4 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
6 #include <linux/ftrace_event.h>
7 #include <linux/ring_buffer.h>
8 #include <linux/trace_clock.h>
9 #include <linux/trace_seq.h>
10 #include <linux/spinlock.h>
11 #include <linux/irq_work.h>
12 #include <linux/debugfs.h>
13 #include <linux/uaccess.h>
14 #include <linux/hardirq.h>
15 #include <linux/kthread.h> /* for self test */
16 #include <linux/kmemcheck.h>
17 #include <linux/module.h>
18 #include <linux/percpu.h>
19 #include <linux/mutex.h>
20 #include <linux/delay.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/hash.h>
24 #include <linux/list.h>
25 #include <linux/cpu.h>
28 #include <asm/local.h>
30 static void update_pages_handler(struct work_struct
*work
);
33 * The ring buffer header is special. We must manually up keep it.
35 int ring_buffer_print_entry_header(struct trace_seq
*s
)
39 ret
= trace_seq_printf(s
, "# compressed entry header\n");
40 ret
= trace_seq_printf(s
, "\ttype_len : 5 bits\n");
41 ret
= trace_seq_printf(s
, "\ttime_delta : 27 bits\n");
42 ret
= trace_seq_printf(s
, "\tarray : 32 bits\n");
43 ret
= trace_seq_printf(s
, "\n");
44 ret
= trace_seq_printf(s
, "\tpadding : type == %d\n",
45 RINGBUF_TYPE_PADDING
);
46 ret
= trace_seq_printf(s
, "\ttime_extend : type == %d\n",
47 RINGBUF_TYPE_TIME_EXTEND
);
48 ret
= trace_seq_printf(s
, "\tdata max type_len == %d\n",
49 RINGBUF_TYPE_DATA_TYPE_LEN_MAX
);
55 * The ring buffer is made up of a list of pages. A separate list of pages is
56 * allocated for each CPU. A writer may only write to a buffer that is
57 * associated with the CPU it is currently executing on. A reader may read
58 * from any per cpu buffer.
60 * The reader is special. For each per cpu buffer, the reader has its own
61 * reader page. When a reader has read the entire reader page, this reader
62 * page is swapped with another page in the ring buffer.
64 * Now, as long as the writer is off the reader page, the reader can do what
65 * ever it wants with that page. The writer will never write to that page
66 * again (as long as it is out of the ring buffer).
68 * Here's some silly ASCII art.
71 * |reader| RING BUFFER
73 * +------+ +---+ +---+ +---+
82 * |reader| RING BUFFER
83 * |page |------------------v
84 * +------+ +---+ +---+ +---+
93 * |reader| RING BUFFER
94 * |page |------------------v
95 * +------+ +---+ +---+ +---+
100 * +------------------------------+
104 * |buffer| RING BUFFER
105 * |page |------------------v
106 * +------+ +---+ +---+ +---+
108 * | New +---+ +---+ +---+
111 * +------------------------------+
114 * After we make this swap, the reader can hand this page off to the splice
115 * code and be done with it. It can even allocate a new page if it needs to
116 * and swap that into the ring buffer.
118 * We will be using cmpxchg soon to make all this lockless.
123 * A fast way to enable or disable all ring buffers is to
124 * call tracing_on or tracing_off. Turning off the ring buffers
125 * prevents all ring buffers from being recorded to.
126 * Turning this switch on, makes it OK to write to the
127 * ring buffer, if the ring buffer is enabled itself.
129 * There's three layers that must be on in order to write
130 * to the ring buffer.
132 * 1) This global flag must be set.
133 * 2) The ring buffer must be enabled for recording.
134 * 3) The per cpu buffer must be enabled for recording.
136 * In case of an anomaly, this global flag has a bit set that
137 * will permantly disable all ring buffers.
141 * Global flag to disable all recording to ring buffers
142 * This has two bits: ON, DISABLED
146 * 0 0 : ring buffers are off
147 * 1 0 : ring buffers are on
148 * X 1 : ring buffers are permanently disabled
152 RB_BUFFERS_ON_BIT
= 0,
153 RB_BUFFERS_DISABLED_BIT
= 1,
157 RB_BUFFERS_ON
= 1 << RB_BUFFERS_ON_BIT
,
158 RB_BUFFERS_DISABLED
= 1 << RB_BUFFERS_DISABLED_BIT
,
161 static unsigned long ring_buffer_flags __read_mostly
= RB_BUFFERS_ON
;
163 /* Used for individual buffers (after the counter) */
164 #define RB_BUFFER_OFF (1 << 20)
166 #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
169 * tracing_off_permanent - permanently disable ring buffers
171 * This function, once called, will disable all ring buffers
174 void tracing_off_permanent(void)
176 set_bit(RB_BUFFERS_DISABLED_BIT
, &ring_buffer_flags
);
179 #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
180 #define RB_ALIGNMENT 4U
181 #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
182 #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */
184 #ifndef CONFIG_HAVE_64BIT_ALIGNED_ACCESS
185 # define RB_FORCE_8BYTE_ALIGNMENT 0
186 # define RB_ARCH_ALIGNMENT RB_ALIGNMENT
188 # define RB_FORCE_8BYTE_ALIGNMENT 1
189 # define RB_ARCH_ALIGNMENT 8U
192 #define RB_ALIGN_DATA __aligned(RB_ARCH_ALIGNMENT)
194 /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
195 #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
198 RB_LEN_TIME_EXTEND
= 8,
199 RB_LEN_TIME_STAMP
= 16,
202 #define skip_time_extend(event) \
203 ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND))
205 static inline int rb_null_event(struct ring_buffer_event
*event
)
207 return event
->type_len
== RINGBUF_TYPE_PADDING
&& !event
->time_delta
;
210 static void rb_event_set_padding(struct ring_buffer_event
*event
)
212 /* padding has a NULL time_delta */
213 event
->type_len
= RINGBUF_TYPE_PADDING
;
214 event
->time_delta
= 0;
218 rb_event_data_length(struct ring_buffer_event
*event
)
223 length
= event
->type_len
* RB_ALIGNMENT
;
225 length
= event
->array
[0];
226 return length
+ RB_EVNT_HDR_SIZE
;
230 * Return the length of the given event. Will return
231 * the length of the time extend if the event is a
234 static inline unsigned
235 rb_event_length(struct ring_buffer_event
*event
)
237 switch (event
->type_len
) {
238 case RINGBUF_TYPE_PADDING
:
239 if (rb_null_event(event
))
242 return event
->array
[0] + RB_EVNT_HDR_SIZE
;
244 case RINGBUF_TYPE_TIME_EXTEND
:
245 return RB_LEN_TIME_EXTEND
;
247 case RINGBUF_TYPE_TIME_STAMP
:
248 return RB_LEN_TIME_STAMP
;
250 case RINGBUF_TYPE_DATA
:
251 return rb_event_data_length(event
);
260 * Return total length of time extend and data,
261 * or just the event length for all other events.
263 static inline unsigned
264 rb_event_ts_length(struct ring_buffer_event
*event
)
268 if (event
->type_len
== RINGBUF_TYPE_TIME_EXTEND
) {
269 /* time extends include the data event after it */
270 len
= RB_LEN_TIME_EXTEND
;
271 event
= skip_time_extend(event
);
273 return len
+ rb_event_length(event
);
277 * ring_buffer_event_length - return the length of the event
278 * @event: the event to get the length of
280 * Returns the size of the data load of a data event.
281 * If the event is something other than a data event, it
282 * returns the size of the event itself. With the exception
283 * of a TIME EXTEND, where it still returns the size of the
284 * data load of the data event after it.
286 unsigned ring_buffer_event_length(struct ring_buffer_event
*event
)
290 if (event
->type_len
== RINGBUF_TYPE_TIME_EXTEND
)
291 event
= skip_time_extend(event
);
293 length
= rb_event_length(event
);
294 if (event
->type_len
> RINGBUF_TYPE_DATA_TYPE_LEN_MAX
)
296 length
-= RB_EVNT_HDR_SIZE
;
297 if (length
> RB_MAX_SMALL_DATA
+ sizeof(event
->array
[0]))
298 length
-= sizeof(event
->array
[0]);
301 EXPORT_SYMBOL_GPL(ring_buffer_event_length
);
303 /* inline for ring buffer fast paths */
305 rb_event_data(struct ring_buffer_event
*event
)
307 if (event
->type_len
== RINGBUF_TYPE_TIME_EXTEND
)
308 event
= skip_time_extend(event
);
309 BUG_ON(event
->type_len
> RINGBUF_TYPE_DATA_TYPE_LEN_MAX
);
310 /* If length is in len field, then array[0] has the data */
312 return (void *)&event
->array
[0];
313 /* Otherwise length is in array[0] and array[1] has the data */
314 return (void *)&event
->array
[1];
318 * ring_buffer_event_data - return the data of the event
319 * @event: the event to get the data from
321 void *ring_buffer_event_data(struct ring_buffer_event
*event
)
323 return rb_event_data(event
);
325 EXPORT_SYMBOL_GPL(ring_buffer_event_data
);
327 #define for_each_buffer_cpu(buffer, cpu) \
328 for_each_cpu(cpu, buffer->cpumask)
331 #define TS_MASK ((1ULL << TS_SHIFT) - 1)
332 #define TS_DELTA_TEST (~TS_MASK)
334 /* Flag when events were overwritten */
335 #define RB_MISSED_EVENTS (1 << 31)
336 /* Missed count stored at end */
337 #define RB_MISSED_STORED (1 << 30)
339 struct buffer_data_page
{
340 u64 time_stamp
; /* page time stamp */
341 local_t commit
; /* write committed index */
342 unsigned char data
[] RB_ALIGN_DATA
; /* data of buffer page */
346 * Note, the buffer_page list must be first. The buffer pages
347 * are allocated in cache lines, which means that each buffer
348 * page will be at the beginning of a cache line, and thus
349 * the least significant bits will be zero. We use this to
350 * add flags in the list struct pointers, to make the ring buffer
354 struct list_head list
; /* list of buffer pages */
355 local_t write
; /* index for next write */
356 unsigned read
; /* index for next read */
357 local_t entries
; /* entries on this page */
358 unsigned long real_end
; /* real end of data */
359 struct buffer_data_page
*page
; /* Actual data page */
363 * The buffer page counters, write and entries, must be reset
364 * atomically when crossing page boundaries. To synchronize this
365 * update, two counters are inserted into the number. One is
366 * the actual counter for the write position or count on the page.
368 * The other is a counter of updaters. Before an update happens
369 * the update partition of the counter is incremented. This will
370 * allow the updater to update the counter atomically.
372 * The counter is 20 bits, and the state data is 12.
374 #define RB_WRITE_MASK 0xfffff
375 #define RB_WRITE_INTCNT (1 << 20)
377 static void rb_init_page(struct buffer_data_page
*bpage
)
379 local_set(&bpage
->commit
, 0);
383 * ring_buffer_page_len - the size of data on the page.
384 * @page: The page to read
386 * Returns the amount of data on the page, including buffer page header.
388 size_t ring_buffer_page_len(void *page
)
390 return local_read(&((struct buffer_data_page
*)page
)->commit
)
395 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
398 static void free_buffer_page(struct buffer_page
*bpage
)
400 free_page((unsigned long)bpage
->page
);
405 * We need to fit the time_stamp delta into 27 bits.
407 static inline int test_time_stamp(u64 delta
)
409 if (delta
& TS_DELTA_TEST
)
414 #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
416 /* Max payload is BUF_PAGE_SIZE - header (8bytes) */
417 #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
419 int ring_buffer_print_page_header(struct trace_seq
*s
)
421 struct buffer_data_page field
;
424 ret
= trace_seq_printf(s
, "\tfield: u64 timestamp;\t"
425 "offset:0;\tsize:%u;\tsigned:%u;\n",
426 (unsigned int)sizeof(field
.time_stamp
),
427 (unsigned int)is_signed_type(u64
));
429 ret
= trace_seq_printf(s
, "\tfield: local_t commit;\t"
430 "offset:%u;\tsize:%u;\tsigned:%u;\n",
431 (unsigned int)offsetof(typeof(field
), commit
),
432 (unsigned int)sizeof(field
.commit
),
433 (unsigned int)is_signed_type(long));
435 ret
= trace_seq_printf(s
, "\tfield: int overwrite;\t"
436 "offset:%u;\tsize:%u;\tsigned:%u;\n",
437 (unsigned int)offsetof(typeof(field
), commit
),
439 (unsigned int)is_signed_type(long));
441 ret
= trace_seq_printf(s
, "\tfield: char data;\t"
442 "offset:%u;\tsize:%u;\tsigned:%u;\n",
443 (unsigned int)offsetof(typeof(field
), data
),
444 (unsigned int)BUF_PAGE_SIZE
,
445 (unsigned int)is_signed_type(char));
451 struct irq_work work
;
452 wait_queue_head_t waiters
;
453 bool waiters_pending
;
457 * head_page == tail_page && head == tail then buffer is empty.
459 struct ring_buffer_per_cpu
{
461 atomic_t record_disabled
;
462 struct ring_buffer
*buffer
;
463 raw_spinlock_t reader_lock
; /* serialize readers */
464 arch_spinlock_t lock
;
465 struct lock_class_key lock_key
;
466 unsigned int nr_pages
;
467 struct list_head
*pages
;
468 struct buffer_page
*head_page
; /* read from head */
469 struct buffer_page
*tail_page
; /* write to tail */
470 struct buffer_page
*commit_page
; /* committed pages */
471 struct buffer_page
*reader_page
;
472 unsigned long lost_events
;
473 unsigned long last_overrun
;
474 local_t entries_bytes
;
477 local_t commit_overrun
;
478 local_t dropped_events
;
482 unsigned long read_bytes
;
485 /* ring buffer pages to update, > 0 to add, < 0 to remove */
486 int nr_pages_to_update
;
487 struct list_head new_pages
; /* new pages to add */
488 struct work_struct update_pages_work
;
489 struct completion update_done
;
491 struct rb_irq_work irq_work
;
497 atomic_t record_disabled
;
498 atomic_t resize_disabled
;
499 cpumask_var_t cpumask
;
501 struct lock_class_key
*reader_lock_key
;
505 struct ring_buffer_per_cpu
**buffers
;
507 #ifdef CONFIG_HOTPLUG_CPU
508 struct notifier_block cpu_notify
;
512 struct rb_irq_work irq_work
;
515 struct ring_buffer_iter
{
516 struct ring_buffer_per_cpu
*cpu_buffer
;
518 struct buffer_page
*head_page
;
519 struct buffer_page
*cache_reader_page
;
520 unsigned long cache_read
;
525 * rb_wake_up_waiters - wake up tasks waiting for ring buffer input
527 * Schedules a delayed work to wake up any task that is blocked on the
528 * ring buffer waiters queue.
530 static void rb_wake_up_waiters(struct irq_work
*work
)
532 struct rb_irq_work
*rbwork
= container_of(work
, struct rb_irq_work
, work
);
534 wake_up_all(&rbwork
->waiters
);
538 * ring_buffer_wait - wait for input to the ring buffer
539 * @buffer: buffer to wait on
540 * @cpu: the cpu buffer to wait on
542 * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
543 * as data is added to any of the @buffer's cpu buffers. Otherwise
544 * it will wait for data to be added to a specific cpu buffer.
546 void ring_buffer_wait(struct ring_buffer
*buffer
, int cpu
)
548 struct ring_buffer_per_cpu
*cpu_buffer
;
550 struct rb_irq_work
*work
;
553 * Depending on what the caller is waiting for, either any
554 * data in any cpu buffer, or a specific buffer, put the
555 * caller on the appropriate wait queue.
557 if (cpu
== RING_BUFFER_ALL_CPUS
)
558 work
= &buffer
->irq_work
;
560 cpu_buffer
= buffer
->buffers
[cpu
];
561 work
= &cpu_buffer
->irq_work
;
565 prepare_to_wait(&work
->waiters
, &wait
, TASK_INTERRUPTIBLE
);
568 * The events can happen in critical sections where
569 * checking a work queue can cause deadlocks.
570 * After adding a task to the queue, this flag is set
571 * only to notify events to try to wake up the queue
574 * We don't clear it even if the buffer is no longer
575 * empty. The flag only causes the next event to run
576 * irq_work to do the work queue wake up. The worse
577 * that can happen if we race with !trace_empty() is that
578 * an event will cause an irq_work to try to wake up
581 * There's no reason to protect this flag either, as
582 * the work queue and irq_work logic will do the necessary
583 * synchronization for the wake ups. The only thing
584 * that is necessary is that the wake up happens after
585 * a task has been queued. It's OK for spurious wake ups.
587 work
->waiters_pending
= true;
589 if ((cpu
== RING_BUFFER_ALL_CPUS
&& ring_buffer_empty(buffer
)) ||
590 (cpu
!= RING_BUFFER_ALL_CPUS
&& ring_buffer_empty_cpu(buffer
, cpu
)))
593 finish_wait(&work
->waiters
, &wait
);
597 * ring_buffer_poll_wait - poll on buffer input
598 * @buffer: buffer to wait on
599 * @cpu: the cpu buffer to wait on
600 * @filp: the file descriptor
601 * @poll_table: The poll descriptor
603 * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
604 * as data is added to any of the @buffer's cpu buffers. Otherwise
605 * it will wait for data to be added to a specific cpu buffer.
607 * Returns POLLIN | POLLRDNORM if data exists in the buffers,
610 int ring_buffer_poll_wait(struct ring_buffer
*buffer
, int cpu
,
611 struct file
*filp
, poll_table
*poll_table
)
613 struct ring_buffer_per_cpu
*cpu_buffer
;
614 struct rb_irq_work
*work
;
616 if ((cpu
== RING_BUFFER_ALL_CPUS
&& !ring_buffer_empty(buffer
)) ||
617 (cpu
!= RING_BUFFER_ALL_CPUS
&& !ring_buffer_empty_cpu(buffer
, cpu
)))
618 return POLLIN
| POLLRDNORM
;
620 if (cpu
== RING_BUFFER_ALL_CPUS
)
621 work
= &buffer
->irq_work
;
623 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
626 cpu_buffer
= buffer
->buffers
[cpu
];
627 work
= &cpu_buffer
->irq_work
;
630 work
->waiters_pending
= true;
631 poll_wait(filp
, &work
->waiters
, poll_table
);
633 if ((cpu
== RING_BUFFER_ALL_CPUS
&& !ring_buffer_empty(buffer
)) ||
634 (cpu
!= RING_BUFFER_ALL_CPUS
&& !ring_buffer_empty_cpu(buffer
, cpu
)))
635 return POLLIN
| POLLRDNORM
;
639 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
640 #define RB_WARN_ON(b, cond) \
642 int _____ret = unlikely(cond); \
644 if (__same_type(*(b), struct ring_buffer_per_cpu)) { \
645 struct ring_buffer_per_cpu *__b = \
647 atomic_inc(&__b->buffer->record_disabled); \
649 atomic_inc(&b->record_disabled); \
655 /* Up this if you want to test the TIME_EXTENTS and normalization */
656 #define DEBUG_SHIFT 0
658 static inline u64
rb_time_stamp(struct ring_buffer
*buffer
)
660 /* shift to debug/test normalization and TIME_EXTENTS */
661 return buffer
->clock() << DEBUG_SHIFT
;
664 u64
ring_buffer_time_stamp(struct ring_buffer
*buffer
, int cpu
)
668 preempt_disable_notrace();
669 time
= rb_time_stamp(buffer
);
670 preempt_enable_no_resched_notrace();
674 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp
);
676 void ring_buffer_normalize_time_stamp(struct ring_buffer
*buffer
,
679 /* Just stupid testing the normalize function and deltas */
682 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp
);
685 * Making the ring buffer lockless makes things tricky.
686 * Although writes only happen on the CPU that they are on,
687 * and they only need to worry about interrupts. Reads can
690 * The reader page is always off the ring buffer, but when the
691 * reader finishes with a page, it needs to swap its page with
692 * a new one from the buffer. The reader needs to take from
693 * the head (writes go to the tail). But if a writer is in overwrite
694 * mode and wraps, it must push the head page forward.
696 * Here lies the problem.
698 * The reader must be careful to replace only the head page, and
699 * not another one. As described at the top of the file in the
700 * ASCII art, the reader sets its old page to point to the next
701 * page after head. It then sets the page after head to point to
702 * the old reader page. But if the writer moves the head page
703 * during this operation, the reader could end up with the tail.
705 * We use cmpxchg to help prevent this race. We also do something
706 * special with the page before head. We set the LSB to 1.
708 * When the writer must push the page forward, it will clear the
709 * bit that points to the head page, move the head, and then set
710 * the bit that points to the new head page.
712 * We also don't want an interrupt coming in and moving the head
713 * page on another writer. Thus we use the second LSB to catch
716 * head->list->prev->next bit 1 bit 0
719 * Points to head page 0 1
722 * Note we can not trust the prev pointer of the head page, because:
724 * +----+ +-----+ +-----+
725 * | |------>| T |---X--->| N |
727 * +----+ +-----+ +-----+
730 * +----------| R |----------+ |
734 * Key: ---X--> HEAD flag set in pointer
739 * (see __rb_reserve_next() to see where this happens)
741 * What the above shows is that the reader just swapped out
742 * the reader page with a page in the buffer, but before it
743 * could make the new header point back to the new page added
744 * it was preempted by a writer. The writer moved forward onto
745 * the new page added by the reader and is about to move forward
748 * You can see, it is legitimate for the previous pointer of
749 * the head (or any page) not to point back to itself. But only
753 #define RB_PAGE_NORMAL 0UL
754 #define RB_PAGE_HEAD 1UL
755 #define RB_PAGE_UPDATE 2UL
758 #define RB_FLAG_MASK 3UL
760 /* PAGE_MOVED is not part of the mask */
761 #define RB_PAGE_MOVED 4UL
764 * rb_list_head - remove any bit
766 static struct list_head
*rb_list_head(struct list_head
*list
)
768 unsigned long val
= (unsigned long)list
;
770 return (struct list_head
*)(val
& ~RB_FLAG_MASK
);
774 * rb_is_head_page - test if the given page is the head page
776 * Because the reader may move the head_page pointer, we can
777 * not trust what the head page is (it may be pointing to
778 * the reader page). But if the next page is a header page,
779 * its flags will be non zero.
782 rb_is_head_page(struct ring_buffer_per_cpu
*cpu_buffer
,
783 struct buffer_page
*page
, struct list_head
*list
)
787 val
= (unsigned long)list
->next
;
789 if ((val
& ~RB_FLAG_MASK
) != (unsigned long)&page
->list
)
790 return RB_PAGE_MOVED
;
792 return val
& RB_FLAG_MASK
;
798 * The unique thing about the reader page, is that, if the
799 * writer is ever on it, the previous pointer never points
800 * back to the reader page.
802 static int rb_is_reader_page(struct buffer_page
*page
)
804 struct list_head
*list
= page
->list
.prev
;
806 return rb_list_head(list
->next
) != &page
->list
;
810 * rb_set_list_to_head - set a list_head to be pointing to head.
812 static void rb_set_list_to_head(struct ring_buffer_per_cpu
*cpu_buffer
,
813 struct list_head
*list
)
817 ptr
= (unsigned long *)&list
->next
;
818 *ptr
|= RB_PAGE_HEAD
;
819 *ptr
&= ~RB_PAGE_UPDATE
;
823 * rb_head_page_activate - sets up head page
825 static void rb_head_page_activate(struct ring_buffer_per_cpu
*cpu_buffer
)
827 struct buffer_page
*head
;
829 head
= cpu_buffer
->head_page
;
834 * Set the previous list pointer to have the HEAD flag.
836 rb_set_list_to_head(cpu_buffer
, head
->list
.prev
);
839 static void rb_list_head_clear(struct list_head
*list
)
841 unsigned long *ptr
= (unsigned long *)&list
->next
;
843 *ptr
&= ~RB_FLAG_MASK
;
847 * rb_head_page_dactivate - clears head page ptr (for free list)
850 rb_head_page_deactivate(struct ring_buffer_per_cpu
*cpu_buffer
)
852 struct list_head
*hd
;
854 /* Go through the whole list and clear any pointers found. */
855 rb_list_head_clear(cpu_buffer
->pages
);
857 list_for_each(hd
, cpu_buffer
->pages
)
858 rb_list_head_clear(hd
);
861 static int rb_head_page_set(struct ring_buffer_per_cpu
*cpu_buffer
,
862 struct buffer_page
*head
,
863 struct buffer_page
*prev
,
864 int old_flag
, int new_flag
)
866 struct list_head
*list
;
867 unsigned long val
= (unsigned long)&head
->list
;
872 val
&= ~RB_FLAG_MASK
;
874 ret
= cmpxchg((unsigned long *)&list
->next
,
875 val
| old_flag
, val
| new_flag
);
877 /* check if the reader took the page */
878 if ((ret
& ~RB_FLAG_MASK
) != val
)
879 return RB_PAGE_MOVED
;
881 return ret
& RB_FLAG_MASK
;
884 static int rb_head_page_set_update(struct ring_buffer_per_cpu
*cpu_buffer
,
885 struct buffer_page
*head
,
886 struct buffer_page
*prev
,
889 return rb_head_page_set(cpu_buffer
, head
, prev
,
890 old_flag
, RB_PAGE_UPDATE
);
893 static int rb_head_page_set_head(struct ring_buffer_per_cpu
*cpu_buffer
,
894 struct buffer_page
*head
,
895 struct buffer_page
*prev
,
898 return rb_head_page_set(cpu_buffer
, head
, prev
,
899 old_flag
, RB_PAGE_HEAD
);
902 static int rb_head_page_set_normal(struct ring_buffer_per_cpu
*cpu_buffer
,
903 struct buffer_page
*head
,
904 struct buffer_page
*prev
,
907 return rb_head_page_set(cpu_buffer
, head
, prev
,
908 old_flag
, RB_PAGE_NORMAL
);
911 static inline void rb_inc_page(struct ring_buffer_per_cpu
*cpu_buffer
,
912 struct buffer_page
**bpage
)
914 struct list_head
*p
= rb_list_head((*bpage
)->list
.next
);
916 *bpage
= list_entry(p
, struct buffer_page
, list
);
919 static struct buffer_page
*
920 rb_set_head_page(struct ring_buffer_per_cpu
*cpu_buffer
)
922 struct buffer_page
*head
;
923 struct buffer_page
*page
;
924 struct list_head
*list
;
927 if (RB_WARN_ON(cpu_buffer
, !cpu_buffer
->head_page
))
931 list
= cpu_buffer
->pages
;
932 if (RB_WARN_ON(cpu_buffer
, rb_list_head(list
->prev
->next
) != list
))
935 page
= head
= cpu_buffer
->head_page
;
937 * It is possible that the writer moves the header behind
938 * where we started, and we miss in one loop.
939 * A second loop should grab the header, but we'll do
940 * three loops just because I'm paranoid.
942 for (i
= 0; i
< 3; i
++) {
944 if (rb_is_head_page(cpu_buffer
, page
, page
->list
.prev
)) {
945 cpu_buffer
->head_page
= page
;
948 rb_inc_page(cpu_buffer
, &page
);
949 } while (page
!= head
);
952 RB_WARN_ON(cpu_buffer
, 1);
957 static int rb_head_page_replace(struct buffer_page
*old
,
958 struct buffer_page
*new)
960 unsigned long *ptr
= (unsigned long *)&old
->list
.prev
->next
;
964 val
= *ptr
& ~RB_FLAG_MASK
;
967 ret
= cmpxchg(ptr
, val
, (unsigned long)&new->list
);
973 * rb_tail_page_update - move the tail page forward
975 * Returns 1 if moved tail page, 0 if someone else did.
977 static int rb_tail_page_update(struct ring_buffer_per_cpu
*cpu_buffer
,
978 struct buffer_page
*tail_page
,
979 struct buffer_page
*next_page
)
981 struct buffer_page
*old_tail
;
982 unsigned long old_entries
;
983 unsigned long old_write
;
987 * The tail page now needs to be moved forward.
989 * We need to reset the tail page, but without messing
990 * with possible erasing of data brought in by interrupts
991 * that have moved the tail page and are currently on it.
993 * We add a counter to the write field to denote this.
995 old_write
= local_add_return(RB_WRITE_INTCNT
, &next_page
->write
);
996 old_entries
= local_add_return(RB_WRITE_INTCNT
, &next_page
->entries
);
999 * Just make sure we have seen our old_write and synchronize
1000 * with any interrupts that come in.
1005 * If the tail page is still the same as what we think
1006 * it is, then it is up to us to update the tail
1009 if (tail_page
== cpu_buffer
->tail_page
) {
1010 /* Zero the write counter */
1011 unsigned long val
= old_write
& ~RB_WRITE_MASK
;
1012 unsigned long eval
= old_entries
& ~RB_WRITE_MASK
;
1015 * This will only succeed if an interrupt did
1016 * not come in and change it. In which case, we
1017 * do not want to modify it.
1019 * We add (void) to let the compiler know that we do not care
1020 * about the return value of these functions. We use the
1021 * cmpxchg to only update if an interrupt did not already
1022 * do it for us. If the cmpxchg fails, we don't care.
1024 (void)local_cmpxchg(&next_page
->write
, old_write
, val
);
1025 (void)local_cmpxchg(&next_page
->entries
, old_entries
, eval
);
1028 * No need to worry about races with clearing out the commit.
1029 * it only can increment when a commit takes place. But that
1030 * only happens in the outer most nested commit.
1032 local_set(&next_page
->page
->commit
, 0);
1034 old_tail
= cmpxchg(&cpu_buffer
->tail_page
,
1035 tail_page
, next_page
);
1037 if (old_tail
== tail_page
)
1044 static int rb_check_bpage(struct ring_buffer_per_cpu
*cpu_buffer
,
1045 struct buffer_page
*bpage
)
1047 unsigned long val
= (unsigned long)bpage
;
1049 if (RB_WARN_ON(cpu_buffer
, val
& RB_FLAG_MASK
))
1056 * rb_check_list - make sure a pointer to a list has the last bits zero
1058 static int rb_check_list(struct ring_buffer_per_cpu
*cpu_buffer
,
1059 struct list_head
*list
)
1061 if (RB_WARN_ON(cpu_buffer
, rb_list_head(list
->prev
) != list
->prev
))
1063 if (RB_WARN_ON(cpu_buffer
, rb_list_head(list
->next
) != list
->next
))
1069 * check_pages - integrity check of buffer pages
1070 * @cpu_buffer: CPU buffer with pages to test
1072 * As a safety measure we check to make sure the data pages have not
1075 static int rb_check_pages(struct ring_buffer_per_cpu
*cpu_buffer
)
1077 struct list_head
*head
= cpu_buffer
->pages
;
1078 struct buffer_page
*bpage
, *tmp
;
1080 /* Reset the head page if it exists */
1081 if (cpu_buffer
->head_page
)
1082 rb_set_head_page(cpu_buffer
);
1084 rb_head_page_deactivate(cpu_buffer
);
1086 if (RB_WARN_ON(cpu_buffer
, head
->next
->prev
!= head
))
1088 if (RB_WARN_ON(cpu_buffer
, head
->prev
->next
!= head
))
1091 if (rb_check_list(cpu_buffer
, head
))
1094 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
1095 if (RB_WARN_ON(cpu_buffer
,
1096 bpage
->list
.next
->prev
!= &bpage
->list
))
1098 if (RB_WARN_ON(cpu_buffer
,
1099 bpage
->list
.prev
->next
!= &bpage
->list
))
1101 if (rb_check_list(cpu_buffer
, &bpage
->list
))
1105 rb_head_page_activate(cpu_buffer
);
1110 static int __rb_allocate_pages(int nr_pages
, struct list_head
*pages
, int cpu
)
1113 struct buffer_page
*bpage
, *tmp
;
1115 for (i
= 0; i
< nr_pages
; i
++) {
1118 * __GFP_NORETRY flag makes sure that the allocation fails
1119 * gracefully without invoking oom-killer and the system is
1122 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
1123 GFP_KERNEL
| __GFP_NORETRY
,
1128 list_add(&bpage
->list
, pages
);
1130 page
= alloc_pages_node(cpu_to_node(cpu
),
1131 GFP_KERNEL
| __GFP_NORETRY
, 0);
1134 bpage
->page
= page_address(page
);
1135 rb_init_page(bpage
->page
);
1141 list_for_each_entry_safe(bpage
, tmp
, pages
, list
) {
1142 list_del_init(&bpage
->list
);
1143 free_buffer_page(bpage
);
1149 static int rb_allocate_pages(struct ring_buffer_per_cpu
*cpu_buffer
,
1156 if (__rb_allocate_pages(nr_pages
, &pages
, cpu_buffer
->cpu
))
1160 * The ring buffer page list is a circular list that does not
1161 * start and end with a list head. All page list items point to
1164 cpu_buffer
->pages
= pages
.next
;
1167 cpu_buffer
->nr_pages
= nr_pages
;
1169 rb_check_pages(cpu_buffer
);
1174 static struct ring_buffer_per_cpu
*
1175 rb_allocate_cpu_buffer(struct ring_buffer
*buffer
, int nr_pages
, int cpu
)
1177 struct ring_buffer_per_cpu
*cpu_buffer
;
1178 struct buffer_page
*bpage
;
1182 cpu_buffer
= kzalloc_node(ALIGN(sizeof(*cpu_buffer
), cache_line_size()),
1183 GFP_KERNEL
, cpu_to_node(cpu
));
1187 cpu_buffer
->cpu
= cpu
;
1188 cpu_buffer
->buffer
= buffer
;
1189 raw_spin_lock_init(&cpu_buffer
->reader_lock
);
1190 lockdep_set_class(&cpu_buffer
->reader_lock
, buffer
->reader_lock_key
);
1191 cpu_buffer
->lock
= (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
1192 INIT_WORK(&cpu_buffer
->update_pages_work
, update_pages_handler
);
1193 init_completion(&cpu_buffer
->update_done
);
1194 init_irq_work(&cpu_buffer
->irq_work
.work
, rb_wake_up_waiters
);
1195 init_waitqueue_head(&cpu_buffer
->irq_work
.waiters
);
1197 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
1198 GFP_KERNEL
, cpu_to_node(cpu
));
1200 goto fail_free_buffer
;
1202 rb_check_bpage(cpu_buffer
, bpage
);
1204 cpu_buffer
->reader_page
= bpage
;
1205 page
= alloc_pages_node(cpu_to_node(cpu
), GFP_KERNEL
, 0);
1207 goto fail_free_reader
;
1208 bpage
->page
= page_address(page
);
1209 rb_init_page(bpage
->page
);
1211 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
1212 INIT_LIST_HEAD(&cpu_buffer
->new_pages
);
1214 ret
= rb_allocate_pages(cpu_buffer
, nr_pages
);
1216 goto fail_free_reader
;
1218 cpu_buffer
->head_page
1219 = list_entry(cpu_buffer
->pages
, struct buffer_page
, list
);
1220 cpu_buffer
->tail_page
= cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
1222 rb_head_page_activate(cpu_buffer
);
1227 free_buffer_page(cpu_buffer
->reader_page
);
1234 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu
*cpu_buffer
)
1236 struct list_head
*head
= cpu_buffer
->pages
;
1237 struct buffer_page
*bpage
, *tmp
;
1239 free_buffer_page(cpu_buffer
->reader_page
);
1241 rb_head_page_deactivate(cpu_buffer
);
1244 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
1245 list_del_init(&bpage
->list
);
1246 free_buffer_page(bpage
);
1248 bpage
= list_entry(head
, struct buffer_page
, list
);
1249 free_buffer_page(bpage
);
1255 #ifdef CONFIG_HOTPLUG_CPU
1256 static int rb_cpu_notify(struct notifier_block
*self
,
1257 unsigned long action
, void *hcpu
);
1261 * ring_buffer_alloc - allocate a new ring_buffer
1262 * @size: the size in bytes per cpu that is needed.
1263 * @flags: attributes to set for the ring buffer.
1265 * Currently the only flag that is available is the RB_FL_OVERWRITE
1266 * flag. This flag means that the buffer will overwrite old data
1267 * when the buffer wraps. If this flag is not set, the buffer will
1268 * drop data when the tail hits the head.
1270 struct ring_buffer
*__ring_buffer_alloc(unsigned long size
, unsigned flags
,
1271 struct lock_class_key
*key
)
1273 struct ring_buffer
*buffer
;
1277 /* keep it in its own cache line */
1278 buffer
= kzalloc(ALIGN(sizeof(*buffer
), cache_line_size()),
1283 if (!alloc_cpumask_var(&buffer
->cpumask
, GFP_KERNEL
))
1284 goto fail_free_buffer
;
1286 nr_pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
1287 buffer
->flags
= flags
;
1288 buffer
->clock
= trace_clock_local
;
1289 buffer
->reader_lock_key
= key
;
1291 init_irq_work(&buffer
->irq_work
.work
, rb_wake_up_waiters
);
1292 init_waitqueue_head(&buffer
->irq_work
.waiters
);
1294 /* need at least two pages */
1299 * In case of non-hotplug cpu, if the ring-buffer is allocated
1300 * in early initcall, it will not be notified of secondary cpus.
1301 * In that off case, we need to allocate for all possible cpus.
1303 #ifdef CONFIG_HOTPLUG_CPU
1305 cpumask_copy(buffer
->cpumask
, cpu_online_mask
);
1307 cpumask_copy(buffer
->cpumask
, cpu_possible_mask
);
1309 buffer
->cpus
= nr_cpu_ids
;
1311 bsize
= sizeof(void *) * nr_cpu_ids
;
1312 buffer
->buffers
= kzalloc(ALIGN(bsize
, cache_line_size()),
1314 if (!buffer
->buffers
)
1315 goto fail_free_cpumask
;
1317 for_each_buffer_cpu(buffer
, cpu
) {
1318 buffer
->buffers
[cpu
] =
1319 rb_allocate_cpu_buffer(buffer
, nr_pages
, cpu
);
1320 if (!buffer
->buffers
[cpu
])
1321 goto fail_free_buffers
;
1324 #ifdef CONFIG_HOTPLUG_CPU
1325 buffer
->cpu_notify
.notifier_call
= rb_cpu_notify
;
1326 buffer
->cpu_notify
.priority
= 0;
1327 register_cpu_notifier(&buffer
->cpu_notify
);
1331 mutex_init(&buffer
->mutex
);
1336 for_each_buffer_cpu(buffer
, cpu
) {
1337 if (buffer
->buffers
[cpu
])
1338 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
1340 kfree(buffer
->buffers
);
1343 free_cpumask_var(buffer
->cpumask
);
1350 EXPORT_SYMBOL_GPL(__ring_buffer_alloc
);
1353 * ring_buffer_free - free a ring buffer.
1354 * @buffer: the buffer to free.
1357 ring_buffer_free(struct ring_buffer
*buffer
)
1363 #ifdef CONFIG_HOTPLUG_CPU
1364 unregister_cpu_notifier(&buffer
->cpu_notify
);
1367 for_each_buffer_cpu(buffer
, cpu
)
1368 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
1372 kfree(buffer
->buffers
);
1373 free_cpumask_var(buffer
->cpumask
);
1377 EXPORT_SYMBOL_GPL(ring_buffer_free
);
1379 void ring_buffer_set_clock(struct ring_buffer
*buffer
,
1382 buffer
->clock
= clock
;
1385 static void rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
);
1387 static inline unsigned long rb_page_entries(struct buffer_page
*bpage
)
1389 return local_read(&bpage
->entries
) & RB_WRITE_MASK
;
1392 static inline unsigned long rb_page_write(struct buffer_page
*bpage
)
1394 return local_read(&bpage
->write
) & RB_WRITE_MASK
;
1398 rb_remove_pages(struct ring_buffer_per_cpu
*cpu_buffer
, unsigned int nr_pages
)
1400 struct list_head
*tail_page
, *to_remove
, *next_page
;
1401 struct buffer_page
*to_remove_page
, *tmp_iter_page
;
1402 struct buffer_page
*last_page
, *first_page
;
1403 unsigned int nr_removed
;
1404 unsigned long head_bit
;
1409 raw_spin_lock_irq(&cpu_buffer
->reader_lock
);
1410 atomic_inc(&cpu_buffer
->record_disabled
);
1412 * We don't race with the readers since we have acquired the reader
1413 * lock. We also don't race with writers after disabling recording.
1414 * This makes it easy to figure out the first and the last page to be
1415 * removed from the list. We unlink all the pages in between including
1416 * the first and last pages. This is done in a busy loop so that we
1417 * lose the least number of traces.
1418 * The pages are freed after we restart recording and unlock readers.
1420 tail_page
= &cpu_buffer
->tail_page
->list
;
1423 * tail page might be on reader page, we remove the next page
1424 * from the ring buffer
1426 if (cpu_buffer
->tail_page
== cpu_buffer
->reader_page
)
1427 tail_page
= rb_list_head(tail_page
->next
);
1428 to_remove
= tail_page
;
1430 /* start of pages to remove */
1431 first_page
= list_entry(rb_list_head(to_remove
->next
),
1432 struct buffer_page
, list
);
1434 for (nr_removed
= 0; nr_removed
< nr_pages
; nr_removed
++) {
1435 to_remove
= rb_list_head(to_remove
)->next
;
1436 head_bit
|= (unsigned long)to_remove
& RB_PAGE_HEAD
;
1439 next_page
= rb_list_head(to_remove
)->next
;
1442 * Now we remove all pages between tail_page and next_page.
1443 * Make sure that we have head_bit value preserved for the
1446 tail_page
->next
= (struct list_head
*)((unsigned long)next_page
|
1448 next_page
= rb_list_head(next_page
);
1449 next_page
->prev
= tail_page
;
1451 /* make sure pages points to a valid page in the ring buffer */
1452 cpu_buffer
->pages
= next_page
;
1454 /* update head page */
1456 cpu_buffer
->head_page
= list_entry(next_page
,
1457 struct buffer_page
, list
);
1460 * change read pointer to make sure any read iterators reset
1463 cpu_buffer
->read
= 0;
1465 /* pages are removed, resume tracing and then free the pages */
1466 atomic_dec(&cpu_buffer
->record_disabled
);
1467 raw_spin_unlock_irq(&cpu_buffer
->reader_lock
);
1469 RB_WARN_ON(cpu_buffer
, list_empty(cpu_buffer
->pages
));
1471 /* last buffer page to remove */
1472 last_page
= list_entry(rb_list_head(to_remove
), struct buffer_page
,
1474 tmp_iter_page
= first_page
;
1477 to_remove_page
= tmp_iter_page
;
1478 rb_inc_page(cpu_buffer
, &tmp_iter_page
);
1480 /* update the counters */
1481 page_entries
= rb_page_entries(to_remove_page
);
1484 * If something was added to this page, it was full
1485 * since it is not the tail page. So we deduct the
1486 * bytes consumed in ring buffer from here.
1487 * Increment overrun to account for the lost events.
1489 local_add(page_entries
, &cpu_buffer
->overrun
);
1490 local_sub(BUF_PAGE_SIZE
, &cpu_buffer
->entries_bytes
);
1494 * We have already removed references to this list item, just
1495 * free up the buffer_page and its page
1497 free_buffer_page(to_remove_page
);
1500 } while (to_remove_page
!= last_page
);
1502 RB_WARN_ON(cpu_buffer
, nr_removed
);
1504 return nr_removed
== 0;
1508 rb_insert_pages(struct ring_buffer_per_cpu
*cpu_buffer
)
1510 struct list_head
*pages
= &cpu_buffer
->new_pages
;
1511 int retries
, success
;
1513 raw_spin_lock_irq(&cpu_buffer
->reader_lock
);
1515 * We are holding the reader lock, so the reader page won't be swapped
1516 * in the ring buffer. Now we are racing with the writer trying to
1517 * move head page and the tail page.
1518 * We are going to adapt the reader page update process where:
1519 * 1. We first splice the start and end of list of new pages between
1520 * the head page and its previous page.
1521 * 2. We cmpxchg the prev_page->next to point from head page to the
1522 * start of new pages list.
1523 * 3. Finally, we update the head->prev to the end of new list.
1525 * We will try this process 10 times, to make sure that we don't keep
1531 struct list_head
*head_page
, *prev_page
, *r
;
1532 struct list_head
*last_page
, *first_page
;
1533 struct list_head
*head_page_with_bit
;
1535 head_page
= &rb_set_head_page(cpu_buffer
)->list
;
1538 prev_page
= head_page
->prev
;
1540 first_page
= pages
->next
;
1541 last_page
= pages
->prev
;
1543 head_page_with_bit
= (struct list_head
*)
1544 ((unsigned long)head_page
| RB_PAGE_HEAD
);
1546 last_page
->next
= head_page_with_bit
;
1547 first_page
->prev
= prev_page
;
1549 r
= cmpxchg(&prev_page
->next
, head_page_with_bit
, first_page
);
1551 if (r
== head_page_with_bit
) {
1553 * yay, we replaced the page pointer to our new list,
1554 * now, we just have to update to head page's prev
1555 * pointer to point to end of list
1557 head_page
->prev
= last_page
;
1564 INIT_LIST_HEAD(pages
);
1566 * If we weren't successful in adding in new pages, warn and stop
1569 RB_WARN_ON(cpu_buffer
, !success
);
1570 raw_spin_unlock_irq(&cpu_buffer
->reader_lock
);
1572 /* free pages if they weren't inserted */
1574 struct buffer_page
*bpage
, *tmp
;
1575 list_for_each_entry_safe(bpage
, tmp
, &cpu_buffer
->new_pages
,
1577 list_del_init(&bpage
->list
);
1578 free_buffer_page(bpage
);
1584 static void rb_update_pages(struct ring_buffer_per_cpu
*cpu_buffer
)
1588 if (cpu_buffer
->nr_pages_to_update
> 0)
1589 success
= rb_insert_pages(cpu_buffer
);
1591 success
= rb_remove_pages(cpu_buffer
,
1592 -cpu_buffer
->nr_pages_to_update
);
1595 cpu_buffer
->nr_pages
+= cpu_buffer
->nr_pages_to_update
;
1598 static void update_pages_handler(struct work_struct
*work
)
1600 struct ring_buffer_per_cpu
*cpu_buffer
= container_of(work
,
1601 struct ring_buffer_per_cpu
, update_pages_work
);
1602 rb_update_pages(cpu_buffer
);
1603 complete(&cpu_buffer
->update_done
);
1607 * ring_buffer_resize - resize the ring buffer
1608 * @buffer: the buffer to resize.
1609 * @size: the new size.
1611 * Minimum size is 2 * BUF_PAGE_SIZE.
1613 * Returns 0 on success and < 0 on failure.
1615 int ring_buffer_resize(struct ring_buffer
*buffer
, unsigned long size
,
1618 struct ring_buffer_per_cpu
*cpu_buffer
;
1623 * Always succeed at resizing a non-existent buffer:
1628 /* Make sure the requested buffer exists */
1629 if (cpu_id
!= RING_BUFFER_ALL_CPUS
&&
1630 !cpumask_test_cpu(cpu_id
, buffer
->cpumask
))
1633 size
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
1634 size
*= BUF_PAGE_SIZE
;
1636 /* we need a minimum of two pages */
1637 if (size
< BUF_PAGE_SIZE
* 2)
1638 size
= BUF_PAGE_SIZE
* 2;
1640 nr_pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
1643 * Don't succeed if resizing is disabled, as a reader might be
1644 * manipulating the ring buffer and is expecting a sane state while
1647 if (atomic_read(&buffer
->resize_disabled
))
1650 /* prevent another thread from changing buffer sizes */
1651 mutex_lock(&buffer
->mutex
);
1653 if (cpu_id
== RING_BUFFER_ALL_CPUS
) {
1654 /* calculate the pages to update */
1655 for_each_buffer_cpu(buffer
, cpu
) {
1656 cpu_buffer
= buffer
->buffers
[cpu
];
1658 cpu_buffer
->nr_pages_to_update
= nr_pages
-
1659 cpu_buffer
->nr_pages
;
1661 * nothing more to do for removing pages or no update
1663 if (cpu_buffer
->nr_pages_to_update
<= 0)
1666 * to add pages, make sure all new pages can be
1667 * allocated without receiving ENOMEM
1669 INIT_LIST_HEAD(&cpu_buffer
->new_pages
);
1670 if (__rb_allocate_pages(cpu_buffer
->nr_pages_to_update
,
1671 &cpu_buffer
->new_pages
, cpu
)) {
1672 /* not enough memory for new pages */
1680 * Fire off all the required work handlers
1681 * We can't schedule on offline CPUs, but it's not necessary
1682 * since we can change their buffer sizes without any race.
1684 for_each_buffer_cpu(buffer
, cpu
) {
1685 cpu_buffer
= buffer
->buffers
[cpu
];
1686 if (!cpu_buffer
->nr_pages_to_update
)
1689 /* The update must run on the CPU that is being updated. */
1691 if (cpu
== smp_processor_id() || !cpu_online(cpu
)) {
1692 rb_update_pages(cpu_buffer
);
1693 cpu_buffer
->nr_pages_to_update
= 0;
1696 * Can not disable preemption for schedule_work_on()
1700 schedule_work_on(cpu
,
1701 &cpu_buffer
->update_pages_work
);
1707 /* wait for all the updates to complete */
1708 for_each_buffer_cpu(buffer
, cpu
) {
1709 cpu_buffer
= buffer
->buffers
[cpu
];
1710 if (!cpu_buffer
->nr_pages_to_update
)
1713 if (cpu_online(cpu
))
1714 wait_for_completion(&cpu_buffer
->update_done
);
1715 cpu_buffer
->nr_pages_to_update
= 0;
1720 /* Make sure this CPU has been intitialized */
1721 if (!cpumask_test_cpu(cpu_id
, buffer
->cpumask
))
1724 cpu_buffer
= buffer
->buffers
[cpu_id
];
1726 if (nr_pages
== cpu_buffer
->nr_pages
)
1729 cpu_buffer
->nr_pages_to_update
= nr_pages
-
1730 cpu_buffer
->nr_pages
;
1732 INIT_LIST_HEAD(&cpu_buffer
->new_pages
);
1733 if (cpu_buffer
->nr_pages_to_update
> 0 &&
1734 __rb_allocate_pages(cpu_buffer
->nr_pages_to_update
,
1735 &cpu_buffer
->new_pages
, cpu_id
)) {
1743 /* The update must run on the CPU that is being updated. */
1744 if (cpu_id
== smp_processor_id() || !cpu_online(cpu_id
))
1745 rb_update_pages(cpu_buffer
);
1748 * Can not disable preemption for schedule_work_on()
1752 schedule_work_on(cpu_id
,
1753 &cpu_buffer
->update_pages_work
);
1754 wait_for_completion(&cpu_buffer
->update_done
);
1759 cpu_buffer
->nr_pages_to_update
= 0;
1765 * The ring buffer resize can happen with the ring buffer
1766 * enabled, so that the update disturbs the tracing as little
1767 * as possible. But if the buffer is disabled, we do not need
1768 * to worry about that, and we can take the time to verify
1769 * that the buffer is not corrupt.
1771 if (atomic_read(&buffer
->record_disabled
)) {
1772 atomic_inc(&buffer
->record_disabled
);
1774 * Even though the buffer was disabled, we must make sure
1775 * that it is truly disabled before calling rb_check_pages.
1776 * There could have been a race between checking
1777 * record_disable and incrementing it.
1779 synchronize_sched();
1780 for_each_buffer_cpu(buffer
, cpu
) {
1781 cpu_buffer
= buffer
->buffers
[cpu
];
1782 rb_check_pages(cpu_buffer
);
1784 atomic_dec(&buffer
->record_disabled
);
1787 mutex_unlock(&buffer
->mutex
);
1791 for_each_buffer_cpu(buffer
, cpu
) {
1792 struct buffer_page
*bpage
, *tmp
;
1794 cpu_buffer
= buffer
->buffers
[cpu
];
1795 cpu_buffer
->nr_pages_to_update
= 0;
1797 if (list_empty(&cpu_buffer
->new_pages
))
1800 list_for_each_entry_safe(bpage
, tmp
, &cpu_buffer
->new_pages
,
1802 list_del_init(&bpage
->list
);
1803 free_buffer_page(bpage
);
1806 mutex_unlock(&buffer
->mutex
);
1809 EXPORT_SYMBOL_GPL(ring_buffer_resize
);
1811 void ring_buffer_change_overwrite(struct ring_buffer
*buffer
, int val
)
1813 mutex_lock(&buffer
->mutex
);
1815 buffer
->flags
|= RB_FL_OVERWRITE
;
1817 buffer
->flags
&= ~RB_FL_OVERWRITE
;
1818 mutex_unlock(&buffer
->mutex
);
1820 EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite
);
1822 static inline void *
1823 __rb_data_page_index(struct buffer_data_page
*bpage
, unsigned index
)
1825 return bpage
->data
+ index
;
1828 static inline void *__rb_page_index(struct buffer_page
*bpage
, unsigned index
)
1830 return bpage
->page
->data
+ index
;
1833 static inline struct ring_buffer_event
*
1834 rb_reader_event(struct ring_buffer_per_cpu
*cpu_buffer
)
1836 return __rb_page_index(cpu_buffer
->reader_page
,
1837 cpu_buffer
->reader_page
->read
);
1840 static inline struct ring_buffer_event
*
1841 rb_iter_head_event(struct ring_buffer_iter
*iter
)
1843 return __rb_page_index(iter
->head_page
, iter
->head
);
1846 static inline unsigned rb_page_commit(struct buffer_page
*bpage
)
1848 return local_read(&bpage
->page
->commit
);
1851 /* Size is determined by what has been committed */
1852 static inline unsigned rb_page_size(struct buffer_page
*bpage
)
1854 return rb_page_commit(bpage
);
1857 static inline unsigned
1858 rb_commit_index(struct ring_buffer_per_cpu
*cpu_buffer
)
1860 return rb_page_commit(cpu_buffer
->commit_page
);
1863 static inline unsigned
1864 rb_event_index(struct ring_buffer_event
*event
)
1866 unsigned long addr
= (unsigned long)event
;
1868 return (addr
& ~PAGE_MASK
) - BUF_PAGE_HDR_SIZE
;
1872 rb_event_is_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
1873 struct ring_buffer_event
*event
)
1875 unsigned long addr
= (unsigned long)event
;
1876 unsigned long index
;
1878 index
= rb_event_index(event
);
1881 return cpu_buffer
->commit_page
->page
== (void *)addr
&&
1882 rb_commit_index(cpu_buffer
) == index
;
1886 rb_set_commit_to_write(struct ring_buffer_per_cpu
*cpu_buffer
)
1888 unsigned long max_count
;
1891 * We only race with interrupts and NMIs on this CPU.
1892 * If we own the commit event, then we can commit
1893 * all others that interrupted us, since the interruptions
1894 * are in stack format (they finish before they come
1895 * back to us). This allows us to do a simple loop to
1896 * assign the commit to the tail.
1899 max_count
= cpu_buffer
->nr_pages
* 100;
1901 while (cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
) {
1902 if (RB_WARN_ON(cpu_buffer
, !(--max_count
)))
1904 if (RB_WARN_ON(cpu_buffer
,
1905 rb_is_reader_page(cpu_buffer
->tail_page
)))
1907 local_set(&cpu_buffer
->commit_page
->page
->commit
,
1908 rb_page_write(cpu_buffer
->commit_page
));
1909 rb_inc_page(cpu_buffer
, &cpu_buffer
->commit_page
);
1910 cpu_buffer
->write_stamp
=
1911 cpu_buffer
->commit_page
->page
->time_stamp
;
1912 /* add barrier to keep gcc from optimizing too much */
1915 while (rb_commit_index(cpu_buffer
) !=
1916 rb_page_write(cpu_buffer
->commit_page
)) {
1918 local_set(&cpu_buffer
->commit_page
->page
->commit
,
1919 rb_page_write(cpu_buffer
->commit_page
));
1920 RB_WARN_ON(cpu_buffer
,
1921 local_read(&cpu_buffer
->commit_page
->page
->commit
) &
1926 /* again, keep gcc from optimizing */
1930 * If an interrupt came in just after the first while loop
1931 * and pushed the tail page forward, we will be left with
1932 * a dangling commit that will never go forward.
1934 if (unlikely(cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
))
1938 static void rb_reset_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
1940 cpu_buffer
->read_stamp
= cpu_buffer
->reader_page
->page
->time_stamp
;
1941 cpu_buffer
->reader_page
->read
= 0;
1944 static void rb_inc_iter(struct ring_buffer_iter
*iter
)
1946 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
1949 * The iterator could be on the reader page (it starts there).
1950 * But the head could have moved, since the reader was
1951 * found. Check for this case and assign the iterator
1952 * to the head page instead of next.
1954 if (iter
->head_page
== cpu_buffer
->reader_page
)
1955 iter
->head_page
= rb_set_head_page(cpu_buffer
);
1957 rb_inc_page(cpu_buffer
, &iter
->head_page
);
1959 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
1963 /* Slow path, do not inline */
1964 static noinline
struct ring_buffer_event
*
1965 rb_add_time_stamp(struct ring_buffer_event
*event
, u64 delta
)
1967 event
->type_len
= RINGBUF_TYPE_TIME_EXTEND
;
1969 /* Not the first event on the page? */
1970 if (rb_event_index(event
)) {
1971 event
->time_delta
= delta
& TS_MASK
;
1972 event
->array
[0] = delta
>> TS_SHIFT
;
1974 /* nope, just zero it */
1975 event
->time_delta
= 0;
1976 event
->array
[0] = 0;
1979 return skip_time_extend(event
);
1983 * rb_update_event - update event type and data
1984 * @event: the even to update
1985 * @type: the type of event
1986 * @length: the size of the event field in the ring buffer
1988 * Update the type and data fields of the event. The length
1989 * is the actual size that is written to the ring buffer,
1990 * and with this, we can determine what to place into the
1994 rb_update_event(struct ring_buffer_per_cpu
*cpu_buffer
,
1995 struct ring_buffer_event
*event
, unsigned length
,
1996 int add_timestamp
, u64 delta
)
1998 /* Only a commit updates the timestamp */
1999 if (unlikely(!rb_event_is_commit(cpu_buffer
, event
)))
2003 * If we need to add a timestamp, then we
2004 * add it to the start of the resevered space.
2006 if (unlikely(add_timestamp
)) {
2007 event
= rb_add_time_stamp(event
, delta
);
2008 length
-= RB_LEN_TIME_EXTEND
;
2012 event
->time_delta
= delta
;
2013 length
-= RB_EVNT_HDR_SIZE
;
2014 if (length
> RB_MAX_SMALL_DATA
|| RB_FORCE_8BYTE_ALIGNMENT
) {
2015 event
->type_len
= 0;
2016 event
->array
[0] = length
;
2018 event
->type_len
= DIV_ROUND_UP(length
, RB_ALIGNMENT
);
2022 * rb_handle_head_page - writer hit the head page
2024 * Returns: +1 to retry page
2029 rb_handle_head_page(struct ring_buffer_per_cpu
*cpu_buffer
,
2030 struct buffer_page
*tail_page
,
2031 struct buffer_page
*next_page
)
2033 struct buffer_page
*new_head
;
2038 entries
= rb_page_entries(next_page
);
2041 * The hard part is here. We need to move the head
2042 * forward, and protect against both readers on
2043 * other CPUs and writers coming in via interrupts.
2045 type
= rb_head_page_set_update(cpu_buffer
, next_page
, tail_page
,
2049 * type can be one of four:
2050 * NORMAL - an interrupt already moved it for us
2051 * HEAD - we are the first to get here.
2052 * UPDATE - we are the interrupt interrupting
2054 * MOVED - a reader on another CPU moved the next
2055 * pointer to its reader page. Give up
2062 * We changed the head to UPDATE, thus
2063 * it is our responsibility to update
2066 local_add(entries
, &cpu_buffer
->overrun
);
2067 local_sub(BUF_PAGE_SIZE
, &cpu_buffer
->entries_bytes
);
2070 * The entries will be zeroed out when we move the
2074 /* still more to do */
2077 case RB_PAGE_UPDATE
:
2079 * This is an interrupt that interrupt the
2080 * previous update. Still more to do.
2083 case RB_PAGE_NORMAL
:
2085 * An interrupt came in before the update
2086 * and processed this for us.
2087 * Nothing left to do.
2092 * The reader is on another CPU and just did
2093 * a swap with our next_page.
2098 RB_WARN_ON(cpu_buffer
, 1); /* WTF??? */
2103 * Now that we are here, the old head pointer is
2104 * set to UPDATE. This will keep the reader from
2105 * swapping the head page with the reader page.
2106 * The reader (on another CPU) will spin till
2109 * We just need to protect against interrupts
2110 * doing the job. We will set the next pointer
2111 * to HEAD. After that, we set the old pointer
2112 * to NORMAL, but only if it was HEAD before.
2113 * otherwise we are an interrupt, and only
2114 * want the outer most commit to reset it.
2116 new_head
= next_page
;
2117 rb_inc_page(cpu_buffer
, &new_head
);
2119 ret
= rb_head_page_set_head(cpu_buffer
, new_head
, next_page
,
2123 * Valid returns are:
2124 * HEAD - an interrupt came in and already set it.
2125 * NORMAL - One of two things:
2126 * 1) We really set it.
2127 * 2) A bunch of interrupts came in and moved
2128 * the page forward again.
2132 case RB_PAGE_NORMAL
:
2136 RB_WARN_ON(cpu_buffer
, 1);
2141 * It is possible that an interrupt came in,
2142 * set the head up, then more interrupts came in
2143 * and moved it again. When we get back here,
2144 * the page would have been set to NORMAL but we
2145 * just set it back to HEAD.
2147 * How do you detect this? Well, if that happened
2148 * the tail page would have moved.
2150 if (ret
== RB_PAGE_NORMAL
) {
2152 * If the tail had moved passed next, then we need
2153 * to reset the pointer.
2155 if (cpu_buffer
->tail_page
!= tail_page
&&
2156 cpu_buffer
->tail_page
!= next_page
)
2157 rb_head_page_set_normal(cpu_buffer
, new_head
,
2163 * If this was the outer most commit (the one that
2164 * changed the original pointer from HEAD to UPDATE),
2165 * then it is up to us to reset it to NORMAL.
2167 if (type
== RB_PAGE_HEAD
) {
2168 ret
= rb_head_page_set_normal(cpu_buffer
, next_page
,
2171 if (RB_WARN_ON(cpu_buffer
,
2172 ret
!= RB_PAGE_UPDATE
))
2179 static unsigned rb_calculate_event_length(unsigned length
)
2181 struct ring_buffer_event event
; /* Used only for sizeof array */
2183 /* zero length can cause confusions */
2187 if (length
> RB_MAX_SMALL_DATA
|| RB_FORCE_8BYTE_ALIGNMENT
)
2188 length
+= sizeof(event
.array
[0]);
2190 length
+= RB_EVNT_HDR_SIZE
;
2191 length
= ALIGN(length
, RB_ARCH_ALIGNMENT
);
2197 rb_reset_tail(struct ring_buffer_per_cpu
*cpu_buffer
,
2198 struct buffer_page
*tail_page
,
2199 unsigned long tail
, unsigned long length
)
2201 struct ring_buffer_event
*event
;
2204 * Only the event that crossed the page boundary
2205 * must fill the old tail_page with padding.
2207 if (tail
>= BUF_PAGE_SIZE
) {
2209 * If the page was filled, then we still need
2210 * to update the real_end. Reset it to zero
2211 * and the reader will ignore it.
2213 if (tail
== BUF_PAGE_SIZE
)
2214 tail_page
->real_end
= 0;
2216 local_sub(length
, &tail_page
->write
);
2220 event
= __rb_page_index(tail_page
, tail
);
2221 kmemcheck_annotate_bitfield(event
, bitfield
);
2223 /* account for padding bytes */
2224 local_add(BUF_PAGE_SIZE
- tail
, &cpu_buffer
->entries_bytes
);
2227 * Save the original length to the meta data.
2228 * This will be used by the reader to add lost event
2231 tail_page
->real_end
= tail
;
2234 * If this event is bigger than the minimum size, then
2235 * we need to be careful that we don't subtract the
2236 * write counter enough to allow another writer to slip
2238 * We put in a discarded commit instead, to make sure
2239 * that this space is not used again.
2241 * If we are less than the minimum size, we don't need to
2244 if (tail
> (BUF_PAGE_SIZE
- RB_EVNT_MIN_SIZE
)) {
2245 /* No room for any events */
2247 /* Mark the rest of the page with padding */
2248 rb_event_set_padding(event
);
2250 /* Set the write back to the previous setting */
2251 local_sub(length
, &tail_page
->write
);
2255 /* Put in a discarded event */
2256 event
->array
[0] = (BUF_PAGE_SIZE
- tail
) - RB_EVNT_HDR_SIZE
;
2257 event
->type_len
= RINGBUF_TYPE_PADDING
;
2258 /* time delta must be non zero */
2259 event
->time_delta
= 1;
2261 /* Set write to end of buffer */
2262 length
= (tail
+ length
) - BUF_PAGE_SIZE
;
2263 local_sub(length
, &tail_page
->write
);
2267 * This is the slow path, force gcc not to inline it.
2269 static noinline
struct ring_buffer_event
*
2270 rb_move_tail(struct ring_buffer_per_cpu
*cpu_buffer
,
2271 unsigned long length
, unsigned long tail
,
2272 struct buffer_page
*tail_page
, u64 ts
)
2274 struct buffer_page
*commit_page
= cpu_buffer
->commit_page
;
2275 struct ring_buffer
*buffer
= cpu_buffer
->buffer
;
2276 struct buffer_page
*next_page
;
2279 next_page
= tail_page
;
2281 rb_inc_page(cpu_buffer
, &next_page
);
2284 * If for some reason, we had an interrupt storm that made
2285 * it all the way around the buffer, bail, and warn
2288 if (unlikely(next_page
== commit_page
)) {
2289 local_inc(&cpu_buffer
->commit_overrun
);
2294 * This is where the fun begins!
2296 * We are fighting against races between a reader that
2297 * could be on another CPU trying to swap its reader
2298 * page with the buffer head.
2300 * We are also fighting against interrupts coming in and
2301 * moving the head or tail on us as well.
2303 * If the next page is the head page then we have filled
2304 * the buffer, unless the commit page is still on the
2307 if (rb_is_head_page(cpu_buffer
, next_page
, &tail_page
->list
)) {
2310 * If the commit is not on the reader page, then
2311 * move the header page.
2313 if (!rb_is_reader_page(cpu_buffer
->commit_page
)) {
2315 * If we are not in overwrite mode,
2316 * this is easy, just stop here.
2318 if (!(buffer
->flags
& RB_FL_OVERWRITE
)) {
2319 local_inc(&cpu_buffer
->dropped_events
);
2323 ret
= rb_handle_head_page(cpu_buffer
,
2332 * We need to be careful here too. The
2333 * commit page could still be on the reader
2334 * page. We could have a small buffer, and
2335 * have filled up the buffer with events
2336 * from interrupts and such, and wrapped.
2338 * Note, if the tail page is also the on the
2339 * reader_page, we let it move out.
2341 if (unlikely((cpu_buffer
->commit_page
!=
2342 cpu_buffer
->tail_page
) &&
2343 (cpu_buffer
->commit_page
==
2344 cpu_buffer
->reader_page
))) {
2345 local_inc(&cpu_buffer
->commit_overrun
);
2351 ret
= rb_tail_page_update(cpu_buffer
, tail_page
, next_page
);
2354 * Nested commits always have zero deltas, so
2355 * just reread the time stamp
2357 ts
= rb_time_stamp(buffer
);
2358 next_page
->page
->time_stamp
= ts
;
2363 rb_reset_tail(cpu_buffer
, tail_page
, tail
, length
);
2365 /* fail and let the caller try again */
2366 return ERR_PTR(-EAGAIN
);
2370 rb_reset_tail(cpu_buffer
, tail_page
, tail
, length
);
2375 static struct ring_buffer_event
*
2376 __rb_reserve_next(struct ring_buffer_per_cpu
*cpu_buffer
,
2377 unsigned long length
, u64 ts
,
2378 u64 delta
, int add_timestamp
)
2380 struct buffer_page
*tail_page
;
2381 struct ring_buffer_event
*event
;
2382 unsigned long tail
, write
;
2385 * If the time delta since the last event is too big to
2386 * hold in the time field of the event, then we append a
2387 * TIME EXTEND event ahead of the data event.
2389 if (unlikely(add_timestamp
))
2390 length
+= RB_LEN_TIME_EXTEND
;
2392 tail_page
= cpu_buffer
->tail_page
;
2393 write
= local_add_return(length
, &tail_page
->write
);
2395 /* set write to only the index of the write */
2396 write
&= RB_WRITE_MASK
;
2397 tail
= write
- length
;
2399 /* See if we shot pass the end of this buffer page */
2400 if (unlikely(write
> BUF_PAGE_SIZE
))
2401 return rb_move_tail(cpu_buffer
, length
, tail
,
2404 /* We reserved something on the buffer */
2406 event
= __rb_page_index(tail_page
, tail
);
2407 kmemcheck_annotate_bitfield(event
, bitfield
);
2408 rb_update_event(cpu_buffer
, event
, length
, add_timestamp
, delta
);
2410 local_inc(&tail_page
->entries
);
2413 * If this is the first commit on the page, then update
2417 tail_page
->page
->time_stamp
= ts
;
2419 /* account for these added bytes */
2420 local_add(length
, &cpu_buffer
->entries_bytes
);
2426 rb_try_to_discard(struct ring_buffer_per_cpu
*cpu_buffer
,
2427 struct ring_buffer_event
*event
)
2429 unsigned long new_index
, old_index
;
2430 struct buffer_page
*bpage
;
2431 unsigned long index
;
2434 new_index
= rb_event_index(event
);
2435 old_index
= new_index
+ rb_event_ts_length(event
);
2436 addr
= (unsigned long)event
;
2439 bpage
= cpu_buffer
->tail_page
;
2441 if (bpage
->page
== (void *)addr
&& rb_page_write(bpage
) == old_index
) {
2442 unsigned long write_mask
=
2443 local_read(&bpage
->write
) & ~RB_WRITE_MASK
;
2444 unsigned long event_length
= rb_event_length(event
);
2446 * This is on the tail page. It is possible that
2447 * a write could come in and move the tail page
2448 * and write to the next page. That is fine
2449 * because we just shorten what is on this page.
2451 old_index
+= write_mask
;
2452 new_index
+= write_mask
;
2453 index
= local_cmpxchg(&bpage
->write
, old_index
, new_index
);
2454 if (index
== old_index
) {
2455 /* update counters */
2456 local_sub(event_length
, &cpu_buffer
->entries_bytes
);
2461 /* could not discard */
2465 static void rb_start_commit(struct ring_buffer_per_cpu
*cpu_buffer
)
2467 local_inc(&cpu_buffer
->committing
);
2468 local_inc(&cpu_buffer
->commits
);
2471 static inline void rb_end_commit(struct ring_buffer_per_cpu
*cpu_buffer
)
2473 unsigned long commits
;
2475 if (RB_WARN_ON(cpu_buffer
,
2476 !local_read(&cpu_buffer
->committing
)))
2480 commits
= local_read(&cpu_buffer
->commits
);
2481 /* synchronize with interrupts */
2483 if (local_read(&cpu_buffer
->committing
) == 1)
2484 rb_set_commit_to_write(cpu_buffer
);
2486 local_dec(&cpu_buffer
->committing
);
2488 /* synchronize with interrupts */
2492 * Need to account for interrupts coming in between the
2493 * updating of the commit page and the clearing of the
2494 * committing counter.
2496 if (unlikely(local_read(&cpu_buffer
->commits
) != commits
) &&
2497 !local_read(&cpu_buffer
->committing
)) {
2498 local_inc(&cpu_buffer
->committing
);
2503 static struct ring_buffer_event
*
2504 rb_reserve_next_event(struct ring_buffer
*buffer
,
2505 struct ring_buffer_per_cpu
*cpu_buffer
,
2506 unsigned long length
)
2508 struct ring_buffer_event
*event
;
2514 rb_start_commit(cpu_buffer
);
2516 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2518 * Due to the ability to swap a cpu buffer from a buffer
2519 * it is possible it was swapped before we committed.
2520 * (committing stops a swap). We check for it here and
2521 * if it happened, we have to fail the write.
2524 if (unlikely(ACCESS_ONCE(cpu_buffer
->buffer
) != buffer
)) {
2525 local_dec(&cpu_buffer
->committing
);
2526 local_dec(&cpu_buffer
->commits
);
2531 length
= rb_calculate_event_length(length
);
2537 * We allow for interrupts to reenter here and do a trace.
2538 * If one does, it will cause this original code to loop
2539 * back here. Even with heavy interrupts happening, this
2540 * should only happen a few times in a row. If this happens
2541 * 1000 times in a row, there must be either an interrupt
2542 * storm or we have something buggy.
2545 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 1000))
2548 ts
= rb_time_stamp(cpu_buffer
->buffer
);
2549 diff
= ts
- cpu_buffer
->write_stamp
;
2551 /* make sure this diff is calculated here */
2554 /* Did the write stamp get updated already? */
2555 if (likely(ts
>= cpu_buffer
->write_stamp
)) {
2557 if (unlikely(test_time_stamp(delta
))) {
2558 int local_clock_stable
= 1;
2559 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
2560 local_clock_stable
= sched_clock_stable
;
2562 WARN_ONCE(delta
> (1ULL << 59),
2563 KERN_WARNING
"Delta way too big! %llu ts=%llu write stamp = %llu\n%s",
2564 (unsigned long long)delta
,
2565 (unsigned long long)ts
,
2566 (unsigned long long)cpu_buffer
->write_stamp
,
2567 local_clock_stable
? "" :
2568 "If you just came from a suspend/resume,\n"
2569 "please switch to the trace global clock:\n"
2570 " echo global > /sys/kernel/debug/tracing/trace_clock\n");
2575 event
= __rb_reserve_next(cpu_buffer
, length
, ts
,
2576 delta
, add_timestamp
);
2577 if (unlikely(PTR_ERR(event
) == -EAGAIN
))
2586 rb_end_commit(cpu_buffer
);
2590 #ifdef CONFIG_TRACING
2593 * The lock and unlock are done within a preempt disable section.
2594 * The current_context per_cpu variable can only be modified
2595 * by the current task between lock and unlock. But it can
2596 * be modified more than once via an interrupt. To pass this
2597 * information from the lock to the unlock without having to
2598 * access the 'in_interrupt()' functions again (which do show
2599 * a bit of overhead in something as critical as function tracing,
2600 * we use a bitmask trick.
2602 * bit 0 = NMI context
2603 * bit 1 = IRQ context
2604 * bit 2 = SoftIRQ context
2605 * bit 3 = normal context.
2607 * This works because this is the order of contexts that can
2608 * preempt other contexts. A SoftIRQ never preempts an IRQ
2611 * When the context is determined, the corresponding bit is
2612 * checked and set (if it was set, then a recursion of that context
2615 * On unlock, we need to clear this bit. To do so, just subtract
2616 * 1 from the current_context and AND it to itself.
2620 * 101 & 100 = 100 (clearing bit zero)
2623 * 1010 & 1001 = 1000 (clearing bit 1)
2625 * The least significant bit can be cleared this way, and it
2626 * just so happens that it is the same bit corresponding to
2627 * the current context.
2629 static DEFINE_PER_CPU(unsigned int, current_context
);
2631 static __always_inline
int trace_recursive_lock(void)
2633 unsigned int val
= this_cpu_read(current_context
);
2636 if (in_interrupt()) {
2646 if (unlikely(val
& (1 << bit
)))
2650 this_cpu_write(current_context
, val
);
2655 static __always_inline
void trace_recursive_unlock(void)
2657 unsigned int val
= this_cpu_read(current_context
);
2660 val
&= this_cpu_read(current_context
);
2661 this_cpu_write(current_context
, val
);
2666 #define trace_recursive_lock() (0)
2667 #define trace_recursive_unlock() do { } while (0)
2672 * ring_buffer_lock_reserve - reserve a part of the buffer
2673 * @buffer: the ring buffer to reserve from
2674 * @length: the length of the data to reserve (excluding event header)
2676 * Returns a reseverd event on the ring buffer to copy directly to.
2677 * The user of this interface will need to get the body to write into
2678 * and can use the ring_buffer_event_data() interface.
2680 * The length is the length of the data needed, not the event length
2681 * which also includes the event header.
2683 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
2684 * If NULL is returned, then nothing has been allocated or locked.
2686 struct ring_buffer_event
*
2687 ring_buffer_lock_reserve(struct ring_buffer
*buffer
, unsigned long length
)
2689 struct ring_buffer_per_cpu
*cpu_buffer
;
2690 struct ring_buffer_event
*event
;
2693 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
2696 /* If we are tracing schedule, we don't want to recurse */
2697 preempt_disable_notrace();
2699 if (atomic_read(&buffer
->record_disabled
))
2702 if (trace_recursive_lock())
2705 cpu
= raw_smp_processor_id();
2707 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2710 cpu_buffer
= buffer
->buffers
[cpu
];
2712 if (atomic_read(&cpu_buffer
->record_disabled
))
2715 if (length
> BUF_MAX_DATA_SIZE
)
2718 event
= rb_reserve_next_event(buffer
, cpu_buffer
, length
);
2725 trace_recursive_unlock();
2728 preempt_enable_notrace();
2731 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve
);
2734 rb_update_write_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
2735 struct ring_buffer_event
*event
)
2740 * The event first in the commit queue updates the
2743 if (rb_event_is_commit(cpu_buffer
, event
)) {
2745 * A commit event that is first on a page
2746 * updates the write timestamp with the page stamp
2748 if (!rb_event_index(event
))
2749 cpu_buffer
->write_stamp
=
2750 cpu_buffer
->commit_page
->page
->time_stamp
;
2751 else if (event
->type_len
== RINGBUF_TYPE_TIME_EXTEND
) {
2752 delta
= event
->array
[0];
2754 delta
+= event
->time_delta
;
2755 cpu_buffer
->write_stamp
+= delta
;
2757 cpu_buffer
->write_stamp
+= event
->time_delta
;
2761 static void rb_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
2762 struct ring_buffer_event
*event
)
2764 local_inc(&cpu_buffer
->entries
);
2765 rb_update_write_stamp(cpu_buffer
, event
);
2766 rb_end_commit(cpu_buffer
);
2769 static __always_inline
void
2770 rb_wakeups(struct ring_buffer
*buffer
, struct ring_buffer_per_cpu
*cpu_buffer
)
2772 if (buffer
->irq_work
.waiters_pending
) {
2773 buffer
->irq_work
.waiters_pending
= false;
2774 /* irq_work_queue() supplies it's own memory barriers */
2775 irq_work_queue(&buffer
->irq_work
.work
);
2778 if (cpu_buffer
->irq_work
.waiters_pending
) {
2779 cpu_buffer
->irq_work
.waiters_pending
= false;
2780 /* irq_work_queue() supplies it's own memory barriers */
2781 irq_work_queue(&cpu_buffer
->irq_work
.work
);
2786 * ring_buffer_unlock_commit - commit a reserved
2787 * @buffer: The buffer to commit to
2788 * @event: The event pointer to commit.
2790 * This commits the data to the ring buffer, and releases any locks held.
2792 * Must be paired with ring_buffer_lock_reserve.
2794 int ring_buffer_unlock_commit(struct ring_buffer
*buffer
,
2795 struct ring_buffer_event
*event
)
2797 struct ring_buffer_per_cpu
*cpu_buffer
;
2798 int cpu
= raw_smp_processor_id();
2800 cpu_buffer
= buffer
->buffers
[cpu
];
2802 rb_commit(cpu_buffer
, event
);
2804 rb_wakeups(buffer
, cpu_buffer
);
2806 trace_recursive_unlock();
2808 preempt_enable_notrace();
2812 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit
);
2814 static inline void rb_event_discard(struct ring_buffer_event
*event
)
2816 if (event
->type_len
== RINGBUF_TYPE_TIME_EXTEND
)
2817 event
= skip_time_extend(event
);
2819 /* array[0] holds the actual length for the discarded event */
2820 event
->array
[0] = rb_event_data_length(event
) - RB_EVNT_HDR_SIZE
;
2821 event
->type_len
= RINGBUF_TYPE_PADDING
;
2822 /* time delta must be non zero */
2823 if (!event
->time_delta
)
2824 event
->time_delta
= 1;
2828 * Decrement the entries to the page that an event is on.
2829 * The event does not even need to exist, only the pointer
2830 * to the page it is on. This may only be called before the commit
2834 rb_decrement_entry(struct ring_buffer_per_cpu
*cpu_buffer
,
2835 struct ring_buffer_event
*event
)
2837 unsigned long addr
= (unsigned long)event
;
2838 struct buffer_page
*bpage
= cpu_buffer
->commit_page
;
2839 struct buffer_page
*start
;
2843 /* Do the likely case first */
2844 if (likely(bpage
->page
== (void *)addr
)) {
2845 local_dec(&bpage
->entries
);
2850 * Because the commit page may be on the reader page we
2851 * start with the next page and check the end loop there.
2853 rb_inc_page(cpu_buffer
, &bpage
);
2856 if (bpage
->page
== (void *)addr
) {
2857 local_dec(&bpage
->entries
);
2860 rb_inc_page(cpu_buffer
, &bpage
);
2861 } while (bpage
!= start
);
2863 /* commit not part of this buffer?? */
2864 RB_WARN_ON(cpu_buffer
, 1);
2868 * ring_buffer_commit_discard - discard an event that has not been committed
2869 * @buffer: the ring buffer
2870 * @event: non committed event to discard
2872 * Sometimes an event that is in the ring buffer needs to be ignored.
2873 * This function lets the user discard an event in the ring buffer
2874 * and then that event will not be read later.
2876 * This function only works if it is called before the the item has been
2877 * committed. It will try to free the event from the ring buffer
2878 * if another event has not been added behind it.
2880 * If another event has been added behind it, it will set the event
2881 * up as discarded, and perform the commit.
2883 * If this function is called, do not call ring_buffer_unlock_commit on
2886 void ring_buffer_discard_commit(struct ring_buffer
*buffer
,
2887 struct ring_buffer_event
*event
)
2889 struct ring_buffer_per_cpu
*cpu_buffer
;
2892 /* The event is discarded regardless */
2893 rb_event_discard(event
);
2895 cpu
= smp_processor_id();
2896 cpu_buffer
= buffer
->buffers
[cpu
];
2899 * This must only be called if the event has not been
2900 * committed yet. Thus we can assume that preemption
2901 * is still disabled.
2903 RB_WARN_ON(buffer
, !local_read(&cpu_buffer
->committing
));
2905 rb_decrement_entry(cpu_buffer
, event
);
2906 if (rb_try_to_discard(cpu_buffer
, event
))
2910 * The commit is still visible by the reader, so we
2911 * must still update the timestamp.
2913 rb_update_write_stamp(cpu_buffer
, event
);
2915 rb_end_commit(cpu_buffer
);
2917 trace_recursive_unlock();
2919 preempt_enable_notrace();
2922 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit
);
2925 * ring_buffer_write - write data to the buffer without reserving
2926 * @buffer: The ring buffer to write to.
2927 * @length: The length of the data being written (excluding the event header)
2928 * @data: The data to write to the buffer.
2930 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
2931 * one function. If you already have the data to write to the buffer, it
2932 * may be easier to simply call this function.
2934 * Note, like ring_buffer_lock_reserve, the length is the length of the data
2935 * and not the length of the event which would hold the header.
2937 int ring_buffer_write(struct ring_buffer
*buffer
,
2938 unsigned long length
,
2941 struct ring_buffer_per_cpu
*cpu_buffer
;
2942 struct ring_buffer_event
*event
;
2947 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
2950 preempt_disable_notrace();
2952 if (atomic_read(&buffer
->record_disabled
))
2955 cpu
= raw_smp_processor_id();
2957 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2960 cpu_buffer
= buffer
->buffers
[cpu
];
2962 if (atomic_read(&cpu_buffer
->record_disabled
))
2965 if (length
> BUF_MAX_DATA_SIZE
)
2968 event
= rb_reserve_next_event(buffer
, cpu_buffer
, length
);
2972 body
= rb_event_data(event
);
2974 memcpy(body
, data
, length
);
2976 rb_commit(cpu_buffer
, event
);
2978 rb_wakeups(buffer
, cpu_buffer
);
2982 preempt_enable_notrace();
2986 EXPORT_SYMBOL_GPL(ring_buffer_write
);
2988 static int rb_per_cpu_empty(struct ring_buffer_per_cpu
*cpu_buffer
)
2990 struct buffer_page
*reader
= cpu_buffer
->reader_page
;
2991 struct buffer_page
*head
= rb_set_head_page(cpu_buffer
);
2992 struct buffer_page
*commit
= cpu_buffer
->commit_page
;
2994 /* In case of error, head will be NULL */
2995 if (unlikely(!head
))
2998 return reader
->read
== rb_page_commit(reader
) &&
2999 (commit
== reader
||
3001 head
->read
== rb_page_commit(commit
)));
3005 * ring_buffer_record_disable - stop all writes into the buffer
3006 * @buffer: The ring buffer to stop writes to.
3008 * This prevents all writes to the buffer. Any attempt to write
3009 * to the buffer after this will fail and return NULL.
3011 * The caller should call synchronize_sched() after this.
3013 void ring_buffer_record_disable(struct ring_buffer
*buffer
)
3015 atomic_inc(&buffer
->record_disabled
);
3017 EXPORT_SYMBOL_GPL(ring_buffer_record_disable
);
3020 * ring_buffer_record_enable - enable writes to the buffer
3021 * @buffer: The ring buffer to enable writes
3023 * Note, multiple disables will need the same number of enables
3024 * to truly enable the writing (much like preempt_disable).
3026 void ring_buffer_record_enable(struct ring_buffer
*buffer
)
3028 atomic_dec(&buffer
->record_disabled
);
3030 EXPORT_SYMBOL_GPL(ring_buffer_record_enable
);
3033 * ring_buffer_record_off - stop all writes into the buffer
3034 * @buffer: The ring buffer to stop writes to.
3036 * This prevents all writes to the buffer. Any attempt to write
3037 * to the buffer after this will fail and return NULL.
3039 * This is different than ring_buffer_record_disable() as
3040 * it works like an on/off switch, where as the disable() version
3041 * must be paired with a enable().
3043 void ring_buffer_record_off(struct ring_buffer
*buffer
)
3046 unsigned int new_rd
;
3049 rd
= atomic_read(&buffer
->record_disabled
);
3050 new_rd
= rd
| RB_BUFFER_OFF
;
3051 } while (atomic_cmpxchg(&buffer
->record_disabled
, rd
, new_rd
) != rd
);
3053 EXPORT_SYMBOL_GPL(ring_buffer_record_off
);
3056 * ring_buffer_record_on - restart writes into the buffer
3057 * @buffer: The ring buffer to start writes to.
3059 * This enables all writes to the buffer that was disabled by
3060 * ring_buffer_record_off().
3062 * This is different than ring_buffer_record_enable() as
3063 * it works like an on/off switch, where as the enable() version
3064 * must be paired with a disable().
3066 void ring_buffer_record_on(struct ring_buffer
*buffer
)
3069 unsigned int new_rd
;
3072 rd
= atomic_read(&buffer
->record_disabled
);
3073 new_rd
= rd
& ~RB_BUFFER_OFF
;
3074 } while (atomic_cmpxchg(&buffer
->record_disabled
, rd
, new_rd
) != rd
);
3076 EXPORT_SYMBOL_GPL(ring_buffer_record_on
);
3079 * ring_buffer_record_is_on - return true if the ring buffer can write
3080 * @buffer: The ring buffer to see if write is enabled
3082 * Returns true if the ring buffer is in a state that it accepts writes.
3084 int ring_buffer_record_is_on(struct ring_buffer
*buffer
)
3086 return !atomic_read(&buffer
->record_disabled
);
3090 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
3091 * @buffer: The ring buffer to stop writes to.
3092 * @cpu: The CPU buffer to stop
3094 * This prevents all writes to the buffer. Any attempt to write
3095 * to the buffer after this will fail and return NULL.
3097 * The caller should call synchronize_sched() after this.
3099 void ring_buffer_record_disable_cpu(struct ring_buffer
*buffer
, int cpu
)
3101 struct ring_buffer_per_cpu
*cpu_buffer
;
3103 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3106 cpu_buffer
= buffer
->buffers
[cpu
];
3107 atomic_inc(&cpu_buffer
->record_disabled
);
3109 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu
);
3112 * ring_buffer_record_enable_cpu - enable writes to the buffer
3113 * @buffer: The ring buffer to enable writes
3114 * @cpu: The CPU to enable.
3116 * Note, multiple disables will need the same number of enables
3117 * to truly enable the writing (much like preempt_disable).
3119 void ring_buffer_record_enable_cpu(struct ring_buffer
*buffer
, int cpu
)
3121 struct ring_buffer_per_cpu
*cpu_buffer
;
3123 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3126 cpu_buffer
= buffer
->buffers
[cpu
];
3127 atomic_dec(&cpu_buffer
->record_disabled
);
3129 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu
);
3132 * The total entries in the ring buffer is the running counter
3133 * of entries entered into the ring buffer, minus the sum of
3134 * the entries read from the ring buffer and the number of
3135 * entries that were overwritten.
3137 static inline unsigned long
3138 rb_num_of_entries(struct ring_buffer_per_cpu
*cpu_buffer
)
3140 return local_read(&cpu_buffer
->entries
) -
3141 (local_read(&cpu_buffer
->overrun
) + cpu_buffer
->read
);
3145 * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer
3146 * @buffer: The ring buffer
3147 * @cpu: The per CPU buffer to read from.
3149 u64
ring_buffer_oldest_event_ts(struct ring_buffer
*buffer
, int cpu
)
3151 unsigned long flags
;
3152 struct ring_buffer_per_cpu
*cpu_buffer
;
3153 struct buffer_page
*bpage
;
3156 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3159 cpu_buffer
= buffer
->buffers
[cpu
];
3160 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
3162 * if the tail is on reader_page, oldest time stamp is on the reader
3165 if (cpu_buffer
->tail_page
== cpu_buffer
->reader_page
)
3166 bpage
= cpu_buffer
->reader_page
;
3168 bpage
= rb_set_head_page(cpu_buffer
);
3170 ret
= bpage
->page
->time_stamp
;
3171 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
3175 EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts
);
3178 * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer
3179 * @buffer: The ring buffer
3180 * @cpu: The per CPU buffer to read from.
3182 unsigned long ring_buffer_bytes_cpu(struct ring_buffer
*buffer
, int cpu
)
3184 struct ring_buffer_per_cpu
*cpu_buffer
;
3187 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3190 cpu_buffer
= buffer
->buffers
[cpu
];
3191 ret
= local_read(&cpu_buffer
->entries_bytes
) - cpu_buffer
->read_bytes
;
3195 EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu
);
3198 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
3199 * @buffer: The ring buffer
3200 * @cpu: The per CPU buffer to get the entries from.
3202 unsigned long ring_buffer_entries_cpu(struct ring_buffer
*buffer
, int cpu
)
3204 struct ring_buffer_per_cpu
*cpu_buffer
;
3206 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3209 cpu_buffer
= buffer
->buffers
[cpu
];
3211 return rb_num_of_entries(cpu_buffer
);
3213 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu
);
3216 * ring_buffer_overrun_cpu - get the number of overruns caused by the ring
3217 * buffer wrapping around (only if RB_FL_OVERWRITE is on).
3218 * @buffer: The ring buffer
3219 * @cpu: The per CPU buffer to get the number of overruns from
3221 unsigned long ring_buffer_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
3223 struct ring_buffer_per_cpu
*cpu_buffer
;
3226 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3229 cpu_buffer
= buffer
->buffers
[cpu
];
3230 ret
= local_read(&cpu_buffer
->overrun
);
3234 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu
);
3237 * ring_buffer_commit_overrun_cpu - get the number of overruns caused by
3238 * commits failing due to the buffer wrapping around while there are uncommitted
3239 * events, such as during an interrupt storm.
3240 * @buffer: The ring buffer
3241 * @cpu: The per CPU buffer to get the number of overruns from
3244 ring_buffer_commit_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
3246 struct ring_buffer_per_cpu
*cpu_buffer
;
3249 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3252 cpu_buffer
= buffer
->buffers
[cpu
];
3253 ret
= local_read(&cpu_buffer
->commit_overrun
);
3257 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu
);
3260 * ring_buffer_dropped_events_cpu - get the number of dropped events caused by
3261 * the ring buffer filling up (only if RB_FL_OVERWRITE is off).
3262 * @buffer: The ring buffer
3263 * @cpu: The per CPU buffer to get the number of overruns from
3266 ring_buffer_dropped_events_cpu(struct ring_buffer
*buffer
, int cpu
)
3268 struct ring_buffer_per_cpu
*cpu_buffer
;
3271 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3274 cpu_buffer
= buffer
->buffers
[cpu
];
3275 ret
= local_read(&cpu_buffer
->dropped_events
);
3279 EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu
);
3282 * ring_buffer_read_events_cpu - get the number of events successfully read
3283 * @buffer: The ring buffer
3284 * @cpu: The per CPU buffer to get the number of events read
3287 ring_buffer_read_events_cpu(struct ring_buffer
*buffer
, int cpu
)
3289 struct ring_buffer_per_cpu
*cpu_buffer
;
3291 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3294 cpu_buffer
= buffer
->buffers
[cpu
];
3295 return cpu_buffer
->read
;
3297 EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu
);
3300 * ring_buffer_entries - get the number of entries in a buffer
3301 * @buffer: The ring buffer
3303 * Returns the total number of entries in the ring buffer
3306 unsigned long ring_buffer_entries(struct ring_buffer
*buffer
)
3308 struct ring_buffer_per_cpu
*cpu_buffer
;
3309 unsigned long entries
= 0;
3312 /* if you care about this being correct, lock the buffer */
3313 for_each_buffer_cpu(buffer
, cpu
) {
3314 cpu_buffer
= buffer
->buffers
[cpu
];
3315 entries
+= rb_num_of_entries(cpu_buffer
);
3320 EXPORT_SYMBOL_GPL(ring_buffer_entries
);
3323 * ring_buffer_overruns - get the number of overruns in buffer
3324 * @buffer: The ring buffer
3326 * Returns the total number of overruns in the ring buffer
3329 unsigned long ring_buffer_overruns(struct ring_buffer
*buffer
)
3331 struct ring_buffer_per_cpu
*cpu_buffer
;
3332 unsigned long overruns
= 0;
3335 /* if you care about this being correct, lock the buffer */
3336 for_each_buffer_cpu(buffer
, cpu
) {
3337 cpu_buffer
= buffer
->buffers
[cpu
];
3338 overruns
+= local_read(&cpu_buffer
->overrun
);
3343 EXPORT_SYMBOL_GPL(ring_buffer_overruns
);
3345 static void rb_iter_reset(struct ring_buffer_iter
*iter
)
3347 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
3349 /* Iterator usage is expected to have record disabled */
3350 if (list_empty(&cpu_buffer
->reader_page
->list
)) {
3351 iter
->head_page
= rb_set_head_page(cpu_buffer
);
3352 if (unlikely(!iter
->head_page
))
3354 iter
->head
= iter
->head_page
->read
;
3356 iter
->head_page
= cpu_buffer
->reader_page
;
3357 iter
->head
= cpu_buffer
->reader_page
->read
;
3360 iter
->read_stamp
= cpu_buffer
->read_stamp
;
3362 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
3363 iter
->cache_reader_page
= cpu_buffer
->reader_page
;
3364 iter
->cache_read
= cpu_buffer
->read
;
3368 * ring_buffer_iter_reset - reset an iterator
3369 * @iter: The iterator to reset
3371 * Resets the iterator, so that it will start from the beginning
3374 void ring_buffer_iter_reset(struct ring_buffer_iter
*iter
)
3376 struct ring_buffer_per_cpu
*cpu_buffer
;
3377 unsigned long flags
;
3382 cpu_buffer
= iter
->cpu_buffer
;
3384 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
3385 rb_iter_reset(iter
);
3386 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
3388 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset
);
3391 * ring_buffer_iter_empty - check if an iterator has no more to read
3392 * @iter: The iterator to check
3394 int ring_buffer_iter_empty(struct ring_buffer_iter
*iter
)
3396 struct ring_buffer_per_cpu
*cpu_buffer
;
3398 cpu_buffer
= iter
->cpu_buffer
;
3400 return iter
->head_page
== cpu_buffer
->commit_page
&&
3401 iter
->head
== rb_commit_index(cpu_buffer
);
3403 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty
);
3406 rb_update_read_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
3407 struct ring_buffer_event
*event
)
3411 switch (event
->type_len
) {
3412 case RINGBUF_TYPE_PADDING
:
3415 case RINGBUF_TYPE_TIME_EXTEND
:
3416 delta
= event
->array
[0];
3418 delta
+= event
->time_delta
;
3419 cpu_buffer
->read_stamp
+= delta
;
3422 case RINGBUF_TYPE_TIME_STAMP
:
3423 /* FIXME: not implemented */
3426 case RINGBUF_TYPE_DATA
:
3427 cpu_buffer
->read_stamp
+= event
->time_delta
;
3437 rb_update_iter_read_stamp(struct ring_buffer_iter
*iter
,
3438 struct ring_buffer_event
*event
)
3442 switch (event
->type_len
) {
3443 case RINGBUF_TYPE_PADDING
:
3446 case RINGBUF_TYPE_TIME_EXTEND
:
3447 delta
= event
->array
[0];
3449 delta
+= event
->time_delta
;
3450 iter
->read_stamp
+= delta
;
3453 case RINGBUF_TYPE_TIME_STAMP
:
3454 /* FIXME: not implemented */
3457 case RINGBUF_TYPE_DATA
:
3458 iter
->read_stamp
+= event
->time_delta
;
3467 static struct buffer_page
*
3468 rb_get_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
3470 struct buffer_page
*reader
= NULL
;
3471 unsigned long overwrite
;
3472 unsigned long flags
;
3476 local_irq_save(flags
);
3477 arch_spin_lock(&cpu_buffer
->lock
);
3481 * This should normally only loop twice. But because the
3482 * start of the reader inserts an empty page, it causes
3483 * a case where we will loop three times. There should be no
3484 * reason to loop four times (that I know of).
3486 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 3)) {
3491 reader
= cpu_buffer
->reader_page
;
3493 /* If there's more to read, return this page */
3494 if (cpu_buffer
->reader_page
->read
< rb_page_size(reader
))
3497 /* Never should we have an index greater than the size */
3498 if (RB_WARN_ON(cpu_buffer
,
3499 cpu_buffer
->reader_page
->read
> rb_page_size(reader
)))
3502 /* check if we caught up to the tail */
3504 if (cpu_buffer
->commit_page
== cpu_buffer
->reader_page
)
3507 /* Don't bother swapping if the ring buffer is empty */
3508 if (rb_num_of_entries(cpu_buffer
) == 0)
3512 * Reset the reader page to size zero.
3514 local_set(&cpu_buffer
->reader_page
->write
, 0);
3515 local_set(&cpu_buffer
->reader_page
->entries
, 0);
3516 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
3517 cpu_buffer
->reader_page
->real_end
= 0;
3521 * Splice the empty reader page into the list around the head.
3523 reader
= rb_set_head_page(cpu_buffer
);
3526 cpu_buffer
->reader_page
->list
.next
= rb_list_head(reader
->list
.next
);
3527 cpu_buffer
->reader_page
->list
.prev
= reader
->list
.prev
;
3530 * cpu_buffer->pages just needs to point to the buffer, it
3531 * has no specific buffer page to point to. Lets move it out
3532 * of our way so we don't accidentally swap it.
3534 cpu_buffer
->pages
= reader
->list
.prev
;
3536 /* The reader page will be pointing to the new head */
3537 rb_set_list_to_head(cpu_buffer
, &cpu_buffer
->reader_page
->list
);
3540 * We want to make sure we read the overruns after we set up our
3541 * pointers to the next object. The writer side does a
3542 * cmpxchg to cross pages which acts as the mb on the writer
3543 * side. Note, the reader will constantly fail the swap
3544 * while the writer is updating the pointers, so this
3545 * guarantees that the overwrite recorded here is the one we
3546 * want to compare with the last_overrun.
3549 overwrite
= local_read(&(cpu_buffer
->overrun
));
3552 * Here's the tricky part.
3554 * We need to move the pointer past the header page.
3555 * But we can only do that if a writer is not currently
3556 * moving it. The page before the header page has the
3557 * flag bit '1' set if it is pointing to the page we want.
3558 * but if the writer is in the process of moving it
3559 * than it will be '2' or already moved '0'.
3562 ret
= rb_head_page_replace(reader
, cpu_buffer
->reader_page
);
3565 * If we did not convert it, then we must try again.
3571 * Yeah! We succeeded in replacing the page.
3573 * Now make the new head point back to the reader page.
3575 rb_list_head(reader
->list
.next
)->prev
= &cpu_buffer
->reader_page
->list
;
3576 rb_inc_page(cpu_buffer
, &cpu_buffer
->head_page
);
3578 /* Finally update the reader page to the new head */
3579 cpu_buffer
->reader_page
= reader
;
3580 rb_reset_reader_page(cpu_buffer
);
3582 if (overwrite
!= cpu_buffer
->last_overrun
) {
3583 cpu_buffer
->lost_events
= overwrite
- cpu_buffer
->last_overrun
;
3584 cpu_buffer
->last_overrun
= overwrite
;
3590 arch_spin_unlock(&cpu_buffer
->lock
);
3591 local_irq_restore(flags
);
3596 static void rb_advance_reader(struct ring_buffer_per_cpu
*cpu_buffer
)
3598 struct ring_buffer_event
*event
;
3599 struct buffer_page
*reader
;
3602 reader
= rb_get_reader_page(cpu_buffer
);
3604 /* This function should not be called when buffer is empty */
3605 if (RB_WARN_ON(cpu_buffer
, !reader
))
3608 event
= rb_reader_event(cpu_buffer
);
3610 if (event
->type_len
<= RINGBUF_TYPE_DATA_TYPE_LEN_MAX
)
3613 rb_update_read_stamp(cpu_buffer
, event
);
3615 length
= rb_event_length(event
);
3616 cpu_buffer
->reader_page
->read
+= length
;
3619 static void rb_advance_iter(struct ring_buffer_iter
*iter
)
3621 struct ring_buffer_per_cpu
*cpu_buffer
;
3622 struct ring_buffer_event
*event
;
3625 cpu_buffer
= iter
->cpu_buffer
;
3628 * Check if we are at the end of the buffer.
3630 if (iter
->head
>= rb_page_size(iter
->head_page
)) {
3631 /* discarded commits can make the page empty */
3632 if (iter
->head_page
== cpu_buffer
->commit_page
)
3638 event
= rb_iter_head_event(iter
);
3640 length
= rb_event_length(event
);
3643 * This should not be called to advance the header if we are
3644 * at the tail of the buffer.
3646 if (RB_WARN_ON(cpu_buffer
,
3647 (iter
->head_page
== cpu_buffer
->commit_page
) &&
3648 (iter
->head
+ length
> rb_commit_index(cpu_buffer
))))
3651 rb_update_iter_read_stamp(iter
, event
);
3653 iter
->head
+= length
;
3655 /* check for end of page padding */
3656 if ((iter
->head
>= rb_page_size(iter
->head_page
)) &&
3657 (iter
->head_page
!= cpu_buffer
->commit_page
))
3661 static int rb_lost_events(struct ring_buffer_per_cpu
*cpu_buffer
)
3663 return cpu_buffer
->lost_events
;
3666 static struct ring_buffer_event
*
3667 rb_buffer_peek(struct ring_buffer_per_cpu
*cpu_buffer
, u64
*ts
,
3668 unsigned long *lost_events
)
3670 struct ring_buffer_event
*event
;
3671 struct buffer_page
*reader
;
3676 * We repeat when a time extend is encountered.
3677 * Since the time extend is always attached to a data event,
3678 * we should never loop more than once.
3679 * (We never hit the following condition more than twice).
3681 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 2))
3684 reader
= rb_get_reader_page(cpu_buffer
);
3688 event
= rb_reader_event(cpu_buffer
);
3690 switch (event
->type_len
) {
3691 case RINGBUF_TYPE_PADDING
:
3692 if (rb_null_event(event
))
3693 RB_WARN_ON(cpu_buffer
, 1);
3695 * Because the writer could be discarding every
3696 * event it creates (which would probably be bad)
3697 * if we were to go back to "again" then we may never
3698 * catch up, and will trigger the warn on, or lock
3699 * the box. Return the padding, and we will release
3700 * the current locks, and try again.
3704 case RINGBUF_TYPE_TIME_EXTEND
:
3705 /* Internal data, OK to advance */
3706 rb_advance_reader(cpu_buffer
);
3709 case RINGBUF_TYPE_TIME_STAMP
:
3710 /* FIXME: not implemented */
3711 rb_advance_reader(cpu_buffer
);
3714 case RINGBUF_TYPE_DATA
:
3716 *ts
= cpu_buffer
->read_stamp
+ event
->time_delta
;
3717 ring_buffer_normalize_time_stamp(cpu_buffer
->buffer
,
3718 cpu_buffer
->cpu
, ts
);
3721 *lost_events
= rb_lost_events(cpu_buffer
);
3730 EXPORT_SYMBOL_GPL(ring_buffer_peek
);
3732 static struct ring_buffer_event
*
3733 rb_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
3735 struct ring_buffer
*buffer
;
3736 struct ring_buffer_per_cpu
*cpu_buffer
;
3737 struct ring_buffer_event
*event
;
3740 cpu_buffer
= iter
->cpu_buffer
;
3741 buffer
= cpu_buffer
->buffer
;
3744 * Check if someone performed a consuming read to
3745 * the buffer. A consuming read invalidates the iterator
3746 * and we need to reset the iterator in this case.
3748 if (unlikely(iter
->cache_read
!= cpu_buffer
->read
||
3749 iter
->cache_reader_page
!= cpu_buffer
->reader_page
))
3750 rb_iter_reset(iter
);
3753 if (ring_buffer_iter_empty(iter
))
3757 * We repeat when a time extend is encountered.
3758 * Since the time extend is always attached to a data event,
3759 * we should never loop more than once.
3760 * (We never hit the following condition more than twice).
3762 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 2))
3765 if (rb_per_cpu_empty(cpu_buffer
))
3768 if (iter
->head
>= local_read(&iter
->head_page
->page
->commit
)) {
3773 event
= rb_iter_head_event(iter
);
3775 switch (event
->type_len
) {
3776 case RINGBUF_TYPE_PADDING
:
3777 if (rb_null_event(event
)) {
3781 rb_advance_iter(iter
);
3784 case RINGBUF_TYPE_TIME_EXTEND
:
3785 /* Internal data, OK to advance */
3786 rb_advance_iter(iter
);
3789 case RINGBUF_TYPE_TIME_STAMP
:
3790 /* FIXME: not implemented */
3791 rb_advance_iter(iter
);
3794 case RINGBUF_TYPE_DATA
:
3796 *ts
= iter
->read_stamp
+ event
->time_delta
;
3797 ring_buffer_normalize_time_stamp(buffer
,
3798 cpu_buffer
->cpu
, ts
);
3808 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek
);
3810 static inline int rb_ok_to_lock(void)
3813 * If an NMI die dumps out the content of the ring buffer
3814 * do not grab locks. We also permanently disable the ring
3815 * buffer too. A one time deal is all you get from reading
3816 * the ring buffer from an NMI.
3818 if (likely(!in_nmi()))
3821 tracing_off_permanent();
3826 * ring_buffer_peek - peek at the next event to be read
3827 * @buffer: The ring buffer to read
3828 * @cpu: The cpu to peak at
3829 * @ts: The timestamp counter of this event.
3830 * @lost_events: a variable to store if events were lost (may be NULL)
3832 * This will return the event that will be read next, but does
3833 * not consume the data.
3835 struct ring_buffer_event
*
3836 ring_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
,
3837 unsigned long *lost_events
)
3839 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
3840 struct ring_buffer_event
*event
;
3841 unsigned long flags
;
3844 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3847 dolock
= rb_ok_to_lock();
3849 local_irq_save(flags
);
3851 raw_spin_lock(&cpu_buffer
->reader_lock
);
3852 event
= rb_buffer_peek(cpu_buffer
, ts
, lost_events
);
3853 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
)
3854 rb_advance_reader(cpu_buffer
);
3856 raw_spin_unlock(&cpu_buffer
->reader_lock
);
3857 local_irq_restore(flags
);
3859 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
)
3866 * ring_buffer_iter_peek - peek at the next event to be read
3867 * @iter: The ring buffer iterator
3868 * @ts: The timestamp counter of this event.
3870 * This will return the event that will be read next, but does
3871 * not increment the iterator.
3873 struct ring_buffer_event
*
3874 ring_buffer_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
3876 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
3877 struct ring_buffer_event
*event
;
3878 unsigned long flags
;
3881 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
3882 event
= rb_iter_peek(iter
, ts
);
3883 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
3885 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
)
3892 * ring_buffer_consume - return an event and consume it
3893 * @buffer: The ring buffer to get the next event from
3894 * @cpu: the cpu to read the buffer from
3895 * @ts: a variable to store the timestamp (may be NULL)
3896 * @lost_events: a variable to store if events were lost (may be NULL)
3898 * Returns the next event in the ring buffer, and that event is consumed.
3899 * Meaning, that sequential reads will keep returning a different event,
3900 * and eventually empty the ring buffer if the producer is slower.
3902 struct ring_buffer_event
*
3903 ring_buffer_consume(struct ring_buffer
*buffer
, int cpu
, u64
*ts
,
3904 unsigned long *lost_events
)
3906 struct ring_buffer_per_cpu
*cpu_buffer
;
3907 struct ring_buffer_event
*event
= NULL
;
3908 unsigned long flags
;
3911 dolock
= rb_ok_to_lock();
3914 /* might be called in atomic */
3917 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3920 cpu_buffer
= buffer
->buffers
[cpu
];
3921 local_irq_save(flags
);
3923 raw_spin_lock(&cpu_buffer
->reader_lock
);
3925 event
= rb_buffer_peek(cpu_buffer
, ts
, lost_events
);
3927 cpu_buffer
->lost_events
= 0;
3928 rb_advance_reader(cpu_buffer
);
3932 raw_spin_unlock(&cpu_buffer
->reader_lock
);
3933 local_irq_restore(flags
);
3938 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
)
3943 EXPORT_SYMBOL_GPL(ring_buffer_consume
);
3946 * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer
3947 * @buffer: The ring buffer to read from
3948 * @cpu: The cpu buffer to iterate over
3950 * This performs the initial preparations necessary to iterate
3951 * through the buffer. Memory is allocated, buffer recording
3952 * is disabled, and the iterator pointer is returned to the caller.
3954 * Disabling buffer recordng prevents the reading from being
3955 * corrupted. This is not a consuming read, so a producer is not
3958 * After a sequence of ring_buffer_read_prepare calls, the user is
3959 * expected to make at least one call to ring_buffer_prepare_sync.
3960 * Afterwards, ring_buffer_read_start is invoked to get things going
3963 * This overall must be paired with ring_buffer_finish.
3965 struct ring_buffer_iter
*
3966 ring_buffer_read_prepare(struct ring_buffer
*buffer
, int cpu
)
3968 struct ring_buffer_per_cpu
*cpu_buffer
;
3969 struct ring_buffer_iter
*iter
;
3971 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3974 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
3978 cpu_buffer
= buffer
->buffers
[cpu
];
3980 iter
->cpu_buffer
= cpu_buffer
;
3982 atomic_inc(&buffer
->resize_disabled
);
3983 atomic_inc(&cpu_buffer
->record_disabled
);
3987 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare
);
3990 * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls
3992 * All previously invoked ring_buffer_read_prepare calls to prepare
3993 * iterators will be synchronized. Afterwards, read_buffer_read_start
3994 * calls on those iterators are allowed.
3997 ring_buffer_read_prepare_sync(void)
3999 synchronize_sched();
4001 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync
);
4004 * ring_buffer_read_start - start a non consuming read of the buffer
4005 * @iter: The iterator returned by ring_buffer_read_prepare
4007 * This finalizes the startup of an iteration through the buffer.
4008 * The iterator comes from a call to ring_buffer_read_prepare and
4009 * an intervening ring_buffer_read_prepare_sync must have been
4012 * Must be paired with ring_buffer_finish.
4015 ring_buffer_read_start(struct ring_buffer_iter
*iter
)
4017 struct ring_buffer_per_cpu
*cpu_buffer
;
4018 unsigned long flags
;
4023 cpu_buffer
= iter
->cpu_buffer
;
4025 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
4026 arch_spin_lock(&cpu_buffer
->lock
);
4027 rb_iter_reset(iter
);
4028 arch_spin_unlock(&cpu_buffer
->lock
);
4029 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
4031 EXPORT_SYMBOL_GPL(ring_buffer_read_start
);
4034 * ring_buffer_finish - finish reading the iterator of the buffer
4035 * @iter: The iterator retrieved by ring_buffer_start
4037 * This re-enables the recording to the buffer, and frees the
4041 ring_buffer_read_finish(struct ring_buffer_iter
*iter
)
4043 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
4044 unsigned long flags
;
4047 * Ring buffer is disabled from recording, here's a good place
4048 * to check the integrity of the ring buffer.
4049 * Must prevent readers from trying to read, as the check
4050 * clears the HEAD page and readers require it.
4052 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
4053 rb_check_pages(cpu_buffer
);
4054 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
4056 atomic_dec(&cpu_buffer
->record_disabled
);
4057 atomic_dec(&cpu_buffer
->buffer
->resize_disabled
);
4060 EXPORT_SYMBOL_GPL(ring_buffer_read_finish
);
4063 * ring_buffer_read - read the next item in the ring buffer by the iterator
4064 * @iter: The ring buffer iterator
4065 * @ts: The time stamp of the event read.
4067 * This reads the next event in the ring buffer and increments the iterator.
4069 struct ring_buffer_event
*
4070 ring_buffer_read(struct ring_buffer_iter
*iter
, u64
*ts
)
4072 struct ring_buffer_event
*event
;
4073 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
4074 unsigned long flags
;
4076 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
4078 event
= rb_iter_peek(iter
, ts
);
4082 if (event
->type_len
== RINGBUF_TYPE_PADDING
)
4085 rb_advance_iter(iter
);
4087 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
4091 EXPORT_SYMBOL_GPL(ring_buffer_read
);
4094 * ring_buffer_size - return the size of the ring buffer (in bytes)
4095 * @buffer: The ring buffer.
4097 unsigned long ring_buffer_size(struct ring_buffer
*buffer
, int cpu
)
4100 * Earlier, this method returned
4101 * BUF_PAGE_SIZE * buffer->nr_pages
4102 * Since the nr_pages field is now removed, we have converted this to
4103 * return the per cpu buffer value.
4105 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
4108 return BUF_PAGE_SIZE
* buffer
->buffers
[cpu
]->nr_pages
;
4110 EXPORT_SYMBOL_GPL(ring_buffer_size
);
4113 rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
)
4115 rb_head_page_deactivate(cpu_buffer
);
4117 cpu_buffer
->head_page
4118 = list_entry(cpu_buffer
->pages
, struct buffer_page
, list
);
4119 local_set(&cpu_buffer
->head_page
->write
, 0);
4120 local_set(&cpu_buffer
->head_page
->entries
, 0);
4121 local_set(&cpu_buffer
->head_page
->page
->commit
, 0);
4123 cpu_buffer
->head_page
->read
= 0;
4125 cpu_buffer
->tail_page
= cpu_buffer
->head_page
;
4126 cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
4128 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
4129 INIT_LIST_HEAD(&cpu_buffer
->new_pages
);
4130 local_set(&cpu_buffer
->reader_page
->write
, 0);
4131 local_set(&cpu_buffer
->reader_page
->entries
, 0);
4132 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
4133 cpu_buffer
->reader_page
->read
= 0;
4135 local_set(&cpu_buffer
->entries_bytes
, 0);
4136 local_set(&cpu_buffer
->overrun
, 0);
4137 local_set(&cpu_buffer
->commit_overrun
, 0);
4138 local_set(&cpu_buffer
->dropped_events
, 0);
4139 local_set(&cpu_buffer
->entries
, 0);
4140 local_set(&cpu_buffer
->committing
, 0);
4141 local_set(&cpu_buffer
->commits
, 0);
4142 cpu_buffer
->read
= 0;
4143 cpu_buffer
->read_bytes
= 0;
4145 cpu_buffer
->write_stamp
= 0;
4146 cpu_buffer
->read_stamp
= 0;
4148 cpu_buffer
->lost_events
= 0;
4149 cpu_buffer
->last_overrun
= 0;
4151 rb_head_page_activate(cpu_buffer
);
4155 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
4156 * @buffer: The ring buffer to reset a per cpu buffer of
4157 * @cpu: The CPU buffer to be reset
4159 void ring_buffer_reset_cpu(struct ring_buffer
*buffer
, int cpu
)
4161 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
4162 unsigned long flags
;
4164 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
4167 atomic_inc(&buffer
->resize_disabled
);
4168 atomic_inc(&cpu_buffer
->record_disabled
);
4170 /* Make sure all commits have finished */
4171 synchronize_sched();
4173 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
4175 if (RB_WARN_ON(cpu_buffer
, local_read(&cpu_buffer
->committing
)))
4178 arch_spin_lock(&cpu_buffer
->lock
);
4180 rb_reset_cpu(cpu_buffer
);
4182 arch_spin_unlock(&cpu_buffer
->lock
);
4185 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
4187 atomic_dec(&cpu_buffer
->record_disabled
);
4188 atomic_dec(&buffer
->resize_disabled
);
4190 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu
);
4193 * ring_buffer_reset - reset a ring buffer
4194 * @buffer: The ring buffer to reset all cpu buffers
4196 void ring_buffer_reset(struct ring_buffer
*buffer
)
4200 for_each_buffer_cpu(buffer
, cpu
)
4201 ring_buffer_reset_cpu(buffer
, cpu
);
4203 EXPORT_SYMBOL_GPL(ring_buffer_reset
);
4206 * rind_buffer_empty - is the ring buffer empty?
4207 * @buffer: The ring buffer to test
4209 int ring_buffer_empty(struct ring_buffer
*buffer
)
4211 struct ring_buffer_per_cpu
*cpu_buffer
;
4212 unsigned long flags
;
4217 dolock
= rb_ok_to_lock();
4219 /* yes this is racy, but if you don't like the race, lock the buffer */
4220 for_each_buffer_cpu(buffer
, cpu
) {
4221 cpu_buffer
= buffer
->buffers
[cpu
];
4222 local_irq_save(flags
);
4224 raw_spin_lock(&cpu_buffer
->reader_lock
);
4225 ret
= rb_per_cpu_empty(cpu_buffer
);
4227 raw_spin_unlock(&cpu_buffer
->reader_lock
);
4228 local_irq_restore(flags
);
4236 EXPORT_SYMBOL_GPL(ring_buffer_empty
);
4239 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
4240 * @buffer: The ring buffer
4241 * @cpu: The CPU buffer to test
4243 int ring_buffer_empty_cpu(struct ring_buffer
*buffer
, int cpu
)
4245 struct ring_buffer_per_cpu
*cpu_buffer
;
4246 unsigned long flags
;
4250 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
4253 dolock
= rb_ok_to_lock();
4255 cpu_buffer
= buffer
->buffers
[cpu
];
4256 local_irq_save(flags
);
4258 raw_spin_lock(&cpu_buffer
->reader_lock
);
4259 ret
= rb_per_cpu_empty(cpu_buffer
);
4261 raw_spin_unlock(&cpu_buffer
->reader_lock
);
4262 local_irq_restore(flags
);
4266 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu
);
4268 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4270 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
4271 * @buffer_a: One buffer to swap with
4272 * @buffer_b: The other buffer to swap with
4274 * This function is useful for tracers that want to take a "snapshot"
4275 * of a CPU buffer and has another back up buffer lying around.
4276 * it is expected that the tracer handles the cpu buffer not being
4277 * used at the moment.
4279 int ring_buffer_swap_cpu(struct ring_buffer
*buffer_a
,
4280 struct ring_buffer
*buffer_b
, int cpu
)
4282 struct ring_buffer_per_cpu
*cpu_buffer_a
;
4283 struct ring_buffer_per_cpu
*cpu_buffer_b
;
4286 if (!cpumask_test_cpu(cpu
, buffer_a
->cpumask
) ||
4287 !cpumask_test_cpu(cpu
, buffer_b
->cpumask
))
4290 cpu_buffer_a
= buffer_a
->buffers
[cpu
];
4291 cpu_buffer_b
= buffer_b
->buffers
[cpu
];
4293 /* At least make sure the two buffers are somewhat the same */
4294 if (cpu_buffer_a
->nr_pages
!= cpu_buffer_b
->nr_pages
)
4299 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
4302 if (atomic_read(&buffer_a
->record_disabled
))
4305 if (atomic_read(&buffer_b
->record_disabled
))
4308 if (atomic_read(&cpu_buffer_a
->record_disabled
))
4311 if (atomic_read(&cpu_buffer_b
->record_disabled
))
4315 * We can't do a synchronize_sched here because this
4316 * function can be called in atomic context.
4317 * Normally this will be called from the same CPU as cpu.
4318 * If not it's up to the caller to protect this.
4320 atomic_inc(&cpu_buffer_a
->record_disabled
);
4321 atomic_inc(&cpu_buffer_b
->record_disabled
);
4324 if (local_read(&cpu_buffer_a
->committing
))
4326 if (local_read(&cpu_buffer_b
->committing
))
4329 buffer_a
->buffers
[cpu
] = cpu_buffer_b
;
4330 buffer_b
->buffers
[cpu
] = cpu_buffer_a
;
4332 cpu_buffer_b
->buffer
= buffer_a
;
4333 cpu_buffer_a
->buffer
= buffer_b
;
4338 atomic_dec(&cpu_buffer_a
->record_disabled
);
4339 atomic_dec(&cpu_buffer_b
->record_disabled
);
4343 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu
);
4344 #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */
4347 * ring_buffer_alloc_read_page - allocate a page to read from buffer
4348 * @buffer: the buffer to allocate for.
4350 * This function is used in conjunction with ring_buffer_read_page.
4351 * When reading a full page from the ring buffer, these functions
4352 * can be used to speed up the process. The calling function should
4353 * allocate a few pages first with this function. Then when it
4354 * needs to get pages from the ring buffer, it passes the result
4355 * of this function into ring_buffer_read_page, which will swap
4356 * the page that was allocated, with the read page of the buffer.
4359 * The page allocated, or NULL on error.
4361 void *ring_buffer_alloc_read_page(struct ring_buffer
*buffer
, int cpu
)
4363 struct buffer_data_page
*bpage
;
4366 page
= alloc_pages_node(cpu_to_node(cpu
),
4367 GFP_KERNEL
| __GFP_NORETRY
, 0);
4371 bpage
= page_address(page
);
4373 rb_init_page(bpage
);
4377 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page
);
4380 * ring_buffer_free_read_page - free an allocated read page
4381 * @buffer: the buffer the page was allocate for
4382 * @data: the page to free
4384 * Free a page allocated from ring_buffer_alloc_read_page.
4386 void ring_buffer_free_read_page(struct ring_buffer
*buffer
, void *data
)
4388 free_page((unsigned long)data
);
4390 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page
);
4393 * ring_buffer_read_page - extract a page from the ring buffer
4394 * @buffer: buffer to extract from
4395 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
4396 * @len: amount to extract
4397 * @cpu: the cpu of the buffer to extract
4398 * @full: should the extraction only happen when the page is full.
4400 * This function will pull out a page from the ring buffer and consume it.
4401 * @data_page must be the address of the variable that was returned
4402 * from ring_buffer_alloc_read_page. This is because the page might be used
4403 * to swap with a page in the ring buffer.
4406 * rpage = ring_buffer_alloc_read_page(buffer);
4409 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
4411 * process_page(rpage, ret);
4413 * When @full is set, the function will not return true unless
4414 * the writer is off the reader page.
4416 * Note: it is up to the calling functions to handle sleeps and wakeups.
4417 * The ring buffer can be used anywhere in the kernel and can not
4418 * blindly call wake_up. The layer that uses the ring buffer must be
4419 * responsible for that.
4422 * >=0 if data has been transferred, returns the offset of consumed data.
4423 * <0 if no data has been transferred.
4425 int ring_buffer_read_page(struct ring_buffer
*buffer
,
4426 void **data_page
, size_t len
, int cpu
, int full
)
4428 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
4429 struct ring_buffer_event
*event
;
4430 struct buffer_data_page
*bpage
;
4431 struct buffer_page
*reader
;
4432 unsigned long missed_events
;
4433 unsigned long flags
;
4434 unsigned int commit
;
4439 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
4443 * If len is not big enough to hold the page header, then
4444 * we can not copy anything.
4446 if (len
<= BUF_PAGE_HDR_SIZE
)
4449 len
-= BUF_PAGE_HDR_SIZE
;
4458 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
4460 reader
= rb_get_reader_page(cpu_buffer
);
4464 event
= rb_reader_event(cpu_buffer
);
4466 read
= reader
->read
;
4467 commit
= rb_page_commit(reader
);
4469 /* Check if any events were dropped */
4470 missed_events
= cpu_buffer
->lost_events
;
4473 * If this page has been partially read or
4474 * if len is not big enough to read the rest of the page or
4475 * a writer is still on the page, then
4476 * we must copy the data from the page to the buffer.
4477 * Otherwise, we can simply swap the page with the one passed in.
4479 if (read
|| (len
< (commit
- read
)) ||
4480 cpu_buffer
->reader_page
== cpu_buffer
->commit_page
) {
4481 struct buffer_data_page
*rpage
= cpu_buffer
->reader_page
->page
;
4482 unsigned int rpos
= read
;
4483 unsigned int pos
= 0;
4489 if (len
> (commit
- read
))
4490 len
= (commit
- read
);
4492 /* Always keep the time extend and data together */
4493 size
= rb_event_ts_length(event
);
4498 /* save the current timestamp, since the user will need it */
4499 save_timestamp
= cpu_buffer
->read_stamp
;
4501 /* Need to copy one event at a time */
4503 /* We need the size of one event, because
4504 * rb_advance_reader only advances by one event,
4505 * whereas rb_event_ts_length may include the size of
4506 * one or two events.
4507 * We have already ensured there's enough space if this
4508 * is a time extend. */
4509 size
= rb_event_length(event
);
4510 memcpy(bpage
->data
+ pos
, rpage
->data
+ rpos
, size
);
4514 rb_advance_reader(cpu_buffer
);
4515 rpos
= reader
->read
;
4521 event
= rb_reader_event(cpu_buffer
);
4522 /* Always keep the time extend and data together */
4523 size
= rb_event_ts_length(event
);
4524 } while (len
>= size
);
4527 local_set(&bpage
->commit
, pos
);
4528 bpage
->time_stamp
= save_timestamp
;
4530 /* we copied everything to the beginning */
4533 /* update the entry counter */
4534 cpu_buffer
->read
+= rb_page_entries(reader
);
4535 cpu_buffer
->read_bytes
+= BUF_PAGE_SIZE
;
4537 /* swap the pages */
4538 rb_init_page(bpage
);
4539 bpage
= reader
->page
;
4540 reader
->page
= *data_page
;
4541 local_set(&reader
->write
, 0);
4542 local_set(&reader
->entries
, 0);
4547 * Use the real_end for the data size,
4548 * This gives us a chance to store the lost events
4551 if (reader
->real_end
)
4552 local_set(&bpage
->commit
, reader
->real_end
);
4556 cpu_buffer
->lost_events
= 0;
4558 commit
= local_read(&bpage
->commit
);
4560 * Set a flag in the commit field if we lost events
4562 if (missed_events
) {
4563 /* If there is room at the end of the page to save the
4564 * missed events, then record it there.
4566 if (BUF_PAGE_SIZE
- commit
>= sizeof(missed_events
)) {
4567 memcpy(&bpage
->data
[commit
], &missed_events
,
4568 sizeof(missed_events
));
4569 local_add(RB_MISSED_STORED
, &bpage
->commit
);
4570 commit
+= sizeof(missed_events
);
4572 local_add(RB_MISSED_EVENTS
, &bpage
->commit
);
4576 * This page may be off to user land. Zero it out here.
4578 if (commit
< BUF_PAGE_SIZE
)
4579 memset(&bpage
->data
[commit
], 0, BUF_PAGE_SIZE
- commit
);
4582 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
4587 EXPORT_SYMBOL_GPL(ring_buffer_read_page
);
4589 #ifdef CONFIG_HOTPLUG_CPU
4590 static int rb_cpu_notify(struct notifier_block
*self
,
4591 unsigned long action
, void *hcpu
)
4593 struct ring_buffer
*buffer
=
4594 container_of(self
, struct ring_buffer
, cpu_notify
);
4595 long cpu
= (long)hcpu
;
4596 int cpu_i
, nr_pages_same
;
4597 unsigned int nr_pages
;
4600 case CPU_UP_PREPARE
:
4601 case CPU_UP_PREPARE_FROZEN
:
4602 if (cpumask_test_cpu(cpu
, buffer
->cpumask
))
4607 /* check if all cpu sizes are same */
4608 for_each_buffer_cpu(buffer
, cpu_i
) {
4609 /* fill in the size from first enabled cpu */
4611 nr_pages
= buffer
->buffers
[cpu_i
]->nr_pages
;
4612 if (nr_pages
!= buffer
->buffers
[cpu_i
]->nr_pages
) {
4617 /* allocate minimum pages, user can later expand it */
4620 buffer
->buffers
[cpu
] =
4621 rb_allocate_cpu_buffer(buffer
, nr_pages
, cpu
);
4622 if (!buffer
->buffers
[cpu
]) {
4623 WARN(1, "failed to allocate ring buffer on CPU %ld\n",
4628 cpumask_set_cpu(cpu
, buffer
->cpumask
);
4630 case CPU_DOWN_PREPARE
:
4631 case CPU_DOWN_PREPARE_FROZEN
:
4634 * If we were to free the buffer, then the user would
4635 * lose any trace that was in the buffer.
4645 #ifdef CONFIG_RING_BUFFER_STARTUP_TEST
4647 * This is a basic integrity check of the ring buffer.
4648 * Late in the boot cycle this test will run when configured in.
4649 * It will kick off a thread per CPU that will go into a loop
4650 * writing to the per cpu ring buffer various sizes of data.
4651 * Some of the data will be large items, some small.
4653 * Another thread is created that goes into a spin, sending out
4654 * IPIs to the other CPUs to also write into the ring buffer.
4655 * this is to test the nesting ability of the buffer.
4657 * Basic stats are recorded and reported. If something in the
4658 * ring buffer should happen that's not expected, a big warning
4659 * is displayed and all ring buffers are disabled.
4661 static struct task_struct
*rb_threads
[NR_CPUS
] __initdata
;
4663 struct rb_test_data
{
4664 struct ring_buffer
*buffer
;
4665 unsigned long events
;
4666 unsigned long bytes_written
;
4667 unsigned long bytes_alloc
;
4668 unsigned long bytes_dropped
;
4669 unsigned long events_nested
;
4670 unsigned long bytes_written_nested
;
4671 unsigned long bytes_alloc_nested
;
4672 unsigned long bytes_dropped_nested
;
4673 int min_size_nested
;
4674 int max_size_nested
;
4681 static struct rb_test_data rb_data
[NR_CPUS
] __initdata
;
4684 #define RB_TEST_BUFFER_SIZE 1048576
4686 static char rb_string
[] __initdata
=
4687 "abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\"
4688 "?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890"
4689 "!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv";
4691 static bool rb_test_started __initdata
;
4698 static __init
int rb_write_something(struct rb_test_data
*data
, bool nested
)
4700 struct ring_buffer_event
*event
;
4701 struct rb_item
*item
;
4708 /* Have nested writes different that what is written */
4709 cnt
= data
->cnt
+ (nested
? 27 : 0);
4711 /* Multiply cnt by ~e, to make some unique increment */
4712 size
= (data
->cnt
* 68 / 25) % (sizeof(rb_string
) - 1);
4714 len
= size
+ sizeof(struct rb_item
);
4716 started
= rb_test_started
;
4717 /* read rb_test_started before checking buffer enabled */
4720 event
= ring_buffer_lock_reserve(data
->buffer
, len
);
4722 /* Ignore dropped events before test starts. */
4725 data
->bytes_dropped
+= len
;
4727 data
->bytes_dropped_nested
+= len
;
4732 event_len
= ring_buffer_event_length(event
);
4734 if (RB_WARN_ON(data
->buffer
, event_len
< len
))
4737 item
= ring_buffer_event_data(event
);
4739 memcpy(item
->str
, rb_string
, size
);
4742 data
->bytes_alloc_nested
+= event_len
;
4743 data
->bytes_written_nested
+= len
;
4744 data
->events_nested
++;
4745 if (!data
->min_size_nested
|| len
< data
->min_size_nested
)
4746 data
->min_size_nested
= len
;
4747 if (len
> data
->max_size_nested
)
4748 data
->max_size_nested
= len
;
4750 data
->bytes_alloc
+= event_len
;
4751 data
->bytes_written
+= len
;
4753 if (!data
->min_size
|| len
< data
->min_size
)
4754 data
->max_size
= len
;
4755 if (len
> data
->max_size
)
4756 data
->max_size
= len
;
4760 ring_buffer_unlock_commit(data
->buffer
, event
);
4765 static __init
int rb_test(void *arg
)
4767 struct rb_test_data
*data
= arg
;
4769 while (!kthread_should_stop()) {
4770 rb_write_something(data
, false);
4773 set_current_state(TASK_INTERRUPTIBLE
);
4774 /* Now sleep between a min of 100-300us and a max of 1ms */
4775 usleep_range(((data
->cnt
% 3) + 1) * 100, 1000);
4781 static __init
void rb_ipi(void *ignore
)
4783 struct rb_test_data
*data
;
4784 int cpu
= smp_processor_id();
4786 data
= &rb_data
[cpu
];
4787 rb_write_something(data
, true);
4790 static __init
int rb_hammer_test(void *arg
)
4792 while (!kthread_should_stop()) {
4794 /* Send an IPI to all cpus to write data! */
4795 smp_call_function(rb_ipi
, NULL
, 1);
4796 /* No sleep, but for non preempt, let others run */
4803 static __init
int test_ringbuffer(void)
4805 struct task_struct
*rb_hammer
;
4806 struct ring_buffer
*buffer
;
4810 pr_info("Running ring buffer tests...\n");
4812 buffer
= ring_buffer_alloc(RB_TEST_BUFFER_SIZE
, RB_FL_OVERWRITE
);
4813 if (WARN_ON(!buffer
))
4816 /* Disable buffer so that threads can't write to it yet */
4817 ring_buffer_record_off(buffer
);
4819 for_each_online_cpu(cpu
) {
4820 rb_data
[cpu
].buffer
= buffer
;
4821 rb_data
[cpu
].cpu
= cpu
;
4822 rb_data
[cpu
].cnt
= cpu
;
4823 rb_threads
[cpu
] = kthread_create(rb_test
, &rb_data
[cpu
],
4824 "rbtester/%d", cpu
);
4825 if (WARN_ON(!rb_threads
[cpu
])) {
4826 pr_cont("FAILED\n");
4831 kthread_bind(rb_threads
[cpu
], cpu
);
4832 wake_up_process(rb_threads
[cpu
]);
4835 /* Now create the rb hammer! */
4836 rb_hammer
= kthread_run(rb_hammer_test
, NULL
, "rbhammer");
4837 if (WARN_ON(!rb_hammer
)) {
4838 pr_cont("FAILED\n");
4843 ring_buffer_record_on(buffer
);
4845 * Show buffer is enabled before setting rb_test_started.
4846 * Yes there's a small race window where events could be
4847 * dropped and the thread wont catch it. But when a ring
4848 * buffer gets enabled, there will always be some kind of
4849 * delay before other CPUs see it. Thus, we don't care about
4850 * those dropped events. We care about events dropped after
4851 * the threads see that the buffer is active.
4854 rb_test_started
= true;
4856 set_current_state(TASK_INTERRUPTIBLE
);
4857 /* Just run for 10 seconds */;
4858 schedule_timeout(10 * HZ
);
4860 kthread_stop(rb_hammer
);
4863 for_each_online_cpu(cpu
) {
4864 if (!rb_threads
[cpu
])
4866 kthread_stop(rb_threads
[cpu
]);
4869 ring_buffer_free(buffer
);
4874 pr_info("finished\n");
4875 for_each_online_cpu(cpu
) {
4876 struct ring_buffer_event
*event
;
4877 struct rb_test_data
*data
= &rb_data
[cpu
];
4878 struct rb_item
*item
;
4879 unsigned long total_events
;
4880 unsigned long total_dropped
;
4881 unsigned long total_written
;
4882 unsigned long total_alloc
;
4883 unsigned long total_read
= 0;
4884 unsigned long total_size
= 0;
4885 unsigned long total_len
= 0;
4886 unsigned long total_lost
= 0;
4889 int small_event_size
;
4893 total_events
= data
->events
+ data
->events_nested
;
4894 total_written
= data
->bytes_written
+ data
->bytes_written_nested
;
4895 total_alloc
= data
->bytes_alloc
+ data
->bytes_alloc_nested
;
4896 total_dropped
= data
->bytes_dropped
+ data
->bytes_dropped_nested
;
4898 big_event_size
= data
->max_size
+ data
->max_size_nested
;
4899 small_event_size
= data
->min_size
+ data
->min_size_nested
;
4901 pr_info("CPU %d:\n", cpu
);
4902 pr_info(" events: %ld\n", total_events
);
4903 pr_info(" dropped bytes: %ld\n", total_dropped
);
4904 pr_info(" alloced bytes: %ld\n", total_alloc
);
4905 pr_info(" written bytes: %ld\n", total_written
);
4906 pr_info(" biggest event: %d\n", big_event_size
);
4907 pr_info(" smallest event: %d\n", small_event_size
);
4909 if (RB_WARN_ON(buffer
, total_dropped
))
4914 while ((event
= ring_buffer_consume(buffer
, cpu
, NULL
, &lost
))) {
4916 item
= ring_buffer_event_data(event
);
4917 total_len
+= ring_buffer_event_length(event
);
4918 total_size
+= item
->size
+ sizeof(struct rb_item
);
4919 if (memcmp(&item
->str
[0], rb_string
, item
->size
) != 0) {
4920 pr_info("FAILED!\n");
4921 pr_info("buffer had: %.*s\n", item
->size
, item
->str
);
4922 pr_info("expected: %.*s\n", item
->size
, rb_string
);
4923 RB_WARN_ON(buffer
, 1);
4934 pr_info(" read events: %ld\n", total_read
);
4935 pr_info(" lost events: %ld\n", total_lost
);
4936 pr_info(" total events: %ld\n", total_lost
+ total_read
);
4937 pr_info(" recorded len bytes: %ld\n", total_len
);
4938 pr_info(" recorded size bytes: %ld\n", total_size
);
4940 pr_info(" With dropped events, record len and size may not match\n"
4941 " alloced and written from above\n");
4943 if (RB_WARN_ON(buffer
, total_len
!= total_alloc
||
4944 total_size
!= total_written
))
4947 if (RB_WARN_ON(buffer
, total_lost
+ total_read
!= total_events
))
4953 pr_info("Ring buffer PASSED!\n");
4955 ring_buffer_free(buffer
);
4959 late_initcall(test_ringbuffer
);
4960 #endif /* CONFIG_RING_BUFFER_STARTUP_TEST */