free_buffer:
322 kfree(cpu_buffer);
323 return NULL;
324}
325
326static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
327{
328 struct list_head *head = &cpu_buffer->pages;
329 struct buffer_page *page, *tmp;
330
331 list_del_init(&cpu_buffer->reader_page->list);
332 free_buffer_page(cpu_buffer->reader_page);
333
334 list_for_each_entry_safe(page, tmp, head, list) {
335 list_del_init(&page->list);
336 free_buffer_page(page);
337 }
338 kfree(cpu_buffer);
339}
340
341/*
342 * Causes compile errors if the struct buffer_page gets bigger
343 * than the struct page.
344 */
345extern int ring_buffer_page_too_big(void);
346
347/**
348 * ring_buffer_alloc - allocate a new ring_buffer
349 * @size: the size in bytes that is needed.
350 * @flags: attributes to set for the ring buffer.
351 *
352 * Currently the only flag that is available is the RB_FL_OVERWRITE
353 * flag. This flag means that the buffer will overwrite old data
354 * when the buffer wraps. If this flag is not set, the buffer will
355 * drop data when the tail hits the head.
356 */
357struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags)
358{
359 struct ring_buffer *buffer;
360 int bsize;
361 int cpu;
362
363 /* Paranoid! Optimizes out when all is well */
364 if (sizeof(struct buffer_page) > sizeof(struct page))
365 ring_buffer_page_too_big();
366
367
368 /* keep it in its own cache line */
369 buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
370 GFP_KERNEL);
371 if (!buffer)
372 return NULL;
373
374 buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
375 buffer->flags = flags;
376
377 /* need at least two pages */
378 if (buffer->pages == 1)
379 buffer->pages++;
380
381 buffer->cpumask = cpu_possible_map;
382 buffer->cpus = nr_cpu_ids;
383
384 bsize = sizeof(void *) * nr_cpu_ids;
385 buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
386 GFP_KERNEL);
387 if (!buffer->buffers)
388 goto fail_free_buffer;
389
390 for_each_buffer_cpu(buffer, cpu) {
391 buffer->buffers[cpu] =
392 rb_allocate_cpu_buffer(buffer, cpu);
393 if (!buffer->buffers[cpu])
394 goto fail_free_buffers;
395 }
396
397 mutex_init(&buffer->mutex);
398
399 return buffer;
400
401 fail_free_buffers:
402 for_each_buffer_cpu(buffer, cpu) {
403 if (buffer->buffers[cpu])
404 rb_free_cpu_buffer(buffer->buffers[cpu]);
405 }
406 kfree(buffer->buffers);
407
408 fail_free_buffer:
409 kfree(buffer);
410 return NULL;
411}
412
413/**
414 * ring_buffer_free - free a ring buffer.
415 * @buffer: the buffer to free.
416 */
417void
418ring_buffer_free(struct ring_buffer *buffer)
419{
420 int cpu;
421
422 for_each_buffer_cpu(buffer, cpu)
423 rb_free_cpu_buffer(buffer->buffers[cpu]);
424
425 kfree(buffer);
426}
427
428static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
429
430static void
431rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages)
432{
433 struct buffer_page *page;
434 struct list_head *p;
435 unsigned i;
436
437 atomic_inc(&cpu_buffer->record_disabled);
438 synchronize_sched();
439
440 for (i = 0; i < nr_pages; i++) {
441 BUG_ON(list_empty(&cpu_buffer->pages));
442 p = cpu_buffer->pages.next;
443 page = list_entry(p, struct buffer_page, list);
444 list_del_init(&page->list);
445 free_buffer_page(page);
446 }
447 BUG_ON(list_empty(&cpu_buffer->pages));
448
449 rb_reset_cpu(cpu_buffer);
450
451 rb_check_pages(cpu_buffer);
452
453 atomic_dec(&cpu_buffer->record_disabled);
454
455}
456
457static void
458rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer,
459 struct list_head *pages, unsigned nr_pages)
460{
461 struct buffer_page *page;
462 struct list_head *p;
463 unsigned i;
464
465 atomic_inc(&cpu_buffer->record_disabled);
466 synchronize_sched();
467
468 for (i = 0; i < nr_pages; i++) {
469 BUG_ON(list_empty(pages));
470 p = pages->next;
471 page = list_entry(p, struct buffer_page, list);
472 list_del_init(&page->list);
473 list_add_tail(&page->list, &cpu_buffer->pages);
474 }
475 rb_reset_cpu(cpu_buffer);
476
477 rb_check_pages(cpu_buffer);
478
479 atomic_dec(&cpu_buffer->record_disabled);
480}
481
482/**
483 * ring_buffer_resize - resize the ring buffer
484 * @buffer: the buffer to resize.
485 * @size: the new size.
486 *
487 * The tracer is responsible for making sure that the buffer is
488 * not being used while changing the size.
489 * Note: We may be able to change the above requirement by using
490 * RCU synchronizations.
491 *
492 * Minimum size is 2 * BUF_PAGE_SIZE.
493 *
494 * Returns -1 on failure.
495 */
496int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
497{
498 struct ring_buffer_per_cpu *cpu_buffer;
499 unsigned nr_pages, rm_pages, new_pages;
500 struct buffer_page *page, *tmp;
501 unsigned long buffer_size;
502 unsigned long addr;
503 LIST_HEAD(pages);
504 int i, cpu;
505
506 size = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
507 size *= BUF_PAGE_SIZE;
508 buffer_size = buffer->pages * BUF_PAGE_SIZE;
509
510 /* we need a minimum of two pages */
511 if (size < BUF_PAGE_SIZE * 2)
512 size = BUF_PAGE_SIZE * 2;
513
514 if (size == buffer_size)
515 return size;
516
517 mutex_lock(&buffer->mutex);
518
519 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
520
521 if (size < buffer_size) {
522
523 /* easy case, just free pages */
524 BUG_ON(nr_pages >= buffer->pages);
525
526 rm_pages = buffer->pages - nr_pages;
527
528 for_each_buffer_cpu(buffer, cpu) {
529 cpu_buffer = buffer->buffers[cpu];
530 rb_remove_pages(cpu_buffer, rm_pages);
531 }
532 goto out;
533 }
534
535 /*
536 * This is a bit more difficult. We only want to add pages
537 * when we can allocate enough for all CPUs. We do this
538 * by allocating all the pages and storing them on a local
539 * link list. If we succeed in our allocation, then we
540 * add these pages to the cpu_buffers. Otherwise we just free
541 * them all and return -ENOMEM;
542 */
543 BUG_ON(nr_pages <= buffer->pages);
544 new_pages = nr_pages - buffer->pages;
545
546 for_each_buffer_cpu(buffer, cpu) {
547 for (i = 0; i < new_pages; i++) {
548 page = kzalloc_node(ALIGN(sizeof(*page),
549 cache_line_size()),
550 GFP_KERNEL, cpu_to_node(cpu));
551 if (!page)
552 goto free_pages;
553 list_add(&page->list, &pages);
554 addr = __get_free_page(GFP_KERNEL);
555 if (!addr)
556 goto free_pages;
557 page->page = (void *)addr;
558 }
559 }
560
561 for_each_buffer_cpu(buffer, cpu) {
562 cpu_buffer = buffer->buffers[cpu];
563 rb_insert_pages(cpu_buffer, &pages, new_pages);
564 }
565
566 BUG_ON(!list_empty(&pages));
567
568 out:
569 buffer->pages = nr_pages;
570 mutex_unlock(&buffer->mutex);
571
572 return size;
573
574 free_pages:
575 list_for_each_entry_safe(page, tmp, &pages, list) {
576 list_del_init(&page->list);
577 free_buffer_page(page);
578 }
579 return -ENOMEM;
580}
581
582static inline int rb_null_event(struct ring_buffer_event *event)
583{
584 return event->type == RINGBUF_TYPE_PADDING;
585}
586
587static inline void *__rb_page_index(struct buffer_page *page, unsigned index)
588{
589 return page->page + index;
590}
591
592static inline struct ring_buffer_event *
593rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
594{
595 return __rb_page_index(cpu_buffer->reader_page,
596 cpu_buffer->reader_page->read);
597}
598
599static inline struct ring_buffer_event *
600rb_head_event(struct ring_buffer_per_cpu *cpu_buffer)
601{
602 return __rb_page_index(cpu_buffer->head_page,
603 cpu_buffer->head_page->read);
604}
605
606static inline struct ring_buffer_event *
607rb_iter_head_event(struct ring_buffer_iter *iter)
608{
609 return __rb_page_index(iter->head_page, iter->head);
610}
611
612static inline unsigned rb_page_write(struct buffer_page *bpage)
613{
614 return local_read(&bpage->write);
615}
616
617static inline unsigned rb_page_commit(struct buffer_page *bpage)
618{
619 return local_read(&bpage->commit);
620}
621
622/* Size is determined by what has been commited */
623static inline unsigned rb_page_size(struct buffer_page *bpage)
624{
625 return rb_page_commit(bpage);
626}
627
628static inline unsigned
629rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
630{
631 return rb_page_commit(cpu_buffer->commit_page);
632}
633
634static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer)
635{
636 return rb_page_commit(cpu_buffer->head_page);
637}
638
639/*
640 * When the tail hits the head and the buffer is in overwrite mode,
641 * the head jumps to the next page and all content on the previous
642 * page is discarded. But before doing so, we update the overrun
643 * variable of the buffer.
644 */
645static void rb_update_overflow(struct ring_buffer_per_cpu *cpu_buffer)
646{
647 struct ring_buffer_event *event;
648 unsigned long head;
649
650 for (head = 0; head < rb_head_size(cpu_buffer);
651 head += rb_event_length(event)) {
652
653 event = __rb_page_index(cpu_buffer->head_page, head);
654 BUG_ON(rb_null_event(event));
655 /* Only count data entries */
656 if (event->type != RINGBUF_TYPE_DATA)
657 continue;
658 cpu_buffer->overrun++;
659 cpu_buffer->entries--;
660 }
661}
662
663static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer,
664 struct buffer_page **page)
665{
666 struct list_head *p = (*page)->list.next;
667
668 if (p == &cpu_buffer->pages)
669 p = p->next;
670
671 *page = list_entry(p, struct buffer_page, list);
672}
673
674static inline unsigned
675rb_event_index(struct ring_buffer_event *event)
676{
677 unsigned long addr = (unsigned long)event;
678
679 return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE);
680}
681
682static inline int
683rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
684 struct ring_buffer_event *event)
685{
686 unsigned long addr = (unsigned long)event;
687 unsigned long index;
688
689 index = rb_event_index(event);
690 addr &= PAGE_MASK;
691
692 return cpu_buffer->commit_page->page == (void *)addr &&
693 rb_commit_index(cpu_buffer) == index;
694}
695
696static inline void
697rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer,
698 struct ring_buffer_event *event)
699{
700 unsigned long addr = (unsigned long)event;
701 unsigned long index;
702
703 index = rb_event_index(event);
704 addr &= PAGE_MASK;
705
706 while (cpu_buffer->commit_page->page != (void *)addr) {
707 RB_WARN_ON(cpu_buffer,
708 cpu_buffer->commit_page == cpu_buffer->tail_page);
709 cpu_buffer->commit_page->commit =
710 cpu_buffer->commit_page->write;
711 rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
712 cpu_buffer->write_stamp = cpu_buffer->commit_page->time_stamp;
713 }
714
715 /* Now set the commit to the event's index */
716 local_set(&cpu_buffer->commit_page->commit, index);
717}
718
719static inline void
720rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
721{
722 /*
723 * We only race with interrupts and NMIs on this CPU.
724 * If we own the commit event, then we can commit
725 * all others that interrupted us, since the interruptions
726 * are in stack format (they finish before they come
727 * back to us). This allows us to do a simple loop to
728 * assign the commit to the tail.
729 */
730 while (cpu_buffer->commit_page != cpu_buffer->tail_page) {
731 cpu_buffer->commit_page->commit =
732 cpu_buffer->commit_page->write;
733 rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
734 cpu_buffer->write_stamp = cpu_buffer->commit_page->time_stamp;
735 /* add barrier to keep gcc from optimizing too much */
736 barrier();
737 }
738 while (rb_commit_index(cpu_buffer) !=
739 rb_page_write(cpu_buffer->commit_page)) {
740 cpu_buffer->commit_page->commit =
741 cpu_buffer->commit_page->write;
742 barrier();
743 }
744}
745
746static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
747{
748 cpu_buffer->read_stamp = cpu_buffer->reader_page->time_stamp;
749 cpu_buffer->reader_page->read = 0;
750}
751
752static inline void rb_inc_iter(struct ring_buffer_iter *iter)
753{
754 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
755
756 /*
757 * The iterator could be on the reader page (it starts there).
758 * But the head could have moved, since the reader was
759 * found. Check for this case and assign the iterator
760 * to the head page instead of next.
761 */
762 if (iter->head_page == cpu_buffer->reader_page)
763 iter->head_page = cpu_buffer->head_page;
764 else
765 rb_inc_page(cpu_buffer, &iter->head_page);
766
767 iter->read_stamp = iter->head_page->time_stamp;
768 iter->head = 0;
769}
770
771/**
772 * ring_buffer_update_event - update event type and data
773 * @event: the even to update
774 * @type: the type of event
775 * @length: the size of the event field in the ring buffer
776 *
777 * Update the type and data fields of the event. The length
778 * is the actual size that is written to the ring buffer,
779 * and with this, we can determine what to place into the
780 * data field.
781 */
782static inline void
783rb_update_event(struct ring_buffer_event *event,
784 unsigned type, unsigned length)
785{
786 event->type = type;
787
788 switch (type) {
789
790 case RINGBUF_TYPE_PADDING:
791 break;
792
793 case RINGBUF_TYPE_TIME_EXTEND:
794 event->len =
795 (RB_LEN_TIME_EXTEND + (RB_ALIGNMENT-1))
796 >> RB_ALIGNMENT_SHIFT;
797 break;
798
799 case RINGBUF_TYPE_TIME_STAMP:
800 event->len =
801 (RB_LEN_TIME_STAMP + (RB_ALIGNMENT-1))
802 >> RB_ALIGNMENT_SHIFT;
803 break;
804
805 case RINGBUF_TYPE_DATA:
806 length -= RB_EVNT_HDR_SIZE;
807 if (length > RB_MAX_SMALL_DATA) {
808 event->len = 0;
809 event->array[0] = length;
810 } else
811 event->len =
812 (length + (RB_ALIGNMENT-1))
813 >> RB_ALIGNMENT_SHIFT;
814 break;
815 default:
816 BUG();
817 }
818}
819
820static inline unsigned rb_calculate_event_length(unsigned length)
821{
822 struct ring_buffer_event event; /* Used only for sizeof array */
823
824 /* zero length can cause confusions */
825 if (!length)
826 length = 1;
827
828 if (length > RB_MAX_SMALL_DATA)
829 length += sizeof(event.array[0]);
830
831 length += RB_EVNT_HDR_SIZE;
832 length = ALIGN(length, RB_ALIGNMENT);
833
834 return length;
835}
836
837static struct ring_buffer_event *
838__rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
839 unsigned type, unsigned long length, u64 *ts)
840{
841 struct buffer_page *tail_page, *head_page, *reader_page;
842 unsigned long tail, write;
843 struct ring_buffer *buffer = cpu_buffer->buffer;
844 struct ring_buffer_event *event;
845 unsigned long flags;
846
847 tail_page = cpu_buffer->tail_page;
848 write = local_add_return(length, &tail_page->write);
849 tail = write - length;
850
851 /* See if we shot pass the end of this buffer page */
852 if (write > BUF_PAGE_SIZE) {
853 struct buffer_page *next_page = tail_page;
854
855 spin_lock_irqsave(&cpu_buffer->lock, flags);
856
857 rb_inc_page(cpu_buffer, &next_page);
858
859 head_page = cpu_buffer->head_page;
860 reader_page = cpu_buffer->reader_page;
861
862 /* we grabbed the lock before incrementing */
863 RB_WARN_ON(cpu_buffer, next_page == reader_page);
864
865 /*
866 * If for some reason, we had an interrupt storm that made
867 * it all the way around the buffer, bail, and warn
868 * about it.
869 */
870 if (unlikely(next_page == cpu_buffer->commit_page)) {
871 WARN_ON_ONCE(1);
872 goto out_unlock;
873 }
874
875 if (next_page == head_page) {
876 if (!(buffer->flags & RB_FL_OVERWRITE)) {
877 /* reset write */
878 if (tail <= BUF_PAGE_SIZE)
879 local_set(&tail_page->write, tail);
880 goto out_unlock;
881 }
882
883 /* tail_page has not moved yet? */
884 if (tail_page == cpu_buffer->tail_page) {
885 /* count overflows */
886 rb_update_overflow(cpu_buffer);
887
888 rb_inc_page(cpu_buffer, &head_page);
889 cpu_buffer->head_page = head_page;
890 cpu_buffer->head_page->read = 0;
891 }
892 }
893
894 /*
895 * If the tail page is still the same as what we think
896 * it is, then it is up to us to update the tail
897 * pointer.
898 */
899 if (tail_page == cpu_buffer->tail_page) {
900 local_set(&next_page->write, 0);
901 local_set(&next_page->commit, 0);
902 cpu_buffer->tail_page = next_page;
903
904 /* reread the time stamp */
905 *ts = ring_buffer_time_stamp(cpu_buffer->cpu);
906 cpu_buffer->tail_page->time_stamp = *ts;
907 }
908
909 /*
910 * The actual tail page has moved forward.
911 */
912 if (tail < BUF_PAGE_SIZE) {
913 /* Mark the rest of the page with padding */
914 event = __rb_page_index(tail_page, tail);
915 event->type = RINGBUF_TYPE_PADDING;
916 }
917
918 if (tail <= BUF_PAGE_SIZE)
919 /* Set the write back to the previous setting */
920 local_set(&tail_page->write, tail);
921
922 /*
923 * If this was a commit entry that failed,
924 * increment that too
925 */
926 if (tail_page == cpu_buffer->commit_page &&
927 tail == rb_commit_index(cpu_buffer)) {
928 rb_set_commit_to_write(cpu_buffer);
929 }
930
931 spin_unlock_irqrestore(&cpu_buffer->lock, flags);
932
933 /* fail and let the caller try again */
934 return ERR_PTR(-EAGAIN);
935 }
936
937 /* We reserved something on the buffer */
938
939 BUG_ON(write > BUF_PAGE_SIZE);
940
941 event = __rb_page_index(tail_page, tail);
942 rb_update_event(event, type, length);
943
944 /*
945 * If this is a commit and the tail is zero, then update
946 * this page's time stamp.
947 */
948 if (!tail && rb_is_commit(cpu_buffer, event))
949 cpu_buffer->commit_page->time_stamp = *ts;
950
951 return event;
952
953 out_unlock:
954 spin_unlock_irqrestore(&cpu_buffer->lock, flags);
955 return NULL;
956}
957
958static int
959rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer,
960 u64 *ts, u64 *delta)
961{
962 struct ring_buffer_event *event;
963 static int once;
964 int ret;
965
966 if (unlikely(*delta > (1ULL << 59) && !once++)) {
967 printk(KERN_WARNING "Delta way too big! %llu"
968 " ts=%llu write stamp = %llu\n",
969 (unsigned long long)*delta,
970 (unsigned long long)*ts,
971 (unsigned long long)cpu_buffer->write_stamp);
972 WARN_ON(1);
973 }
974
975 /*
976 * The delta is too big, we to add a
977 * new timestamp.
978 */
979 event = __rb_reserve_next(cpu_buffer,
980 RINGBUF_TYPE_TIME_EXTEND,
981 RB_LEN_TIME_EXTEND,
982 ts);
983 if (!event)
984 return -EBUSY;
985
986 if (PTR_ERR(event) == -EAGAIN)
987 return -EAGAIN;
988
989 /* Only a commited time event can update the write stamp */
990 if (rb_is_commit(cpu_buffer, event)) {
991 /*
992 * If this is the first on the page, then we need to
993 * update the page itself, and just put in a zero.
994 */
995 if (rb_event_index(event)) {
996 event->time_delta = *delta & TS_MASK;
997 event->array[0] = *delta >> TS_SHIFT;
998 } else {
999 cpu_buffer->commit_page->time_stamp = *ts;
1000 event->time_delta = 0;
1001 event->array[0] = 0;
1002 }
1003 cpu_buffer->write_stamp = *ts;
1004 /* let the caller know this was the commit */
1005 ret = 1;
1006 } else {
1007 /* Darn, this is just wasted space */
1008 event->time_delta = 0;
1009 event->array[0] = 0;
1010 ret = 0;
1011 }
1012
1013 *delta = 0;
1014
1015 return ret;
1016}
1017
1018static struct ring_buffer_event *
1019rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer,
1020 unsigned type, unsigned long length)
1021{
1022 struct ring_buffer_event *event;
1023 u64 ts, delta;
1024 int commit = 0;
1025 int nr_loops = 0;
1026
1027 again:
1028 /*
1029 * We allow for interrupts to reenter here and do a trace.
1030 * If one does, it will cause this original code to loop
1031 * back here. Even with heavy interrupts happening, this
1032 * should only happen a few times in a row. If this happens
1033 * 1000 times in a row, there must be either an interrupt
1034 * storm or we have something buggy.
1035 * Bail!
1036 */
1037 if (unlikely(++nr_loops > 1000)) {
1038 RB_WARN_ON(cpu_buffer, 1);
1039 return NULL;
1040 }
1041
1042 ts = ring_buffer_time_stamp(cpu_buffer->cpu);
1043
1044 /*
1045 * Only the first commit can update the timestamp.
1046 * Yes there is a race here. If an interrupt comes in
1047 * just after the conditional and it traces too, then it
1048 * will also check the deltas. More than one timestamp may
1049 * also be made. But only the entry that did the actual
1050 * commit will be something other than zero.
1051 */
1052 if (cpu_buffer->tail_page == cpu_buffer->commit_page &&
1053 rb_page_write(cpu_buffer->tail_page) ==
1054 rb_commit_index(cpu_buffer)) {
1055
1056 delta = ts - cpu_buffer->write_stamp;
1057
1058 /* make sure this delta is calculated here */
1059 barrier();
1060
1061 /* Did the write stamp get updated already? */
1062 if (unlikely(ts < cpu_buffer->write_stamp))
1063 goto again;
1064
1065 if (test_time_stamp(delta)) {
1066
1067 commit = rb_add_time_stamp(cpu_buffer, &ts, &delta);
1068
1069 if (commit == -EBUSY)
1070 return NULL;
1071
1072 if (commit == -EAGAIN)
1073 goto again;
1074
1075 RB_WARN_ON(cpu_buffer, commit < 0);
1076 }
1077 } else
1078 /* Non commits have zero deltas */
1079 delta = 0;
1080
1081 event = __rb_reserve_next(cpu_buffer, type, length, &ts);
1082 if (PTR_ERR(event) == -EAGAIN)
1083 goto again;
1084
1085 if (!event) {
1086 if (unlikely(commit))
1087 /*
1088 * Ouch! We needed a timestamp and it was commited. But
1089 * we didn't get our event reserved.
1090 */
1091 rb_set_commit_to_write(cpu_buffer);
1092 return NULL;
1093 }
1094
1095 /*
1096 * If the timestamp was commited, make the commit our entry
1097 * now so that we will update it when needed.
1098 */
1099 if (commit)
1100 rb_set_commit_event(cpu_buffer, event);
1101 else if (!rb_is_commit(cpu_buffer, event))
1102 delta = 0;
1103
1104 event->time_delta = delta;
1105
1106 return event;
1107}
1108
1109static DEFINE_PER_CPU(int, rb_need_resched);
1110
1111/**
1112 * ring_buffer_lock_reserve - reserve a part of the buffer
1113 * @buffer: the ring buffer to reserve from
1114 * @length: the length of the data to reserve (excluding event header)
1115 * @flags: a pointer to save the interrupt flags
1116 *
1117 * Returns a reseverd event on the ring buffer to copy directly to.
1118 * The user of this interface will need to get the body to write into
1119 * and can use the ring_buffer_event_data() interface.
1120 *
1121 * The length is the length of the data needed, not the event length
1122 * which also includes the event header.
1123 *
1124 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
1125 * If NULL is returned, then nothing has been allocated or locked.
1126 */
1127struct ring_buffer_event *
1128ring_buffer_lock_reserve(struct ring_buffer *buffer,
1129 unsigned long length,
1130 unsigned long *flags)
1131{
1132 struct ring_buffer_per_cpu *cpu_buffer;
1133 struct ring_buffer_event *event;
1134 int cpu, resched;
1135
1136 if (atomic_read(&buffer->record_disabled))
1137 return NULL;
1138
1139 /* If we are tracing schedule, we don't want to recurse */
1140 resched = need_resched();
1141 preempt_disable_notrace();
1142
1143 cpu = raw_smp_processor_id();
1144
1145 if (!cpu_isset(cpu, buffer->cpumask))
1146 goto out;
1147
1148 cpu_buffer = buffer->buffers[cpu];
1149
1150 if (atomic_read(&cpu_buffer->record_disabled))
1151 goto out;
1152
1153 length = rb_calculate_event_length(length);
1154 if (length > BUF_PAGE_SIZE)
1155 goto out;
1156
1157 event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length);
1158 if (!event)
1159 goto out;
1160
1161 /*
1162 * Need to store resched state on this cpu.
1163 * Only the first needs to.
1164 */
1165
1166 if (preempt_count() == 1)
1167 per_cpu(rb_need_resched, cpu) = resched;
1168
1169 return event;
1170
1171 out:
1172 if (resched)
1173 preempt_enable_notrace();
1174 else
1175 preempt_enable_notrace();
1176 return NULL;
1177}
1178
1179static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
1180 struct ring_buffer_event *event)
1181{
1182 cpu_buffer->entries++;
1183
1184 /* Only process further if we own the commit */
1185 if (!rb_is_commit(cpu_buffer, event))
1186 return;
1187
1188 cpu_buffer->write_stamp += event->time_delta;
1189
1190 rb_set_commit_to_write(cpu_buffer);
1191}
1192
1193/**
1194 * ring_buffer_unlock_commit - commit a reserved
1195 * @buffer: The buffer to commit to
1196 * @event: The event pointer to commit.
1197 * @flags: the interrupt flags received from ring_buffer_lock_reserve.
1198 *
1199 * This commits the data to the ring buffer, and releases any locks held.
1200 *
1201 * Must be paired with ring_buffer_lock_reserve.
1202 */
1203int ring_buffer_unlock_commit(struct ring_buffer *buffer,
1204 struct ring_buffer_event *event,
1205 unsigned long flags)
1206{
1207 struct ring_buffer_per_cpu *cpu_buffer;
1208 int cpu = raw_smp_processor_id();
1209
1210 cpu_buffer = buffer->buffers[cpu];
1211
1212 rb_commit(cpu_buffer, event);
1213
1214 /*
1215 * Only the last preempt count needs to restore preemption.
1216 */
1217 if (preempt_count() == 1) {
1218 if (per_cpu(rb_need_resched, cpu))
1219 preempt_enable_no_resched_notrace();
1220 else
1221 preempt_enable_notrace();
1222 } else
1223 preempt_enable_no_resched_notrace();
1224
1225 return 0;
1226}
1227
1228/**
1229 * ring_buffer_write - write data to the buffer without reserving
1230 * @buffer: The ring buffer to write to.
1231 * @length: The length of the data being written (excluding the event header)
1232 * @data: The data to write to the buffer.
1233 *
1234 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
1235 * one function. If you already have the data to write to the buffer, it
1236 * may be easier to simply call this function.
1237 *
1238 * Note, like ring_buffer_lock_reserve, the length is the length of the data
1239 * and not the length of the event which would hold the header.
1240 */
1241int ring_buffer_write(struct ring_buffer *buffer,
1242 unsigned long length,
1243 void *data)
1244{
1245 struct ring_buffer_per_cpu *cpu_buffer;
1246 struct ring_buffer_event *event;
1247 unsigned long event_length;
1248 void *body;
1249 int ret = -EBUSY;
1250 int cpu, resched;
1251
1252 if (atomic_read(&buffer->record_disabled))
1253 return -EBUSY;
1254
1255 resched = need_resched();
1256 preempt_disable_notrace();
1257
1258 cpu = raw_smp_processor_id();
1259
1260 if (!cpu_isset(cpu, buffer->cpumask))
1261 goto out;
1262
1263 cpu_buffer = buffer->buffers[cpu];
1264
1265 if (atomic_read(&cpu_buffer->record_disabled))
1266 goto out;
1267
1268 event_length = rb_calculate_event_length(length);
1269 event = rb_reserve_next_event(cpu_buffer,
1270 RINGBUF_TYPE_DATA, event_length);
1271 if (!event)
1272 goto out;
1273
1274 body = rb_event_data(event);
1275
1276 memcpy(body, data, length);
1277
1278 rb_commit(cpu_buffer, event);
1279
1280 ret = 0;
1281 out:
1282 if (resched)
1283 preempt_enable_no_resched_notrace();
1284 else
1285 preempt_enable_notrace();
1286
1287 return ret;
1288}
1289
1290static inline int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
1291{
1292 struct buffer_page *reader = cpu_buffer->reader_page;
1293 struct buffer_page *head = cpu_buffer->head_page;
1294 struct buffer_page *commit = cpu_buffer->commit_page;
1295
1296 return reader->read == rb_page_commit(reader) &&
1297 (commit == reader ||
1298 (commit == head &&
1299 head->read == rb_page_commit(commit)));
1300}
1301
1302/**
1303 * ring_buffer_record_disable - stop all writes into the buffer
1304 * @buffer: The ring buffer to stop writes to.
1305 *
1306 * This prevents all writes to the buffer. Any attempt to write
1307 * to the buffer after this will fail and return NULL.
1308 *
1309 * The caller should call synchronize_sched() after this.
1310 */
1311void ring_buffer_record_disable(struct ring_buffer *buffer)
1312{
1313 atomic_inc(&buffer->record_disabled);
1314}
1315
1316/**
1317 * ring_buffer_record_enable - enable writes to the buffer
1318 * @buffer: The ring buffer to enable writes
1319 *
1320 * Note, multiple disables will need the same number of enables
1321 * to truely enable the writing (much like preempt_disable).
1322 */
1323void ring_buffer_record_enable(struct ring_buffer *buffer)
1324{
1325 atomic_dec(&buffer->record_disabled);
1326}
1327
1328/**
1329 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
1330 * @buffer: The ring buffer to stop writes to.
1331 * @cpu: The CPU buffer to stop
1332 *
1333 * This prevents all writes to the buffer. Any attempt to write
1334 * to the buffer after this will fail and return NULL.
1335 *
1336 * The caller should call synchronize_sched() after this.
1337 */
1338void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu)
1339{
1340 struct ring_buffer_per_cpu *cpu_buffer;
1341
1342 if (!cpu_isset(cpu, buffer->cpumask))
1343 return;
1344
1345 cpu_buffer = buffer->buffers[cpu];
1346 atomic_inc(&cpu_buffer->record_disabled);
1347}
1348
1349/**
1350 * ring_buffer_record_enable_cpu - enable writes to the buffer
1351 * @buffer: The ring buffer to enable writes
1352 * @cpu: The CPU to enable.
1353 *
1354 * Note, multiple disables will need the same number of enables
1355 * to truely enable the writing (much like preempt_disable).
1356 */
1357void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu)
1358{
1359 struct ring_buffer_per_cpu *cpu_buffer;
1360
1361 if (!cpu_isset(cpu, buffer->cpumask))
1362 return;
1363
1364 cpu_buffer = buffer->buffers[cpu];
1365 atomic_dec(&cpu_buffer->record_disabled);
1366}
1367
1368/**
1369 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
1370 * @buffer: The ring buffer
1371 * @cpu: The per CPU buffer to get the entries from.
1372 */
1373unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
1374{
1375 struct ring_buffer_per_cpu *cpu_buffer;
1376
1377 if (!cpu_isset(cpu, buffer->cpumask))
1378 return 0;
1379
1380 cpu_buffer = buffer->buffers[cpu];
1381 return cpu_buffer->entries;
1382}
1383
1384/**
1385 * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer
1386 * @buffer: The ring buffer
1387 * @cpu: The per CPU buffer to get the number of overruns from
1388 */
1389unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu)
1390{
1391 struct ring_buffer_per_cpu *cpu_buffer;
1392
1393 if (!cpu_isset(cpu, buffer->cpumask))
1394 return 0;
1395
1396 cpu_buffer = buffer->buffers[cpu];
1397 return cpu_buffer->overrun;
1398}
1399
1400/**
1401 * ring_buffer_entries - get the number of entries in a buffer
1402 * @buffer: The ring buffer
1403 *
1404 * Returns the total number of entries in the ring buffer
1405 * (all CPU entries)
1406 */
1407unsigned long ring_buffer_entries(struct ring_buffer *buffer)
1408{
1409 struct ring_buffer_per_cpu *cpu_buffer;
1410 unsigned long entries = 0;
1411 int cpu;
1412
1413 /* if you care about this being correct, lock the buffer */
1414 for_each_buffer_cpu(buffer, cpu) {
1415 cpu_buffer = buffer->buffers[cpu];
1416 entries += cpu_buffer->entries;
1417 }
1418
1419 return entries;
1420}
1421
1422/**
1423 * ring_buffer_overrun_cpu - get the number of overruns in buffer
1424 * @buffer: The ring buffer
1425 *
1426 * Returns the total number of overruns in the ring buffer
1427 * (all CPU entries)
1428 */
1429unsigned long ring_buffer_overruns(struct ring_buffer *buffer)
1430{
1431 struct ring_buffer_per_cpu *cpu_buffer;
1432 unsigned long overruns = 0;
1433 int cpu;
1434
1435 /* if you care about this being correct, lock the buffer */
1436 for_each_buffer_cpu(buffer, cpu) {
1437 cpu_buffer = buffer->buffers[cpu];
1438 overruns += cpu_buffer->overrun;
1439 }
1440
1441 return overruns;
1442}
1443
1444/**
1445 * ring_buffer_iter_reset - reset an iterator
1446 * @iter: The iterator to reset
1447 *
1448 * Resets the iterator, so that it will start from the beginning
1449 * again.
1450 */
1451void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
1452{
1453 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1454
1455 /* Iterator usage is expected to have record disabled */
1456 if (list_empty(&cpu_buffer->reader_page->list)) {
1457 iter->head_page = cpu_buffer->head_page;
1458 iter->head = cpu_buffer->head_page->read;
1459 } else {
1460 iter->head_page = cpu_buffer->reader_page;
1461 iter->head = cpu_buffer->reader_page->read;
1462 }
1463 if (iter->head)
1464 iter->read_stamp = cpu_buffer->read_stamp;
1465 else
1466 iter->read_stamp = iter->head_page->time_stamp;
1467}
1468
1469/**
1470 * ring_buffer_iter_empty - check if an iterator has no more to read
1471 * @iter: The iterator to check
1472 */
1473int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
1474{
1475 struct ring_buffer_per_cpu *cpu_buffer;
1476
1477 cpu_buffer = iter->cpu_buffer;
1478
1479 return iter->head_page == cpu_buffer->commit_page &&
1480 iter->head == rb_commit_index(cpu_buffer);
1481}
1482
1483static void
1484rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
1485 struct ring_buffer_event *event)
1486{
1487 u64 delta;
1488
1489 switch (event->type) {
1490 case RINGBUF_TYPE_PADDING:
1491 return;
1492
1493 case RINGBUF_TYPE_TIME_EXTEND:
1494 delta = event->array[0];
1495 delta <<= TS_SHIFT;
1496 delta += event->time_delta;
1497 cpu_buffer->read_stamp += delta;
1498 return;
1499
1500 case RINGBUF_TYPE_TIME_STAMP:
1501 /* FIXME: not implemented */
1502 return;
1503
1504 case RINGBUF_TYPE_DATA:
1505 cpu_buffer->read_stamp += event->time_delta;
1506 return;
1507
1508 default:
1509 BUG();
1510 }
1511 return;
1512}
1513
1514static void
1515rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
1516 struct ring_buffer_event *event)
1517{
1518 u64 delta;
1519
1520 switch (event->type) {
1521 case RINGBUF_TYPE_PADDING:
1522 return;
1523
1524 case RINGBUF_TYPE_TIME_EXTEND:
1525 delta = event->array[0];
1526 delta <<= TS_SHIFT;
1527 delta += event->time_delta;
1528 iter->read_stamp += delta;
1529 return;
1530
1531 case RINGBUF_TYPE_TIME_STAMP:
1532 /* FIXME: not implemented */
1533 return;
1534
1535 case RINGBUF_TYPE_DATA:
1536 iter->read_stamp += event->time_delta;
1537 return;
1538
1539 default:
1540 BUG();
1541 }
1542 return;
1543}
1544
1545static struct buffer_page *
1546rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
1547{
1548 struct buffer_page *reader = NULL;
1549 unsigned long flags;
1550 int nr_loops = 0;
1551
1552 spin_lock_irqsave(&cpu_buffer->lock, flags);
1553
1554 again:
1555 /*
1556 * This should normally only loop twice. But because the
1557 * start of the reader inserts an empty page, it causes
1558 * a case where we will loop three times. There should be no
1559 * reason to loop four times (that I know of).
1560 */
1561 if (unlikely(++nr_loops > 3)) {
1562 RB_WARN_ON(cpu_buffer, 1);
1563 reader = NULL;
1564 goto out;
1565 }
1566
1567 reader = cpu_buffer->reader_page;
1568
1569 /* If there's more to read, return this page */
1570 if (cpu_buffer->reader_page->read < rb_page_size(reader))
1571 goto out;
1572
1573 /* Never should we have an index greater than the size */
1574 RB_WARN_ON(cpu_buffer,
1575 cpu_buffer->reader_page->read > rb_page_size(reader));
1576
1577 /* check if we caught up to the tail */
1578 reader = NULL;
1579 if (cpu_buffer->commit_page == cpu_buffer->reader_page)
1580 goto out;
1581
1582 /*
1583 * Splice the empty reader page into the list around the head.
1584 * Reset the reader page to size zero.
1585 */
1586
1587 reader = cpu_buffer->head_page;
1588 cpu_buffer->reader_page->list.next = reader->list.next;
1589 cpu_buffer->reader_page->list.prev = reader->list.prev;
1590
1591 local_set(&cpu_buffer->reader_page->write, 0);
1592 local_set(&cpu_buffer->reader_page->commit, 0);
1593
1594 /* Make the reader page now replace the head */
1595 reader->list.prev->next = &cpu_buffer->reader_page->list;
1596 reader->list.next->prev = &cpu_buffer->reader_page->list;
1597
1598 /*
1599 * If the tail is on the reader, then we must set the head
1600 * to the inserted page, otherwise we set it one before.
1601 */
1602 cpu_buffer->head_page = cpu_buffer->reader_page;
1603
1604 if (cpu_buffer->commit_page != reader)
1605 rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
1606
1607 /* Finally update the reader page to the new head */
1608 cpu_buffer->reader_page = reader;
1609 rb_reset_reader_page(cpu_buffer);
1610
1611 goto again;
1612
1613 out:
1614 spin_unlock_irqrestore(&cpu_buffer->lock, flags);
1615
1616 return reader;
1617}
1618
1619static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
1620{
1621 struct ring_buffer_event *event;
1622 struct buffer_page *reader;
1623 unsigned length;
1624
1625 reader = rb_get_reader_page(cpu_buffer);
1626
1627 /* This function should not be called when buffer is empty */
1628 BUG_ON(!reader);
1629
1630 event = rb_reader_event(cpu_buffer);
1631
1632 if (event->type == RINGBUF_TYPE_DATA)
1633 cpu_buffer->entries--;
1634
1635 rb_update_read_stamp(cpu_buffer, event);
1636
1637 length = rb_event_length(event);
1638 cpu_buffer->reader_page->read += length;
1639}
1640
1641static void rb_advance_iter(struct ring_buffer_iter *iter)
1642{
1643 struct ring_buffer *buffer;
1644 struct ring_buffer_per_cpu *cpu_buffer;
1645 struct ring_buffer_event *event;
1646 unsigned length;
1647
1648 cpu_buffer = iter->cpu_buffer;
1649 buffer = cpu_buffer->buffer;
1650
1651 /*
1652 * Check if we are at the end of the buffer.
1653 */
1654 if (iter->head >= rb_page_size(iter->head_page)) {
1655 BUG_ON(iter->head_page == cpu_buffer->commit_page);
1656 rb_inc_iter(iter);
1657 return;
1658 }
1659
1660 event = rb_iter_head_event(iter);
1661
1662 length = rb_event_length(event);
1663
1664 /*
1665 * This should not be called to advance the header if we are
1666 * at the tail of the buffer.
1667 */
1668 BUG_ON((iter->head_page == cpu_buffer->commit_page) &&
1669 (iter->head + length > rb_commit_index(cpu_buffer)));
1670
1671 rb_update_iter_read_stamp(iter, event);
1672
1673 iter->head += length;
1674
1675 /* check for end of page padding */
1676 if ((iter->head >= rb_page_size(iter->head_page)) &&
1677 (iter->head_page != cpu_buffer->commit_page))
1678 rb_advance_iter(iter);
1679}
1680
1681/**
1682 * ring_buffer_peek - peek at the next event to be read
1683 * @buffer: The ring buffer to read
1684 * @cpu: The cpu to peak at
1685 * @ts: The timestamp counter of this event.
1686 *
1687 * This will return the event that will be read next, but does
1688 * not consume the data.
1689 */
1690struct ring_buffer_event *
1691ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
1692{
1693 struct ring_buffer_per_cpu *cpu_buffer;
1694 struct ring_buffer_event *event;
1695 struct buffer_page *reader;
1696 int nr_loops = 0;
1697
1698 if (!cpu_isset(cpu, buffer->cpumask))
1699 return NULL;
1700
1701 cpu_buffer = buffer->buffers[cpu];
1702
1703 again:
1704 /*
1705 * We repeat when a timestamp is encountered. It is possible
1706 * to get multiple timestamps from an interrupt entering just
1707 * as one timestamp is about to be written. The max times
1708 * that this can happen is the number of nested interrupts we
1709 * can have. Nesting 10 deep of interrupts is clearly
1710 * an anomaly.
1711 */
1712 if (unlikely(++nr_loops > 10)) {
1713 RB_WARN_ON(cpu_buffer, 1);
1714 return NULL;
1715 }
1716
1717 reader = rb_get_reader_page(cpu_buffer);
1718 if (!reader)
1719 return NULL;
1720
1721 event = rb_reader_event(cpu_buffer);
1722
1723 switch (event->type) {
1724 case RINGBUF_TYPE_PADDING:
1725 RB_WARN_ON(cpu_buffer, 1);
1726 rb_advance_reader(cpu_buffer);
1727 return NULL;
1728
1729 case RINGBUF_TYPE_TIME_EXTEND:
1730 /* Internal data, OK to advance */
1731 rb_advance_reader(cpu_buffer);
1732 goto again;
1733
1734 case RINGBUF_TYPE_TIME_STAMP:
1735 /* FIXME: not implemented */
1736 rb_advance_reader(cpu_buffer);
1737 goto again;
1738
1739 case RINGBUF_TYPE_DATA:
1740 if (ts) {
1741 *ts = cpu_buffer->read_stamp + event->time_delta;
1742 ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts);
1743 }
1744 return event;
1745
1746 default:
1747 BUG();
1748 }
1749
1750 return NULL;
1751}
1752
1753/**
1754 * ring_buffer_iter_peek - peek at the next event to be read
1755 * @iter: The ring buffer iterator
1756 * @ts: The timestamp counter of this event.
1757 *
1758 * This will return the event that will be read next, but does
1759 * not increment the iterator.
1760 */
1761struct ring_buffer_event *
1762ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
1763{
1764 struct ring_buffer *buffer;
1765 struct ring_buffer_per_cpu *cpu_buffer;
1766 struct ring_buffer_event *event;
1767 int nr_loops = 0;
1768
1769 if (ring_buffer_iter_empty(iter))
1770 return NULL;
1771
1772 cpu_buffer = iter->cpu_buffer;
1773 buffer = cpu_buffer->buffer;
1774
1775 again:
1776 /*
1777 * We repeat when a timestamp is encountered. It is possible
1778 * to get multiple timestamps from an interrupt entering just
1779 * as one timestamp is about to be written. The max times
1780 * that this can happen is the number of nested interrupts we
1781 * can have. Nesting 10 deep of interrupts is clearly
1782 * an anomaly.
1783 */
1784 if (unlikely(++nr_loops > 10)) {
1785 RB_WARN_ON(cpu_buffer, 1);
1786 return NULL;
1787 }
1788
1789 if (rb_per_cpu_empty(cpu_buffer))
1790 return NULL;
1791
1792 event = rb_iter_head_event(iter);
1793
1794 switch (event->type) {
1795 case RINGBUF_TYPE_PADDING:
1796 rb_inc_iter(iter);
1797 goto again;
1798
1799 case RINGBUF_TYPE_TIME_EXTEND:
1800 /* Internal data, OK to advance */
1801 rb_advance_iter(iter);
1802 goto again;
1803
1804 case RINGBUF_TYPE_TIME_STAMP:
1805 /* FIXME: not implemented */
1806 rb_advance_iter(iter);
1807 goto again;
1808
1809 case RINGBUF_TYPE_DATA:
1810 if (ts) {
1811 *ts = iter->read_stamp + event->time_delta;
1812 ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts);
1813 }
1814 return event;
1815
1816 default:
1817 BUG();
1818 }
1819
1820 return NULL;
1821}
1822
1823/**
1824 * ring_buffer_consume - return an event and consume it
1825 * @buffer: The ring buffer to get the next event from
1826 *
1827 * Returns the next event in the ring buffer, and that event is consumed.
1828 * Meaning, that sequential reads will keep returning a different event,
1829 * and eventually empty the ring buffer if the producer is slower.
1830 */
1831struct ring_buffer_event *
1832ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts)
1833{
1834 struct ring_buffer_per_cpu *cpu_buffer;
1835 struct ring_buffer_event *event;
1836
1837 if (!cpu_isset(cpu, buffer->cpumask))
1838 return NULL;
1839
1840 event = ring_buffer_peek(buffer, cpu, ts);
1841 if (!event)
1842 return NULL;
1843
1844 cpu_buffer = buffer->buffers[cpu];
1845 rb_advance_reader(cpu_buffer);
1846
1847 return event;
1848}
1849
1850/**
1851 * ring_buffer_read_start - start a non consuming read of the buffer
1852 * @buffer: The ring buffer to read from
1853 * @cpu: The cpu buffer to iterate over
1854 *
1855 * This starts up an iteration through the buffer. It also disables
1856 * the recording to the buffer until the reading is finished.
1857 * This prevents the reading from being corrupted. This is not
1858 * a consuming read, so a producer is not expected.
1859 *
1860 * Must be paired with ring_buffer_finish.
1861 */
1862struct ring_buffer_iter *
1863ring_buffer_read_start(struct ring_buffer *buffer, int cpu)
1864{
1865 struct ring_buffer_per_cpu *cpu_buffer;
1866 struct ring_buffer_iter *iter;
1867 unsigned long flags;
1868
1869 if (!cpu_isset(cpu, buffer->cpumask))
1870 return NULL;
1871
1872 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1873 if (!iter)
1874 return NULL;
1875
1876 cpu_buffer = buffer->buffers[cpu];
1877
1878 iter->cpu_buffer = cpu_buffer;
1879
1880 atomic_inc(&cpu_buffer->record_disabled);
1881 synchronize_sched();
1882
1883 spin_lock_irqsave(&cpu_buffer->lock, flags);
1884 ring_buffer_iter_reset(iter);
1885 spin_unlock_irqrestore(&cpu_buffer->lock, flags);
1886
1887 return iter;
1888}
1889
1890/**
1891 * ring_buffer_finish - finish reading the iterator of the buffer
1892 * @iter: The iterator retrieved by ring_buffer_start
1893 *
1894 * This re-enables the recording to the buffer, and frees the
1895 * iterator.
1896 */
1897void
1898ring_buffer_read_finish(struct ring_buffer_iter *iter)
1899{
1900 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1901
1902 atomic_dec(&cpu_buffer->record_disabled);
1903 kfree(iter);
1904}
1905
1906/**
1907 * ring_buffer_read - read the next item in the ring buffer by the iterator
1908 * @iter: The ring buffer iterator
1909 * @ts: The time stamp of the event read.
1910 *
1911 * This reads the next event in the ring buffer and increments the iterator.
1912 */
1913struct ring_buffer_event *
1914ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
1915{
1916 struct ring_buffer_event *event;
1917
1918 event = ring_buffer_iter_peek(iter, ts);
1919 if (!event)
1920 return NULL;
1921
1922 rb_advance_iter(iter);
1923
1924 return event;
1925}
1926
1927/**
1928 * ring_buffer_size - return the size of the ring buffer (in bytes)
1929 * @buffer: The ring buffer.
1930 */
1931unsigned long ring_buffer_size(struct ring_buffer *buffer)
1932{
1933 return BUF_PAGE_SIZE * buffer->pages;
1934}
1935
1936static void
1937rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
1938{
1939 cpu_buffer->head_page
1940 = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
1941 local_set(&cpu_buffer->head_page->write, 0);
1942 local_set(&cpu_buffer->head_page->commit, 0);
1943
1944 cpu_buffer->head_page->read = 0;
1945
1946 cpu_buffer->tail_page = cpu_buffer->head_page;
1947 cpu_buffer->commit_page = cpu_buffer->head_page;
1948
1949 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
1950 local_set(&cpu_buffer->reader_page->write, 0);
1951 local_set(&cpu_buffer->reader_page->commit, 0);
1952 cpu_buffer->reader_page->read = 0;
1953
1954 cpu_buffer->overrun = 0;
1955 cpu_buffer->entries = 0;
1956}
1957
1958/**
1959 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
1960 * @buffer: The ring buffer to reset a per cpu buffer of
1961 * @cpu: The CPU buffer to be reset
1962 */
1963void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
1964{
1965 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
1966 unsigned long flags;
1967
1968 if (!cpu_isset(cpu, buffer->cpumask))
1969 return;
1970
1971 spin_lock_irqsave(&cpu_buffer->lock, flags);
1972
1973 rb_reset_cpu(cpu_buffer);
1974
1975 spin_unlock_irqrestore(&cpu_buffer->lock, flags);
1976}
1977
1978/**
1979 * ring_buffer_reset - reset a ring buffer
1980 * @buffer: The ring buffer to reset all cpu buffers
1981 */
1982void ring_buffer_reset(struct ring_buffer *buffer)
1983{
1984 int cpu;
1985
1986 for_each_buffer_cpu(buffer, cpu)
1987 ring_buffer_reset_cpu(buffer, cpu);
1988}
1989
1990/**
1991 * rind_buffer_empty - is the ring buffer empty?
1992 * @buffer: The ring buffer to test
1993 */
1994int ring_buffer_empty(struct ring_buffer *buffer)
1995{
1996 struct ring_buffer_per_cpu *cpu_buffer;
1997 int cpu;
1998
1999 /* yes this is racy, but if you don't like the race, lock the buffer */
2000 for_each_buffer_cpu(buffer, cpu) {
2001 cpu_buffer = buffer->buffers[cpu];
2002 if (!rb_per_cpu_empty(cpu_buffer))
2003 return 0;
2004 }
2005 return 1;
2006}
2007
2008/**
2009 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
2010 * @buffer: The ring buffer
2011 * @cpu: The CPU buffer to test
2012 */
2013int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu)
2014{
2015 struct ring_buffer_per_cpu *cpu_buffer;
2016
2017 if (!cpu_isset(cpu, buffer->cpumask))
2018 return 1;
2019
2020 cpu_buffer = buffer->buffers[cpu];
2021 return rb_per_cpu_empty(cpu_buffer);
2022}
2023
2024/**
2025 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
2026 * @buffer_a: One buffer to swap with
2027 * @buffer_b: The other buffer to swap with
2028 *
2029 * This function is useful for tracers that want to take a "snapshot"
2030 * of a CPU buffer and has another back up buffer lying around.
2031 * it is expected that the tracer handles the cpu buffer not being
2032 * used at the moment.
2033 */
2034int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
2035 struct ring_buffer *buffer_b, int cpu)
2036{
2037 struct ring_buffer_per_cpu *cpu_buffer_a;
2038 struct ring_buffer_per_cpu *cpu_buffer_b;
2039
2040 if (!cpu_isset(cpu, buffer_a->cpumask) ||
2041 !cpu_isset(cpu, buffer_b->cpumask))
2042 return -EINVAL;
2043
2044 /* At least make sure the two buffers are somewhat the same */
2045 if (buffer_a->size != buffer_b->size ||
2046 buffer_a->pages != buffer_b->pages)
2047 return -EINVAL;
2048
2049 cpu_buffer_a = buffer_a->buffers[cpu];
2050 cpu_buffer_b = buffer_b->buffers[cpu];
2051
2052 /*
2053 * We can't do a synchronize_sched here because this
2054 * function can be called in atomic context.
2055 * Normally this will be called from the same CPU as cpu.
2056 * If not it's up to the caller to protect this.
2057 */
2058 atomic_inc(&cpu_buffer_a->record_disabled);
2059 atomic_inc(&cpu_buffer_b->record_disabled);
2060
2061 buffer_a->buffers[cpu] = cpu_buffer_b;
2062 buffer_b->buffers[cpu] = cpu_buffer_a;
2063
2064 cpu_buffer_b->buffer = buffer_a;
2065 cpu_buffer_a->buffer = buffer_b;
2066
2067 atomic_dec(&cpu_buffer_a->record_disabled);
2068 atomic_dec(&cpu_buffer_b->record_disabled);
2069
2070 return 0;
2071}
2072
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-25 03:31:46 -0500