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 #ifdef CONFIG_MTK_EXTMEM
31 extern void* extmem_malloc_page_align(size_t bytes
);
32 extern void extmem_free(void* mem
);
35 static void update_pages_handler(struct work_struct
*work
);
38 * The ring buffer header is special. We must manually up keep it.
40 int ring_buffer_print_entry_header(struct trace_seq
*s
)
44 ret
= trace_seq_printf(s
, "# compressed entry header\n");
45 ret
= trace_seq_printf(s
, "\ttype_len : 5 bits\n");
46 ret
= trace_seq_printf(s
, "\ttime_delta : 27 bits\n");
47 ret
= trace_seq_printf(s
, "\tarray : 32 bits\n");
48 ret
= trace_seq_printf(s
, "\n");
49 ret
= trace_seq_printf(s
, "\tpadding : type == %d\n",
50 RINGBUF_TYPE_PADDING
);
51 ret
= trace_seq_printf(s
, "\ttime_extend : type == %d\n",
52 RINGBUF_TYPE_TIME_EXTEND
);
53 ret
= trace_seq_printf(s
, "\tdata max type_len == %d\n",
54 RINGBUF_TYPE_DATA_TYPE_LEN_MAX
);
60 * The ring buffer is made up of a list of pages. A separate list of pages is
61 * allocated for each CPU. A writer may only write to a buffer that is
62 * associated with the CPU it is currently executing on. A reader may read
63 * from any per cpu buffer.
65 * The reader is special. For each per cpu buffer, the reader has its own
66 * reader page. When a reader has read the entire reader page, this reader
67 * page is swapped with another page in the ring buffer.
69 * Now, as long as the writer is off the reader page, the reader can do what
70 * ever it wants with that page. The writer will never write to that page
71 * again (as long as it is out of the ring buffer).
73 * Here's some silly ASCII art.
76 * |reader| RING BUFFER
78 * +------+ +---+ +---+ +---+
87 * |reader| RING BUFFER
88 * |page |------------------v
89 * +------+ +---+ +---+ +---+
98 * |reader| RING BUFFER
99 * |page |------------------v
100 * +------+ +---+ +---+ +---+
102 * | +---+ +---+ +---+
105 * +------------------------------+
109 * |buffer| RING BUFFER
110 * |page |------------------v
111 * +------+ +---+ +---+ +---+
113 * | New +---+ +---+ +---+
116 * +------------------------------+
119 * After we make this swap, the reader can hand this page off to the splice
120 * code and be done with it. It can even allocate a new page if it needs to
121 * and swap that into the ring buffer.
123 * We will be using cmpxchg soon to make all this lockless.
128 * A fast way to enable or disable all ring buffers is to
129 * call tracing_on or tracing_off. Turning off the ring buffers
130 * prevents all ring buffers from being recorded to.
131 * Turning this switch on, makes it OK to write to the
132 * ring buffer, if the ring buffer is enabled itself.
134 * There's three layers that must be on in order to write
135 * to the ring buffer.
137 * 1) This global flag must be set.
138 * 2) The ring buffer must be enabled for recording.
139 * 3) The per cpu buffer must be enabled for recording.
141 * In case of an anomaly, this global flag has a bit set that
142 * will permantly disable all ring buffers.
146 * Global flag to disable all recording to ring buffers
147 * This has two bits: ON, DISABLED
151 * 0 0 : ring buffers are off
152 * 1 0 : ring buffers are on
153 * X 1 : ring buffers are permanently disabled
157 RB_BUFFERS_ON_BIT
= 0,
158 RB_BUFFERS_DISABLED_BIT
= 1,
162 RB_BUFFERS_ON
= 1 << RB_BUFFERS_ON_BIT
,
163 RB_BUFFERS_DISABLED
= 1 << RB_BUFFERS_DISABLED_BIT
,
166 static unsigned long ring_buffer_flags __read_mostly
= RB_BUFFERS_ON
;
168 /* Used for individual buffers (after the counter) */
169 #define RB_BUFFER_OFF (1 << 20)
171 #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
174 * tracing_off_permanent - permanently disable ring buffers
176 * This function, once called, will disable all ring buffers
179 void tracing_off_permanent(void)
181 set_bit(RB_BUFFERS_DISABLED_BIT
, &ring_buffer_flags
);
184 #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
185 #define RB_ALIGNMENT 4U
186 #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
187 #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */
189 #ifndef CONFIG_HAVE_64BIT_ALIGNED_ACCESS
190 # define RB_FORCE_8BYTE_ALIGNMENT 0
191 # define RB_ARCH_ALIGNMENT RB_ALIGNMENT
193 # define RB_FORCE_8BYTE_ALIGNMENT 1
194 # define RB_ARCH_ALIGNMENT 8U
197 #define RB_ALIGN_DATA __aligned(RB_ARCH_ALIGNMENT)
199 /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
200 #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
203 RB_LEN_TIME_EXTEND
= 8,
204 RB_LEN_TIME_STAMP
= 16,
207 #define skip_time_extend(event) \
208 ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND))
210 static inline int rb_null_event(struct ring_buffer_event
*event
)
212 return event
->type_len
== RINGBUF_TYPE_PADDING
&& !event
->time_delta
;
215 static void rb_event_set_padding(struct ring_buffer_event
*event
)
217 /* padding has a NULL time_delta */
218 event
->type_len
= RINGBUF_TYPE_PADDING
;
219 event
->time_delta
= 0;
223 rb_event_data_length(struct ring_buffer_event
*event
)
228 length
= event
->type_len
* RB_ALIGNMENT
;
230 length
= event
->array
[0];
231 return length
+ RB_EVNT_HDR_SIZE
;
235 * Return the length of the given event. Will return
236 * the length of the time extend if the event is a
239 static inline unsigned
240 rb_event_length(struct ring_buffer_event
*event
)
242 switch (event
->type_len
) {
243 case RINGBUF_TYPE_PADDING
:
244 if (rb_null_event(event
))
247 return event
->array
[0] + RB_EVNT_HDR_SIZE
;
249 case RINGBUF_TYPE_TIME_EXTEND
:
250 return RB_LEN_TIME_EXTEND
;
252 case RINGBUF_TYPE_TIME_STAMP
:
253 return RB_LEN_TIME_STAMP
;
255 case RINGBUF_TYPE_DATA
:
256 return rb_event_data_length(event
);
265 * Return total length of time extend and data,
266 * or just the event length for all other events.
268 static inline unsigned
269 rb_event_ts_length(struct ring_buffer_event
*event
)
273 if (event
->type_len
== RINGBUF_TYPE_TIME_EXTEND
) {
274 /* time extends include the data event after it */
275 len
= RB_LEN_TIME_EXTEND
;
276 event
= skip_time_extend(event
);
278 return len
+ rb_event_length(event
);
282 * ring_buffer_event_length - return the length of the event
283 * @event: the event to get the length of
285 * Returns the size of the data load of a data event.
286 * If the event is something other than a data event, it
287 * returns the size of the event itself. With the exception
288 * of a TIME EXTEND, where it still returns the size of the
289 * data load of the data event after it.
291 unsigned ring_buffer_event_length(struct ring_buffer_event
*event
)
295 if (event
->type_len
== RINGBUF_TYPE_TIME_EXTEND
)
296 event
= skip_time_extend(event
);
298 length
= rb_event_length(event
);
299 if (event
->type_len
> RINGBUF_TYPE_DATA_TYPE_LEN_MAX
)
301 length
-= RB_EVNT_HDR_SIZE
;
302 if (length
> RB_MAX_SMALL_DATA
+ sizeof(event
->array
[0]))
303 length
-= sizeof(event
->array
[0]);
306 EXPORT_SYMBOL_GPL(ring_buffer_event_length
);
308 /* inline for ring buffer fast paths */
310 rb_event_data(struct ring_buffer_event
*event
)
312 if (event
->type_len
== RINGBUF_TYPE_TIME_EXTEND
)
313 event
= skip_time_extend(event
);
314 BUG_ON(event
->type_len
> RINGBUF_TYPE_DATA_TYPE_LEN_MAX
);
315 /* If length is in len field, then array[0] has the data */
317 return (void *)&event
->array
[0];
318 /* Otherwise length is in array[0] and array[1] has the data */
319 return (void *)&event
->array
[1];
323 * ring_buffer_event_data - return the data of the event
324 * @event: the event to get the data from
326 void *ring_buffer_event_data(struct ring_buffer_event
*event
)
328 return rb_event_data(event
);
330 EXPORT_SYMBOL_GPL(ring_buffer_event_data
);
332 #define for_each_buffer_cpu(buffer, cpu) \
333 for_each_cpu(cpu, buffer->cpumask)
336 #define TS_MASK ((1ULL << TS_SHIFT) - 1)
337 #define TS_DELTA_TEST (~TS_MASK)
339 /* Flag when events were overwritten */
340 #define RB_MISSED_EVENTS (1 << 31)
341 /* Missed count stored at end */
342 #define RB_MISSED_STORED (1 << 30)
344 struct buffer_data_page
{
345 u64 time_stamp
; /* page time stamp */
346 local_t commit
; /* write committed index */
347 unsigned char data
[] RB_ALIGN_DATA
; /* data of buffer page */
351 * Note, the buffer_page list must be first. The buffer pages
352 * are allocated in cache lines, which means that each buffer
353 * page will be at the beginning of a cache line, and thus
354 * the least significant bits will be zero. We use this to
355 * add flags in the list struct pointers, to make the ring buffer
359 struct list_head list
; /* list of buffer pages */
360 local_t write
; /* index for next write */
361 unsigned read
; /* index for next read */
362 local_t entries
; /* entries on this page */
363 unsigned long real_end
; /* real end of data */
364 struct buffer_data_page
*page
; /* Actual data page */
368 * The buffer page counters, write and entries, must be reset
369 * atomically when crossing page boundaries. To synchronize this
370 * update, two counters are inserted into the number. One is
371 * the actual counter for the write position or count on the page.
373 * The other is a counter of updaters. Before an update happens
374 * the update partition of the counter is incremented. This will
375 * allow the updater to update the counter atomically.
377 * The counter is 20 bits, and the state data is 12.
379 #define RB_WRITE_MASK 0xfffff
380 #define RB_WRITE_INTCNT (1 << 20)
382 static void rb_init_page(struct buffer_data_page
*bpage
)
384 local_set(&bpage
->commit
, 0);
388 * ring_buffer_page_len - the size of data on the page.
389 * @page: The page to read
391 * Returns the amount of data on the page, including buffer page header.
393 size_t ring_buffer_page_len(void *page
)
395 return local_read(&((struct buffer_data_page
*)page
)->commit
)
400 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
403 static void free_buffer_page(struct buffer_page
*bpage
)
405 #ifdef CONFIG_MTK_EXTMEM
406 extmem_free((void*) bpage
->page
);
408 free_page((unsigned long)bpage
->page
);
414 * We need to fit the time_stamp delta into 27 bits.
416 static inline int test_time_stamp(u64 delta
)
418 if (delta
& TS_DELTA_TEST
)
423 #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
425 /* Max payload is BUF_PAGE_SIZE - header (8bytes) */
426 #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
428 int ring_buffer_print_page_header(struct trace_seq
*s
)
430 struct buffer_data_page field
;
433 ret
= trace_seq_printf(s
, "\tfield: u64 timestamp;\t"
434 "offset:0;\tsize:%u;\tsigned:%u;\n",
435 (unsigned int)sizeof(field
.time_stamp
),
436 (unsigned int)is_signed_type(u64
));
438 ret
= trace_seq_printf(s
, "\tfield: local_t commit;\t"
439 "offset:%u;\tsize:%u;\tsigned:%u;\n",
440 (unsigned int)offsetof(typeof(field
), commit
),
441 (unsigned int)sizeof(field
.commit
),
442 (unsigned int)is_signed_type(long));
444 ret
= trace_seq_printf(s
, "\tfield: int overwrite;\t"
445 "offset:%u;\tsize:%u;\tsigned:%u;\n",
446 (unsigned int)offsetof(typeof(field
), commit
),
448 (unsigned int)is_signed_type(long));
450 ret
= trace_seq_printf(s
, "\tfield: char data;\t"
451 "offset:%u;\tsize:%u;\tsigned:%u;\n",
452 (unsigned int)offsetof(typeof(field
), data
),
453 (unsigned int)BUF_PAGE_SIZE
,
454 (unsigned int)is_signed_type(char));
460 struct irq_work work
;
461 wait_queue_head_t waiters
;
462 bool waiters_pending
;
466 * head_page == tail_page && head == tail then buffer is empty.
468 struct ring_buffer_per_cpu
{
470 atomic_t record_disabled
;
471 struct ring_buffer
*buffer
;
472 raw_spinlock_t reader_lock
; /* serialize readers */
473 arch_spinlock_t lock
;
474 struct lock_class_key lock_key
;
475 unsigned int nr_pages
;
476 struct list_head
*pages
;
477 struct buffer_page
*head_page
; /* read from head */
478 struct buffer_page
*tail_page
; /* write to tail */
479 struct buffer_page
*commit_page
; /* committed pages */
480 struct buffer_page
*reader_page
;
481 unsigned long lost_events
;
482 unsigned long last_overrun
;
483 local_t entries_bytes
;
486 local_t commit_overrun
;
487 local_t dropped_events
;
491 unsigned long read_bytes
;
494 /* ring buffer pages to update, > 0 to add, < 0 to remove */
495 int nr_pages_to_update
;
496 struct list_head new_pages
; /* new pages to add */
497 struct work_struct update_pages_work
;
498 struct completion update_done
;
500 struct rb_irq_work irq_work
;
506 atomic_t record_disabled
;
507 atomic_t resize_disabled
;
508 cpumask_var_t cpumask
;
510 struct lock_class_key
*reader_lock_key
;
514 struct ring_buffer_per_cpu
**buffers
;
516 #ifdef CONFIG_HOTPLUG_CPU
517 struct notifier_block cpu_notify
;
521 struct rb_irq_work irq_work
;
524 struct ring_buffer_iter
{
525 struct ring_buffer_per_cpu
*cpu_buffer
;
527 struct buffer_page
*head_page
;
528 struct buffer_page
*cache_reader_page
;
529 unsigned long cache_read
;
534 * rb_wake_up_waiters - wake up tasks waiting for ring buffer input
536 * Schedules a delayed work to wake up any task that is blocked on the
537 * ring buffer waiters queue.
539 static void rb_wake_up_waiters(struct irq_work
*work
)
541 struct rb_irq_work
*rbwork
= container_of(work
, struct rb_irq_work
, work
);
543 wake_up_all(&rbwork
->waiters
);
547 * ring_buffer_wait - wait for input to the ring buffer
548 * @buffer: buffer to wait on
549 * @cpu: the cpu buffer to wait on
551 * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
552 * as data is added to any of the @buffer's cpu buffers. Otherwise
553 * it will wait for data to be added to a specific cpu buffer.
555 int ring_buffer_wait(struct ring_buffer
*buffer
, int cpu
)
557 struct ring_buffer_per_cpu
*cpu_buffer
;
559 struct rb_irq_work
*work
;
562 * Depending on what the caller is waiting for, either any
563 * data in any cpu buffer, or a specific buffer, put the
564 * caller on the appropriate wait queue.
566 if (cpu
== RING_BUFFER_ALL_CPUS
)
567 work
= &buffer
->irq_work
;
569 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
571 cpu_buffer
= buffer
->buffers
[cpu
];
572 work
= &cpu_buffer
->irq_work
;
576 prepare_to_wait(&work
->waiters
, &wait
, TASK_INTERRUPTIBLE
);
579 * The events can happen in critical sections where
580 * checking a work queue can cause deadlocks.
581 * After adding a task to the queue, this flag is set
582 * only to notify events to try to wake up the queue
585 * We don't clear it even if the buffer is no longer
586 * empty. The flag only causes the next event to run
587 * irq_work to do the work queue wake up. The worse
588 * that can happen if we race with !trace_empty() is that
589 * an event will cause an irq_work to try to wake up
592 * There's no reason to protect this flag either, as
593 * the work queue and irq_work logic will do the necessary
594 * synchronization for the wake ups. The only thing
595 * that is necessary is that the wake up happens after
596 * a task has been queued. It's OK for spurious wake ups.
598 work
->waiters_pending
= true;
600 if ((cpu
== RING_BUFFER_ALL_CPUS
&& ring_buffer_empty(buffer
)) ||
601 (cpu
!= RING_BUFFER_ALL_CPUS
&& ring_buffer_empty_cpu(buffer
, cpu
)))
604 finish_wait(&work
->waiters
, &wait
);
609 * ring_buffer_poll_wait - poll on buffer input
610 * @buffer: buffer to wait on
611 * @cpu: the cpu buffer to wait on
612 * @filp: the file descriptor
613 * @poll_table: The poll descriptor
615 * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
616 * as data is added to any of the @buffer's cpu buffers. Otherwise
617 * it will wait for data to be added to a specific cpu buffer.
619 * Returns POLLIN | POLLRDNORM if data exists in the buffers,
622 int ring_buffer_poll_wait(struct ring_buffer
*buffer
, int cpu
,
623 struct file
*filp
, poll_table
*poll_table
)
625 struct ring_buffer_per_cpu
*cpu_buffer
;
626 struct rb_irq_work
*work
;
628 if (cpu
== RING_BUFFER_ALL_CPUS
)
629 work
= &buffer
->irq_work
;
631 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
634 cpu_buffer
= buffer
->buffers
[cpu
];
635 work
= &cpu_buffer
->irq_work
;
638 poll_wait(filp
, &work
->waiters
, poll_table
);
639 work
->waiters_pending
= true;
641 * There's a tight race between setting the waiters_pending and
642 * checking if the ring buffer is empty. Once the waiters_pending bit
643 * is set, the next event will wake the task up, but we can get stuck
644 * if there's only a single event in.
646 * FIXME: Ideally, we need a memory barrier on the writer side as well,
647 * but adding a memory barrier to all events will cause too much of a
648 * performance hit in the fast path. We only need a memory barrier when
649 * the buffer goes from empty to having content. But as this race is
650 * extremely small, and it's not a problem if another event comes in, we
655 if ((cpu
== RING_BUFFER_ALL_CPUS
&& !ring_buffer_empty(buffer
)) ||
656 (cpu
!= RING_BUFFER_ALL_CPUS
&& !ring_buffer_empty_cpu(buffer
, cpu
)))
657 return POLLIN
| POLLRDNORM
;
661 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
662 #define RB_WARN_ON(b, cond) \
664 int _____ret = unlikely(cond); \
666 if (__same_type(*(b), struct ring_buffer_per_cpu)) { \
667 struct ring_buffer_per_cpu *__b = \
669 atomic_inc(&__b->buffer->record_disabled); \
671 atomic_inc(&b->record_disabled); \
677 /* Up this if you want to test the TIME_EXTENTS and normalization */
678 #define DEBUG_SHIFT 0
680 static inline u64
rb_time_stamp(struct ring_buffer
*buffer
)
682 /* shift to debug/test normalization and TIME_EXTENTS */
683 return buffer
->clock() << DEBUG_SHIFT
;
686 u64
ring_buffer_time_stamp(struct ring_buffer
*buffer
, int cpu
)
690 preempt_disable_notrace();
691 time
= rb_time_stamp(buffer
);
692 preempt_enable_no_resched_notrace();
696 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp
);
698 void ring_buffer_normalize_time_stamp(struct ring_buffer
*buffer
,
701 /* Just stupid testing the normalize function and deltas */
704 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp
);
707 * Making the ring buffer lockless makes things tricky.
708 * Although writes only happen on the CPU that they are on,
709 * and they only need to worry about interrupts. Reads can
712 * The reader page is always off the ring buffer, but when the
713 * reader finishes with a page, it needs to swap its page with
714 * a new one from the buffer. The reader needs to take from
715 * the head (writes go to the tail). But if a writer is in overwrite
716 * mode and wraps, it must push the head page forward.
718 * Here lies the problem.
720 * The reader must be careful to replace only the head page, and
721 * not another one. As described at the top of the file in the
722 * ASCII art, the reader sets its old page to point to the next
723 * page after head. It then sets the page after head to point to
724 * the old reader page. But if the writer moves the head page
725 * during this operation, the reader could end up with the tail.
727 * We use cmpxchg to help prevent this race. We also do something
728 * special with the page before head. We set the LSB to 1.
730 * When the writer must push the page forward, it will clear the
731 * bit that points to the head page, move the head, and then set
732 * the bit that points to the new head page.
734 * We also don't want an interrupt coming in and moving the head
735 * page on another writer. Thus we use the second LSB to catch
738 * head->list->prev->next bit 1 bit 0
741 * Points to head page 0 1
744 * Note we can not trust the prev pointer of the head page, because:
746 * +----+ +-----+ +-----+
747 * | |------>| T |---X--->| N |
749 * +----+ +-----+ +-----+
752 * +----------| R |----------+ |
756 * Key: ---X--> HEAD flag set in pointer
761 * (see __rb_reserve_next() to see where this happens)
763 * What the above shows is that the reader just swapped out
764 * the reader page with a page in the buffer, but before it
765 * could make the new header point back to the new page added
766 * it was preempted by a writer. The writer moved forward onto
767 * the new page added by the reader and is about to move forward
770 * You can see, it is legitimate for the previous pointer of
771 * the head (or any page) not to point back to itself. But only
775 #define RB_PAGE_NORMAL 0UL
776 #define RB_PAGE_HEAD 1UL
777 #define RB_PAGE_UPDATE 2UL
780 #define RB_FLAG_MASK 3UL
782 /* PAGE_MOVED is not part of the mask */
783 #define RB_PAGE_MOVED 4UL
786 * rb_list_head - remove any bit
788 static struct list_head
*rb_list_head(struct list_head
*list
)
790 unsigned long val
= (unsigned long)list
;
792 return (struct list_head
*)(val
& ~RB_FLAG_MASK
);
796 * rb_is_head_page - test if the given page is the head page
798 * Because the reader may move the head_page pointer, we can
799 * not trust what the head page is (it may be pointing to
800 * the reader page). But if the next page is a header page,
801 * its flags will be non zero.
804 rb_is_head_page(struct ring_buffer_per_cpu
*cpu_buffer
,
805 struct buffer_page
*page
, struct list_head
*list
)
809 val
= (unsigned long)list
->next
;
811 if ((val
& ~RB_FLAG_MASK
) != (unsigned long)&page
->list
)
812 return RB_PAGE_MOVED
;
814 return val
& RB_FLAG_MASK
;
820 * The unique thing about the reader page, is that, if the
821 * writer is ever on it, the previous pointer never points
822 * back to the reader page.
824 static int rb_is_reader_page(struct buffer_page
*page
)
826 struct list_head
*list
= page
->list
.prev
;
828 return rb_list_head(list
->next
) != &page
->list
;
832 * rb_set_list_to_head - set a list_head to be pointing to head.
834 static void rb_set_list_to_head(struct ring_buffer_per_cpu
*cpu_buffer
,
835 struct list_head
*list
)
839 ptr
= (unsigned long *)&list
->next
;
840 *ptr
|= RB_PAGE_HEAD
;
841 *ptr
&= ~RB_PAGE_UPDATE
;
845 * rb_head_page_activate - sets up head page
847 static void rb_head_page_activate(struct ring_buffer_per_cpu
*cpu_buffer
)
849 struct buffer_page
*head
;
851 head
= cpu_buffer
->head_page
;
856 * Set the previous list pointer to have the HEAD flag.
858 rb_set_list_to_head(cpu_buffer
, head
->list
.prev
);
861 static void rb_list_head_clear(struct list_head
*list
)
863 unsigned long *ptr
= (unsigned long *)&list
->next
;
865 *ptr
&= ~RB_FLAG_MASK
;
869 * rb_head_page_dactivate - clears head page ptr (for free list)
872 rb_head_page_deactivate(struct ring_buffer_per_cpu
*cpu_buffer
)
874 struct list_head
*hd
;
876 /* Go through the whole list and clear any pointers found. */
877 rb_list_head_clear(cpu_buffer
->pages
);
879 list_for_each(hd
, cpu_buffer
->pages
)
880 rb_list_head_clear(hd
);
883 static int rb_head_page_set(struct ring_buffer_per_cpu
*cpu_buffer
,
884 struct buffer_page
*head
,
885 struct buffer_page
*prev
,
886 int old_flag
, int new_flag
)
888 struct list_head
*list
;
889 unsigned long val
= (unsigned long)&head
->list
;
894 val
&= ~RB_FLAG_MASK
;
896 ret
= cmpxchg((unsigned long *)&list
->next
,
897 val
| old_flag
, val
| new_flag
);
899 /* check if the reader took the page */
900 if ((ret
& ~RB_FLAG_MASK
) != val
)
901 return RB_PAGE_MOVED
;
903 return ret
& RB_FLAG_MASK
;
906 static int rb_head_page_set_update(struct ring_buffer_per_cpu
*cpu_buffer
,
907 struct buffer_page
*head
,
908 struct buffer_page
*prev
,
911 return rb_head_page_set(cpu_buffer
, head
, prev
,
912 old_flag
, RB_PAGE_UPDATE
);
915 static int rb_head_page_set_head(struct ring_buffer_per_cpu
*cpu_buffer
,
916 struct buffer_page
*head
,
917 struct buffer_page
*prev
,
920 return rb_head_page_set(cpu_buffer
, head
, prev
,
921 old_flag
, RB_PAGE_HEAD
);
924 static int rb_head_page_set_normal(struct ring_buffer_per_cpu
*cpu_buffer
,
925 struct buffer_page
*head
,
926 struct buffer_page
*prev
,
929 return rb_head_page_set(cpu_buffer
, head
, prev
,
930 old_flag
, RB_PAGE_NORMAL
);
933 static inline void rb_inc_page(struct ring_buffer_per_cpu
*cpu_buffer
,
934 struct buffer_page
**bpage
)
936 struct list_head
*p
= rb_list_head((*bpage
)->list
.next
);
938 *bpage
= list_entry(p
, struct buffer_page
, list
);
941 static struct buffer_page
*
942 rb_set_head_page(struct ring_buffer_per_cpu
*cpu_buffer
)
944 struct buffer_page
*head
;
945 struct buffer_page
*page
;
946 struct list_head
*list
;
949 if (RB_WARN_ON(cpu_buffer
, !cpu_buffer
->head_page
))
953 list
= cpu_buffer
->pages
;
954 if (RB_WARN_ON(cpu_buffer
, rb_list_head(list
->prev
->next
) != list
))
957 page
= head
= cpu_buffer
->head_page
;
959 * It is possible that the writer moves the header behind
960 * where we started, and we miss in one loop.
961 * A second loop should grab the header, but we'll do
962 * three loops just because I'm paranoid.
964 for (i
= 0; i
< 3; i
++) {
966 if (rb_is_head_page(cpu_buffer
, page
, page
->list
.prev
)) {
967 cpu_buffer
->head_page
= page
;
970 rb_inc_page(cpu_buffer
, &page
);
971 } while (page
!= head
);
974 RB_WARN_ON(cpu_buffer
, 1);
979 static int rb_head_page_replace(struct buffer_page
*old
,
980 struct buffer_page
*new)
982 unsigned long *ptr
= (unsigned long *)&old
->list
.prev
->next
;
986 val
= *ptr
& ~RB_FLAG_MASK
;
989 ret
= cmpxchg(ptr
, val
, (unsigned long)&new->list
);
995 * rb_tail_page_update - move the tail page forward
997 * Returns 1 if moved tail page, 0 if someone else did.
999 static int rb_tail_page_update(struct ring_buffer_per_cpu
*cpu_buffer
,
1000 struct buffer_page
*tail_page
,
1001 struct buffer_page
*next_page
)
1003 struct buffer_page
*old_tail
;
1004 unsigned long old_entries
;
1005 unsigned long old_write
;
1009 * The tail page now needs to be moved forward.
1011 * We need to reset the tail page, but without messing
1012 * with possible erasing of data brought in by interrupts
1013 * that have moved the tail page and are currently on it.
1015 * We add a counter to the write field to denote this.
1017 old_write
= local_add_return(RB_WRITE_INTCNT
, &next_page
->write
);
1018 old_entries
= local_add_return(RB_WRITE_INTCNT
, &next_page
->entries
);
1021 * Just make sure we have seen our old_write and synchronize
1022 * with any interrupts that come in.
1027 * If the tail page is still the same as what we think
1028 * it is, then it is up to us to update the tail
1031 if (tail_page
== cpu_buffer
->tail_page
) {
1032 /* Zero the write counter */
1033 unsigned long val
= old_write
& ~RB_WRITE_MASK
;
1034 unsigned long eval
= old_entries
& ~RB_WRITE_MASK
;
1037 * This will only succeed if an interrupt did
1038 * not come in and change it. In which case, we
1039 * do not want to modify it.
1041 * We add (void) to let the compiler know that we do not care
1042 * about the return value of these functions. We use the
1043 * cmpxchg to only update if an interrupt did not already
1044 * do it for us. If the cmpxchg fails, we don't care.
1046 (void)local_cmpxchg(&next_page
->write
, old_write
, val
);
1047 (void)local_cmpxchg(&next_page
->entries
, old_entries
, eval
);
1050 * No need to worry about races with clearing out the commit.
1051 * it only can increment when a commit takes place. But that
1052 * only happens in the outer most nested commit.
1054 local_set(&next_page
->page
->commit
, 0);
1056 old_tail
= cmpxchg(&cpu_buffer
->tail_page
,
1057 tail_page
, next_page
);
1059 if (old_tail
== tail_page
)
1066 static int rb_check_bpage(struct ring_buffer_per_cpu
*cpu_buffer
,
1067 struct buffer_page
*bpage
)
1069 unsigned long val
= (unsigned long)bpage
;
1071 if (RB_WARN_ON(cpu_buffer
, val
& RB_FLAG_MASK
))
1078 * rb_check_list - make sure a pointer to a list has the last bits zero
1080 static int rb_check_list(struct ring_buffer_per_cpu
*cpu_buffer
,
1081 struct list_head
*list
)
1083 if (RB_WARN_ON(cpu_buffer
, rb_list_head(list
->prev
) != list
->prev
))
1085 if (RB_WARN_ON(cpu_buffer
, rb_list_head(list
->next
) != list
->next
))
1091 * check_pages - integrity check of buffer pages
1092 * @cpu_buffer: CPU buffer with pages to test
1094 * As a safety measure we check to make sure the data pages have not
1097 static int rb_check_pages(struct ring_buffer_per_cpu
*cpu_buffer
)
1099 struct list_head
*head
= cpu_buffer
->pages
;
1100 struct buffer_page
*bpage
, *tmp
;
1102 /* Reset the head page if it exists */
1103 if (cpu_buffer
->head_page
)
1104 rb_set_head_page(cpu_buffer
);
1106 rb_head_page_deactivate(cpu_buffer
);
1108 if (RB_WARN_ON(cpu_buffer
, head
->next
->prev
!= head
))
1110 if (RB_WARN_ON(cpu_buffer
, head
->prev
->next
!= head
))
1113 if (rb_check_list(cpu_buffer
, head
))
1116 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
1117 if (RB_WARN_ON(cpu_buffer
,
1118 bpage
->list
.next
->prev
!= &bpage
->list
))
1120 if (RB_WARN_ON(cpu_buffer
,
1121 bpage
->list
.prev
->next
!= &bpage
->list
))
1123 if (rb_check_list(cpu_buffer
, &bpage
->list
))
1127 rb_head_page_activate(cpu_buffer
);
1132 static int __rb_allocate_pages(int nr_pages
, struct list_head
*pages
, int cpu
)
1135 struct buffer_page
*bpage
, *tmp
;
1137 for (i
= 0; i
< nr_pages
; i
++) {
1138 #if !defined (CONFIG_MTK_EXTMEM)
1142 * __GFP_NORETRY flag makes sure that the allocation fails
1143 * gracefully without invoking oom-killer and the system is
1146 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
1147 GFP_KERNEL
| __GFP_NORETRY
,
1152 list_add(&bpage
->list
, pages
);
1154 #ifdef CONFIG_MTK_EXTMEM
1155 bpage
->page
= extmem_malloc_page_align(PAGE_SIZE
);
1156 if(bpage
->page
== NULL
) {
1157 pr_err("%s[%s] ext memory alloc failed!!!\n", __FILE__
, __FUNCTION__
);
1161 page
= alloc_pages_node(cpu_to_node(cpu
),
1162 GFP_KERNEL
| __GFP_NORETRY
, 0);
1165 bpage
->page
= page_address(page
);
1167 rb_init_page(bpage
->page
);
1173 list_for_each_entry_safe(bpage
, tmp
, pages
, list
) {
1174 list_del_init(&bpage
->list
);
1175 free_buffer_page(bpage
);
1181 static int rb_allocate_pages(struct ring_buffer_per_cpu
*cpu_buffer
,
1188 if (__rb_allocate_pages(nr_pages
, &pages
, cpu_buffer
->cpu
))
1192 * The ring buffer page list is a circular list that does not
1193 * start and end with a list head. All page list items point to
1196 cpu_buffer
->pages
= pages
.next
;
1199 cpu_buffer
->nr_pages
= nr_pages
;
1201 rb_check_pages(cpu_buffer
);
1206 static struct ring_buffer_per_cpu
*
1207 rb_allocate_cpu_buffer(struct ring_buffer
*buffer
, int nr_pages
, int cpu
)
1209 struct ring_buffer_per_cpu
*cpu_buffer
;
1210 struct buffer_page
*bpage
;
1211 #if !defined (CONFIG_MTK_EXTMEM)
1216 cpu_buffer
= kzalloc_node(ALIGN(sizeof(*cpu_buffer
), cache_line_size()),
1217 GFP_KERNEL
, cpu_to_node(cpu
));
1221 cpu_buffer
->cpu
= cpu
;
1222 cpu_buffer
->buffer
= buffer
;
1223 raw_spin_lock_init(&cpu_buffer
->reader_lock
);
1224 lockdep_set_class(&cpu_buffer
->reader_lock
, buffer
->reader_lock_key
);
1225 cpu_buffer
->lock
= (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
1226 INIT_WORK(&cpu_buffer
->update_pages_work
, update_pages_handler
);
1227 init_completion(&cpu_buffer
->update_done
);
1228 init_irq_work(&cpu_buffer
->irq_work
.work
, rb_wake_up_waiters
);
1229 init_waitqueue_head(&cpu_buffer
->irq_work
.waiters
);
1231 bpage
= kzalloc_node(ALIGN(sizeof(*bpage
), cache_line_size()),
1232 GFP_KERNEL
, cpu_to_node(cpu
));
1234 goto fail_free_buffer
;
1236 rb_check_bpage(cpu_buffer
, bpage
);
1238 cpu_buffer
->reader_page
= bpage
;
1240 #ifdef CONFIG_MTK_EXTMEM
1241 bpage
->page
= extmem_malloc_page_align(PAGE_SIZE
);
1242 if(bpage
->page
== NULL
)
1243 goto fail_free_reader
;
1245 page
= alloc_pages_node(cpu_to_node(cpu
), GFP_KERNEL
, 0);
1247 goto fail_free_reader
;
1248 bpage
->page
= page_address(page
);
1250 rb_init_page(bpage
->page
);
1252 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
1253 INIT_LIST_HEAD(&cpu_buffer
->new_pages
);
1255 ret
= rb_allocate_pages(cpu_buffer
, nr_pages
);
1257 goto fail_free_reader
;
1259 cpu_buffer
->head_page
1260 = list_entry(cpu_buffer
->pages
, struct buffer_page
, list
);
1261 cpu_buffer
->tail_page
= cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
1263 rb_head_page_activate(cpu_buffer
);
1268 free_buffer_page(cpu_buffer
->reader_page
);
1275 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu
*cpu_buffer
)
1277 struct list_head
*head
= cpu_buffer
->pages
;
1278 struct buffer_page
*bpage
, *tmp
;
1280 free_buffer_page(cpu_buffer
->reader_page
);
1282 rb_head_page_deactivate(cpu_buffer
);
1285 list_for_each_entry_safe(bpage
, tmp
, head
, list
) {
1286 list_del_init(&bpage
->list
);
1287 free_buffer_page(bpage
);
1289 bpage
= list_entry(head
, struct buffer_page
, list
);
1290 free_buffer_page(bpage
);
1296 #ifdef CONFIG_HOTPLUG_CPU
1297 static int rb_cpu_notify(struct notifier_block
*self
,
1298 unsigned long action
, void *hcpu
);
1302 * ring_buffer_alloc - allocate a new ring_buffer
1303 * @size: the size in bytes per cpu that is needed.
1304 * @flags: attributes to set for the ring buffer.
1306 * Currently the only flag that is available is the RB_FL_OVERWRITE
1307 * flag. This flag means that the buffer will overwrite old data
1308 * when the buffer wraps. If this flag is not set, the buffer will
1309 * drop data when the tail hits the head.
1311 struct ring_buffer
*__ring_buffer_alloc(unsigned long size
, unsigned flags
,
1312 struct lock_class_key
*key
)
1314 struct ring_buffer
*buffer
;
1318 /* keep it in its own cache line */
1319 buffer
= kzalloc(ALIGN(sizeof(*buffer
), cache_line_size()),
1324 if (!alloc_cpumask_var(&buffer
->cpumask
, GFP_KERNEL
))
1325 goto fail_free_buffer
;
1327 nr_pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
1328 buffer
->flags
= flags
;
1329 buffer
->clock
= trace_clock_local
;
1330 buffer
->reader_lock_key
= key
;
1332 init_irq_work(&buffer
->irq_work
.work
, rb_wake_up_waiters
);
1333 init_waitqueue_head(&buffer
->irq_work
.waiters
);
1335 /* need at least two pages */
1340 * In case of non-hotplug cpu, if the ring-buffer is allocated
1341 * in early initcall, it will not be notified of secondary cpus.
1342 * In that off case, we need to allocate for all possible cpus.
1344 #ifdef CONFIG_HOTPLUG_CPU
1346 cpumask_copy(buffer
->cpumask
, cpu_online_mask
);
1348 cpumask_copy(buffer
->cpumask
, cpu_possible_mask
);
1350 buffer
->cpus
= nr_cpu_ids
;
1352 bsize
= sizeof(void *) * nr_cpu_ids
;
1353 buffer
->buffers
= kzalloc(ALIGN(bsize
, cache_line_size()),
1355 if (!buffer
->buffers
)
1356 goto fail_free_cpumask
;
1358 for_each_buffer_cpu(buffer
, cpu
) {
1359 buffer
->buffers
[cpu
] =
1360 rb_allocate_cpu_buffer(buffer
, nr_pages
, cpu
);
1361 if (!buffer
->buffers
[cpu
])
1362 goto fail_free_buffers
;
1365 #ifdef CONFIG_HOTPLUG_CPU
1366 buffer
->cpu_notify
.notifier_call
= rb_cpu_notify
;
1367 buffer
->cpu_notify
.priority
= 0;
1368 register_cpu_notifier(&buffer
->cpu_notify
);
1372 mutex_init(&buffer
->mutex
);
1377 for_each_buffer_cpu(buffer
, cpu
) {
1378 if (buffer
->buffers
[cpu
])
1379 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
1381 kfree(buffer
->buffers
);
1384 free_cpumask_var(buffer
->cpumask
);
1391 EXPORT_SYMBOL_GPL(__ring_buffer_alloc
);
1394 * ring_buffer_free - free a ring buffer.
1395 * @buffer: the buffer to free.
1398 ring_buffer_free(struct ring_buffer
*buffer
)
1404 #ifdef CONFIG_HOTPLUG_CPU
1405 unregister_cpu_notifier(&buffer
->cpu_notify
);
1408 for_each_buffer_cpu(buffer
, cpu
)
1409 rb_free_cpu_buffer(buffer
->buffers
[cpu
]);
1413 kfree(buffer
->buffers
);
1414 free_cpumask_var(buffer
->cpumask
);
1418 EXPORT_SYMBOL_GPL(ring_buffer_free
);
1420 void ring_buffer_set_clock(struct ring_buffer
*buffer
,
1423 buffer
->clock
= clock
;
1426 static void rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
);
1428 static inline unsigned long rb_page_entries(struct buffer_page
*bpage
)
1430 return local_read(&bpage
->entries
) & RB_WRITE_MASK
;
1433 static inline unsigned long rb_page_write(struct buffer_page
*bpage
)
1435 return local_read(&bpage
->write
) & RB_WRITE_MASK
;
1439 rb_remove_pages(struct ring_buffer_per_cpu
*cpu_buffer
, unsigned int nr_pages
)
1441 struct list_head
*tail_page
, *to_remove
, *next_page
;
1442 struct buffer_page
*to_remove_page
, *tmp_iter_page
;
1443 struct buffer_page
*last_page
, *first_page
;
1444 unsigned int nr_removed
;
1445 unsigned long head_bit
;
1450 raw_spin_lock_irq(&cpu_buffer
->reader_lock
);
1451 atomic_inc(&cpu_buffer
->record_disabled
);
1453 * We don't race with the readers since we have acquired the reader
1454 * lock. We also don't race with writers after disabling recording.
1455 * This makes it easy to figure out the first and the last page to be
1456 * removed from the list. We unlink all the pages in between including
1457 * the first and last pages. This is done in a busy loop so that we
1458 * lose the least number of traces.
1459 * The pages are freed after we restart recording and unlock readers.
1461 tail_page
= &cpu_buffer
->tail_page
->list
;
1464 * tail page might be on reader page, we remove the next page
1465 * from the ring buffer
1467 if (cpu_buffer
->tail_page
== cpu_buffer
->reader_page
)
1468 tail_page
= rb_list_head(tail_page
->next
);
1469 to_remove
= tail_page
;
1471 /* start of pages to remove */
1472 first_page
= list_entry(rb_list_head(to_remove
->next
),
1473 struct buffer_page
, list
);
1475 for (nr_removed
= 0; nr_removed
< nr_pages
; nr_removed
++) {
1476 to_remove
= rb_list_head(to_remove
)->next
;
1477 head_bit
|= (unsigned long)to_remove
& RB_PAGE_HEAD
;
1480 next_page
= rb_list_head(to_remove
)->next
;
1483 * Now we remove all pages between tail_page and next_page.
1484 * Make sure that we have head_bit value preserved for the
1487 tail_page
->next
= (struct list_head
*)((unsigned long)next_page
|
1489 next_page
= rb_list_head(next_page
);
1490 next_page
->prev
= tail_page
;
1492 /* make sure pages points to a valid page in the ring buffer */
1493 cpu_buffer
->pages
= next_page
;
1495 /* update head page */
1497 cpu_buffer
->head_page
= list_entry(next_page
,
1498 struct buffer_page
, list
);
1501 * change read pointer to make sure any read iterators reset
1504 cpu_buffer
->read
= 0;
1506 /* pages are removed, resume tracing and then free the pages */
1507 atomic_dec(&cpu_buffer
->record_disabled
);
1508 raw_spin_unlock_irq(&cpu_buffer
->reader_lock
);
1510 RB_WARN_ON(cpu_buffer
, list_empty(cpu_buffer
->pages
));
1512 /* last buffer page to remove */
1513 last_page
= list_entry(rb_list_head(to_remove
), struct buffer_page
,
1515 tmp_iter_page
= first_page
;
1518 to_remove_page
= tmp_iter_page
;
1519 rb_inc_page(cpu_buffer
, &tmp_iter_page
);
1521 /* update the counters */
1522 page_entries
= rb_page_entries(to_remove_page
);
1525 * If something was added to this page, it was full
1526 * since it is not the tail page. So we deduct the
1527 * bytes consumed in ring buffer from here.
1528 * Increment overrun to account for the lost events.
1530 local_add(page_entries
, &cpu_buffer
->overrun
);
1531 local_sub(BUF_PAGE_SIZE
, &cpu_buffer
->entries_bytes
);
1535 * We have already removed references to this list item, just
1536 * free up the buffer_page and its page
1538 free_buffer_page(to_remove_page
);
1541 } while (to_remove_page
!= last_page
);
1543 RB_WARN_ON(cpu_buffer
, nr_removed
);
1545 return nr_removed
== 0;
1549 rb_insert_pages(struct ring_buffer_per_cpu
*cpu_buffer
)
1551 struct list_head
*pages
= &cpu_buffer
->new_pages
;
1552 int retries
, success
;
1554 raw_spin_lock_irq(&cpu_buffer
->reader_lock
);
1556 * We are holding the reader lock, so the reader page won't be swapped
1557 * in the ring buffer. Now we are racing with the writer trying to
1558 * move head page and the tail page.
1559 * We are going to adapt the reader page update process where:
1560 * 1. We first splice the start and end of list of new pages between
1561 * the head page and its previous page.
1562 * 2. We cmpxchg the prev_page->next to point from head page to the
1563 * start of new pages list.
1564 * 3. Finally, we update the head->prev to the end of new list.
1566 * We will try this process 10 times, to make sure that we don't keep
1572 struct list_head
*head_page
, *prev_page
, *r
;
1573 struct list_head
*last_page
, *first_page
;
1574 struct list_head
*head_page_with_bit
;
1576 head_page
= &rb_set_head_page(cpu_buffer
)->list
;
1579 prev_page
= head_page
->prev
;
1581 first_page
= pages
->next
;
1582 last_page
= pages
->prev
;
1584 head_page_with_bit
= (struct list_head
*)
1585 ((unsigned long)head_page
| RB_PAGE_HEAD
);
1587 last_page
->next
= head_page_with_bit
;
1588 first_page
->prev
= prev_page
;
1590 r
= cmpxchg(&prev_page
->next
, head_page_with_bit
, first_page
);
1592 if (r
== head_page_with_bit
) {
1594 * yay, we replaced the page pointer to our new list,
1595 * now, we just have to update to head page's prev
1596 * pointer to point to end of list
1598 head_page
->prev
= last_page
;
1605 INIT_LIST_HEAD(pages
);
1607 * If we weren't successful in adding in new pages, warn and stop
1610 RB_WARN_ON(cpu_buffer
, !success
);
1611 raw_spin_unlock_irq(&cpu_buffer
->reader_lock
);
1613 /* free pages if they weren't inserted */
1615 struct buffer_page
*bpage
, *tmp
;
1616 list_for_each_entry_safe(bpage
, tmp
, &cpu_buffer
->new_pages
,
1618 list_del_init(&bpage
->list
);
1619 free_buffer_page(bpage
);
1625 static void rb_update_pages(struct ring_buffer_per_cpu
*cpu_buffer
)
1629 if (cpu_buffer
->nr_pages_to_update
> 0)
1630 success
= rb_insert_pages(cpu_buffer
);
1632 success
= rb_remove_pages(cpu_buffer
,
1633 -cpu_buffer
->nr_pages_to_update
);
1636 cpu_buffer
->nr_pages
+= cpu_buffer
->nr_pages_to_update
;
1639 static void update_pages_handler(struct work_struct
*work
)
1641 struct ring_buffer_per_cpu
*cpu_buffer
= container_of(work
,
1642 struct ring_buffer_per_cpu
, update_pages_work
);
1643 rb_update_pages(cpu_buffer
);
1644 complete(&cpu_buffer
->update_done
);
1648 * ring_buffer_resize - resize the ring buffer
1649 * @buffer: the buffer to resize.
1650 * @size: the new size.
1652 * Minimum size is 2 * BUF_PAGE_SIZE.
1654 * Returns 0 on success and < 0 on failure.
1656 int ring_buffer_resize(struct ring_buffer
*buffer
, unsigned long size
,
1659 struct ring_buffer_per_cpu
*cpu_buffer
;
1664 * Always succeed at resizing a non-existent buffer:
1669 /* Make sure the requested buffer exists */
1670 if (cpu_id
!= RING_BUFFER_ALL_CPUS
&&
1671 !cpumask_test_cpu(cpu_id
, buffer
->cpumask
))
1674 size
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
1675 size
*= BUF_PAGE_SIZE
;
1677 /* we need a minimum of two pages */
1678 if (size
< BUF_PAGE_SIZE
* 2)
1679 size
= BUF_PAGE_SIZE
* 2;
1681 nr_pages
= DIV_ROUND_UP(size
, BUF_PAGE_SIZE
);
1684 * Don't succeed if resizing is disabled, as a reader might be
1685 * manipulating the ring buffer and is expecting a sane state while
1688 if (atomic_read(&buffer
->resize_disabled
))
1691 /* prevent another thread from changing buffer sizes */
1692 mutex_lock(&buffer
->mutex
);
1694 if (cpu_id
== RING_BUFFER_ALL_CPUS
) {
1695 /* calculate the pages to update */
1696 for_each_buffer_cpu(buffer
, cpu
) {
1697 cpu_buffer
= buffer
->buffers
[cpu
];
1699 cpu_buffer
->nr_pages_to_update
= nr_pages
-
1700 cpu_buffer
->nr_pages
;
1702 * nothing more to do for removing pages or no update
1704 if (cpu_buffer
->nr_pages_to_update
<= 0)
1707 * to add pages, make sure all new pages can be
1708 * allocated without receiving ENOMEM
1710 INIT_LIST_HEAD(&cpu_buffer
->new_pages
);
1711 if (__rb_allocate_pages(cpu_buffer
->nr_pages_to_update
,
1712 &cpu_buffer
->new_pages
, cpu
)) {
1713 /* not enough memory for new pages */
1721 * Fire off all the required work handlers
1722 * We can't schedule on offline CPUs, but it's not necessary
1723 * since we can change their buffer sizes without any race.
1725 for_each_buffer_cpu(buffer
, cpu
) {
1726 cpu_buffer
= buffer
->buffers
[cpu
];
1727 if (!cpu_buffer
->nr_pages_to_update
)
1730 /* The update must run on the CPU that is being updated. */
1732 if (cpu
== smp_processor_id() || !cpu_online(cpu
)) {
1733 rb_update_pages(cpu_buffer
);
1734 cpu_buffer
->nr_pages_to_update
= 0;
1737 * Can not disable preemption for schedule_work_on()
1741 schedule_work_on(cpu
,
1742 &cpu_buffer
->update_pages_work
);
1748 /* wait for all the updates to complete */
1749 for_each_buffer_cpu(buffer
, cpu
) {
1750 cpu_buffer
= buffer
->buffers
[cpu
];
1751 if (!cpu_buffer
->nr_pages_to_update
)
1754 if (cpu_online(cpu
))
1755 wait_for_completion(&cpu_buffer
->update_done
);
1756 cpu_buffer
->nr_pages_to_update
= 0;
1761 /* Make sure this CPU has been intitialized */
1762 if (!cpumask_test_cpu(cpu_id
, buffer
->cpumask
))
1765 cpu_buffer
= buffer
->buffers
[cpu_id
];
1767 if (nr_pages
== cpu_buffer
->nr_pages
)
1770 cpu_buffer
->nr_pages_to_update
= nr_pages
-
1771 cpu_buffer
->nr_pages
;
1773 INIT_LIST_HEAD(&cpu_buffer
->new_pages
);
1774 if (cpu_buffer
->nr_pages_to_update
> 0 &&
1775 __rb_allocate_pages(cpu_buffer
->nr_pages_to_update
,
1776 &cpu_buffer
->new_pages
, cpu_id
)) {
1784 /* The update must run on the CPU that is being updated. */
1785 if (cpu_id
== smp_processor_id() || !cpu_online(cpu_id
))
1786 rb_update_pages(cpu_buffer
);
1789 * Can not disable preemption for schedule_work_on()
1793 schedule_work_on(cpu_id
,
1794 &cpu_buffer
->update_pages_work
);
1795 wait_for_completion(&cpu_buffer
->update_done
);
1800 cpu_buffer
->nr_pages_to_update
= 0;
1806 * The ring buffer resize can happen with the ring buffer
1807 * enabled, so that the update disturbs the tracing as little
1808 * as possible. But if the buffer is disabled, we do not need
1809 * to worry about that, and we can take the time to verify
1810 * that the buffer is not corrupt.
1812 if (atomic_read(&buffer
->record_disabled
)) {
1813 atomic_inc(&buffer
->record_disabled
);
1815 * Even though the buffer was disabled, we must make sure
1816 * that it is truly disabled before calling rb_check_pages.
1817 * There could have been a race between checking
1818 * record_disable and incrementing it.
1820 synchronize_sched();
1821 for_each_buffer_cpu(buffer
, cpu
) {
1822 cpu_buffer
= buffer
->buffers
[cpu
];
1823 rb_check_pages(cpu_buffer
);
1825 atomic_dec(&buffer
->record_disabled
);
1828 mutex_unlock(&buffer
->mutex
);
1832 for_each_buffer_cpu(buffer
, cpu
) {
1833 struct buffer_page
*bpage
, *tmp
;
1835 cpu_buffer
= buffer
->buffers
[cpu
];
1836 cpu_buffer
->nr_pages_to_update
= 0;
1838 if (list_empty(&cpu_buffer
->new_pages
))
1841 list_for_each_entry_safe(bpage
, tmp
, &cpu_buffer
->new_pages
,
1843 list_del_init(&bpage
->list
);
1844 free_buffer_page(bpage
);
1847 mutex_unlock(&buffer
->mutex
);
1850 EXPORT_SYMBOL_GPL(ring_buffer_resize
);
1852 void ring_buffer_change_overwrite(struct ring_buffer
*buffer
, int val
)
1854 mutex_lock(&buffer
->mutex
);
1856 buffer
->flags
|= RB_FL_OVERWRITE
;
1858 buffer
->flags
&= ~RB_FL_OVERWRITE
;
1859 mutex_unlock(&buffer
->mutex
);
1861 EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite
);
1863 static inline void *
1864 __rb_data_page_index(struct buffer_data_page
*bpage
, unsigned index
)
1866 return bpage
->data
+ index
;
1869 static inline void *__rb_page_index(struct buffer_page
*bpage
, unsigned index
)
1871 return bpage
->page
->data
+ index
;
1874 static inline struct ring_buffer_event
*
1875 rb_reader_event(struct ring_buffer_per_cpu
*cpu_buffer
)
1877 return __rb_page_index(cpu_buffer
->reader_page
,
1878 cpu_buffer
->reader_page
->read
);
1881 static inline struct ring_buffer_event
*
1882 rb_iter_head_event(struct ring_buffer_iter
*iter
)
1884 return __rb_page_index(iter
->head_page
, iter
->head
);
1887 static inline unsigned rb_page_commit(struct buffer_page
*bpage
)
1889 return local_read(&bpage
->page
->commit
);
1892 /* Size is determined by what has been committed */
1893 static inline unsigned rb_page_size(struct buffer_page
*bpage
)
1895 return rb_page_commit(bpage
);
1898 static inline unsigned
1899 rb_commit_index(struct ring_buffer_per_cpu
*cpu_buffer
)
1901 return rb_page_commit(cpu_buffer
->commit_page
);
1904 static inline unsigned
1905 rb_event_index(struct ring_buffer_event
*event
)
1907 unsigned long addr
= (unsigned long)event
;
1909 return (addr
& ~PAGE_MASK
) - BUF_PAGE_HDR_SIZE
;
1913 rb_event_is_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
1914 struct ring_buffer_event
*event
)
1916 unsigned long addr
= (unsigned long)event
;
1917 unsigned long index
;
1919 index
= rb_event_index(event
);
1922 return cpu_buffer
->commit_page
->page
== (void *)addr
&&
1923 rb_commit_index(cpu_buffer
) == index
;
1927 rb_set_commit_to_write(struct ring_buffer_per_cpu
*cpu_buffer
)
1929 unsigned long max_count
;
1932 * We only race with interrupts and NMIs on this CPU.
1933 * If we own the commit event, then we can commit
1934 * all others that interrupted us, since the interruptions
1935 * are in stack format (they finish before they come
1936 * back to us). This allows us to do a simple loop to
1937 * assign the commit to the tail.
1940 max_count
= cpu_buffer
->nr_pages
* 100;
1942 while (cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
) {
1943 if (RB_WARN_ON(cpu_buffer
, !(--max_count
)))
1945 if (RB_WARN_ON(cpu_buffer
,
1946 rb_is_reader_page(cpu_buffer
->tail_page
)))
1948 local_set(&cpu_buffer
->commit_page
->page
->commit
,
1949 rb_page_write(cpu_buffer
->commit_page
));
1950 rb_inc_page(cpu_buffer
, &cpu_buffer
->commit_page
);
1951 cpu_buffer
->write_stamp
=
1952 cpu_buffer
->commit_page
->page
->time_stamp
;
1953 /* add barrier to keep gcc from optimizing too much */
1956 while (rb_commit_index(cpu_buffer
) !=
1957 rb_page_write(cpu_buffer
->commit_page
)) {
1959 local_set(&cpu_buffer
->commit_page
->page
->commit
,
1960 rb_page_write(cpu_buffer
->commit_page
));
1961 RB_WARN_ON(cpu_buffer
,
1962 local_read(&cpu_buffer
->commit_page
->page
->commit
) &
1967 /* again, keep gcc from optimizing */
1971 * If an interrupt came in just after the first while loop
1972 * and pushed the tail page forward, we will be left with
1973 * a dangling commit that will never go forward.
1975 if (unlikely(cpu_buffer
->commit_page
!= cpu_buffer
->tail_page
))
1979 static void rb_reset_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
1981 cpu_buffer
->read_stamp
= cpu_buffer
->reader_page
->page
->time_stamp
;
1982 cpu_buffer
->reader_page
->read
= 0;
1985 static void rb_inc_iter(struct ring_buffer_iter
*iter
)
1987 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
1990 * The iterator could be on the reader page (it starts there).
1991 * But the head could have moved, since the reader was
1992 * found. Check for this case and assign the iterator
1993 * to the head page instead of next.
1995 if (iter
->head_page
== cpu_buffer
->reader_page
)
1996 iter
->head_page
= rb_set_head_page(cpu_buffer
);
1998 rb_inc_page(cpu_buffer
, &iter
->head_page
);
2000 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
2004 /* Slow path, do not inline */
2005 static noinline
struct ring_buffer_event
*
2006 rb_add_time_stamp(struct ring_buffer_event
*event
, u64 delta
)
2008 event
->type_len
= RINGBUF_TYPE_TIME_EXTEND
;
2010 /* Not the first event on the page? */
2011 if (rb_event_index(event
)) {
2012 event
->time_delta
= delta
& TS_MASK
;
2013 event
->array
[0] = delta
>> TS_SHIFT
;
2015 /* nope, just zero it */
2016 event
->time_delta
= 0;
2017 event
->array
[0] = 0;
2020 return skip_time_extend(event
);
2024 * rb_update_event - update event type and data
2025 * @event: the event to update
2026 * @type: the type of event
2027 * @length: the size of the event field in the ring buffer
2029 * Update the type and data fields of the event. The length
2030 * is the actual size that is written to the ring buffer,
2031 * and with this, we can determine what to place into the
2035 rb_update_event(struct ring_buffer_per_cpu
*cpu_buffer
,
2036 struct ring_buffer_event
*event
, unsigned length
,
2037 int add_timestamp
, u64 delta
)
2039 /* Only a commit updates the timestamp */
2040 if (unlikely(!rb_event_is_commit(cpu_buffer
, event
)))
2044 * If we need to add a timestamp, then we
2045 * add it to the start of the resevered space.
2047 if (unlikely(add_timestamp
)) {
2048 event
= rb_add_time_stamp(event
, delta
);
2049 length
-= RB_LEN_TIME_EXTEND
;
2053 event
->time_delta
= delta
;
2054 length
-= RB_EVNT_HDR_SIZE
;
2055 if (length
> RB_MAX_SMALL_DATA
|| RB_FORCE_8BYTE_ALIGNMENT
) {
2056 event
->type_len
= 0;
2057 event
->array
[0] = length
;
2059 event
->type_len
= DIV_ROUND_UP(length
, RB_ALIGNMENT
);
2063 * rb_handle_head_page - writer hit the head page
2065 * Returns: +1 to retry page
2070 rb_handle_head_page(struct ring_buffer_per_cpu
*cpu_buffer
,
2071 struct buffer_page
*tail_page
,
2072 struct buffer_page
*next_page
)
2074 struct buffer_page
*new_head
;
2079 entries
= rb_page_entries(next_page
);
2082 * The hard part is here. We need to move the head
2083 * forward, and protect against both readers on
2084 * other CPUs and writers coming in via interrupts.
2086 type
= rb_head_page_set_update(cpu_buffer
, next_page
, tail_page
,
2090 * type can be one of four:
2091 * NORMAL - an interrupt already moved it for us
2092 * HEAD - we are the first to get here.
2093 * UPDATE - we are the interrupt interrupting
2095 * MOVED - a reader on another CPU moved the next
2096 * pointer to its reader page. Give up
2103 * We changed the head to UPDATE, thus
2104 * it is our responsibility to update
2107 local_add(entries
, &cpu_buffer
->overrun
);
2108 local_sub(BUF_PAGE_SIZE
, &cpu_buffer
->entries_bytes
);
2111 * The entries will be zeroed out when we move the
2115 /* still more to do */
2118 case RB_PAGE_UPDATE
:
2120 * This is an interrupt that interrupt the
2121 * previous update. Still more to do.
2124 case RB_PAGE_NORMAL
:
2126 * An interrupt came in before the update
2127 * and processed this for us.
2128 * Nothing left to do.
2133 * The reader is on another CPU and just did
2134 * a swap with our next_page.
2139 RB_WARN_ON(cpu_buffer
, 1); /* WTF??? */
2144 * Now that we are here, the old head pointer is
2145 * set to UPDATE. This will keep the reader from
2146 * swapping the head page with the reader page.
2147 * The reader (on another CPU) will spin till
2150 * We just need to protect against interrupts
2151 * doing the job. We will set the next pointer
2152 * to HEAD. After that, we set the old pointer
2153 * to NORMAL, but only if it was HEAD before.
2154 * otherwise we are an interrupt, and only
2155 * want the outer most commit to reset it.
2157 new_head
= next_page
;
2158 rb_inc_page(cpu_buffer
, &new_head
);
2160 ret
= rb_head_page_set_head(cpu_buffer
, new_head
, next_page
,
2164 * Valid returns are:
2165 * HEAD - an interrupt came in and already set it.
2166 * NORMAL - One of two things:
2167 * 1) We really set it.
2168 * 2) A bunch of interrupts came in and moved
2169 * the page forward again.
2173 case RB_PAGE_NORMAL
:
2177 RB_WARN_ON(cpu_buffer
, 1);
2182 * It is possible that an interrupt came in,
2183 * set the head up, then more interrupts came in
2184 * and moved it again. When we get back here,
2185 * the page would have been set to NORMAL but we
2186 * just set it back to HEAD.
2188 * How do you detect this? Well, if that happened
2189 * the tail page would have moved.
2191 if (ret
== RB_PAGE_NORMAL
) {
2193 * If the tail had moved passed next, then we need
2194 * to reset the pointer.
2196 if (cpu_buffer
->tail_page
!= tail_page
&&
2197 cpu_buffer
->tail_page
!= next_page
)
2198 rb_head_page_set_normal(cpu_buffer
, new_head
,
2204 * If this was the outer most commit (the one that
2205 * changed the original pointer from HEAD to UPDATE),
2206 * then it is up to us to reset it to NORMAL.
2208 if (type
== RB_PAGE_HEAD
) {
2209 ret
= rb_head_page_set_normal(cpu_buffer
, next_page
,
2212 if (RB_WARN_ON(cpu_buffer
,
2213 ret
!= RB_PAGE_UPDATE
))
2220 static unsigned rb_calculate_event_length(unsigned length
)
2222 struct ring_buffer_event event
; /* Used only for sizeof array */
2224 /* zero length can cause confusions */
2228 if (length
> RB_MAX_SMALL_DATA
|| RB_FORCE_8BYTE_ALIGNMENT
)
2229 length
+= sizeof(event
.array
[0]);
2231 length
+= RB_EVNT_HDR_SIZE
;
2232 length
= ALIGN(length
, RB_ARCH_ALIGNMENT
);
2238 rb_reset_tail(struct ring_buffer_per_cpu
*cpu_buffer
,
2239 struct buffer_page
*tail_page
,
2240 unsigned long tail
, unsigned long length
)
2242 struct ring_buffer_event
*event
;
2245 * Only the event that crossed the page boundary
2246 * must fill the old tail_page with padding.
2248 if (tail
>= BUF_PAGE_SIZE
) {
2250 * If the page was filled, then we still need
2251 * to update the real_end. Reset it to zero
2252 * and the reader will ignore it.
2254 if (tail
== BUF_PAGE_SIZE
)
2255 tail_page
->real_end
= 0;
2257 local_sub(length
, &tail_page
->write
);
2261 event
= __rb_page_index(tail_page
, tail
);
2262 kmemcheck_annotate_bitfield(event
, bitfield
);
2264 /* account for padding bytes */
2265 local_add(BUF_PAGE_SIZE
- tail
, &cpu_buffer
->entries_bytes
);
2268 * Save the original length to the meta data.
2269 * This will be used by the reader to add lost event
2272 tail_page
->real_end
= tail
;
2275 * If this event is bigger than the minimum size, then
2276 * we need to be careful that we don't subtract the
2277 * write counter enough to allow another writer to slip
2279 * We put in a discarded commit instead, to make sure
2280 * that this space is not used again.
2282 * If we are less than the minimum size, we don't need to
2285 if (tail
> (BUF_PAGE_SIZE
- RB_EVNT_MIN_SIZE
)) {
2286 /* No room for any events */
2288 /* Mark the rest of the page with padding */
2289 rb_event_set_padding(event
);
2291 /* Set the write back to the previous setting */
2292 local_sub(length
, &tail_page
->write
);
2296 /* Put in a discarded event */
2297 event
->array
[0] = (BUF_PAGE_SIZE
- tail
) - RB_EVNT_HDR_SIZE
;
2298 event
->type_len
= RINGBUF_TYPE_PADDING
;
2299 /* time delta must be non zero */
2300 event
->time_delta
= 1;
2302 /* Set write to end of buffer */
2303 length
= (tail
+ length
) - BUF_PAGE_SIZE
;
2304 local_sub(length
, &tail_page
->write
);
2308 * This is the slow path, force gcc not to inline it.
2310 static noinline
struct ring_buffer_event
*
2311 rb_move_tail(struct ring_buffer_per_cpu
*cpu_buffer
,
2312 unsigned long length
, unsigned long tail
,
2313 struct buffer_page
*tail_page
, u64 ts
)
2315 struct buffer_page
*commit_page
= cpu_buffer
->commit_page
;
2316 struct ring_buffer
*buffer
= cpu_buffer
->buffer
;
2317 struct buffer_page
*next_page
;
2320 next_page
= tail_page
;
2322 rb_inc_page(cpu_buffer
, &next_page
);
2325 * If for some reason, we had an interrupt storm that made
2326 * it all the way around the buffer, bail, and warn
2329 if (unlikely(next_page
== commit_page
)) {
2330 local_inc(&cpu_buffer
->commit_overrun
);
2335 * This is where the fun begins!
2337 * We are fighting against races between a reader that
2338 * could be on another CPU trying to swap its reader
2339 * page with the buffer head.
2341 * We are also fighting against interrupts coming in and
2342 * moving the head or tail on us as well.
2344 * If the next page is the head page then we have filled
2345 * the buffer, unless the commit page is still on the
2348 if (rb_is_head_page(cpu_buffer
, next_page
, &tail_page
->list
)) {
2351 * If the commit is not on the reader page, then
2352 * move the header page.
2354 if (!rb_is_reader_page(cpu_buffer
->commit_page
)) {
2356 * If we are not in overwrite mode,
2357 * this is easy, just stop here.
2359 if (!(buffer
->flags
& RB_FL_OVERWRITE
)) {
2360 local_inc(&cpu_buffer
->dropped_events
);
2364 ret
= rb_handle_head_page(cpu_buffer
,
2373 * We need to be careful here too. The
2374 * commit page could still be on the reader
2375 * page. We could have a small buffer, and
2376 * have filled up the buffer with events
2377 * from interrupts and such, and wrapped.
2379 * Note, if the tail page is also the on the
2380 * reader_page, we let it move out.
2382 if (unlikely((cpu_buffer
->commit_page
!=
2383 cpu_buffer
->tail_page
) &&
2384 (cpu_buffer
->commit_page
==
2385 cpu_buffer
->reader_page
))) {
2386 local_inc(&cpu_buffer
->commit_overrun
);
2392 ret
= rb_tail_page_update(cpu_buffer
, tail_page
, next_page
);
2395 * Nested commits always have zero deltas, so
2396 * just reread the time stamp
2398 ts
= rb_time_stamp(buffer
);
2399 next_page
->page
->time_stamp
= ts
;
2404 rb_reset_tail(cpu_buffer
, tail_page
, tail
, length
);
2406 /* fail and let the caller try again */
2407 return ERR_PTR(-EAGAIN
);
2411 rb_reset_tail(cpu_buffer
, tail_page
, tail
, length
);
2416 static struct ring_buffer_event
*
2417 __rb_reserve_next(struct ring_buffer_per_cpu
*cpu_buffer
,
2418 unsigned long length
, u64 ts
,
2419 u64 delta
, int add_timestamp
)
2421 struct buffer_page
*tail_page
;
2422 struct ring_buffer_event
*event
;
2423 unsigned long tail
, write
;
2426 * If the time delta since the last event is too big to
2427 * hold in the time field of the event, then we append a
2428 * TIME EXTEND event ahead of the data event.
2430 if (unlikely(add_timestamp
))
2431 length
+= RB_LEN_TIME_EXTEND
;
2433 tail_page
= cpu_buffer
->tail_page
;
2434 write
= local_add_return(length
, &tail_page
->write
);
2436 /* set write to only the index of the write */
2437 write
&= RB_WRITE_MASK
;
2438 tail
= write
- length
;
2441 * If this is the first commit on the page, then it has the same
2442 * timestamp as the page itself.
2447 /* See if we shot pass the end of this buffer page */
2448 if (unlikely(write
> BUF_PAGE_SIZE
))
2449 return rb_move_tail(cpu_buffer
, length
, tail
,
2452 /* We reserved something on the buffer */
2454 event
= __rb_page_index(tail_page
, tail
);
2455 kmemcheck_annotate_bitfield(event
, bitfield
);
2456 rb_update_event(cpu_buffer
, event
, length
, add_timestamp
, delta
);
2458 local_inc(&tail_page
->entries
);
2461 * If this is the first commit on the page, then update
2465 tail_page
->page
->time_stamp
= ts
;
2467 /* account for these added bytes */
2468 local_add(length
, &cpu_buffer
->entries_bytes
);
2474 rb_try_to_discard(struct ring_buffer_per_cpu
*cpu_buffer
,
2475 struct ring_buffer_event
*event
)
2477 unsigned long new_index
, old_index
;
2478 struct buffer_page
*bpage
;
2479 unsigned long index
;
2482 new_index
= rb_event_index(event
);
2483 old_index
= new_index
+ rb_event_ts_length(event
);
2484 addr
= (unsigned long)event
;
2487 bpage
= cpu_buffer
->tail_page
;
2489 if (bpage
->page
== (void *)addr
&& rb_page_write(bpage
) == old_index
) {
2490 unsigned long write_mask
=
2491 local_read(&bpage
->write
) & ~RB_WRITE_MASK
;
2492 unsigned long event_length
= rb_event_length(event
);
2494 * This is on the tail page. It is possible that
2495 * a write could come in and move the tail page
2496 * and write to the next page. That is fine
2497 * because we just shorten what is on this page.
2499 old_index
+= write_mask
;
2500 new_index
+= write_mask
;
2501 index
= local_cmpxchg(&bpage
->write
, old_index
, new_index
);
2502 if (index
== old_index
) {
2503 /* update counters */
2504 local_sub(event_length
, &cpu_buffer
->entries_bytes
);
2509 /* could not discard */
2513 static void rb_start_commit(struct ring_buffer_per_cpu
*cpu_buffer
)
2515 local_inc(&cpu_buffer
->committing
);
2516 local_inc(&cpu_buffer
->commits
);
2519 static inline void rb_end_commit(struct ring_buffer_per_cpu
*cpu_buffer
)
2521 unsigned long commits
;
2523 if (RB_WARN_ON(cpu_buffer
,
2524 !local_read(&cpu_buffer
->committing
)))
2528 commits
= local_read(&cpu_buffer
->commits
);
2529 /* synchronize with interrupts */
2531 if (local_read(&cpu_buffer
->committing
) == 1)
2532 rb_set_commit_to_write(cpu_buffer
);
2534 local_dec(&cpu_buffer
->committing
);
2536 /* synchronize with interrupts */
2540 * Need to account for interrupts coming in between the
2541 * updating of the commit page and the clearing of the
2542 * committing counter.
2544 if (unlikely(local_read(&cpu_buffer
->commits
) != commits
) &&
2545 !local_read(&cpu_buffer
->committing
)) {
2546 local_inc(&cpu_buffer
->committing
);
2551 static struct ring_buffer_event
*
2552 rb_reserve_next_event(struct ring_buffer
*buffer
,
2553 struct ring_buffer_per_cpu
*cpu_buffer
,
2554 unsigned long length
)
2556 struct ring_buffer_event
*event
;
2562 rb_start_commit(cpu_buffer
);
2564 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2566 * Due to the ability to swap a cpu buffer from a buffer
2567 * it is possible it was swapped before we committed.
2568 * (committing stops a swap). We check for it here and
2569 * if it happened, we have to fail the write.
2572 if (unlikely(ACCESS_ONCE(cpu_buffer
->buffer
) != buffer
)) {
2573 local_dec(&cpu_buffer
->committing
);
2574 local_dec(&cpu_buffer
->commits
);
2579 length
= rb_calculate_event_length(length
);
2585 * We allow for interrupts to reenter here and do a trace.
2586 * If one does, it will cause this original code to loop
2587 * back here. Even with heavy interrupts happening, this
2588 * should only happen a few times in a row. If this happens
2589 * 1000 times in a row, there must be either an interrupt
2590 * storm or we have something buggy.
2593 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 1000))
2596 ts
= rb_time_stamp(cpu_buffer
->buffer
);
2597 diff
= ts
- cpu_buffer
->write_stamp
;
2599 /* make sure this diff is calculated here */
2602 /* Did the write stamp get updated already? */
2603 if (likely(ts
>= cpu_buffer
->write_stamp
)) {
2605 if (unlikely(test_time_stamp(delta
))) {
2606 int local_clock_stable
= 1;
2607 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
2608 local_clock_stable
= sched_clock_stable
;
2610 WARN_ONCE(delta
> (1ULL << 59),
2611 KERN_WARNING
"Delta way too big! %llu ts=%llu write stamp = %llu\n%s",
2612 (unsigned long long)delta
,
2613 (unsigned long long)ts
,
2614 (unsigned long long)cpu_buffer
->write_stamp
,
2615 local_clock_stable
? "" :
2616 "If you just came from a suspend/resume,\n"
2617 "please switch to the trace global clock:\n"
2618 " echo global > /sys/kernel/debug/tracing/trace_clock\n");
2623 event
= __rb_reserve_next(cpu_buffer
, length
, ts
,
2624 delta
, add_timestamp
);
2625 if (unlikely(PTR_ERR(event
) == -EAGAIN
))
2634 rb_end_commit(cpu_buffer
);
2638 #ifdef CONFIG_TRACING
2641 * The lock and unlock are done within a preempt disable section.
2642 * The current_context per_cpu variable can only be modified
2643 * by the current task between lock and unlock. But it can
2644 * be modified more than once via an interrupt. To pass this
2645 * information from the lock to the unlock without having to
2646 * access the 'in_interrupt()' functions again (which do show
2647 * a bit of overhead in something as critical as function tracing,
2648 * we use a bitmask trick.
2650 * bit 0 = NMI context
2651 * bit 1 = IRQ context
2652 * bit 2 = SoftIRQ context
2653 * bit 3 = normal context.
2655 * This works because this is the order of contexts that can
2656 * preempt other contexts. A SoftIRQ never preempts an IRQ
2659 * When the context is determined, the corresponding bit is
2660 * checked and set (if it was set, then a recursion of that context
2663 * On unlock, we need to clear this bit. To do so, just subtract
2664 * 1 from the current_context and AND it to itself.
2668 * 101 & 100 = 100 (clearing bit zero)
2671 * 1010 & 1001 = 1000 (clearing bit 1)
2673 * The least significant bit can be cleared this way, and it
2674 * just so happens that it is the same bit corresponding to
2675 * the current context.
2677 static DEFINE_PER_CPU(unsigned int, current_context
);
2679 static __always_inline
int trace_recursive_lock(void)
2681 unsigned int val
= this_cpu_read(current_context
);
2684 if (in_interrupt()) {
2694 if (unlikely(val
& (1 << bit
)))
2698 this_cpu_write(current_context
, val
);
2703 static __always_inline
void trace_recursive_unlock(void)
2705 unsigned int val
= this_cpu_read(current_context
);
2708 val
&= this_cpu_read(current_context
);
2709 this_cpu_write(current_context
, val
);
2714 #define trace_recursive_lock() (0)
2715 #define trace_recursive_unlock() do { } while (0)
2720 * ring_buffer_lock_reserve - reserve a part of the buffer
2721 * @buffer: the ring buffer to reserve from
2722 * @length: the length of the data to reserve (excluding event header)
2724 * Returns a reseverd event on the ring buffer to copy directly to.
2725 * The user of this interface will need to get the body to write into
2726 * and can use the ring_buffer_event_data() interface.
2728 * The length is the length of the data needed, not the event length
2729 * which also includes the event header.
2731 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
2732 * If NULL is returned, then nothing has been allocated or locked.
2734 struct ring_buffer_event
*
2735 ring_buffer_lock_reserve(struct ring_buffer
*buffer
, unsigned long length
)
2737 struct ring_buffer_per_cpu
*cpu_buffer
;
2738 struct ring_buffer_event
*event
;
2741 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
2744 /* If we are tracing schedule, we don't want to recurse */
2745 preempt_disable_notrace();
2747 if (atomic_read(&buffer
->record_disabled
))
2750 if (trace_recursive_lock())
2753 cpu
= raw_smp_processor_id();
2755 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
2758 cpu_buffer
= buffer
->buffers
[cpu
];
2760 if (atomic_read(&cpu_buffer
->record_disabled
))
2763 if (length
> BUF_MAX_DATA_SIZE
)
2766 event
= rb_reserve_next_event(buffer
, cpu_buffer
, length
);
2773 trace_recursive_unlock();
2776 preempt_enable_notrace();
2779 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve
);
2782 rb_update_write_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
2783 struct ring_buffer_event
*event
)
2788 * The event first in the commit queue updates the
2791 if (rb_event_is_commit(cpu_buffer
, event
)) {
2793 * A commit event that is first on a page
2794 * updates the write timestamp with the page stamp
2796 if (!rb_event_index(event
))
2797 cpu_buffer
->write_stamp
=
2798 cpu_buffer
->commit_page
->page
->time_stamp
;
2799 else if (event
->type_len
== RINGBUF_TYPE_TIME_EXTEND
) {
2800 delta
= event
->array
[0];
2802 delta
+= event
->time_delta
;
2803 cpu_buffer
->write_stamp
+= delta
;
2805 cpu_buffer
->write_stamp
+= event
->time_delta
;
2809 static void rb_commit(struct ring_buffer_per_cpu
*cpu_buffer
,
2810 struct ring_buffer_event
*event
)
2812 local_inc(&cpu_buffer
->entries
);
2813 rb_update_write_stamp(cpu_buffer
, event
);
2814 rb_end_commit(cpu_buffer
);
2817 static __always_inline
void
2818 rb_wakeups(struct ring_buffer
*buffer
, struct ring_buffer_per_cpu
*cpu_buffer
)
2820 if (buffer
->irq_work
.waiters_pending
) {
2821 buffer
->irq_work
.waiters_pending
= false;
2822 /* irq_work_queue() supplies it's own memory barriers */
2823 irq_work_queue(&buffer
->irq_work
.work
);
2826 if (cpu_buffer
->irq_work
.waiters_pending
) {
2827 cpu_buffer
->irq_work
.waiters_pending
= false;
2828 /* irq_work_queue() supplies it's own memory barriers */
2829 irq_work_queue(&cpu_buffer
->irq_work
.work
);
2834 * ring_buffer_unlock_commit - commit a reserved
2835 * @buffer: The buffer to commit to
2836 * @event: The event pointer to commit.
2838 * This commits the data to the ring buffer, and releases any locks held.
2840 * Must be paired with ring_buffer_lock_reserve.
2842 int ring_buffer_unlock_commit(struct ring_buffer
*buffer
,
2843 struct ring_buffer_event
*event
)
2845 struct ring_buffer_per_cpu
*cpu_buffer
;
2846 int cpu
= raw_smp_processor_id();
2848 cpu_buffer
= buffer
->buffers
[cpu
];
2850 rb_commit(cpu_buffer
, event
);
2852 rb_wakeups(buffer
, cpu_buffer
);
2854 trace_recursive_unlock();
2856 preempt_enable_notrace();
2860 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit
);
2862 static inline void rb_event_discard(struct ring_buffer_event
*event
)
2864 if (event
->type_len
== RINGBUF_TYPE_TIME_EXTEND
)
2865 event
= skip_time_extend(event
);
2867 /* array[0] holds the actual length for the discarded event */
2868 event
->array
[0] = rb_event_data_length(event
) - RB_EVNT_HDR_SIZE
;
2869 event
->type_len
= RINGBUF_TYPE_PADDING
;
2870 /* time delta must be non zero */
2871 if (!event
->time_delta
)
2872 event
->time_delta
= 1;
2876 * Decrement the entries to the page that an event is on.
2877 * The event does not even need to exist, only the pointer
2878 * to the page it is on. This may only be called before the commit
2882 rb_decrement_entry(struct ring_buffer_per_cpu
*cpu_buffer
,
2883 struct ring_buffer_event
*event
)
2885 unsigned long addr
= (unsigned long)event
;
2886 struct buffer_page
*bpage
= cpu_buffer
->commit_page
;
2887 struct buffer_page
*start
;
2891 /* Do the likely case first */
2892 if (likely(bpage
->page
== (void *)addr
)) {
2893 local_dec(&bpage
->entries
);
2898 * Because the commit page may be on the reader page we
2899 * start with the next page and check the end loop there.
2901 rb_inc_page(cpu_buffer
, &bpage
);
2904 if (bpage
->page
== (void *)addr
) {
2905 local_dec(&bpage
->entries
);
2908 rb_inc_page(cpu_buffer
, &bpage
);
2909 } while (bpage
!= start
);
2911 /* commit not part of this buffer?? */
2912 RB_WARN_ON(cpu_buffer
, 1);
2916 * ring_buffer_commit_discard - discard an event that has not been committed
2917 * @buffer: the ring buffer
2918 * @event: non committed event to discard
2920 * Sometimes an event that is in the ring buffer needs to be ignored.
2921 * This function lets the user discard an event in the ring buffer
2922 * and then that event will not be read later.
2924 * This function only works if it is called before the the item has been
2925 * committed. It will try to free the event from the ring buffer
2926 * if another event has not been added behind it.
2928 * If another event has been added behind it, it will set the event
2929 * up as discarded, and perform the commit.
2931 * If this function is called, do not call ring_buffer_unlock_commit on
2934 void ring_buffer_discard_commit(struct ring_buffer
*buffer
,
2935 struct ring_buffer_event
*event
)
2937 struct ring_buffer_per_cpu
*cpu_buffer
;
2940 /* The event is discarded regardless */
2941 rb_event_discard(event
);
2943 cpu
= smp_processor_id();
2944 cpu_buffer
= buffer
->buffers
[cpu
];
2947 * This must only be called if the event has not been
2948 * committed yet. Thus we can assume that preemption
2949 * is still disabled.
2951 RB_WARN_ON(buffer
, !local_read(&cpu_buffer
->committing
));
2953 rb_decrement_entry(cpu_buffer
, event
);
2954 if (rb_try_to_discard(cpu_buffer
, event
))
2958 * The commit is still visible by the reader, so we
2959 * must still update the timestamp.
2961 rb_update_write_stamp(cpu_buffer
, event
);
2963 rb_end_commit(cpu_buffer
);
2965 trace_recursive_unlock();
2967 preempt_enable_notrace();
2970 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit
);
2973 * ring_buffer_write - write data to the buffer without reserving
2974 * @buffer: The ring buffer to write to.
2975 * @length: The length of the data being written (excluding the event header)
2976 * @data: The data to write to the buffer.
2978 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
2979 * one function. If you already have the data to write to the buffer, it
2980 * may be easier to simply call this function.
2982 * Note, like ring_buffer_lock_reserve, the length is the length of the data
2983 * and not the length of the event which would hold the header.
2985 int ring_buffer_write(struct ring_buffer
*buffer
,
2986 unsigned long length
,
2989 struct ring_buffer_per_cpu
*cpu_buffer
;
2990 struct ring_buffer_event
*event
;
2995 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
2998 preempt_disable_notrace();
3000 if (atomic_read(&buffer
->record_disabled
))
3003 cpu
= raw_smp_processor_id();
3005 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3008 cpu_buffer
= buffer
->buffers
[cpu
];
3010 if (atomic_read(&cpu_buffer
->record_disabled
))
3013 if (length
> BUF_MAX_DATA_SIZE
)
3016 event
= rb_reserve_next_event(buffer
, cpu_buffer
, length
);
3020 body
= rb_event_data(event
);
3022 memcpy(body
, data
, length
);
3024 rb_commit(cpu_buffer
, event
);
3026 rb_wakeups(buffer
, cpu_buffer
);
3030 preempt_enable_notrace();
3034 EXPORT_SYMBOL_GPL(ring_buffer_write
);
3036 static int rb_per_cpu_empty(struct ring_buffer_per_cpu
*cpu_buffer
)
3038 struct buffer_page
*reader
= cpu_buffer
->reader_page
;
3039 struct buffer_page
*head
= rb_set_head_page(cpu_buffer
);
3040 struct buffer_page
*commit
= cpu_buffer
->commit_page
;
3042 /* In case of error, head will be NULL */
3043 if (unlikely(!head
))
3046 return reader
->read
== rb_page_commit(reader
) &&
3047 (commit
== reader
||
3049 head
->read
== rb_page_commit(commit
)));
3053 * ring_buffer_record_disable - stop all writes into the buffer
3054 * @buffer: The ring buffer to stop writes to.
3056 * This prevents all writes to the buffer. Any attempt to write
3057 * to the buffer after this will fail and return NULL.
3059 * The caller should call synchronize_sched() after this.
3061 void ring_buffer_record_disable(struct ring_buffer
*buffer
)
3063 atomic_inc(&buffer
->record_disabled
);
3065 EXPORT_SYMBOL_GPL(ring_buffer_record_disable
);
3068 * ring_buffer_record_enable - enable writes to the buffer
3069 * @buffer: The ring buffer to enable writes
3071 * Note, multiple disables will need the same number of enables
3072 * to truly enable the writing (much like preempt_disable).
3074 void ring_buffer_record_enable(struct ring_buffer
*buffer
)
3076 atomic_dec(&buffer
->record_disabled
);
3078 EXPORT_SYMBOL_GPL(ring_buffer_record_enable
);
3081 * ring_buffer_record_off - stop all writes into the buffer
3082 * @buffer: The ring buffer to stop writes to.
3084 * This prevents all writes to the buffer. Any attempt to write
3085 * to the buffer after this will fail and return NULL.
3087 * This is different than ring_buffer_record_disable() as
3088 * it works like an on/off switch, where as the disable() version
3089 * must be paired with a enable().
3091 void ring_buffer_record_off(struct ring_buffer
*buffer
)
3094 unsigned int new_rd
;
3097 rd
= atomic_read(&buffer
->record_disabled
);
3098 new_rd
= rd
| RB_BUFFER_OFF
;
3099 } while (atomic_cmpxchg(&buffer
->record_disabled
, rd
, new_rd
) != rd
);
3101 EXPORT_SYMBOL_GPL(ring_buffer_record_off
);
3104 * ring_buffer_record_on - restart writes into the buffer
3105 * @buffer: The ring buffer to start writes to.
3107 * This enables all writes to the buffer that was disabled by
3108 * ring_buffer_record_off().
3110 * This is different than ring_buffer_record_enable() as
3111 * it works like an on/off switch, where as the enable() version
3112 * must be paired with a disable().
3114 void ring_buffer_record_on(struct ring_buffer
*buffer
)
3117 unsigned int new_rd
;
3120 rd
= atomic_read(&buffer
->record_disabled
);
3121 new_rd
= rd
& ~RB_BUFFER_OFF
;
3122 } while (atomic_cmpxchg(&buffer
->record_disabled
, rd
, new_rd
) != rd
);
3124 EXPORT_SYMBOL_GPL(ring_buffer_record_on
);
3127 * ring_buffer_record_is_on - return true if the ring buffer can write
3128 * @buffer: The ring buffer to see if write is enabled
3130 * Returns true if the ring buffer is in a state that it accepts writes.
3132 int ring_buffer_record_is_on(struct ring_buffer
*buffer
)
3134 return !atomic_read(&buffer
->record_disabled
);
3138 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
3139 * @buffer: The ring buffer to stop writes to.
3140 * @cpu: The CPU buffer to stop
3142 * This prevents all writes to the buffer. Any attempt to write
3143 * to the buffer after this will fail and return NULL.
3145 * The caller should call synchronize_sched() after this.
3147 void ring_buffer_record_disable_cpu(struct ring_buffer
*buffer
, int cpu
)
3149 struct ring_buffer_per_cpu
*cpu_buffer
;
3151 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3154 cpu_buffer
= buffer
->buffers
[cpu
];
3155 atomic_inc(&cpu_buffer
->record_disabled
);
3157 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu
);
3160 * ring_buffer_record_enable_cpu - enable writes to the buffer
3161 * @buffer: The ring buffer to enable writes
3162 * @cpu: The CPU to enable.
3164 * Note, multiple disables will need the same number of enables
3165 * to truly enable the writing (much like preempt_disable).
3167 void ring_buffer_record_enable_cpu(struct ring_buffer
*buffer
, int cpu
)
3169 struct ring_buffer_per_cpu
*cpu_buffer
;
3171 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3174 cpu_buffer
= buffer
->buffers
[cpu
];
3175 atomic_dec(&cpu_buffer
->record_disabled
);
3177 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu
);
3180 * The total entries in the ring buffer is the running counter
3181 * of entries entered into the ring buffer, minus the sum of
3182 * the entries read from the ring buffer and the number of
3183 * entries that were overwritten.
3185 static inline unsigned long
3186 rb_num_of_entries(struct ring_buffer_per_cpu
*cpu_buffer
)
3188 return local_read(&cpu_buffer
->entries
) -
3189 (local_read(&cpu_buffer
->overrun
) + cpu_buffer
->read
);
3193 * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer
3194 * @buffer: The ring buffer
3195 * @cpu: The per CPU buffer to read from.
3197 u64
ring_buffer_oldest_event_ts(struct ring_buffer
*buffer
, int cpu
)
3199 unsigned long flags
;
3200 struct ring_buffer_per_cpu
*cpu_buffer
;
3201 struct buffer_page
*bpage
;
3204 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3207 cpu_buffer
= buffer
->buffers
[cpu
];
3208 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
3210 * if the tail is on reader_page, oldest time stamp is on the reader
3213 if (cpu_buffer
->tail_page
== cpu_buffer
->reader_page
)
3214 bpage
= cpu_buffer
->reader_page
;
3216 bpage
= rb_set_head_page(cpu_buffer
);
3218 ret
= bpage
->page
->time_stamp
;
3219 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
3223 EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts
);
3226 * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer
3227 * @buffer: The ring buffer
3228 * @cpu: The per CPU buffer to read from.
3230 unsigned long ring_buffer_bytes_cpu(struct ring_buffer
*buffer
, int cpu
)
3232 struct ring_buffer_per_cpu
*cpu_buffer
;
3235 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3238 cpu_buffer
= buffer
->buffers
[cpu
];
3239 ret
= local_read(&cpu_buffer
->entries_bytes
) - cpu_buffer
->read_bytes
;
3243 EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu
);
3246 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
3247 * @buffer: The ring buffer
3248 * @cpu: The per CPU buffer to get the entries from.
3250 unsigned long ring_buffer_entries_cpu(struct ring_buffer
*buffer
, int cpu
)
3252 struct ring_buffer_per_cpu
*cpu_buffer
;
3254 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3257 cpu_buffer
= buffer
->buffers
[cpu
];
3259 return rb_num_of_entries(cpu_buffer
);
3261 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu
);
3264 * ring_buffer_overrun_cpu - get the number of overruns caused by the ring
3265 * buffer wrapping around (only if RB_FL_OVERWRITE is on).
3266 * @buffer: The ring buffer
3267 * @cpu: The per CPU buffer to get the number of overruns from
3269 unsigned long ring_buffer_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
3271 struct ring_buffer_per_cpu
*cpu_buffer
;
3274 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3277 cpu_buffer
= buffer
->buffers
[cpu
];
3278 ret
= local_read(&cpu_buffer
->overrun
);
3282 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu
);
3285 * ring_buffer_commit_overrun_cpu - get the number of overruns caused by
3286 * commits failing due to the buffer wrapping around while there are uncommitted
3287 * events, such as during an interrupt storm.
3288 * @buffer: The ring buffer
3289 * @cpu: The per CPU buffer to get the number of overruns from
3292 ring_buffer_commit_overrun_cpu(struct ring_buffer
*buffer
, int cpu
)
3294 struct ring_buffer_per_cpu
*cpu_buffer
;
3297 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3300 cpu_buffer
= buffer
->buffers
[cpu
];
3301 ret
= local_read(&cpu_buffer
->commit_overrun
);
3305 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu
);
3308 * ring_buffer_dropped_events_cpu - get the number of dropped events caused by
3309 * the ring buffer filling up (only if RB_FL_OVERWRITE is off).
3310 * @buffer: The ring buffer
3311 * @cpu: The per CPU buffer to get the number of overruns from
3314 ring_buffer_dropped_events_cpu(struct ring_buffer
*buffer
, int cpu
)
3316 struct ring_buffer_per_cpu
*cpu_buffer
;
3319 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3322 cpu_buffer
= buffer
->buffers
[cpu
];
3323 ret
= local_read(&cpu_buffer
->dropped_events
);
3327 EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu
);
3330 * ring_buffer_read_events_cpu - get the number of events successfully read
3331 * @buffer: The ring buffer
3332 * @cpu: The per CPU buffer to get the number of events read
3335 ring_buffer_read_events_cpu(struct ring_buffer
*buffer
, int cpu
)
3337 struct ring_buffer_per_cpu
*cpu_buffer
;
3339 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3342 cpu_buffer
= buffer
->buffers
[cpu
];
3343 return cpu_buffer
->read
;
3345 EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu
);
3348 * ring_buffer_entries - get the number of entries in a buffer
3349 * @buffer: The ring buffer
3351 * Returns the total number of entries in the ring buffer
3354 unsigned long ring_buffer_entries(struct ring_buffer
*buffer
)
3356 struct ring_buffer_per_cpu
*cpu_buffer
;
3357 unsigned long entries
= 0;
3360 /* if you care about this being correct, lock the buffer */
3361 for_each_buffer_cpu(buffer
, cpu
) {
3362 cpu_buffer
= buffer
->buffers
[cpu
];
3363 entries
+= rb_num_of_entries(cpu_buffer
);
3368 EXPORT_SYMBOL_GPL(ring_buffer_entries
);
3371 * ring_buffer_overruns - get the number of overruns in buffer
3372 * @buffer: The ring buffer
3374 * Returns the total number of overruns in the ring buffer
3377 unsigned long ring_buffer_overruns(struct ring_buffer
*buffer
)
3379 struct ring_buffer_per_cpu
*cpu_buffer
;
3380 unsigned long overruns
= 0;
3383 /* if you care about this being correct, lock the buffer */
3384 for_each_buffer_cpu(buffer
, cpu
) {
3385 cpu_buffer
= buffer
->buffers
[cpu
];
3386 overruns
+= local_read(&cpu_buffer
->overrun
);
3391 EXPORT_SYMBOL_GPL(ring_buffer_overruns
);
3393 static void rb_iter_reset(struct ring_buffer_iter
*iter
)
3395 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
3397 /* Iterator usage is expected to have record disabled */
3398 iter
->head_page
= cpu_buffer
->reader_page
;
3399 iter
->head
= cpu_buffer
->reader_page
->read
;
3401 iter
->cache_reader_page
= iter
->head_page
;
3402 iter
->cache_read
= iter
->head
;
3405 iter
->read_stamp
= cpu_buffer
->read_stamp
;
3407 iter
->read_stamp
= iter
->head_page
->page
->time_stamp
;
3411 * ring_buffer_iter_reset - reset an iterator
3412 * @iter: The iterator to reset
3414 * Resets the iterator, so that it will start from the beginning
3417 void ring_buffer_iter_reset(struct ring_buffer_iter
*iter
)
3419 struct ring_buffer_per_cpu
*cpu_buffer
;
3420 unsigned long flags
;
3425 cpu_buffer
= iter
->cpu_buffer
;
3427 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
3428 rb_iter_reset(iter
);
3429 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
3431 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset
);
3434 * ring_buffer_iter_empty - check if an iterator has no more to read
3435 * @iter: The iterator to check
3437 int ring_buffer_iter_empty(struct ring_buffer_iter
*iter
)
3439 struct ring_buffer_per_cpu
*cpu_buffer
;
3441 cpu_buffer
= iter
->cpu_buffer
;
3443 return iter
->head_page
== cpu_buffer
->commit_page
&&
3444 iter
->head
== rb_commit_index(cpu_buffer
);
3446 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty
);
3449 rb_update_read_stamp(struct ring_buffer_per_cpu
*cpu_buffer
,
3450 struct ring_buffer_event
*event
)
3454 switch (event
->type_len
) {
3455 case RINGBUF_TYPE_PADDING
:
3458 case RINGBUF_TYPE_TIME_EXTEND
:
3459 delta
= event
->array
[0];
3461 delta
+= event
->time_delta
;
3462 cpu_buffer
->read_stamp
+= delta
;
3465 case RINGBUF_TYPE_TIME_STAMP
:
3466 /* FIXME: not implemented */
3469 case RINGBUF_TYPE_DATA
:
3470 cpu_buffer
->read_stamp
+= event
->time_delta
;
3480 rb_update_iter_read_stamp(struct ring_buffer_iter
*iter
,
3481 struct ring_buffer_event
*event
)
3485 switch (event
->type_len
) {
3486 case RINGBUF_TYPE_PADDING
:
3489 case RINGBUF_TYPE_TIME_EXTEND
:
3490 delta
= event
->array
[0];
3492 delta
+= event
->time_delta
;
3493 iter
->read_stamp
+= delta
;
3496 case RINGBUF_TYPE_TIME_STAMP
:
3497 /* FIXME: not implemented */
3500 case RINGBUF_TYPE_DATA
:
3501 iter
->read_stamp
+= event
->time_delta
;
3510 static struct buffer_page
*
3511 rb_get_reader_page(struct ring_buffer_per_cpu
*cpu_buffer
)
3513 struct buffer_page
*reader
= NULL
;
3514 unsigned long overwrite
;
3515 unsigned long flags
;
3519 local_irq_save(flags
);
3520 arch_spin_lock(&cpu_buffer
->lock
);
3524 * This should normally only loop twice. But because the
3525 * start of the reader inserts an empty page, it causes
3526 * a case where we will loop three times. There should be no
3527 * reason to loop four times (that I know of).
3529 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 3)) {
3534 reader
= cpu_buffer
->reader_page
;
3536 /* If there's more to read, return this page */
3537 if (cpu_buffer
->reader_page
->read
< rb_page_size(reader
))
3540 /* Never should we have an index greater than the size */
3541 if (RB_WARN_ON(cpu_buffer
,
3542 cpu_buffer
->reader_page
->read
> rb_page_size(reader
)))
3545 /* check if we caught up to the tail */
3547 if (cpu_buffer
->commit_page
== cpu_buffer
->reader_page
)
3550 /* Don't bother swapping if the ring buffer is empty */
3551 if (rb_num_of_entries(cpu_buffer
) == 0)
3555 * Reset the reader page to size zero.
3557 local_set(&cpu_buffer
->reader_page
->write
, 0);
3558 local_set(&cpu_buffer
->reader_page
->entries
, 0);
3559 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
3560 cpu_buffer
->reader_page
->real_end
= 0;
3564 * Splice the empty reader page into the list around the head.
3566 reader
= rb_set_head_page(cpu_buffer
);
3569 cpu_buffer
->reader_page
->list
.next
= rb_list_head(reader
->list
.next
);
3570 cpu_buffer
->reader_page
->list
.prev
= reader
->list
.prev
;
3573 * cpu_buffer->pages just needs to point to the buffer, it
3574 * has no specific buffer page to point to. Lets move it out
3575 * of our way so we don't accidentally swap it.
3577 cpu_buffer
->pages
= reader
->list
.prev
;
3579 /* The reader page will be pointing to the new head */
3580 rb_set_list_to_head(cpu_buffer
, &cpu_buffer
->reader_page
->list
);
3583 * We want to make sure we read the overruns after we set up our
3584 * pointers to the next object. The writer side does a
3585 * cmpxchg to cross pages which acts as the mb on the writer
3586 * side. Note, the reader will constantly fail the swap
3587 * while the writer is updating the pointers, so this
3588 * guarantees that the overwrite recorded here is the one we
3589 * want to compare with the last_overrun.
3592 overwrite
= local_read(&(cpu_buffer
->overrun
));
3595 * Here's the tricky part.
3597 * We need to move the pointer past the header page.
3598 * But we can only do that if a writer is not currently
3599 * moving it. The page before the header page has the
3600 * flag bit '1' set if it is pointing to the page we want.
3601 * but if the writer is in the process of moving it
3602 * than it will be '2' or already moved '0'.
3605 ret
= rb_head_page_replace(reader
, cpu_buffer
->reader_page
);
3608 * If we did not convert it, then we must try again.
3614 * Yeah! We succeeded in replacing the page.
3616 * Now make the new head point back to the reader page.
3618 rb_list_head(reader
->list
.next
)->prev
= &cpu_buffer
->reader_page
->list
;
3619 rb_inc_page(cpu_buffer
, &cpu_buffer
->head_page
);
3621 /* Finally update the reader page to the new head */
3622 cpu_buffer
->reader_page
= reader
;
3623 rb_reset_reader_page(cpu_buffer
);
3625 if (overwrite
!= cpu_buffer
->last_overrun
) {
3626 cpu_buffer
->lost_events
= overwrite
- cpu_buffer
->last_overrun
;
3627 cpu_buffer
->last_overrun
= overwrite
;
3633 arch_spin_unlock(&cpu_buffer
->lock
);
3634 local_irq_restore(flags
);
3639 static void rb_advance_reader(struct ring_buffer_per_cpu
*cpu_buffer
)
3641 struct ring_buffer_event
*event
;
3642 struct buffer_page
*reader
;
3645 reader
= rb_get_reader_page(cpu_buffer
);
3647 /* This function should not be called when buffer is empty */
3648 if (RB_WARN_ON(cpu_buffer
, !reader
))
3651 event
= rb_reader_event(cpu_buffer
);
3653 if (event
->type_len
<= RINGBUF_TYPE_DATA_TYPE_LEN_MAX
)
3656 rb_update_read_stamp(cpu_buffer
, event
);
3658 length
= rb_event_length(event
);
3659 cpu_buffer
->reader_page
->read
+= length
;
3662 static void rb_advance_iter(struct ring_buffer_iter
*iter
)
3664 struct ring_buffer_per_cpu
*cpu_buffer
;
3665 struct ring_buffer_event
*event
;
3668 cpu_buffer
= iter
->cpu_buffer
;
3671 * Check if we are at the end of the buffer.
3673 if (iter
->head
>= rb_page_size(iter
->head_page
)) {
3674 /* discarded commits can make the page empty */
3675 if (iter
->head_page
== cpu_buffer
->commit_page
)
3681 event
= rb_iter_head_event(iter
);
3683 length
= rb_event_length(event
);
3686 * This should not be called to advance the header if we are
3687 * at the tail of the buffer.
3689 if (RB_WARN_ON(cpu_buffer
,
3690 (iter
->head_page
== cpu_buffer
->commit_page
) &&
3691 (iter
->head
+ length
> rb_commit_index(cpu_buffer
))))
3694 rb_update_iter_read_stamp(iter
, event
);
3696 iter
->head
+= length
;
3698 /* check for end of page padding */
3699 if ((iter
->head
>= rb_page_size(iter
->head_page
)) &&
3700 (iter
->head_page
!= cpu_buffer
->commit_page
))
3704 static int rb_lost_events(struct ring_buffer_per_cpu
*cpu_buffer
)
3706 return cpu_buffer
->lost_events
;
3709 static struct ring_buffer_event
*
3710 rb_buffer_peek(struct ring_buffer_per_cpu
*cpu_buffer
, u64
*ts
,
3711 unsigned long *lost_events
)
3713 struct ring_buffer_event
*event
;
3714 struct buffer_page
*reader
;
3719 * We repeat when a time extend is encountered.
3720 * Since the time extend is always attached to a data event,
3721 * we should never loop more than once.
3722 * (We never hit the following condition more than twice).
3724 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 2))
3727 reader
= rb_get_reader_page(cpu_buffer
);
3731 event
= rb_reader_event(cpu_buffer
);
3733 switch (event
->type_len
) {
3734 case RINGBUF_TYPE_PADDING
:
3735 if (rb_null_event(event
))
3736 RB_WARN_ON(cpu_buffer
, 1);
3738 * Because the writer could be discarding every
3739 * event it creates (which would probably be bad)
3740 * if we were to go back to "again" then we may never
3741 * catch up, and will trigger the warn on, or lock
3742 * the box. Return the padding, and we will release
3743 * the current locks, and try again.
3747 case RINGBUF_TYPE_TIME_EXTEND
:
3748 /* Internal data, OK to advance */
3749 rb_advance_reader(cpu_buffer
);
3752 case RINGBUF_TYPE_TIME_STAMP
:
3753 /* FIXME: not implemented */
3754 rb_advance_reader(cpu_buffer
);
3757 case RINGBUF_TYPE_DATA
:
3759 *ts
= cpu_buffer
->read_stamp
+ event
->time_delta
;
3760 ring_buffer_normalize_time_stamp(cpu_buffer
->buffer
,
3761 cpu_buffer
->cpu
, ts
);
3764 *lost_events
= rb_lost_events(cpu_buffer
);
3773 EXPORT_SYMBOL_GPL(ring_buffer_peek
);
3775 static struct ring_buffer_event
*
3776 rb_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
3778 struct ring_buffer
*buffer
;
3779 struct ring_buffer_per_cpu
*cpu_buffer
;
3780 struct ring_buffer_event
*event
;
3783 cpu_buffer
= iter
->cpu_buffer
;
3784 buffer
= cpu_buffer
->buffer
;
3787 * Check if someone performed a consuming read to
3788 * the buffer. A consuming read invalidates the iterator
3789 * and we need to reset the iterator in this case.
3791 if (unlikely(iter
->cache_read
!= cpu_buffer
->read
||
3792 iter
->cache_reader_page
!= cpu_buffer
->reader_page
))
3793 rb_iter_reset(iter
);
3796 if (ring_buffer_iter_empty(iter
))
3800 * We repeat when a time extend is encountered or we hit
3801 * the end of the page. Since the time extend is always attached
3802 * to a data event, we should never loop more than three times.
3803 * Once for going to next page, once on time extend, and
3804 * finally once to get the event.
3805 * (We never hit the following condition more than thrice).
3807 if (RB_WARN_ON(cpu_buffer
, ++nr_loops
> 3))
3810 if (rb_per_cpu_empty(cpu_buffer
))
3813 if (iter
->head
>= local_read(&iter
->head_page
->page
->commit
)) {
3818 event
= rb_iter_head_event(iter
);
3820 switch (event
->type_len
) {
3821 case RINGBUF_TYPE_PADDING
:
3822 if (rb_null_event(event
)) {
3826 rb_advance_iter(iter
);
3829 case RINGBUF_TYPE_TIME_EXTEND
:
3830 /* Internal data, OK to advance */
3831 rb_advance_iter(iter
);
3834 case RINGBUF_TYPE_TIME_STAMP
:
3835 /* FIXME: not implemented */
3836 rb_advance_iter(iter
);
3839 case RINGBUF_TYPE_DATA
:
3841 *ts
= iter
->read_stamp
+ event
->time_delta
;
3842 ring_buffer_normalize_time_stamp(buffer
,
3843 cpu_buffer
->cpu
, ts
);
3853 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek
);
3855 static inline int rb_ok_to_lock(void)
3858 * If an NMI die dumps out the content of the ring buffer
3859 * do not grab locks. We also permanently disable the ring
3860 * buffer too. A one time deal is all you get from reading
3861 * the ring buffer from an NMI.
3863 if (likely(!in_nmi()))
3866 tracing_off_permanent();
3871 * ring_buffer_peek - peek at the next event to be read
3872 * @buffer: The ring buffer to read
3873 * @cpu: The cpu to peak at
3874 * @ts: The timestamp counter of this event.
3875 * @lost_events: a variable to store if events were lost (may be NULL)
3877 * This will return the event that will be read next, but does
3878 * not consume the data.
3880 struct ring_buffer_event
*
3881 ring_buffer_peek(struct ring_buffer
*buffer
, int cpu
, u64
*ts
,
3882 unsigned long *lost_events
)
3884 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
3885 struct ring_buffer_event
*event
;
3886 unsigned long flags
;
3889 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3892 dolock
= rb_ok_to_lock();
3894 local_irq_save(flags
);
3896 raw_spin_lock(&cpu_buffer
->reader_lock
);
3897 event
= rb_buffer_peek(cpu_buffer
, ts
, lost_events
);
3898 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
)
3899 rb_advance_reader(cpu_buffer
);
3901 raw_spin_unlock(&cpu_buffer
->reader_lock
);
3902 local_irq_restore(flags
);
3904 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
)
3911 * ring_buffer_iter_peek - peek at the next event to be read
3912 * @iter: The ring buffer iterator
3913 * @ts: The timestamp counter of this event.
3915 * This will return the event that will be read next, but does
3916 * not increment the iterator.
3918 struct ring_buffer_event
*
3919 ring_buffer_iter_peek(struct ring_buffer_iter
*iter
, u64
*ts
)
3921 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
3922 struct ring_buffer_event
*event
;
3923 unsigned long flags
;
3926 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
3927 event
= rb_iter_peek(iter
, ts
);
3928 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
3930 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
)
3937 * ring_buffer_consume - return an event and consume it
3938 * @buffer: The ring buffer to get the next event from
3939 * @cpu: the cpu to read the buffer from
3940 * @ts: a variable to store the timestamp (may be NULL)
3941 * @lost_events: a variable to store if events were lost (may be NULL)
3943 * Returns the next event in the ring buffer, and that event is consumed.
3944 * Meaning, that sequential reads will keep returning a different event,
3945 * and eventually empty the ring buffer if the producer is slower.
3947 struct ring_buffer_event
*
3948 ring_buffer_consume(struct ring_buffer
*buffer
, int cpu
, u64
*ts
,
3949 unsigned long *lost_events
)
3951 struct ring_buffer_per_cpu
*cpu_buffer
;
3952 struct ring_buffer_event
*event
= NULL
;
3953 unsigned long flags
;
3956 dolock
= rb_ok_to_lock();
3959 /* might be called in atomic */
3962 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
3965 cpu_buffer
= buffer
->buffers
[cpu
];
3966 local_irq_save(flags
);
3968 raw_spin_lock(&cpu_buffer
->reader_lock
);
3970 event
= rb_buffer_peek(cpu_buffer
, ts
, lost_events
);
3972 cpu_buffer
->lost_events
= 0;
3973 rb_advance_reader(cpu_buffer
);
3977 raw_spin_unlock(&cpu_buffer
->reader_lock
);
3978 local_irq_restore(flags
);
3983 if (event
&& event
->type_len
== RINGBUF_TYPE_PADDING
)
3988 EXPORT_SYMBOL_GPL(ring_buffer_consume
);
3991 * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer
3992 * @buffer: The ring buffer to read from
3993 * @cpu: The cpu buffer to iterate over
3995 * This performs the initial preparations necessary to iterate
3996 * through the buffer. Memory is allocated, buffer recording
3997 * is disabled, and the iterator pointer is returned to the caller.
3999 * Disabling buffer recordng prevents the reading from being
4000 * corrupted. This is not a consuming read, so a producer is not
4003 * After a sequence of ring_buffer_read_prepare calls, the user is
4004 * expected to make at least one call to ring_buffer_prepare_sync.
4005 * Afterwards, ring_buffer_read_start is invoked to get things going
4008 * This overall must be paired with ring_buffer_finish.
4010 struct ring_buffer_iter
*
4011 ring_buffer_read_prepare(struct ring_buffer
*buffer
, int cpu
)
4013 struct ring_buffer_per_cpu
*cpu_buffer
;
4014 struct ring_buffer_iter
*iter
;
4016 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
4019 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
4023 cpu_buffer
= buffer
->buffers
[cpu
];
4025 iter
->cpu_buffer
= cpu_buffer
;
4027 atomic_inc(&buffer
->resize_disabled
);
4028 atomic_inc(&cpu_buffer
->record_disabled
);
4032 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare
);
4035 * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls
4037 * All previously invoked ring_buffer_read_prepare calls to prepare
4038 * iterators will be synchronized. Afterwards, read_buffer_read_start
4039 * calls on those iterators are allowed.
4042 ring_buffer_read_prepare_sync(void)
4044 synchronize_sched();
4046 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync
);
4049 * ring_buffer_read_start - start a non consuming read of the buffer
4050 * @iter: The iterator returned by ring_buffer_read_prepare
4052 * This finalizes the startup of an iteration through the buffer.
4053 * The iterator comes from a call to ring_buffer_read_prepare and
4054 * an intervening ring_buffer_read_prepare_sync must have been
4057 * Must be paired with ring_buffer_finish.
4060 ring_buffer_read_start(struct ring_buffer_iter
*iter
)
4062 struct ring_buffer_per_cpu
*cpu_buffer
;
4063 unsigned long flags
;
4068 cpu_buffer
= iter
->cpu_buffer
;
4070 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
4071 arch_spin_lock(&cpu_buffer
->lock
);
4072 rb_iter_reset(iter
);
4073 arch_spin_unlock(&cpu_buffer
->lock
);
4074 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
4076 EXPORT_SYMBOL_GPL(ring_buffer_read_start
);
4079 * ring_buffer_finish - finish reading the iterator of the buffer
4080 * @iter: The iterator retrieved by ring_buffer_start
4082 * This re-enables the recording to the buffer, and frees the
4086 ring_buffer_read_finish(struct ring_buffer_iter
*iter
)
4088 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
4089 unsigned long flags
;
4092 * Ring buffer is disabled from recording, here's a good place
4093 * to check the integrity of the ring buffer.
4094 * Must prevent readers from trying to read, as the check
4095 * clears the HEAD page and readers require it.
4097 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
4098 rb_check_pages(cpu_buffer
);
4099 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
4101 atomic_dec(&cpu_buffer
->record_disabled
);
4102 atomic_dec(&cpu_buffer
->buffer
->resize_disabled
);
4105 EXPORT_SYMBOL_GPL(ring_buffer_read_finish
);
4108 * ring_buffer_read - read the next item in the ring buffer by the iterator
4109 * @iter: The ring buffer iterator
4110 * @ts: The time stamp of the event read.
4112 * This reads the next event in the ring buffer and increments the iterator.
4114 struct ring_buffer_event
*
4115 ring_buffer_read(struct ring_buffer_iter
*iter
, u64
*ts
)
4117 struct ring_buffer_event
*event
;
4118 struct ring_buffer_per_cpu
*cpu_buffer
= iter
->cpu_buffer
;
4119 unsigned long flags
;
4121 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
4123 event
= rb_iter_peek(iter
, ts
);
4127 if (event
->type_len
== RINGBUF_TYPE_PADDING
)
4130 rb_advance_iter(iter
);
4132 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
4136 EXPORT_SYMBOL_GPL(ring_buffer_read
);
4139 * ring_buffer_size - return the size of the ring buffer (in bytes)
4140 * @buffer: The ring buffer.
4142 unsigned long ring_buffer_size(struct ring_buffer
*buffer
, int cpu
)
4145 * Earlier, this method returned
4146 * BUF_PAGE_SIZE * buffer->nr_pages
4147 * Since the nr_pages field is now removed, we have converted this to
4148 * return the per cpu buffer value.
4150 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
4153 return BUF_PAGE_SIZE
* buffer
->buffers
[cpu
]->nr_pages
;
4155 EXPORT_SYMBOL_GPL(ring_buffer_size
);
4158 rb_reset_cpu(struct ring_buffer_per_cpu
*cpu_buffer
)
4160 rb_head_page_deactivate(cpu_buffer
);
4162 cpu_buffer
->head_page
4163 = list_entry(cpu_buffer
->pages
, struct buffer_page
, list
);
4164 local_set(&cpu_buffer
->head_page
->write
, 0);
4165 local_set(&cpu_buffer
->head_page
->entries
, 0);
4166 local_set(&cpu_buffer
->head_page
->page
->commit
, 0);
4168 cpu_buffer
->head_page
->read
= 0;
4170 cpu_buffer
->tail_page
= cpu_buffer
->head_page
;
4171 cpu_buffer
->commit_page
= cpu_buffer
->head_page
;
4173 INIT_LIST_HEAD(&cpu_buffer
->reader_page
->list
);
4174 INIT_LIST_HEAD(&cpu_buffer
->new_pages
);
4175 local_set(&cpu_buffer
->reader_page
->write
, 0);
4176 local_set(&cpu_buffer
->reader_page
->entries
, 0);
4177 local_set(&cpu_buffer
->reader_page
->page
->commit
, 0);
4178 cpu_buffer
->reader_page
->read
= 0;
4180 local_set(&cpu_buffer
->entries_bytes
, 0);
4181 local_set(&cpu_buffer
->overrun
, 0);
4182 local_set(&cpu_buffer
->commit_overrun
, 0);
4183 local_set(&cpu_buffer
->dropped_events
, 0);
4184 local_set(&cpu_buffer
->entries
, 0);
4185 local_set(&cpu_buffer
->committing
, 0);
4186 local_set(&cpu_buffer
->commits
, 0);
4187 cpu_buffer
->read
= 0;
4188 cpu_buffer
->read_bytes
= 0;
4190 cpu_buffer
->write_stamp
= 0;
4191 cpu_buffer
->read_stamp
= 0;
4193 cpu_buffer
->lost_events
= 0;
4194 cpu_buffer
->last_overrun
= 0;
4196 rb_head_page_activate(cpu_buffer
);
4200 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
4201 * @buffer: The ring buffer to reset a per cpu buffer of
4202 * @cpu: The CPU buffer to be reset
4204 void ring_buffer_reset_cpu(struct ring_buffer
*buffer
, int cpu
)
4206 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
4207 unsigned long flags
;
4209 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
4212 atomic_inc(&buffer
->resize_disabled
);
4213 atomic_inc(&cpu_buffer
->record_disabled
);
4215 /* Make sure all commits have finished */
4216 synchronize_sched();
4218 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
4220 if (RB_WARN_ON(cpu_buffer
, local_read(&cpu_buffer
->committing
)))
4223 arch_spin_lock(&cpu_buffer
->lock
);
4225 rb_reset_cpu(cpu_buffer
);
4227 arch_spin_unlock(&cpu_buffer
->lock
);
4230 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
4232 atomic_dec(&cpu_buffer
->record_disabled
);
4233 atomic_dec(&buffer
->resize_disabled
);
4235 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu
);
4238 * ring_buffer_reset - reset a ring buffer
4239 * @buffer: The ring buffer to reset all cpu buffers
4241 void ring_buffer_reset(struct ring_buffer
*buffer
)
4245 for_each_buffer_cpu(buffer
, cpu
)
4246 ring_buffer_reset_cpu(buffer
, cpu
);
4248 EXPORT_SYMBOL_GPL(ring_buffer_reset
);
4251 * rind_buffer_empty - is the ring buffer empty?
4252 * @buffer: The ring buffer to test
4254 int ring_buffer_empty(struct ring_buffer
*buffer
)
4256 struct ring_buffer_per_cpu
*cpu_buffer
;
4257 unsigned long flags
;
4262 dolock
= rb_ok_to_lock();
4264 /* yes this is racy, but if you don't like the race, lock the buffer */
4265 for_each_buffer_cpu(buffer
, cpu
) {
4266 cpu_buffer
= buffer
->buffers
[cpu
];
4267 local_irq_save(flags
);
4269 raw_spin_lock(&cpu_buffer
->reader_lock
);
4270 ret
= rb_per_cpu_empty(cpu_buffer
);
4272 raw_spin_unlock(&cpu_buffer
->reader_lock
);
4273 local_irq_restore(flags
);
4281 EXPORT_SYMBOL_GPL(ring_buffer_empty
);
4284 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
4285 * @buffer: The ring buffer
4286 * @cpu: The CPU buffer to test
4288 int ring_buffer_empty_cpu(struct ring_buffer
*buffer
, int cpu
)
4290 struct ring_buffer_per_cpu
*cpu_buffer
;
4291 unsigned long flags
;
4295 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
4298 dolock
= rb_ok_to_lock();
4300 cpu_buffer
= buffer
->buffers
[cpu
];
4301 local_irq_save(flags
);
4303 raw_spin_lock(&cpu_buffer
->reader_lock
);
4304 ret
= rb_per_cpu_empty(cpu_buffer
);
4306 raw_spin_unlock(&cpu_buffer
->reader_lock
);
4307 local_irq_restore(flags
);
4311 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu
);
4313 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4315 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
4316 * @buffer_a: One buffer to swap with
4317 * @buffer_b: The other buffer to swap with
4319 * This function is useful for tracers that want to take a "snapshot"
4320 * of a CPU buffer and has another back up buffer lying around.
4321 * it is expected that the tracer handles the cpu buffer not being
4322 * used at the moment.
4324 int ring_buffer_swap_cpu(struct ring_buffer
*buffer_a
,
4325 struct ring_buffer
*buffer_b
, int cpu
)
4327 struct ring_buffer_per_cpu
*cpu_buffer_a
;
4328 struct ring_buffer_per_cpu
*cpu_buffer_b
;
4331 if (!cpumask_test_cpu(cpu
, buffer_a
->cpumask
) ||
4332 !cpumask_test_cpu(cpu
, buffer_b
->cpumask
))
4335 cpu_buffer_a
= buffer_a
->buffers
[cpu
];
4336 cpu_buffer_b
= buffer_b
->buffers
[cpu
];
4338 /* At least make sure the two buffers are somewhat the same */
4339 if (cpu_buffer_a
->nr_pages
!= cpu_buffer_b
->nr_pages
)
4344 if (ring_buffer_flags
!= RB_BUFFERS_ON
)
4347 if (atomic_read(&buffer_a
->record_disabled
))
4350 if (atomic_read(&buffer_b
->record_disabled
))
4353 if (atomic_read(&cpu_buffer_a
->record_disabled
))
4356 if (atomic_read(&cpu_buffer_b
->record_disabled
))
4360 * We can't do a synchronize_sched here because this
4361 * function can be called in atomic context.
4362 * Normally this will be called from the same CPU as cpu.
4363 * If not it's up to the caller to protect this.
4365 atomic_inc(&cpu_buffer_a
->record_disabled
);
4366 atomic_inc(&cpu_buffer_b
->record_disabled
);
4369 if (local_read(&cpu_buffer_a
->committing
))
4371 if (local_read(&cpu_buffer_b
->committing
))
4374 buffer_a
->buffers
[cpu
] = cpu_buffer_b
;
4375 buffer_b
->buffers
[cpu
] = cpu_buffer_a
;
4377 cpu_buffer_b
->buffer
= buffer_a
;
4378 cpu_buffer_a
->buffer
= buffer_b
;
4383 atomic_dec(&cpu_buffer_a
->record_disabled
);
4384 atomic_dec(&cpu_buffer_b
->record_disabled
);
4388 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu
);
4389 #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */
4392 * ring_buffer_alloc_read_page - allocate a page to read from buffer
4393 * @buffer: the buffer to allocate for.
4395 * This function is used in conjunction with ring_buffer_read_page.
4396 * When reading a full page from the ring buffer, these functions
4397 * can be used to speed up the process. The calling function should
4398 * allocate a few pages first with this function. Then when it
4399 * needs to get pages from the ring buffer, it passes the result
4400 * of this function into ring_buffer_read_page, which will swap
4401 * the page that was allocated, with the read page of the buffer.
4404 * The page allocated, or NULL on error.
4406 void *ring_buffer_alloc_read_page(struct ring_buffer
*buffer
, int cpu
)
4408 struct buffer_data_page
*bpage
;
4411 page
= alloc_pages_node(cpu_to_node(cpu
),
4412 GFP_KERNEL
| __GFP_NORETRY
, 0);
4416 bpage
= page_address(page
);
4418 rb_init_page(bpage
);
4422 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page
);
4425 * ring_buffer_free_read_page - free an allocated read page
4426 * @buffer: the buffer the page was allocate for
4427 * @data: the page to free
4429 * Free a page allocated from ring_buffer_alloc_read_page.
4431 void ring_buffer_free_read_page(struct ring_buffer
*buffer
, void *data
)
4433 free_page((unsigned long)data
);
4435 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page
);
4438 * ring_buffer_read_page - extract a page from the ring buffer
4439 * @buffer: buffer to extract from
4440 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
4441 * @len: amount to extract
4442 * @cpu: the cpu of the buffer to extract
4443 * @full: should the extraction only happen when the page is full.
4445 * This function will pull out a page from the ring buffer and consume it.
4446 * @data_page must be the address of the variable that was returned
4447 * from ring_buffer_alloc_read_page. This is because the page might be used
4448 * to swap with a page in the ring buffer.
4451 * rpage = ring_buffer_alloc_read_page(buffer);
4454 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
4456 * process_page(rpage, ret);
4458 * When @full is set, the function will not return true unless
4459 * the writer is off the reader page.
4461 * Note: it is up to the calling functions to handle sleeps and wakeups.
4462 * The ring buffer can be used anywhere in the kernel and can not
4463 * blindly call wake_up. The layer that uses the ring buffer must be
4464 * responsible for that.
4467 * >=0 if data has been transferred, returns the offset of consumed data.
4468 * <0 if no data has been transferred.
4470 int ring_buffer_read_page(struct ring_buffer
*buffer
,
4471 void **data_page
, size_t len
, int cpu
, int full
)
4473 struct ring_buffer_per_cpu
*cpu_buffer
= buffer
->buffers
[cpu
];
4474 struct ring_buffer_event
*event
;
4475 struct buffer_data_page
*bpage
;
4476 struct buffer_page
*reader
;
4477 unsigned long missed_events
;
4478 unsigned long flags
;
4479 unsigned int commit
;
4484 if (!cpumask_test_cpu(cpu
, buffer
->cpumask
))
4488 * If len is not big enough to hold the page header, then
4489 * we can not copy anything.
4491 if (len
<= BUF_PAGE_HDR_SIZE
)
4494 len
-= BUF_PAGE_HDR_SIZE
;
4503 raw_spin_lock_irqsave(&cpu_buffer
->reader_lock
, flags
);
4505 reader
= rb_get_reader_page(cpu_buffer
);
4509 event
= rb_reader_event(cpu_buffer
);
4511 read
= reader
->read
;
4512 commit
= rb_page_commit(reader
);
4514 /* Check if any events were dropped */
4515 missed_events
= cpu_buffer
->lost_events
;
4518 * If this page has been partially read or
4519 * if len is not big enough to read the rest of the page or
4520 * a writer is still on the page, then
4521 * we must copy the data from the page to the buffer.
4522 * Otherwise, we can simply swap the page with the one passed in.
4524 if (read
|| (len
< (commit
- read
)) ||
4525 cpu_buffer
->reader_page
== cpu_buffer
->commit_page
) {
4526 struct buffer_data_page
*rpage
= cpu_buffer
->reader_page
->page
;
4527 unsigned int rpos
= read
;
4528 unsigned int pos
= 0;
4534 if (len
> (commit
- read
))
4535 len
= (commit
- read
);
4537 /* Always keep the time extend and data together */
4538 size
= rb_event_ts_length(event
);
4543 /* save the current timestamp, since the user will need it */
4544 save_timestamp
= cpu_buffer
->read_stamp
;
4546 /* Need to copy one event at a time */
4548 /* We need the size of one event, because
4549 * rb_advance_reader only advances by one event,
4550 * whereas rb_event_ts_length may include the size of
4551 * one or two events.
4552 * We have already ensured there's enough space if this
4553 * is a time extend. */
4554 size
= rb_event_length(event
);
4555 memcpy(bpage
->data
+ pos
, rpage
->data
+ rpos
, size
);
4559 rb_advance_reader(cpu_buffer
);
4560 rpos
= reader
->read
;
4566 event
= rb_reader_event(cpu_buffer
);
4567 /* Always keep the time extend and data together */
4568 size
= rb_event_ts_length(event
);
4569 } while (len
>= size
);
4572 local_set(&bpage
->commit
, pos
);
4573 bpage
->time_stamp
= save_timestamp
;
4575 /* we copied everything to the beginning */
4578 /* update the entry counter */
4579 cpu_buffer
->read
+= rb_page_entries(reader
);
4580 cpu_buffer
->read_bytes
+= BUF_PAGE_SIZE
;
4582 /* swap the pages */
4583 rb_init_page(bpage
);
4584 bpage
= reader
->page
;
4585 reader
->page
= *data_page
;
4586 local_set(&reader
->write
, 0);
4587 local_set(&reader
->entries
, 0);
4592 * Use the real_end for the data size,
4593 * This gives us a chance to store the lost events
4596 if (reader
->real_end
)
4597 local_set(&bpage
->commit
, reader
->real_end
);
4601 cpu_buffer
->lost_events
= 0;
4603 commit
= local_read(&bpage
->commit
);
4605 * Set a flag in the commit field if we lost events
4607 if (missed_events
) {
4608 /* If there is room at the end of the page to save the
4609 * missed events, then record it there.
4611 if (BUF_PAGE_SIZE
- commit
>= sizeof(missed_events
)) {
4612 memcpy(&bpage
->data
[commit
], &missed_events
,
4613 sizeof(missed_events
));
4614 local_add(RB_MISSED_STORED
, &bpage
->commit
);
4615 commit
+= sizeof(missed_events
);
4617 local_add(RB_MISSED_EVENTS
, &bpage
->commit
);
4621 * This page may be off to user land. Zero it out here.
4623 if (commit
< BUF_PAGE_SIZE
)
4624 memset(&bpage
->data
[commit
], 0, BUF_PAGE_SIZE
- commit
);
4627 raw_spin_unlock_irqrestore(&cpu_buffer
->reader_lock
, flags
);
4632 EXPORT_SYMBOL_GPL(ring_buffer_read_page
);
4634 #ifdef CONFIG_HOTPLUG_CPU
4635 static int rb_cpu_notify(struct notifier_block
*self
,
4636 unsigned long action
, void *hcpu
)
4638 struct ring_buffer
*buffer
=
4639 container_of(self
, struct ring_buffer
, cpu_notify
);
4640 long cpu
= (long)hcpu
;
4641 int cpu_i
, nr_pages_same
;
4642 unsigned int nr_pages
;
4645 case CPU_UP_PREPARE
:
4646 case CPU_UP_PREPARE_FROZEN
:
4647 if (cpumask_test_cpu(cpu
, buffer
->cpumask
))
4652 /* check if all cpu sizes are same */
4653 for_each_buffer_cpu(buffer
, cpu_i
) {
4654 /* fill in the size from first enabled cpu */
4656 nr_pages
= buffer
->buffers
[cpu_i
]->nr_pages
;
4657 if (nr_pages
!= buffer
->buffers
[cpu_i
]->nr_pages
) {
4662 /* allocate minimum pages, user can later expand it */
4665 buffer
->buffers
[cpu
] =
4666 rb_allocate_cpu_buffer(buffer
, nr_pages
, cpu
);
4667 if (!buffer
->buffers
[cpu
]) {
4668 WARN(1, "failed to allocate ring buffer on CPU %ld\n",
4673 cpumask_set_cpu(cpu
, buffer
->cpumask
);
4675 case CPU_DOWN_PREPARE
:
4676 case CPU_DOWN_PREPARE_FROZEN
:
4679 * If we were to free the buffer, then the user would
4680 * lose any trace that was in the buffer.
4690 #ifdef CONFIG_RING_BUFFER_STARTUP_TEST
4692 * This is a basic integrity check of the ring buffer.
4693 * Late in the boot cycle this test will run when configured in.
4694 * It will kick off a thread per CPU that will go into a loop
4695 * writing to the per cpu ring buffer various sizes of data.
4696 * Some of the data will be large items, some small.
4698 * Another thread is created that goes into a spin, sending out
4699 * IPIs to the other CPUs to also write into the ring buffer.
4700 * this is to test the nesting ability of the buffer.
4702 * Basic stats are recorded and reported. If something in the
4703 * ring buffer should happen that's not expected, a big warning
4704 * is displayed and all ring buffers are disabled.
4706 static struct task_struct
*rb_threads
[NR_CPUS
] __initdata
;
4708 struct rb_test_data
{
4709 struct ring_buffer
*buffer
;
4710 unsigned long events
;
4711 unsigned long bytes_written
;
4712 unsigned long bytes_alloc
;
4713 unsigned long bytes_dropped
;
4714 unsigned long events_nested
;
4715 unsigned long bytes_written_nested
;
4716 unsigned long bytes_alloc_nested
;
4717 unsigned long bytes_dropped_nested
;
4718 int min_size_nested
;
4719 int max_size_nested
;
4726 static struct rb_test_data rb_data
[NR_CPUS
] __initdata
;
4729 #define RB_TEST_BUFFER_SIZE 1048576
4731 static char rb_string
[] __initdata
=
4732 "abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\"
4733 "?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890"
4734 "!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv";
4736 static bool rb_test_started __initdata
;
4743 static __init
int rb_write_something(struct rb_test_data
*data
, bool nested
)
4745 struct ring_buffer_event
*event
;
4746 struct rb_item
*item
;
4753 /* Have nested writes different that what is written */
4754 cnt
= data
->cnt
+ (nested
? 27 : 0);
4756 /* Multiply cnt by ~e, to make some unique increment */
4757 size
= (data
->cnt
* 68 / 25) % (sizeof(rb_string
) - 1);
4759 len
= size
+ sizeof(struct rb_item
);
4761 started
= rb_test_started
;
4762 /* read rb_test_started before checking buffer enabled */
4765 event
= ring_buffer_lock_reserve(data
->buffer
, len
);
4767 /* Ignore dropped events before test starts. */
4770 data
->bytes_dropped
+= len
;
4772 data
->bytes_dropped_nested
+= len
;
4777 event_len
= ring_buffer_event_length(event
);
4779 if (RB_WARN_ON(data
->buffer
, event_len
< len
))
4782 item
= ring_buffer_event_data(event
);
4784 memcpy(item
->str
, rb_string
, size
);
4787 data
->bytes_alloc_nested
+= event_len
;
4788 data
->bytes_written_nested
+= len
;
4789 data
->events_nested
++;
4790 if (!data
->min_size_nested
|| len
< data
->min_size_nested
)
4791 data
->min_size_nested
= len
;
4792 if (len
> data
->max_size_nested
)
4793 data
->max_size_nested
= len
;
4795 data
->bytes_alloc
+= event_len
;
4796 data
->bytes_written
+= len
;
4798 if (!data
->min_size
|| len
< data
->min_size
)
4799 data
->max_size
= len
;
4800 if (len
> data
->max_size
)
4801 data
->max_size
= len
;
4805 ring_buffer_unlock_commit(data
->buffer
, event
);
4810 static __init
int rb_test(void *arg
)
4812 struct rb_test_data
*data
= arg
;
4814 while (!kthread_should_stop()) {
4815 rb_write_something(data
, false);
4818 set_current_state(TASK_INTERRUPTIBLE
);
4819 /* Now sleep between a min of 100-300us and a max of 1ms */
4820 usleep_range(((data
->cnt
% 3) + 1) * 100, 1000);
4826 static __init
void rb_ipi(void *ignore
)
4828 struct rb_test_data
*data
;
4829 int cpu
= smp_processor_id();
4831 data
= &rb_data
[cpu
];
4832 rb_write_something(data
, true);
4835 static __init
int rb_hammer_test(void *arg
)
4837 while (!kthread_should_stop()) {
4839 /* Send an IPI to all cpus to write data! */
4840 smp_call_function(rb_ipi
, NULL
, 1);
4841 /* No sleep, but for non preempt, let others run */
4848 static __init
int test_ringbuffer(void)
4850 struct task_struct
*rb_hammer
;
4851 struct ring_buffer
*buffer
;
4855 pr_info("Running ring buffer tests...\n");
4857 buffer
= ring_buffer_alloc(RB_TEST_BUFFER_SIZE
, RB_FL_OVERWRITE
);
4858 if (WARN_ON(!buffer
))
4861 /* Disable buffer so that threads can't write to it yet */
4862 ring_buffer_record_off(buffer
);
4864 for_each_online_cpu(cpu
) {
4865 rb_data
[cpu
].buffer
= buffer
;
4866 rb_data
[cpu
].cpu
= cpu
;
4867 rb_data
[cpu
].cnt
= cpu
;
4868 rb_threads
[cpu
] = kthread_create(rb_test
, &rb_data
[cpu
],
4869 "rbtester/%d", cpu
);
4870 if (WARN_ON(!rb_threads
[cpu
])) {
4871 pr_cont("FAILED\n");
4876 kthread_bind(rb_threads
[cpu
], cpu
);
4877 wake_up_process(rb_threads
[cpu
]);
4880 /* Now create the rb hammer! */
4881 rb_hammer
= kthread_run(rb_hammer_test
, NULL
, "rbhammer");
4882 if (WARN_ON(!rb_hammer
)) {
4883 pr_cont("FAILED\n");
4888 ring_buffer_record_on(buffer
);
4890 * Show buffer is enabled before setting rb_test_started.
4891 * Yes there's a small race window where events could be
4892 * dropped and the thread wont catch it. But when a ring
4893 * buffer gets enabled, there will always be some kind of
4894 * delay before other CPUs see it. Thus, we don't care about
4895 * those dropped events. We care about events dropped after
4896 * the threads see that the buffer is active.
4899 rb_test_started
= true;
4901 set_current_state(TASK_INTERRUPTIBLE
);
4902 /* Just run for 10 seconds */;
4903 schedule_timeout(10 * HZ
);
4905 kthread_stop(rb_hammer
);
4908 for_each_online_cpu(cpu
) {
4909 if (!rb_threads
[cpu
])
4911 kthread_stop(rb_threads
[cpu
]);
4914 ring_buffer_free(buffer
);
4919 pr_info("finished\n");
4920 for_each_online_cpu(cpu
) {
4921 struct ring_buffer_event
*event
;
4922 struct rb_test_data
*data
= &rb_data
[cpu
];
4923 struct rb_item
*item
;
4924 unsigned long total_events
;
4925 unsigned long total_dropped
;
4926 unsigned long total_written
;
4927 unsigned long total_alloc
;
4928 unsigned long total_read
= 0;
4929 unsigned long total_size
= 0;
4930 unsigned long total_len
= 0;
4931 unsigned long total_lost
= 0;
4934 int small_event_size
;
4938 total_events
= data
->events
+ data
->events_nested
;
4939 total_written
= data
->bytes_written
+ data
->bytes_written_nested
;
4940 total_alloc
= data
->bytes_alloc
+ data
->bytes_alloc_nested
;
4941 total_dropped
= data
->bytes_dropped
+ data
->bytes_dropped_nested
;
4943 big_event_size
= data
->max_size
+ data
->max_size_nested
;
4944 small_event_size
= data
->min_size
+ data
->min_size_nested
;
4946 pr_info("CPU %d:\n", cpu
);
4947 pr_info(" events: %ld\n", total_events
);
4948 pr_info(" dropped bytes: %ld\n", total_dropped
);
4949 pr_info(" alloced bytes: %ld\n", total_alloc
);
4950 pr_info(" written bytes: %ld\n", total_written
);
4951 pr_info(" biggest event: %d\n", big_event_size
);
4952 pr_info(" smallest event: %d\n", small_event_size
);
4954 if (RB_WARN_ON(buffer
, total_dropped
))
4959 while ((event
= ring_buffer_consume(buffer
, cpu
, NULL
, &lost
))) {
4961 item
= ring_buffer_event_data(event
);
4962 total_len
+= ring_buffer_event_length(event
);
4963 total_size
+= item
->size
+ sizeof(struct rb_item
);
4964 if (memcmp(&item
->str
[0], rb_string
, item
->size
) != 0) {
4965 pr_info("FAILED!\n");
4966 pr_info("buffer had: %.*s\n", item
->size
, item
->str
);
4967 pr_info("expected: %.*s\n", item
->size
, rb_string
);
4968 RB_WARN_ON(buffer
, 1);
4979 pr_info(" read events: %ld\n", total_read
);
4980 pr_info(" lost events: %ld\n", total_lost
);
4981 pr_info(" total events: %ld\n", total_lost
+ total_read
);
4982 pr_info(" recorded len bytes: %ld\n", total_len
);
4983 pr_info(" recorded size bytes: %ld\n", total_size
);
4985 pr_info(" With dropped events, record len and size may not match\n"
4986 " alloced and written from above\n");
4988 if (RB_WARN_ON(buffer
, total_len
!= total_alloc
||
4989 total_size
!= total_written
))
4992 if (RB_WARN_ON(buffer
, total_lost
+ total_read
!= total_events
))
4998 pr_info("Ring buffer PASSED!\n");
5000 ring_buffer_free(buffer
);
5004 late_initcall(test_ringbuffer
);
5005 #endif /* CONFIG_RING_BUFFER_STARTUP_TEST */