1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Generic ring buffer
4 *
5 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
6 */
7 #include <linux/trace_events.h>
8 #include <linux/ring_buffer.h>
9 #include <linux/trace_clock.h>
10 #include <linux/sched/clock.h>
11 #include <linux/trace_seq.h>
12 #include <linux/spinlock.h>
13 #include <linux/irq_work.h>
14 #include <linux/security.h>
15 #include <linux/uaccess.h>
16 #include <linux/hardirq.h>
17 #include <linux/kthread.h> /* for self test */
18 #include <linux/module.h>
19 #include <linux/percpu.h>
20 #include <linux/mutex.h>
21 #include <linux/delay.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/list.h>
26 #include <linux/cpu.h>
27 #include <linux/oom.h>
28
29 #include <asm/local.h>
30
31 static void update_pages_handler(struct work_struct *work);
32
33 /*
34 * The ring buffer header is special. We must manually up keep it.
35 */
ring_buffer_print_entry_header(struct trace_seq * s)36 int ring_buffer_print_entry_header(struct trace_seq *s)
37 {
38 trace_seq_puts(s, "# compressed entry header\n");
39 trace_seq_puts(s, "\ttype_len : 5 bits\n");
40 trace_seq_puts(s, "\ttime_delta : 27 bits\n");
41 trace_seq_puts(s, "\tarray : 32 bits\n");
42 trace_seq_putc(s, '\n');
43 trace_seq_printf(s, "\tpadding : type == %d\n",
44 RINGBUF_TYPE_PADDING);
45 trace_seq_printf(s, "\ttime_extend : type == %d\n",
46 RINGBUF_TYPE_TIME_EXTEND);
47 trace_seq_printf(s, "\ttime_stamp : type == %d\n",
48 RINGBUF_TYPE_TIME_STAMP);
49 trace_seq_printf(s, "\tdata max type_len == %d\n",
50 RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
51
52 return !trace_seq_has_overflowed(s);
53 }
54
55 /*
56 * The ring buffer is made up of a list of pages. A separate list of pages is
57 * allocated for each CPU. A writer may only write to a buffer that is
58 * associated with the CPU it is currently executing on. A reader may read
59 * from any per cpu buffer.
60 *
61 * The reader is special. For each per cpu buffer, the reader has its own
62 * reader page. When a reader has read the entire reader page, this reader
63 * page is swapped with another page in the ring buffer.
64 *
65 * Now, as long as the writer is off the reader page, the reader can do what
66 * ever it wants with that page. The writer will never write to that page
67 * again (as long as it is out of the ring buffer).
68 *
69 * Here's some silly ASCII art.
70 *
71 * +------+
72 * |reader| RING BUFFER
73 * |page |
74 * +------+ +---+ +---+ +---+
75 * | |-->| |-->| |
76 * +---+ +---+ +---+
77 * ^ |
78 * | |
79 * +---------------+
80 *
81 *
82 * +------+
83 * |reader| RING BUFFER
84 * |page |------------------v
85 * +------+ +---+ +---+ +---+
86 * | |-->| |-->| |
87 * +---+ +---+ +---+
88 * ^ |
89 * | |
90 * +---------------+
91 *
92 *
93 * +------+
94 * |reader| RING BUFFER
95 * |page |------------------v
96 * +------+ +---+ +---+ +---+
97 * ^ | |-->| |-->| |
98 * | +---+ +---+ +---+
99 * | |
100 * | |
101 * +------------------------------+
102 *
103 *
104 * +------+
105 * |buffer| RING BUFFER
106 * |page |------------------v
107 * +------+ +---+ +---+ +---+
108 * ^ | | | |-->| |
109 * | New +---+ +---+ +---+
110 * | Reader------^ |
111 * | page |
112 * +------------------------------+
113 *
114 *
115 * After we make this swap, the reader can hand this page off to the splice
116 * code and be done with it. It can even allocate a new page if it needs to
117 * and swap that into the ring buffer.
118 *
119 * We will be using cmpxchg soon to make all this lockless.
120 *
121 */
122
123 /* Used for individual buffers (after the counter) */
124 #define RB_BUFFER_OFF (1 << 20)
125
126 #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
127
128 #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
129 #define RB_ALIGNMENT 4U
130 #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
131 #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */
132 #define RB_ALIGN_DATA __aligned(RB_ALIGNMENT)
133
134 /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
135 #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
136
137 enum {
138 RB_LEN_TIME_EXTEND = 8,
139 RB_LEN_TIME_STAMP = 8,
140 };
141
142 #define skip_time_extend(event) \
143 ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND))
144
145 #define extended_time(event) \
146 (event->type_len >= RINGBUF_TYPE_TIME_EXTEND)
147
rb_null_event(struct ring_buffer_event * event)148 static inline int rb_null_event(struct ring_buffer_event *event)
149 {
150 return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta;
151 }
152
rb_event_set_padding(struct ring_buffer_event * event)153 static void rb_event_set_padding(struct ring_buffer_event *event)
154 {
155 /* padding has a NULL time_delta */
156 event->type_len = RINGBUF_TYPE_PADDING;
157 event->time_delta = 0;
158 }
159
160 static unsigned
rb_event_data_length(struct ring_buffer_event * event)161 rb_event_data_length(struct ring_buffer_event *event)
162 {
163 unsigned length;
164
165 if (event->type_len)
166 length = event->type_len * RB_ALIGNMENT;
167 else
168 length = event->array[0];
169 return length + RB_EVNT_HDR_SIZE;
170 }
171
172 /*
173 * Return the length of the given event. Will return
174 * the length of the time extend if the event is a
175 * time extend.
176 */
177 static inline unsigned
rb_event_length(struct ring_buffer_event * event)178 rb_event_length(struct ring_buffer_event *event)
179 {
180 switch (event->type_len) {
181 case RINGBUF_TYPE_PADDING:
182 if (rb_null_event(event))
183 /* undefined */
184 return -1;
185 return event->array[0] + RB_EVNT_HDR_SIZE;
186
187 case RINGBUF_TYPE_TIME_EXTEND:
188 return RB_LEN_TIME_EXTEND;
189
190 case RINGBUF_TYPE_TIME_STAMP:
191 return RB_LEN_TIME_STAMP;
192
193 case RINGBUF_TYPE_DATA:
194 return rb_event_data_length(event);
195 default:
196 BUG();
197 }
198 /* not hit */
199 return 0;
200 }
201
202 /*
203 * Return total length of time extend and data,
204 * or just the event length for all other events.
205 */
206 static inline unsigned
rb_event_ts_length(struct ring_buffer_event * event)207 rb_event_ts_length(struct ring_buffer_event *event)
208 {
209 unsigned len = 0;
210
211 if (extended_time(event)) {
212 /* time extends include the data event after it */
213 len = RB_LEN_TIME_EXTEND;
214 event = skip_time_extend(event);
215 }
216 return len + rb_event_length(event);
217 }
218
219 /**
220 * ring_buffer_event_length - return the length of the event
221 * @event: the event to get the length of
222 *
223 * Returns the size of the data load of a data event.
224 * If the event is something other than a data event, it
225 * returns the size of the event itself. With the exception
226 * of a TIME EXTEND, where it still returns the size of the
227 * data load of the data event after it.
228 */
ring_buffer_event_length(struct ring_buffer_event * event)229 unsigned ring_buffer_event_length(struct ring_buffer_event *event)
230 {
231 unsigned length;
232
233 if (extended_time(event))
234 event = skip_time_extend(event);
235
236 length = rb_event_length(event);
237 if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
238 return length;
239 length -= RB_EVNT_HDR_SIZE;
240 if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0]))
241 length -= sizeof(event->array[0]);
242 return length;
243 }
244 EXPORT_SYMBOL_GPL(ring_buffer_event_length);
245
246 /* inline for ring buffer fast paths */
247 static __always_inline void *
rb_event_data(struct ring_buffer_event * event)248 rb_event_data(struct ring_buffer_event *event)
249 {
250 if (extended_time(event))
251 event = skip_time_extend(event);
252 BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
253 /* If length is in len field, then array[0] has the data */
254 if (event->type_len)
255 return (void *)&event->array[0];
256 /* Otherwise length is in array[0] and array[1] has the data */
257 return (void *)&event->array[1];
258 }
259
260 /**
261 * ring_buffer_event_data - return the data of the event
262 * @event: the event to get the data from
263 */
ring_buffer_event_data(struct ring_buffer_event * event)264 void *ring_buffer_event_data(struct ring_buffer_event *event)
265 {
266 return rb_event_data(event);
267 }
268 EXPORT_SYMBOL_GPL(ring_buffer_event_data);
269
270 #define for_each_buffer_cpu(buffer, cpu) \
271 for_each_cpu(cpu, buffer->cpumask)
272
273 #define TS_SHIFT 27
274 #define TS_MASK ((1ULL << TS_SHIFT) - 1)
275 #define TS_DELTA_TEST (~TS_MASK)
276
277 /**
278 * ring_buffer_event_time_stamp - return the event's extended timestamp
279 * @event: the event to get the timestamp of
280 *
281 * Returns the extended timestamp associated with a data event.
282 * An extended time_stamp is a 64-bit timestamp represented
283 * internally in a special way that makes the best use of space
284 * contained within a ring buffer event. This function decodes
285 * it and maps it to a straight u64 value.
286 */
ring_buffer_event_time_stamp(struct ring_buffer_event * event)287 u64 ring_buffer_event_time_stamp(struct ring_buffer_event *event)
288 {
289 u64 ts;
290
291 ts = event->array[0];
292 ts <<= TS_SHIFT;
293 ts += event->time_delta;
294
295 return ts;
296 }
297
298 /* Flag when events were overwritten */
299 #define RB_MISSED_EVENTS (1 << 31)
300 /* Missed count stored at end */
301 #define RB_MISSED_STORED (1 << 30)
302
303 #define RB_MISSED_FLAGS (RB_MISSED_EVENTS|RB_MISSED_STORED)
304
305 struct buffer_data_page {
306 u64 time_stamp; /* page time stamp */
307 local_t commit; /* write committed index */
308 unsigned char data[] RB_ALIGN_DATA; /* data of buffer page */
309 };
310
311 /*
312 * Note, the buffer_page list must be first. The buffer pages
313 * are allocated in cache lines, which means that each buffer
314 * page will be at the beginning of a cache line, and thus
315 * the least significant bits will be zero. We use this to
316 * add flags in the list struct pointers, to make the ring buffer
317 * lockless.
318 */
319 struct buffer_page {
320 struct list_head list; /* list of buffer pages */
321 local_t write; /* index for next write */
322 unsigned read; /* index for next read */
323 local_t entries; /* entries on this page */
324 unsigned long real_end; /* real end of data */
325 struct buffer_data_page *page; /* Actual data page */
326 };
327
328 /*
329 * The buffer page counters, write and entries, must be reset
330 * atomically when crossing page boundaries. To synchronize this
331 * update, two counters are inserted into the number. One is
332 * the actual counter for the write position or count on the page.
333 *
334 * The other is a counter of updaters. Before an update happens
335 * the update partition of the counter is incremented. This will
336 * allow the updater to update the counter atomically.
337 *
338 * The counter is 20 bits, and the state data is 12.
339 */
340 #define RB_WRITE_MASK 0xfffff
341 #define RB_WRITE_INTCNT (1 << 20)
342
rb_init_page(struct buffer_data_page * bpage)343 static void rb_init_page(struct buffer_data_page *bpage)
344 {
345 local_set(&bpage->commit, 0);
346 }
347
348 /*
349 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
350 * this issue out.
351 */
free_buffer_page(struct buffer_page * bpage)352 static void free_buffer_page(struct buffer_page *bpage)
353 {
354 free_page((unsigned long)bpage->page);
355 kfree(bpage);
356 }
357
358 /*
359 * We need to fit the time_stamp delta into 27 bits.
360 */
test_time_stamp(u64 delta)361 static inline int test_time_stamp(u64 delta)
362 {
363 if (delta & TS_DELTA_TEST)
364 return 1;
365 return 0;
366 }
367
368 #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
369
370 /* Max payload is BUF_PAGE_SIZE - header (8bytes) */
371 #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
372
ring_buffer_print_page_header(struct trace_seq * s)373 int ring_buffer_print_page_header(struct trace_seq *s)
374 {
375 struct buffer_data_page field;
376
377 trace_seq_printf(s, "\tfield: u64 timestamp;\t"
378 "offset:0;\tsize:%u;\tsigned:%u;\n",
379 (unsigned int)sizeof(field.time_stamp),
380 (unsigned int)is_signed_type(u64));
381
382 trace_seq_printf(s, "\tfield: local_t commit;\t"
383 "offset:%u;\tsize:%u;\tsigned:%u;\n",
384 (unsigned int)offsetof(typeof(field), commit),
385 (unsigned int)sizeof(field.commit),
386 (unsigned int)is_signed_type(long));
387
388 trace_seq_printf(s, "\tfield: int overwrite;\t"
389 "offset:%u;\tsize:%u;\tsigned:%u;\n",
390 (unsigned int)offsetof(typeof(field), commit),
391 1,
392 (unsigned int)is_signed_type(long));
393
394 trace_seq_printf(s, "\tfield: char data;\t"
395 "offset:%u;\tsize:%u;\tsigned:%u;\n",
396 (unsigned int)offsetof(typeof(field), data),
397 (unsigned int)BUF_PAGE_SIZE,
398 (unsigned int)is_signed_type(char));
399
400 return !trace_seq_has_overflowed(s);
401 }
402
403 struct rb_irq_work {
404 struct irq_work work;
405 wait_queue_head_t waiters;
406 wait_queue_head_t full_waiters;
407 bool waiters_pending;
408 bool full_waiters_pending;
409 bool wakeup_full;
410 };
411
412 /*
413 * Structure to hold event state and handle nested events.
414 */
415 struct rb_event_info {
416 u64 ts;
417 u64 delta;
418 unsigned long length;
419 struct buffer_page *tail_page;
420 int add_timestamp;
421 };
422
423 /*
424 * Used for which event context the event is in.
425 * NMI = 0
426 * IRQ = 1
427 * SOFTIRQ = 2
428 * NORMAL = 3
429 *
430 * See trace_recursive_lock() comment below for more details.
431 */
432 enum {
433 RB_CTX_NMI,
434 RB_CTX_IRQ,
435 RB_CTX_SOFTIRQ,
436 RB_CTX_NORMAL,
437 RB_CTX_MAX
438 };
439
440 /*
441 * head_page == tail_page && head == tail then buffer is empty.
442 */
443 struct ring_buffer_per_cpu {
444 int cpu;
445 atomic_t record_disabled;
446 struct ring_buffer *buffer;
447 raw_spinlock_t reader_lock; /* serialize readers */
448 arch_spinlock_t lock;
449 struct lock_class_key lock_key;
450 struct buffer_data_page *free_page;
451 unsigned long nr_pages;
452 unsigned int current_context;
453 struct list_head *pages;
454 struct buffer_page *head_page; /* read from head */
455 struct buffer_page *tail_page; /* write to tail */
456 struct buffer_page *commit_page; /* committed pages */
457 struct buffer_page *reader_page;
458 unsigned long lost_events;
459 unsigned long last_overrun;
460 unsigned long nest;
461 local_t entries_bytes;
462 local_t entries;
463 local_t overrun;
464 local_t commit_overrun;
465 local_t dropped_events;
466 local_t committing;
467 local_t commits;
468 local_t pages_touched;
469 local_t pages_read;
470 long last_pages_touch;
471 size_t shortest_full;
472 unsigned long read;
473 unsigned long read_bytes;
474 u64 write_stamp;
475 u64 read_stamp;
476 /* ring buffer pages to update, > 0 to add, < 0 to remove */
477 long nr_pages_to_update;
478 struct list_head new_pages; /* new pages to add */
479 struct work_struct update_pages_work;
480 struct completion update_done;
481
482 struct rb_irq_work irq_work;
483 };
484
485 struct ring_buffer {
486 unsigned flags;
487 int cpus;
488 atomic_t record_disabled;
489 atomic_t resize_disabled;
490 cpumask_var_t cpumask;
491
492 struct lock_class_key *reader_lock_key;
493
494 struct mutex mutex;
495
496 struct ring_buffer_per_cpu **buffers;
497
498 struct hlist_node node;
499 u64 (*clock)(void);
500
501 struct rb_irq_work irq_work;
502 bool time_stamp_abs;
503 };
504
505 struct ring_buffer_iter {
506 struct ring_buffer_per_cpu *cpu_buffer;
507 unsigned long head;
508 struct buffer_page *head_page;
509 struct buffer_page *cache_reader_page;
510 unsigned long cache_read;
511 u64 read_stamp;
512 };
513
514 /**
515 * ring_buffer_nr_pages - get the number of buffer pages in the ring buffer
516 * @buffer: The ring_buffer to get the number of pages from
517 * @cpu: The cpu of the ring_buffer to get the number of pages from
518 *
519 * Returns the number of pages used by a per_cpu buffer of the ring buffer.
520 */
ring_buffer_nr_pages(struct ring_buffer * buffer,int cpu)521 size_t ring_buffer_nr_pages(struct ring_buffer *buffer, int cpu)
522 {
523 return buffer->buffers[cpu]->nr_pages;
524 }
525
526 /**
527 * ring_buffer_nr_pages_dirty - get the number of used pages in the ring buffer
528 * @buffer: The ring_buffer to get the number of pages from
529 * @cpu: The cpu of the ring_buffer to get the number of pages from
530 *
531 * Returns the number of pages that have content in the ring buffer.
532 */
ring_buffer_nr_dirty_pages(struct ring_buffer * buffer,int cpu)533 size_t ring_buffer_nr_dirty_pages(struct ring_buffer *buffer, int cpu)
534 {
535 size_t read;
536 size_t cnt;
537
538 read = local_read(&buffer->buffers[cpu]->pages_read);
539 cnt = local_read(&buffer->buffers[cpu]->pages_touched);
540 /* The reader can read an empty page, but not more than that */
541 if (cnt < read) {
542 WARN_ON_ONCE(read > cnt + 1);
543 return 0;
544 }
545
546 return cnt - read;
547 }
548
549 /*
550 * rb_wake_up_waiters - wake up tasks waiting for ring buffer input
551 *
552 * Schedules a delayed work to wake up any task that is blocked on the
553 * ring buffer waiters queue.
554 */
rb_wake_up_waiters(struct irq_work * work)555 static void rb_wake_up_waiters(struct irq_work *work)
556 {
557 struct rb_irq_work *rbwork = container_of(work, struct rb_irq_work, work);
558
559 wake_up_all(&rbwork->waiters);
560 if (rbwork->wakeup_full) {
561 rbwork->wakeup_full = false;
562 wake_up_all(&rbwork->full_waiters);
563 }
564 }
565
566 /**
567 * ring_buffer_wait - wait for input to the ring buffer
568 * @buffer: buffer to wait on
569 * @cpu: the cpu buffer to wait on
570 * @full: wait until a full page is available, if @cpu != RING_BUFFER_ALL_CPUS
571 *
572 * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
573 * as data is added to any of the @buffer's cpu buffers. Otherwise
574 * it will wait for data to be added to a specific cpu buffer.
575 */
ring_buffer_wait(struct ring_buffer * buffer,int cpu,int full)576 int ring_buffer_wait(struct ring_buffer *buffer, int cpu, int full)
577 {
578 struct ring_buffer_per_cpu *uninitialized_var(cpu_buffer);
579 DEFINE_WAIT(wait);
580 struct rb_irq_work *work;
581 int ret = 0;
582
583 /*
584 * Depending on what the caller is waiting for, either any
585 * data in any cpu buffer, or a specific buffer, put the
586 * caller on the appropriate wait queue.
587 */
588 if (cpu == RING_BUFFER_ALL_CPUS) {
589 work = &buffer->irq_work;
590 /* Full only makes sense on per cpu reads */
591 full = 0;
592 } else {
593 if (!cpumask_test_cpu(cpu, buffer->cpumask))
594 return -ENODEV;
595 cpu_buffer = buffer->buffers[cpu];
596 work = &cpu_buffer->irq_work;
597 }
598
599
600 while (true) {
601 if (full)
602 prepare_to_wait(&work->full_waiters, &wait, TASK_INTERRUPTIBLE);
603 else
604 prepare_to_wait(&work->waiters, &wait, TASK_INTERRUPTIBLE);
605
606 /*
607 * The events can happen in critical sections where
608 * checking a work queue can cause deadlocks.
609 * After adding a task to the queue, this flag is set
610 * only to notify events to try to wake up the queue
611 * using irq_work.
612 *
613 * We don't clear it even if the buffer is no longer
614 * empty. The flag only causes the next event to run
615 * irq_work to do the work queue wake up. The worse
616 * that can happen if we race with !trace_empty() is that
617 * an event will cause an irq_work to try to wake up
618 * an empty queue.
619 *
620 * There's no reason to protect this flag either, as
621 * the work queue and irq_work logic will do the necessary
622 * synchronization for the wake ups. The only thing
623 * that is necessary is that the wake up happens after
624 * a task has been queued. It's OK for spurious wake ups.
625 */
626 if (full)
627 work->full_waiters_pending = true;
628 else
629 work->waiters_pending = true;
630
631 if (signal_pending(current)) {
632 ret = -EINTR;
633 break;
634 }
635
636 if (cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer))
637 break;
638
639 if (cpu != RING_BUFFER_ALL_CPUS &&
640 !ring_buffer_empty_cpu(buffer, cpu)) {
641 unsigned long flags;
642 bool pagebusy;
643 size_t nr_pages;
644 size_t dirty;
645
646 if (!full)
647 break;
648
649 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
650 pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page;
651 nr_pages = cpu_buffer->nr_pages;
652 dirty = ring_buffer_nr_dirty_pages(buffer, cpu);
653 if (!cpu_buffer->shortest_full ||
654 cpu_buffer->shortest_full < full)
655 cpu_buffer->shortest_full = full;
656 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
657 if (!pagebusy &&
658 (!nr_pages || (dirty * 100) > full * nr_pages))
659 break;
660 }
661
662 schedule();
663 }
664
665 if (full)
666 finish_wait(&work->full_waiters, &wait);
667 else
668 finish_wait(&work->waiters, &wait);
669
670 return ret;
671 }
672
673 /**
674 * ring_buffer_poll_wait - poll on buffer input
675 * @buffer: buffer to wait on
676 * @cpu: the cpu buffer to wait on
677 * @filp: the file descriptor
678 * @poll_table: The poll descriptor
679 *
680 * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
681 * as data is added to any of the @buffer's cpu buffers. Otherwise
682 * it will wait for data to be added to a specific cpu buffer.
683 *
684 * Returns EPOLLIN | EPOLLRDNORM if data exists in the buffers,
685 * zero otherwise.
686 */
ring_buffer_poll_wait(struct ring_buffer * buffer,int cpu,struct file * filp,poll_table * poll_table)687 __poll_t ring_buffer_poll_wait(struct ring_buffer *buffer, int cpu,
688 struct file *filp, poll_table *poll_table)
689 {
690 struct ring_buffer_per_cpu *cpu_buffer;
691 struct rb_irq_work *work;
692
693 if (cpu == RING_BUFFER_ALL_CPUS)
694 work = &buffer->irq_work;
695 else {
696 if (!cpumask_test_cpu(cpu, buffer->cpumask))
697 return -EINVAL;
698
699 cpu_buffer = buffer->buffers[cpu];
700 work = &cpu_buffer->irq_work;
701 }
702
703 poll_wait(filp, &work->waiters, poll_table);
704 work->waiters_pending = true;
705 /*
706 * There's a tight race between setting the waiters_pending and
707 * checking if the ring buffer is empty. Once the waiters_pending bit
708 * is set, the next event will wake the task up, but we can get stuck
709 * if there's only a single event in.
710 *
711 * FIXME: Ideally, we need a memory barrier on the writer side as well,
712 * but adding a memory barrier to all events will cause too much of a
713 * performance hit in the fast path. We only need a memory barrier when
714 * the buffer goes from empty to having content. But as this race is
715 * extremely small, and it's not a problem if another event comes in, we
716 * will fix it later.
717 */
718 smp_mb();
719
720 if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) ||
721 (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu)))
722 return EPOLLIN | EPOLLRDNORM;
723 return 0;
724 }
725
726 /* buffer may be either ring_buffer or ring_buffer_per_cpu */
727 #define RB_WARN_ON(b, cond) \
728 ({ \
729 int _____ret = unlikely(cond); \
730 if (_____ret) { \
731 if (__same_type(*(b), struct ring_buffer_per_cpu)) { \
732 struct ring_buffer_per_cpu *__b = \
733 (void *)b; \
734 atomic_inc(&__b->buffer->record_disabled); \
735 } else \
736 atomic_inc(&b->record_disabled); \
737 WARN_ON(1); \
738 } \
739 _____ret; \
740 })
741
742 /* Up this if you want to test the TIME_EXTENTS and normalization */
743 #define DEBUG_SHIFT 0
744
rb_time_stamp(struct ring_buffer * buffer)745 static inline u64 rb_time_stamp(struct ring_buffer *buffer)
746 {
747 /* shift to debug/test normalization and TIME_EXTENTS */
748 return buffer->clock() << DEBUG_SHIFT;
749 }
750
ring_buffer_time_stamp(struct ring_buffer * buffer,int cpu)751 u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu)
752 {
753 u64 time;
754
755 preempt_disable_notrace();
756 time = rb_time_stamp(buffer);
757 preempt_enable_notrace();
758
759 return time;
760 }
761 EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
762
ring_buffer_normalize_time_stamp(struct ring_buffer * buffer,int cpu,u64 * ts)763 void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer,
764 int cpu, u64 *ts)
765 {
766 /* Just stupid testing the normalize function and deltas */
767 *ts >>= DEBUG_SHIFT;
768 }
769 EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
770
771 /*
772 * Making the ring buffer lockless makes things tricky.
773 * Although writes only happen on the CPU that they are on,
774 * and they only need to worry about interrupts. Reads can
775 * happen on any CPU.
776 *
777 * The reader page is always off the ring buffer, but when the
778 * reader finishes with a page, it needs to swap its page with
779 * a new one from the buffer. The reader needs to take from
780 * the head (writes go to the tail). But if a writer is in overwrite
781 * mode and wraps, it must push the head page forward.
782 *
783 * Here lies the problem.
784 *
785 * The reader must be careful to replace only the head page, and
786 * not another one. As described at the top of the file in the
787 * ASCII art, the reader sets its old page to point to the next
788 * page after head. It then sets the page after head to point to
789 * the old reader page. But if the writer moves the head page
790 * during this operation, the reader could end up with the tail.
791 *
792 * We use cmpxchg to help prevent this race. We also do something
793 * special with the page before head. We set the LSB to 1.
794 *
795 * When the writer must push the page forward, it will clear the
796 * bit that points to the head page, move the head, and then set
797 * the bit that points to the new head page.
798 *
799 * We also don't want an interrupt coming in and moving the head
800 * page on another writer. Thus we use the second LSB to catch
801 * that too. Thus:
802 *
803 * head->list->prev->next bit 1 bit 0
804 * ------- -------
805 * Normal page 0 0
806 * Points to head page 0 1
807 * New head page 1 0
808 *
809 * Note we can not trust the prev pointer of the head page, because:
810 *
811 * +----+ +-----+ +-----+
812 * | |------>| T |---X--->| N |
813 * | |<------| | | |
814 * +----+ +-----+ +-----+
815 * ^ ^ |
816 * | +-----+ | |
817 * +----------| R |----------+ |
818 * | |<-----------+
819 * +-----+
820 *
821 * Key: ---X--> HEAD flag set in pointer
822 * T Tail page
823 * R Reader page
824 * N Next page
825 *
826 * (see __rb_reserve_next() to see where this happens)
827 *
828 * What the above shows is that the reader just swapped out
829 * the reader page with a page in the buffer, but before it
830 * could make the new header point back to the new page added
831 * it was preempted by a writer. The writer moved forward onto
832 * the new page added by the reader and is about to move forward
833 * again.
834 *
835 * You can see, it is legitimate for the previous pointer of
836 * the head (or any page) not to point back to itself. But only
837 * temporarily.
838 */
839
840 #define RB_PAGE_NORMAL 0UL
841 #define RB_PAGE_HEAD 1UL
842 #define RB_PAGE_UPDATE 2UL
843
844
845 #define RB_FLAG_MASK 3UL
846
847 /* PAGE_MOVED is not part of the mask */
848 #define RB_PAGE_MOVED 4UL
849
850 /*
851 * rb_list_head - remove any bit
852 */
rb_list_head(struct list_head * list)853 static struct list_head *rb_list_head(struct list_head *list)
854 {
855 unsigned long val = (unsigned long)list;
856
857 return (struct list_head *)(val & ~RB_FLAG_MASK);
858 }
859
860 /*
861 * rb_is_head_page - test if the given page is the head page
862 *
863 * Because the reader may move the head_page pointer, we can
864 * not trust what the head page is (it may be pointing to
865 * the reader page). But if the next page is a header page,
866 * its flags will be non zero.
867 */
868 static inline int
rb_is_head_page(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * page,struct list_head * list)869 rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer,
870 struct buffer_page *page, struct list_head *list)
871 {
872 unsigned long val;
873
874 val = (unsigned long)list->next;
875
876 if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list)
877 return RB_PAGE_MOVED;
878
879 return val & RB_FLAG_MASK;
880 }
881
882 /*
883 * rb_is_reader_page
884 *
885 * The unique thing about the reader page, is that, if the
886 * writer is ever on it, the previous pointer never points
887 * back to the reader page.
888 */
rb_is_reader_page(struct buffer_page * page)889 static bool rb_is_reader_page(struct buffer_page *page)
890 {
891 struct list_head *list = page->list.prev;
892
893 return rb_list_head(list->next) != &page->list;
894 }
895
896 /*
897 * rb_set_list_to_head - set a list_head to be pointing to head.
898 */
rb_set_list_to_head(struct ring_buffer_per_cpu * cpu_buffer,struct list_head * list)899 static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer,
900 struct list_head *list)
901 {
902 unsigned long *ptr;
903
904 ptr = (unsigned long *)&list->next;
905 *ptr |= RB_PAGE_HEAD;
906 *ptr &= ~RB_PAGE_UPDATE;
907 }
908
909 /*
910 * rb_head_page_activate - sets up head page
911 */
rb_head_page_activate(struct ring_buffer_per_cpu * cpu_buffer)912 static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer)
913 {
914 struct buffer_page *head;
915
916 head = cpu_buffer->head_page;
917 if (!head)
918 return;
919
920 /*
921 * Set the previous list pointer to have the HEAD flag.
922 */
923 rb_set_list_to_head(cpu_buffer, head->list.prev);
924 }
925
rb_list_head_clear(struct list_head * list)926 static void rb_list_head_clear(struct list_head *list)
927 {
928 unsigned long *ptr = (unsigned long *)&list->next;
929
930 *ptr &= ~RB_FLAG_MASK;
931 }
932
933 /*
934 * rb_head_page_deactivate - clears head page ptr (for free list)
935 */
936 static void
rb_head_page_deactivate(struct ring_buffer_per_cpu * cpu_buffer)937 rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer)
938 {
939 struct list_head *hd;
940
941 /* Go through the whole list and clear any pointers found. */
942 rb_list_head_clear(cpu_buffer->pages);
943
944 list_for_each(hd, cpu_buffer->pages)
945 rb_list_head_clear(hd);
946 }
947
rb_head_page_set(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * head,struct buffer_page * prev,int old_flag,int new_flag)948 static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer,
949 struct buffer_page *head,
950 struct buffer_page *prev,
951 int old_flag, int new_flag)
952 {
953 struct list_head *list;
954 unsigned long val = (unsigned long)&head->list;
955 unsigned long ret;
956
957 list = &prev->list;
958
959 val &= ~RB_FLAG_MASK;
960
961 ret = cmpxchg((unsigned long *)&list->next,
962 val | old_flag, val | new_flag);
963
964 /* check if the reader took the page */
965 if ((ret & ~RB_FLAG_MASK) != val)
966 return RB_PAGE_MOVED;
967
968 return ret & RB_FLAG_MASK;
969 }
970
rb_head_page_set_update(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * head,struct buffer_page * prev,int old_flag)971 static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer,
972 struct buffer_page *head,
973 struct buffer_page *prev,
974 int old_flag)
975 {
976 return rb_head_page_set(cpu_buffer, head, prev,
977 old_flag, RB_PAGE_UPDATE);
978 }
979
rb_head_page_set_head(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * head,struct buffer_page * prev,int old_flag)980 static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer,
981 struct buffer_page *head,
982 struct buffer_page *prev,
983 int old_flag)
984 {
985 return rb_head_page_set(cpu_buffer, head, prev,
986 old_flag, RB_PAGE_HEAD);
987 }
988
rb_head_page_set_normal(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * head,struct buffer_page * prev,int old_flag)989 static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer,
990 struct buffer_page *head,
991 struct buffer_page *prev,
992 int old_flag)
993 {
994 return rb_head_page_set(cpu_buffer, head, prev,
995 old_flag, RB_PAGE_NORMAL);
996 }
997
rb_inc_page(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page ** bpage)998 static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer,
999 struct buffer_page **bpage)
1000 {
1001 struct list_head *p = rb_list_head((*bpage)->list.next);
1002
1003 *bpage = list_entry(p, struct buffer_page, list);
1004 }
1005
1006 static struct buffer_page *
rb_set_head_page(struct ring_buffer_per_cpu * cpu_buffer)1007 rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer)
1008 {
1009 struct buffer_page *head;
1010 struct buffer_page *page;
1011 struct list_head *list;
1012 int i;
1013
1014 if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page))
1015 return NULL;
1016
1017 /* sanity check */
1018 list = cpu_buffer->pages;
1019 if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list))
1020 return NULL;
1021
1022 page = head = cpu_buffer->head_page;
1023 /*
1024 * It is possible that the writer moves the header behind
1025 * where we started, and we miss in one loop.
1026 * A second loop should grab the header, but we'll do
1027 * three loops just because I'm paranoid.
1028 */
1029 for (i = 0; i < 3; i++) {
1030 do {
1031 if (rb_is_head_page(cpu_buffer, page, page->list.prev)) {
1032 cpu_buffer->head_page = page;
1033 return page;
1034 }
1035 rb_inc_page(cpu_buffer, &page);
1036 } while (page != head);
1037 }
1038
1039 RB_WARN_ON(cpu_buffer, 1);
1040
1041 return NULL;
1042 }
1043
rb_head_page_replace(struct buffer_page * old,struct buffer_page * new)1044 static int rb_head_page_replace(struct buffer_page *old,
1045 struct buffer_page *new)
1046 {
1047 unsigned long *ptr = (unsigned long *)&old->list.prev->next;
1048 unsigned long val;
1049 unsigned long ret;
1050
1051 val = *ptr & ~RB_FLAG_MASK;
1052 val |= RB_PAGE_HEAD;
1053
1054 ret = cmpxchg(ptr, val, (unsigned long)&new->list);
1055
1056 return ret == val;
1057 }
1058
1059 /*
1060 * rb_tail_page_update - move the tail page forward
1061 */
rb_tail_page_update(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * tail_page,struct buffer_page * next_page)1062 static void rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer,
1063 struct buffer_page *tail_page,
1064 struct buffer_page *next_page)
1065 {
1066 unsigned long old_entries;
1067 unsigned long old_write;
1068
1069 /*
1070 * The tail page now needs to be moved forward.
1071 *
1072 * We need to reset the tail page, but without messing
1073 * with possible erasing of data brought in by interrupts
1074 * that have moved the tail page and are currently on it.
1075 *
1076 * We add a counter to the write field to denote this.
1077 */
1078 old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write);
1079 old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries);
1080
1081 local_inc(&cpu_buffer->pages_touched);
1082 /*
1083 * Just make sure we have seen our old_write and synchronize
1084 * with any interrupts that come in.
1085 */
1086 barrier();
1087
1088 /*
1089 * If the tail page is still the same as what we think
1090 * it is, then it is up to us to update the tail
1091 * pointer.
1092 */
1093 if (tail_page == READ_ONCE(cpu_buffer->tail_page)) {
1094 /* Zero the write counter */
1095 unsigned long val = old_write & ~RB_WRITE_MASK;
1096 unsigned long eval = old_entries & ~RB_WRITE_MASK;
1097
1098 /*
1099 * This will only succeed if an interrupt did
1100 * not come in and change it. In which case, we
1101 * do not want to modify it.
1102 *
1103 * We add (void) to let the compiler know that we do not care
1104 * about the return value of these functions. We use the
1105 * cmpxchg to only update if an interrupt did not already
1106 * do it for us. If the cmpxchg fails, we don't care.
1107 */
1108 (void)local_cmpxchg(&next_page->write, old_write, val);
1109 (void)local_cmpxchg(&next_page->entries, old_entries, eval);
1110
1111 /*
1112 * No need to worry about races with clearing out the commit.
1113 * it only can increment when a commit takes place. But that
1114 * only happens in the outer most nested commit.
1115 */
1116 local_set(&next_page->page->commit, 0);
1117
1118 /* Again, either we update tail_page or an interrupt does */
1119 (void)cmpxchg(&cpu_buffer->tail_page, tail_page, next_page);
1120 }
1121 }
1122
rb_check_bpage(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * bpage)1123 static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer,
1124 struct buffer_page *bpage)
1125 {
1126 unsigned long val = (unsigned long)bpage;
1127
1128 if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK))
1129 return 1;
1130
1131 return 0;
1132 }
1133
1134 /**
1135 * rb_check_list - make sure a pointer to a list has the last bits zero
1136 */
rb_check_list(struct ring_buffer_per_cpu * cpu_buffer,struct list_head * list)1137 static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer,
1138 struct list_head *list)
1139 {
1140 if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev))
1141 return 1;
1142 if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next))
1143 return 1;
1144 return 0;
1145 }
1146
1147 /**
1148 * rb_check_pages - integrity check of buffer pages
1149 * @cpu_buffer: CPU buffer with pages to test
1150 *
1151 * As a safety measure we check to make sure the data pages have not
1152 * been corrupted.
1153 */
rb_check_pages(struct ring_buffer_per_cpu * cpu_buffer)1154 static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
1155 {
1156 struct list_head *head = cpu_buffer->pages;
1157 struct buffer_page *bpage, *tmp;
1158
1159 /* Reset the head page if it exists */
1160 if (cpu_buffer->head_page)
1161 rb_set_head_page(cpu_buffer);
1162
1163 rb_head_page_deactivate(cpu_buffer);
1164
1165 if (RB_WARN_ON(cpu_buffer, head->next->prev != head))
1166 return -1;
1167 if (RB_WARN_ON(cpu_buffer, head->prev->next != head))
1168 return -1;
1169
1170 if (rb_check_list(cpu_buffer, head))
1171 return -1;
1172
1173 list_for_each_entry_safe(bpage, tmp, head, list) {
1174 if (RB_WARN_ON(cpu_buffer,
1175 bpage->list.next->prev != &bpage->list))
1176 return -1;
1177 if (RB_WARN_ON(cpu_buffer,
1178 bpage->list.prev->next != &bpage->list))
1179 return -1;
1180 if (rb_check_list(cpu_buffer, &bpage->list))
1181 return -1;
1182 }
1183
1184 rb_head_page_activate(cpu_buffer);
1185
1186 return 0;
1187 }
1188
__rb_allocate_pages(long nr_pages,struct list_head * pages,int cpu)1189 static int __rb_allocate_pages(long nr_pages, struct list_head *pages, int cpu)
1190 {
1191 struct buffer_page *bpage, *tmp;
1192 bool user_thread = current->mm != NULL;
1193 gfp_t mflags;
1194 long i;
1195
1196 /*
1197 * Check if the available memory is there first.
1198 * Note, si_mem_available() only gives us a rough estimate of available
1199 * memory. It may not be accurate. But we don't care, we just want
1200 * to prevent doing any allocation when it is obvious that it is
1201 * not going to succeed.
1202 */
1203 i = si_mem_available();
1204 if (i < nr_pages)
1205 return -ENOMEM;
1206
1207 /*
1208 * __GFP_RETRY_MAYFAIL flag makes sure that the allocation fails
1209 * gracefully without invoking oom-killer and the system is not
1210 * destabilized.
1211 */
1212 mflags = GFP_KERNEL | __GFP_RETRY_MAYFAIL;
1213
1214 /*
1215 * If a user thread allocates too much, and si_mem_available()
1216 * reports there's enough memory, even though there is not.
1217 * Make sure the OOM killer kills this thread. This can happen
1218 * even with RETRY_MAYFAIL because another task may be doing
1219 * an allocation after this task has taken all memory.
1220 * This is the task the OOM killer needs to take out during this
1221 * loop, even if it was triggered by an allocation somewhere else.
1222 */
1223 if (user_thread)
1224 set_current_oom_origin();
1225 for (i = 0; i < nr_pages; i++) {
1226 struct page *page;
1227
1228 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
1229 mflags, cpu_to_node(cpu));
1230 if (!bpage)
1231 goto free_pages;
1232
1233 list_add(&bpage->list, pages);
1234
1235 page = alloc_pages_node(cpu_to_node(cpu), mflags, 0);
1236 if (!page)
1237 goto free_pages;
1238 bpage->page = page_address(page);
1239 rb_init_page(bpage->page);
1240
1241 if (user_thread && fatal_signal_pending(current))
1242 goto free_pages;
1243 }
1244 if (user_thread)
1245 clear_current_oom_origin();
1246
1247 return 0;
1248
1249 free_pages:
1250 list_for_each_entry_safe(bpage, tmp, pages, list) {
1251 list_del_init(&bpage->list);
1252 free_buffer_page(bpage);
1253 }
1254 if (user_thread)
1255 clear_current_oom_origin();
1256
1257 return -ENOMEM;
1258 }
1259
rb_allocate_pages(struct ring_buffer_per_cpu * cpu_buffer,unsigned long nr_pages)1260 static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
1261 unsigned long nr_pages)
1262 {
1263 LIST_HEAD(pages);
1264
1265 WARN_ON(!nr_pages);
1266
1267 if (__rb_allocate_pages(nr_pages, &pages, cpu_buffer->cpu))
1268 return -ENOMEM;
1269
1270 /*
1271 * The ring buffer page list is a circular list that does not
1272 * start and end with a list head. All page list items point to
1273 * other pages.
1274 */
1275 cpu_buffer->pages = pages.next;
1276 list_del(&pages);
1277
1278 cpu_buffer->nr_pages = nr_pages;
1279
1280 rb_check_pages(cpu_buffer);
1281
1282 return 0;
1283 }
1284
1285 static struct ring_buffer_per_cpu *
rb_allocate_cpu_buffer(struct ring_buffer * buffer,long nr_pages,int cpu)1286 rb_allocate_cpu_buffer(struct ring_buffer *buffer, long nr_pages, int cpu)
1287 {
1288 struct ring_buffer_per_cpu *cpu_buffer;
1289 struct buffer_page *bpage;
1290 struct page *page;
1291 int ret;
1292
1293 cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
1294 GFP_KERNEL, cpu_to_node(cpu));
1295 if (!cpu_buffer)
1296 return NULL;
1297
1298 cpu_buffer->cpu = cpu;
1299 cpu_buffer->buffer = buffer;
1300 raw_spin_lock_init(&cpu_buffer->reader_lock);
1301 lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key);
1302 cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1303 INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler);
1304 init_completion(&cpu_buffer->update_done);
1305 init_irq_work(&cpu_buffer->irq_work.work, rb_wake_up_waiters);
1306 init_waitqueue_head(&cpu_buffer->irq_work.waiters);
1307 init_waitqueue_head(&cpu_buffer->irq_work.full_waiters);
1308
1309 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
1310 GFP_KERNEL, cpu_to_node(cpu));
1311 if (!bpage)
1312 goto fail_free_buffer;
1313
1314 rb_check_bpage(cpu_buffer, bpage);
1315
1316 cpu_buffer->reader_page = bpage;
1317 page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0);
1318 if (!page)
1319 goto fail_free_reader;
1320 bpage->page = page_address(page);
1321 rb_init_page(bpage->page);
1322
1323 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
1324 INIT_LIST_HEAD(&cpu_buffer->new_pages);
1325
1326 ret = rb_allocate_pages(cpu_buffer, nr_pages);
1327 if (ret < 0)
1328 goto fail_free_reader;
1329
1330 cpu_buffer->head_page
1331 = list_entry(cpu_buffer->pages, struct buffer_page, list);
1332 cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
1333
1334 rb_head_page_activate(cpu_buffer);
1335
1336 return cpu_buffer;
1337
1338 fail_free_reader:
1339 free_buffer_page(cpu_buffer->reader_page);
1340
1341 fail_free_buffer:
1342 kfree(cpu_buffer);
1343 return NULL;
1344 }
1345
rb_free_cpu_buffer(struct ring_buffer_per_cpu * cpu_buffer)1346 static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
1347 {
1348 struct list_head *head = cpu_buffer->pages;
1349 struct buffer_page *bpage, *tmp;
1350
1351 free_buffer_page(cpu_buffer->reader_page);
1352
1353 rb_head_page_deactivate(cpu_buffer);
1354
1355 if (head) {
1356 list_for_each_entry_safe(bpage, tmp, head, list) {
1357 list_del_init(&bpage->list);
1358 free_buffer_page(bpage);
1359 }
1360 bpage = list_entry(head, struct buffer_page, list);
1361 free_buffer_page(bpage);
1362 }
1363
1364 kfree(cpu_buffer);
1365 }
1366
1367 /**
1368 * __ring_buffer_alloc - allocate a new ring_buffer
1369 * @size: the size in bytes per cpu that is needed.
1370 * @flags: attributes to set for the ring buffer.
1371 *
1372 * Currently the only flag that is available is the RB_FL_OVERWRITE
1373 * flag. This flag means that the buffer will overwrite old data
1374 * when the buffer wraps. If this flag is not set, the buffer will
1375 * drop data when the tail hits the head.
1376 */
__ring_buffer_alloc(unsigned long size,unsigned flags,struct lock_class_key * key)1377 struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags,
1378 struct lock_class_key *key)
1379 {
1380 struct ring_buffer *buffer;
1381 long nr_pages;
1382 int bsize;
1383 int cpu;
1384 int ret;
1385
1386 /* keep it in its own cache line */
1387 buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
1388 GFP_KERNEL);
1389 if (!buffer)
1390 return NULL;
1391
1392 if (!zalloc_cpumask_var(&buffer->cpumask, GFP_KERNEL))
1393 goto fail_free_buffer;
1394
1395 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
1396 buffer->flags = flags;
1397 buffer->clock = trace_clock_local;
1398 buffer->reader_lock_key = key;
1399
1400 init_irq_work(&buffer->irq_work.work, rb_wake_up_waiters);
1401 init_waitqueue_head(&buffer->irq_work.waiters);
1402
1403 /* need at least two pages */
1404 if (nr_pages < 2)
1405 nr_pages = 2;
1406
1407 buffer->cpus = nr_cpu_ids;
1408
1409 bsize = sizeof(void *) * nr_cpu_ids;
1410 buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
1411 GFP_KERNEL);
1412 if (!buffer->buffers)
1413 goto fail_free_cpumask;
1414
1415 cpu = raw_smp_processor_id();
1416 cpumask_set_cpu(cpu, buffer->cpumask);
1417 buffer->buffers[cpu] = rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
1418 if (!buffer->buffers[cpu])
1419 goto fail_free_buffers;
1420
1421 ret = cpuhp_state_add_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node);
1422 if (ret < 0)
1423 goto fail_free_buffers;
1424
1425 mutex_init(&buffer->mutex);
1426
1427 return buffer;
1428
1429 fail_free_buffers:
1430 for_each_buffer_cpu(buffer, cpu) {
1431 if (buffer->buffers[cpu])
1432 rb_free_cpu_buffer(buffer->buffers[cpu]);
1433 }
1434 kfree(buffer->buffers);
1435
1436 fail_free_cpumask:
1437 free_cpumask_var(buffer->cpumask);
1438
1439 fail_free_buffer:
1440 kfree(buffer);
1441 return NULL;
1442 }
1443 EXPORT_SYMBOL_GPL(__ring_buffer_alloc);
1444
1445 /**
1446 * ring_buffer_free - free a ring buffer.
1447 * @buffer: the buffer to free.
1448 */
1449 void
ring_buffer_free(struct ring_buffer * buffer)1450 ring_buffer_free(struct ring_buffer *buffer)
1451 {
1452 int cpu;
1453
1454 cpuhp_state_remove_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node);
1455
1456 for_each_buffer_cpu(buffer, cpu)
1457 rb_free_cpu_buffer(buffer->buffers[cpu]);
1458
1459 kfree(buffer->buffers);
1460 free_cpumask_var(buffer->cpumask);
1461
1462 kfree(buffer);
1463 }
1464 EXPORT_SYMBOL_GPL(ring_buffer_free);
1465
ring_buffer_set_clock(struct ring_buffer * buffer,u64 (* clock)(void))1466 void ring_buffer_set_clock(struct ring_buffer *buffer,
1467 u64 (*clock)(void))
1468 {
1469 buffer->clock = clock;
1470 }
1471
ring_buffer_set_time_stamp_abs(struct ring_buffer * buffer,bool abs)1472 void ring_buffer_set_time_stamp_abs(struct ring_buffer *buffer, bool abs)
1473 {
1474 buffer->time_stamp_abs = abs;
1475 }
1476
ring_buffer_time_stamp_abs(struct ring_buffer * buffer)1477 bool ring_buffer_time_stamp_abs(struct ring_buffer *buffer)
1478 {
1479 return buffer->time_stamp_abs;
1480 }
1481
1482 static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
1483
rb_page_entries(struct buffer_page * bpage)1484 static inline unsigned long rb_page_entries(struct buffer_page *bpage)
1485 {
1486 return local_read(&bpage->entries) & RB_WRITE_MASK;
1487 }
1488
rb_page_write(struct buffer_page * bpage)1489 static inline unsigned long rb_page_write(struct buffer_page *bpage)
1490 {
1491 return local_read(&bpage->write) & RB_WRITE_MASK;
1492 }
1493
1494 static int
rb_remove_pages(struct ring_buffer_per_cpu * cpu_buffer,unsigned long nr_pages)1495 rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned long nr_pages)
1496 {
1497 struct list_head *tail_page, *to_remove, *next_page;
1498 struct buffer_page *to_remove_page, *tmp_iter_page;
1499 struct buffer_page *last_page, *first_page;
1500 unsigned long nr_removed;
1501 unsigned long head_bit;
1502 int page_entries;
1503
1504 head_bit = 0;
1505
1506 raw_spin_lock_irq(&cpu_buffer->reader_lock);
1507 atomic_inc(&cpu_buffer->record_disabled);
1508 /*
1509 * We don't race with the readers since we have acquired the reader
1510 * lock. We also don't race with writers after disabling recording.
1511 * This makes it easy to figure out the first and the last page to be
1512 * removed from the list. We unlink all the pages in between including
1513 * the first and last pages. This is done in a busy loop so that we
1514 * lose the least number of traces.
1515 * The pages are freed after we restart recording and unlock readers.
1516 */
1517 tail_page = &cpu_buffer->tail_page->list;
1518
1519 /*
1520 * tail page might be on reader page, we remove the next page
1521 * from the ring buffer
1522 */
1523 if (cpu_buffer->tail_page == cpu_buffer->reader_page)
1524 tail_page = rb_list_head(tail_page->next);
1525 to_remove = tail_page;
1526
1527 /* start of pages to remove */
1528 first_page = list_entry(rb_list_head(to_remove->next),
1529 struct buffer_page, list);
1530
1531 for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) {
1532 to_remove = rb_list_head(to_remove)->next;
1533 head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD;
1534 }
1535
1536 next_page = rb_list_head(to_remove)->next;
1537
1538 /*
1539 * Now we remove all pages between tail_page and next_page.
1540 * Make sure that we have head_bit value preserved for the
1541 * next page
1542 */
1543 tail_page->next = (struct list_head *)((unsigned long)next_page |
1544 head_bit);
1545 next_page = rb_list_head(next_page);
1546 next_page->prev = tail_page;
1547
1548 /* make sure pages points to a valid page in the ring buffer */
1549 cpu_buffer->pages = next_page;
1550
1551 /* update head page */
1552 if (head_bit)
1553 cpu_buffer->head_page = list_entry(next_page,
1554 struct buffer_page, list);
1555
1556 /*
1557 * change read pointer to make sure any read iterators reset
1558 * themselves
1559 */
1560 cpu_buffer->read = 0;
1561
1562 /* pages are removed, resume tracing and then free the pages */
1563 atomic_dec(&cpu_buffer->record_disabled);
1564 raw_spin_unlock_irq(&cpu_buffer->reader_lock);
1565
1566 RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages));
1567
1568 /* last buffer page to remove */
1569 last_page = list_entry(rb_list_head(to_remove), struct buffer_page,
1570 list);
1571 tmp_iter_page = first_page;
1572
1573 do {
1574 cond_resched();
1575
1576 to_remove_page = tmp_iter_page;
1577 rb_inc_page(cpu_buffer, &tmp_iter_page);
1578
1579 /* update the counters */
1580 page_entries = rb_page_entries(to_remove_page);
1581 if (page_entries) {
1582 /*
1583 * If something was added to this page, it was full
1584 * since it is not the tail page. So we deduct the
1585 * bytes consumed in ring buffer from here.
1586 * Increment overrun to account for the lost events.
1587 */
1588 local_add(page_entries, &cpu_buffer->overrun);
1589 local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
1590 }
1591
1592 /*
1593 * We have already removed references to this list item, just
1594 * free up the buffer_page and its page
1595 */
1596 free_buffer_page(to_remove_page);
1597 nr_removed--;
1598
1599 } while (to_remove_page != last_page);
1600
1601 RB_WARN_ON(cpu_buffer, nr_removed);
1602
1603 return nr_removed == 0;
1604 }
1605
1606 static int
rb_insert_pages(struct ring_buffer_per_cpu * cpu_buffer)1607 rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer)
1608 {
1609 struct list_head *pages = &cpu_buffer->new_pages;
1610 int retries, success;
1611
1612 raw_spin_lock_irq(&cpu_buffer->reader_lock);
1613 /*
1614 * We are holding the reader lock, so the reader page won't be swapped
1615 * in the ring buffer. Now we are racing with the writer trying to
1616 * move head page and the tail page.
1617 * We are going to adapt the reader page update process where:
1618 * 1. We first splice the start and end of list of new pages between
1619 * the head page and its previous page.
1620 * 2. We cmpxchg the prev_page->next to point from head page to the
1621 * start of new pages list.
1622 * 3. Finally, we update the head->prev to the end of new list.
1623 *
1624 * We will try this process 10 times, to make sure that we don't keep
1625 * spinning.
1626 */
1627 retries = 10;
1628 success = 0;
1629 while (retries--) {
1630 struct list_head *head_page, *prev_page, *r;
1631 struct list_head *last_page, *first_page;
1632 struct list_head *head_page_with_bit;
1633
1634 head_page = &rb_set_head_page(cpu_buffer)->list;
1635 if (!head_page)
1636 break;
1637 prev_page = head_page->prev;
1638
1639 first_page = pages->next;
1640 last_page = pages->prev;
1641
1642 head_page_with_bit = (struct list_head *)
1643 ((unsigned long)head_page | RB_PAGE_HEAD);
1644
1645 last_page->next = head_page_with_bit;
1646 first_page->prev = prev_page;
1647
1648 r = cmpxchg(&prev_page->next, head_page_with_bit, first_page);
1649
1650 if (r == head_page_with_bit) {
1651 /*
1652 * yay, we replaced the page pointer to our new list,
1653 * now, we just have to update to head page's prev
1654 * pointer to point to end of list
1655 */
1656 head_page->prev = last_page;
1657 success = 1;
1658 break;
1659 }
1660 }
1661
1662 if (success)
1663 INIT_LIST_HEAD(pages);
1664 /*
1665 * If we weren't successful in adding in new pages, warn and stop
1666 * tracing
1667 */
1668 RB_WARN_ON(cpu_buffer, !success);
1669 raw_spin_unlock_irq(&cpu_buffer->reader_lock);
1670
1671 /* free pages if they weren't inserted */
1672 if (!success) {
1673 struct buffer_page *bpage, *tmp;
1674 list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
1675 list) {
1676 list_del_init(&bpage->list);
1677 free_buffer_page(bpage);
1678 }
1679 }
1680 return success;
1681 }
1682
rb_update_pages(struct ring_buffer_per_cpu * cpu_buffer)1683 static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer)
1684 {
1685 int success;
1686
1687 if (cpu_buffer->nr_pages_to_update > 0)
1688 success = rb_insert_pages(cpu_buffer);
1689 else
1690 success = rb_remove_pages(cpu_buffer,
1691 -cpu_buffer->nr_pages_to_update);
1692
1693 if (success)
1694 cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update;
1695 }
1696
update_pages_handler(struct work_struct * work)1697 static void update_pages_handler(struct work_struct *work)
1698 {
1699 struct ring_buffer_per_cpu *cpu_buffer = container_of(work,
1700 struct ring_buffer_per_cpu, update_pages_work);
1701 rb_update_pages(cpu_buffer);
1702 complete(&cpu_buffer->update_done);
1703 }
1704
1705 /**
1706 * ring_buffer_resize - resize the ring buffer
1707 * @buffer: the buffer to resize.
1708 * @size: the new size.
1709 * @cpu_id: the cpu buffer to resize
1710 *
1711 * Minimum size is 2 * BUF_PAGE_SIZE.
1712 *
1713 * Returns 0 on success and < 0 on failure.
1714 */
ring_buffer_resize(struct ring_buffer * buffer,unsigned long size,int cpu_id)1715 int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size,
1716 int cpu_id)
1717 {
1718 struct ring_buffer_per_cpu *cpu_buffer;
1719 unsigned long nr_pages;
1720 int cpu, err = 0;
1721
1722 /*
1723 * Always succeed at resizing a non-existent buffer:
1724 */
1725 if (!buffer)
1726 return size;
1727
1728 /* Make sure the requested buffer exists */
1729 if (cpu_id != RING_BUFFER_ALL_CPUS &&
1730 !cpumask_test_cpu(cpu_id, buffer->cpumask))
1731 return size;
1732
1733 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
1734
1735 /* we need a minimum of two pages */
1736 if (nr_pages < 2)
1737 nr_pages = 2;
1738
1739 size = nr_pages * BUF_PAGE_SIZE;
1740
1741 /*
1742 * Don't succeed if resizing is disabled, as a reader might be
1743 * manipulating the ring buffer and is expecting a sane state while
1744 * this is true.
1745 */
1746 if (atomic_read(&buffer->resize_disabled))
1747 return -EBUSY;
1748
1749 /* prevent another thread from changing buffer sizes */
1750 mutex_lock(&buffer->mutex);
1751
1752 if (cpu_id == RING_BUFFER_ALL_CPUS) {
1753 /* calculate the pages to update */
1754 for_each_buffer_cpu(buffer, cpu) {
1755 cpu_buffer = buffer->buffers[cpu];
1756
1757 cpu_buffer->nr_pages_to_update = nr_pages -
1758 cpu_buffer->nr_pages;
1759 /*
1760 * nothing more to do for removing pages or no update
1761 */
1762 if (cpu_buffer->nr_pages_to_update <= 0)
1763 continue;
1764 /*
1765 * to add pages, make sure all new pages can be
1766 * allocated without receiving ENOMEM
1767 */
1768 INIT_LIST_HEAD(&cpu_buffer->new_pages);
1769 if (__rb_allocate_pages(cpu_buffer->nr_pages_to_update,
1770 &cpu_buffer->new_pages, cpu)) {
1771 /* not enough memory for new pages */
1772 err = -ENOMEM;
1773 goto out_err;
1774 }
1775 }
1776
1777 get_online_cpus();
1778 /*
1779 * Fire off all the required work handlers
1780 * We can't schedule on offline CPUs, but it's not necessary
1781 * since we can change their buffer sizes without any race.
1782 */
1783 for_each_buffer_cpu(buffer, cpu) {
1784 cpu_buffer = buffer->buffers[cpu];
1785 if (!cpu_buffer->nr_pages_to_update)
1786 continue;
1787
1788 /* Can't run something on an offline CPU. */
1789 if (!cpu_online(cpu)) {
1790 rb_update_pages(cpu_buffer);
1791 cpu_buffer->nr_pages_to_update = 0;
1792 } else {
1793 schedule_work_on(cpu,
1794 &cpu_buffer->update_pages_work);
1795 }
1796 }
1797
1798 /* wait for all the updates to complete */
1799 for_each_buffer_cpu(buffer, cpu) {
1800 cpu_buffer = buffer->buffers[cpu];
1801 if (!cpu_buffer->nr_pages_to_update)
1802 continue;
1803
1804 if (cpu_online(cpu))
1805 wait_for_completion(&cpu_buffer->update_done);
1806 cpu_buffer->nr_pages_to_update = 0;
1807 }
1808
1809 put_online_cpus();
1810 } else {
1811 /* Make sure this CPU has been initialized */
1812 if (!cpumask_test_cpu(cpu_id, buffer->cpumask))
1813 goto out;
1814
1815 cpu_buffer = buffer->buffers[cpu_id];
1816
1817 if (nr_pages == cpu_buffer->nr_pages)
1818 goto out;
1819
1820 cpu_buffer->nr_pages_to_update = nr_pages -
1821 cpu_buffer->nr_pages;
1822
1823 INIT_LIST_HEAD(&cpu_buffer->new_pages);
1824 if (cpu_buffer->nr_pages_to_update > 0 &&
1825 __rb_allocate_pages(cpu_buffer->nr_pages_to_update,
1826 &cpu_buffer->new_pages, cpu_id)) {
1827 err = -ENOMEM;
1828 goto out_err;
1829 }
1830
1831 get_online_cpus();
1832
1833 /* Can't run something on an offline CPU. */
1834 if (!cpu_online(cpu_id))
1835 rb_update_pages(cpu_buffer);
1836 else {
1837 schedule_work_on(cpu_id,
1838 &cpu_buffer->update_pages_work);
1839 wait_for_completion(&cpu_buffer->update_done);
1840 }
1841
1842 cpu_buffer->nr_pages_to_update = 0;
1843 put_online_cpus();
1844 }
1845
1846 out:
1847 /*
1848 * The ring buffer resize can happen with the ring buffer
1849 * enabled, so that the update disturbs the tracing as little
1850 * as possible. But if the buffer is disabled, we do not need
1851 * to worry about that, and we can take the time to verify
1852 * that the buffer is not corrupt.
1853 */
1854 if (atomic_read(&buffer->record_disabled)) {
1855 atomic_inc(&buffer->record_disabled);
1856 /*
1857 * Even though the buffer was disabled, we must make sure
1858 * that it is truly disabled before calling rb_check_pages.
1859 * There could have been a race between checking
1860 * record_disable and incrementing it.
1861 */
1862 synchronize_rcu();
1863 for_each_buffer_cpu(buffer, cpu) {
1864 cpu_buffer = buffer->buffers[cpu];
1865 rb_check_pages(cpu_buffer);
1866 }
1867 atomic_dec(&buffer->record_disabled);
1868 }
1869
1870 mutex_unlock(&buffer->mutex);
1871 return size;
1872
1873 out_err:
1874 for_each_buffer_cpu(buffer, cpu) {
1875 struct buffer_page *bpage, *tmp;
1876
1877 cpu_buffer = buffer->buffers[cpu];
1878 cpu_buffer->nr_pages_to_update = 0;
1879
1880 if (list_empty(&cpu_buffer->new_pages))
1881 continue;
1882
1883 list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
1884 list) {
1885 list_del_init(&bpage->list);
1886 free_buffer_page(bpage);
1887 }
1888 }
1889 mutex_unlock(&buffer->mutex);
1890 return err;
1891 }
1892 EXPORT_SYMBOL_GPL(ring_buffer_resize);
1893
ring_buffer_change_overwrite(struct ring_buffer * buffer,int val)1894 void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val)
1895 {
1896 mutex_lock(&buffer->mutex);
1897 if (val)
1898 buffer->flags |= RB_FL_OVERWRITE;
1899 else
1900 buffer->flags &= ~RB_FL_OVERWRITE;
1901 mutex_unlock(&buffer->mutex);
1902 }
1903 EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite);
1904
__rb_page_index(struct buffer_page * bpage,unsigned index)1905 static __always_inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
1906 {
1907 return bpage->page->data + index;
1908 }
1909
1910 static __always_inline struct ring_buffer_event *
rb_reader_event(struct ring_buffer_per_cpu * cpu_buffer)1911 rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
1912 {
1913 return __rb_page_index(cpu_buffer->reader_page,
1914 cpu_buffer->reader_page->read);
1915 }
1916
1917 static __always_inline struct ring_buffer_event *
rb_iter_head_event(struct ring_buffer_iter * iter)1918 rb_iter_head_event(struct ring_buffer_iter *iter)
1919 {
1920 return __rb_page_index(iter->head_page, iter->head);
1921 }
1922
rb_page_commit(struct buffer_page * bpage)1923 static __always_inline unsigned rb_page_commit(struct buffer_page *bpage)
1924 {
1925 return local_read(&bpage->page->commit);
1926 }
1927
1928 /* Size is determined by what has been committed */
rb_page_size(struct buffer_page * bpage)1929 static __always_inline unsigned rb_page_size(struct buffer_page *bpage)
1930 {
1931 return rb_page_commit(bpage);
1932 }
1933
1934 static __always_inline unsigned
rb_commit_index(struct ring_buffer_per_cpu * cpu_buffer)1935 rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
1936 {
1937 return rb_page_commit(cpu_buffer->commit_page);
1938 }
1939
1940 static __always_inline unsigned
rb_event_index(struct ring_buffer_event * event)1941 rb_event_index(struct ring_buffer_event *event)
1942 {
1943 unsigned long addr = (unsigned long)event;
1944
1945 return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE;
1946 }
1947
rb_inc_iter(struct ring_buffer_iter * iter)1948 static void rb_inc_iter(struct ring_buffer_iter *iter)
1949 {
1950 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
1951
1952 /*
1953 * The iterator could be on the reader page (it starts there).
1954 * But the head could have moved, since the reader was
1955 * found. Check for this case and assign the iterator
1956 * to the head page instead of next.
1957 */
1958 if (iter->head_page == cpu_buffer->reader_page)
1959 iter->head_page = rb_set_head_page(cpu_buffer);
1960 else
1961 rb_inc_page(cpu_buffer, &iter->head_page);
1962
1963 iter->read_stamp = iter->head_page->page->time_stamp;
1964 iter->head = 0;
1965 }
1966
1967 /*
1968 * rb_handle_head_page - writer hit the head page
1969 *
1970 * Returns: +1 to retry page
1971 * 0 to continue
1972 * -1 on error
1973 */
1974 static int
rb_handle_head_page(struct ring_buffer_per_cpu * cpu_buffer,struct buffer_page * tail_page,struct buffer_page * next_page)1975 rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer,
1976 struct buffer_page *tail_page,
1977 struct buffer_page *next_page)
1978 {
1979 struct buffer_page *new_head;
1980 int entries;
1981 int type;
1982 int ret;
1983
1984 entries = rb_page_entries(next_page);
1985
1986 /*
1987 * The hard part is here. We need to move the head
1988 * forward, and protect against both readers on
1989 * other CPUs and writers coming in via interrupts.
1990 */
1991 type = rb_head_page_set_update(cpu_buffer, next_page, tail_page,
1992 RB_PAGE_HEAD);
1993
1994 /*
1995 * type can be one of four:
1996 * NORMAL - an interrupt already moved it for us
1997 * HEAD - we are the first to get here.
1998 * UPDATE - we are the interrupt interrupting
1999 * a current move.
2000 * MOVED - a reader on another CPU moved the next
2001 * pointer to its reader page. Give up
2002 * and try again.
2003 */
2004
2005 switch (type) {
2006 case RB_PAGE_HEAD:
2007 /*
2008 * We changed the head to UPDATE, thus
2009 * it is our responsibility to update
2010 * the counters.
2011 */
2012 local_add(entries, &cpu_buffer->overrun);
2013 local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
2014
2015 /*
2016 * The entries will be zeroed out when we move the
2017 * tail page.
2018 */
2019
2020 /* still more to do */
2021 break;
2022
2023 case RB_PAGE_UPDATE:
2024 /*
2025 * This is an interrupt that interrupt the
2026 * previous update. Still more to do.
2027 */
2028 break;
2029 case RB_PAGE_NORMAL:
2030 /*
2031 * An interrupt came in before the update
2032 * and processed this for us.
2033 * Nothing left to do.
2034 */
2035 return 1;
2036 case RB_PAGE_MOVED:
2037 /*
2038 * The reader is on another CPU and just did
2039 * a swap with our next_page.
2040 * Try again.
2041 */
2042 return 1;
2043 default:
2044 RB_WARN_ON(cpu_buffer, 1); /* WTF??? */
2045 return -1;
2046 }
2047
2048 /*
2049 * Now that we are here, the old head pointer is
2050 * set to UPDATE. This will keep the reader from
2051 * swapping the head page with the reader page.
2052 * The reader (on another CPU) will spin till
2053 * we are finished.
2054 *
2055 * We just need to protect against interrupts
2056 * doing the job. We will set the next pointer
2057 * to HEAD. After that, we set the old pointer
2058 * to NORMAL, but only if it was HEAD before.
2059 * otherwise we are an interrupt, and only
2060 * want the outer most commit to reset it.
2061 */
2062 new_head = next_page;
2063 rb_inc_page(cpu_buffer, &new_head);
2064
2065 ret = rb_head_page_set_head(cpu_buffer, new_head, next_page,
2066 RB_PAGE_NORMAL);
2067
2068 /*
2069 * Valid returns are:
2070 * HEAD - an interrupt came in and already set it.
2071 * NORMAL - One of two things:
2072 * 1) We really set it.
2073 * 2) A bunch of interrupts came in and moved
2074 * the page forward again.
2075 */
2076 switch (ret) {
2077 case RB_PAGE_HEAD:
2078 case RB_PAGE_NORMAL:
2079 /* OK */
2080 break;
2081 default:
2082 RB_WARN_ON(cpu_buffer, 1);
2083 return -1;
2084 }
2085
2086 /*
2087 * It is possible that an interrupt came in,
2088 * set the head up, then more interrupts came in
2089 * and moved it again. When we get back here,
2090 * the page would have been set to NORMAL but we
2091 * just set it back to HEAD.
2092 *
2093 * How do you detect this? Well, if that happened
2094 * the tail page would have moved.
2095 */
2096 if (ret == RB_PAGE_NORMAL) {
2097 struct buffer_page *buffer_tail_page;
2098
2099 buffer_tail_page = READ_ONCE(cpu_buffer->tail_page);
2100 /*
2101 * If the tail had moved passed next, then we need
2102 * to reset the pointer.
2103 */
2104 if (buffer_tail_page != tail_page &&
2105 buffer_tail_page != next_page)
2106 rb_head_page_set_normal(cpu_buffer, new_head,
2107 next_page,
2108 RB_PAGE_HEAD);
2109 }
2110
2111 /*
2112 * If this was the outer most commit (the one that
2113 * changed the original pointer from HEAD to UPDATE),
2114 * then it is up to us to reset it to NORMAL.
2115 */
2116 if (type == RB_PAGE_HEAD) {
2117 ret = rb_head_page_set_normal(cpu_buffer, next_page,
2118 tail_page,
2119 RB_PAGE_UPDATE);
2120 if (RB_WARN_ON(cpu_buffer,
2121 ret != RB_PAGE_UPDATE))
2122 return -1;
2123 }
2124
2125 return 0;
2126 }
2127
2128 static inline void
rb_reset_tail(struct ring_buffer_per_cpu * cpu_buffer,unsigned long tail,struct rb_event_info * info)2129 rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer,
2130 unsigned long tail, struct rb_event_info *info)
2131 {
2132 struct buffer_page *tail_page = info->tail_page;
2133 struct ring_buffer_event *event;
2134 unsigned long length = info->length;
2135
2136 /*
2137 * Only the event that crossed the page boundary
2138 * must fill the old tail_page with padding.
2139 */
2140 if (tail >= BUF_PAGE_SIZE) {
2141 /*
2142 * If the page was filled, then we still need
2143 * to update the real_end. Reset it to zero
2144 * and the reader will ignore it.
2145 */
2146 if (tail == BUF_PAGE_SIZE)
2147 tail_page->real_end = 0;
2148
2149 local_sub(length, &tail_page->write);
2150 return;
2151 }
2152
2153 event = __rb_page_index(tail_page, tail);
2154
2155 /* account for padding bytes */
2156 local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes);
2157
2158 /*
2159 * Save the original length to the meta data.
2160 * This will be used by the reader to add lost event
2161 * counter.
2162 */
2163 tail_page->real_end = tail;
2164
2165 /*
2166 * If this event is bigger than the minimum size, then
2167 * we need to be careful that we don't subtract the
2168 * write counter enough to allow another writer to slip
2169 * in on this page.
2170 * We put in a discarded commit instead, to make sure
2171 * that this space is not used again.
2172 *
2173 * If we are less than the minimum size, we don't need to
2174 * worry about it.
2175 */
2176 if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) {
2177 /* No room for any events */
2178
2179 /* Mark the rest of the page with padding */
2180 rb_event_set_padding(event);
2181
2182 /* Set the write back to the previous setting */
2183 local_sub(length, &tail_page->write);
2184 return;
2185 }
2186
2187 /* Put in a discarded event */
2188 event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE;
2189 event->type_len = RINGBUF_TYPE_PADDING;
2190 /* time delta must be non zero */
2191 event->time_delta = 1;
2192
2193 /* Set write to end of buffer */
2194 length = (tail + length) - BUF_PAGE_SIZE;
2195 local_sub(length, &tail_page->write);
2196 }
2197
2198 static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer);
2199
2200 /*
2201 * This is the slow path, force gcc not to inline it.
2202 */
2203 static noinline struct ring_buffer_event *
rb_move_tail(struct ring_buffer_per_cpu * cpu_buffer,unsigned long tail,struct rb_event_info * info)2204 rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer,
2205 unsigned long tail, struct rb_event_info *info)
2206 {
2207 struct buffer_page *tail_page = info->tail_page;
2208 struct buffer_page *commit_page = cpu_buffer->commit_page;
2209 struct ring_buffer *buffer = cpu_buffer->buffer;
2210 struct buffer_page *next_page;
2211 int ret;
2212
2213 next_page = tail_page;
2214
2215 rb_inc_page(cpu_buffer, &next_page);
2216
2217 /*
2218 * If for some reason, we had an interrupt storm that made
2219 * it all the way around the buffer, bail, and warn
2220 * about it.
2221 */
2222 if (unlikely(next_page == commit_page)) {
2223 local_inc(&cpu_buffer->commit_overrun);
2224 goto out_reset;
2225 }
2226
2227 /*
2228 * This is where the fun begins!
2229 *
2230 * We are fighting against races between a reader that
2231 * could be on another CPU trying to swap its reader
2232 * page with the buffer head.
2233 *
2234 * We are also fighting against interrupts coming in and
2235 * moving the head or tail on us as well.
2236 *
2237 * If the next page is the head page then we have filled
2238 * the buffer, unless the commit page is still on the
2239 * reader page.
2240 */
2241 if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) {
2242
2243 /*
2244 * If the commit is not on the reader page, then
2245 * move the header page.
2246 */
2247 if (!rb_is_reader_page(cpu_buffer->commit_page)) {
2248 /*
2249 * If we are not in overwrite mode,
2250 * this is easy, just stop here.
2251 */
2252 if (!(buffer->flags & RB_FL_OVERWRITE)) {
2253 local_inc(&cpu_buffer->dropped_events);
2254 goto out_reset;
2255 }
2256
2257 ret = rb_handle_head_page(cpu_buffer,
2258 tail_page,
2259 next_page);
2260 if (ret < 0)
2261 goto out_reset;
2262 if (ret)
2263 goto out_again;
2264 } else {
2265 /*
2266 * We need to be careful here too. The
2267 * commit page could still be on the reader
2268 * page. We could have a small buffer, and
2269 * have filled up the buffer with events
2270 * from interrupts and such, and wrapped.
2271 *
2272 * Note, if the tail page is also the on the
2273 * reader_page, we let it move out.
2274 */
2275 if (unlikely((cpu_buffer->commit_page !=
2276 cpu_buffer->tail_page) &&
2277 (cpu_buffer->commit_page ==
2278 cpu_buffer->reader_page))) {
2279 local_inc(&cpu_buffer->commit_overrun);
2280 goto out_reset;
2281 }
2282 }
2283 }
2284
2285 rb_tail_page_update(cpu_buffer, tail_page, next_page);
2286
2287 out_again:
2288
2289 rb_reset_tail(cpu_buffer, tail, info);
2290
2291 /* Commit what we have for now. */
2292 rb_end_commit(cpu_buffer);
2293 /* rb_end_commit() decs committing */
2294 local_inc(&cpu_buffer->committing);
2295
2296 /* fail and let the caller try again */
2297 return ERR_PTR(-EAGAIN);
2298
2299 out_reset:
2300 /* reset write */
2301 rb_reset_tail(cpu_buffer, tail, info);
2302
2303 return NULL;
2304 }
2305
2306 /* Slow path, do not inline */
2307 static noinline struct ring_buffer_event *
rb_add_time_stamp(struct ring_buffer_event * event,u64 delta,bool abs)2308 rb_add_time_stamp(struct ring_buffer_event *event, u64 delta, bool abs)
2309 {
2310 if (abs)
2311 event->type_len = RINGBUF_TYPE_TIME_STAMP;
2312 else
2313 event->type_len = RINGBUF_TYPE_TIME_EXTEND;
2314
2315 /* Not the first event on the page, or not delta? */
2316 if (abs || rb_event_index(event)) {
2317 event->time_delta = delta & TS_MASK;
2318 event->array[0] = delta >> TS_SHIFT;
2319 } else {
2320 /* nope, just zero it */
2321 event->time_delta = 0;
2322 event->array[0] = 0;
2323 }
2324
2325 return skip_time_extend(event);
2326 }
2327
2328 static inline bool rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
2329 struct ring_buffer_event *event);
2330
2331 /**
2332 * rb_update_event - update event type and data
2333 * @event: the event to update
2334 * @type: the type of event
2335 * @length: the size of the event field in the ring buffer
2336 *
2337 * Update the type and data fields of the event. The length
2338 * is the actual size that is written to the ring buffer,
2339 * and with this, we can determine what to place into the
2340 * data field.
2341 */
2342 static void
rb_update_event(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event,struct rb_event_info * info)2343 rb_update_event(struct ring_buffer_per_cpu *cpu_buffer,
2344 struct ring_buffer_event *event,
2345 struct rb_event_info *info)
2346 {
2347 unsigned length = info->length;
2348 u64 delta = info->delta;
2349
2350 /* Only a commit updates the timestamp */
2351 if (unlikely(!rb_event_is_commit(cpu_buffer, event)))
2352 delta = 0;
2353
2354 /*
2355 * If we need to add a timestamp, then we
2356 * add it to the start of the reserved space.
2357 */
2358 if (unlikely(info->add_timestamp)) {
2359 bool abs = ring_buffer_time_stamp_abs(cpu_buffer->buffer);
2360
2361 event = rb_add_time_stamp(event, info->delta, abs);
2362 length -= RB_LEN_TIME_EXTEND;
2363 delta = 0;
2364 }
2365
2366 event->time_delta = delta;
2367 length -= RB_EVNT_HDR_SIZE;
2368 if (length > RB_MAX_SMALL_DATA) {
2369 event->type_len = 0;
2370 event->array[0] = length;
2371 } else
2372 event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT);
2373 }
2374
rb_calculate_event_length(unsigned length)2375 static unsigned rb_calculate_event_length(unsigned length)
2376 {
2377 struct ring_buffer_event event; /* Used only for sizeof array */
2378
2379 /* zero length can cause confusions */
2380 if (!length)
2381 length++;
2382
2383 if (length > RB_MAX_SMALL_DATA)
2384 length += sizeof(event.array[0]);
2385
2386 length += RB_EVNT_HDR_SIZE;
2387 length = ALIGN(length, RB_ALIGNMENT);
2388
2389 /*
2390 * In case the time delta is larger than the 27 bits for it
2391 * in the header, we need to add a timestamp. If another
2392 * event comes in when trying to discard this one to increase
2393 * the length, then the timestamp will be added in the allocated
2394 * space of this event. If length is bigger than the size needed
2395 * for the TIME_EXTEND, then padding has to be used. The events
2396 * length must be either RB_LEN_TIME_EXTEND, or greater than or equal
2397 * to RB_LEN_TIME_EXTEND + 8, as 8 is the minimum size for padding.
2398 * As length is a multiple of 4, we only need to worry if it
2399 * is 12 (RB_LEN_TIME_EXTEND + 4).
2400 */
2401 if (length == RB_LEN_TIME_EXTEND + RB_ALIGNMENT)
2402 length += RB_ALIGNMENT;
2403
2404 return length;
2405 }
2406
2407 #ifndef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
sched_clock_stable(void)2408 static inline bool sched_clock_stable(void)
2409 {
2410 return true;
2411 }
2412 #endif
2413
2414 static inline int
rb_try_to_discard(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)2415 rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer,
2416 struct ring_buffer_event *event)
2417 {
2418 unsigned long new_index, old_index;
2419 struct buffer_page *bpage;
2420 unsigned long index;
2421 unsigned long addr;
2422
2423 new_index = rb_event_index(event);
2424 old_index = new_index + rb_event_ts_length(event);
2425 addr = (unsigned long)event;
2426 addr &= PAGE_MASK;
2427
2428 bpage = READ_ONCE(cpu_buffer->tail_page);
2429
2430 if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) {
2431 unsigned long write_mask =
2432 local_read(&bpage->write) & ~RB_WRITE_MASK;
2433 unsigned long event_length = rb_event_length(event);
2434 /*
2435 * This is on the tail page. It is possible that
2436 * a write could come in and move the tail page
2437 * and write to the next page. That is fine
2438 * because we just shorten what is on this page.
2439 */
2440 old_index += write_mask;
2441 new_index += write_mask;
2442 index = local_cmpxchg(&bpage->write, old_index, new_index);
2443 if (index == old_index) {
2444 /* update counters */
2445 local_sub(event_length, &cpu_buffer->entries_bytes);
2446 return 1;
2447 }
2448 }
2449
2450 /* could not discard */
2451 return 0;
2452 }
2453
rb_start_commit(struct ring_buffer_per_cpu * cpu_buffer)2454 static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer)
2455 {
2456 local_inc(&cpu_buffer->committing);
2457 local_inc(&cpu_buffer->commits);
2458 }
2459
2460 static __always_inline void
rb_set_commit_to_write(struct ring_buffer_per_cpu * cpu_buffer)2461 rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
2462 {
2463 unsigned long max_count;
2464
2465 /*
2466 * We only race with interrupts and NMIs on this CPU.
2467 * If we own the commit event, then we can commit
2468 * all others that interrupted us, since the interruptions
2469 * are in stack format (they finish before they come
2470 * back to us). This allows us to do a simple loop to
2471 * assign the commit to the tail.
2472 */
2473 again:
2474 max_count = cpu_buffer->nr_pages * 100;
2475
2476 while (cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)) {
2477 if (RB_WARN_ON(cpu_buffer, !(--max_count)))
2478 return;
2479 if (RB_WARN_ON(cpu_buffer,
2480 rb_is_reader_page(cpu_buffer->tail_page)))
2481 return;
2482 local_set(&cpu_buffer->commit_page->page->commit,
2483 rb_page_write(cpu_buffer->commit_page));
2484 rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
2485 /* Only update the write stamp if the page has an event */
2486 if (rb_page_write(cpu_buffer->commit_page))
2487 cpu_buffer->write_stamp =
2488 cpu_buffer->commit_page->page->time_stamp;
2489 /* add barrier to keep gcc from optimizing too much */
2490 barrier();
2491 }
2492 while (rb_commit_index(cpu_buffer) !=
2493 rb_page_write(cpu_buffer->commit_page)) {
2494
2495 local_set(&cpu_buffer->commit_page->page->commit,
2496 rb_page_write(cpu_buffer->commit_page));
2497 RB_WARN_ON(cpu_buffer,
2498 local_read(&cpu_buffer->commit_page->page->commit) &
2499 ~RB_WRITE_MASK);
2500 barrier();
2501 }
2502
2503 /* again, keep gcc from optimizing */
2504 barrier();
2505
2506 /*
2507 * If an interrupt came in just after the first while loop
2508 * and pushed the tail page forward, we will be left with
2509 * a dangling commit that will never go forward.
2510 */
2511 if (unlikely(cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)))
2512 goto again;
2513 }
2514
rb_end_commit(struct ring_buffer_per_cpu * cpu_buffer)2515 static __always_inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer)
2516 {
2517 unsigned long commits;
2518
2519 if (RB_WARN_ON(cpu_buffer,
2520 !local_read(&cpu_buffer->committing)))
2521 return;
2522
2523 again:
2524 commits = local_read(&cpu_buffer->commits);
2525 /* synchronize with interrupts */
2526 barrier();
2527 if (local_read(&cpu_buffer->committing) == 1)
2528 rb_set_commit_to_write(cpu_buffer);
2529
2530 local_dec(&cpu_buffer->committing);
2531
2532 /* synchronize with interrupts */
2533 barrier();
2534
2535 /*
2536 * Need to account for interrupts coming in between the
2537 * updating of the commit page and the clearing of the
2538 * committing counter.
2539 */
2540 if (unlikely(local_read(&cpu_buffer->commits) != commits) &&
2541 !local_read(&cpu_buffer->committing)) {
2542 local_inc(&cpu_buffer->committing);
2543 goto again;
2544 }
2545 }
2546
rb_event_discard(struct ring_buffer_event * event)2547 static inline void rb_event_discard(struct ring_buffer_event *event)
2548 {
2549 if (extended_time(event))
2550 event = skip_time_extend(event);
2551
2552 /* array[0] holds the actual length for the discarded event */
2553 event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE;
2554 event->type_len = RINGBUF_TYPE_PADDING;
2555 /* time delta must be non zero */
2556 if (!event->time_delta)
2557 event->time_delta = 1;
2558 }
2559
2560 static __always_inline bool
rb_event_is_commit(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)2561 rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
2562 struct ring_buffer_event *event)
2563 {
2564 unsigned long addr = (unsigned long)event;
2565 unsigned long index;
2566
2567 index = rb_event_index(event);
2568 addr &= PAGE_MASK;
2569
2570 return cpu_buffer->commit_page->page == (void *)addr &&
2571 rb_commit_index(cpu_buffer) == index;
2572 }
2573
2574 static __always_inline void
rb_update_write_stamp(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)2575 rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer,
2576 struct ring_buffer_event *event)
2577 {
2578 u64 delta;
2579
2580 /*
2581 * The event first in the commit queue updates the
2582 * time stamp.
2583 */
2584 if (rb_event_is_commit(cpu_buffer, event)) {
2585 /*
2586 * A commit event that is first on a page
2587 * updates the write timestamp with the page stamp
2588 */
2589 if (!rb_event_index(event))
2590 cpu_buffer->write_stamp =
2591 cpu_buffer->commit_page->page->time_stamp;
2592 else if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) {
2593 delta = ring_buffer_event_time_stamp(event);
2594 cpu_buffer->write_stamp += delta;
2595 } else if (event->type_len == RINGBUF_TYPE_TIME_STAMP) {
2596 delta = ring_buffer_event_time_stamp(event);
2597 cpu_buffer->write_stamp = delta;
2598 } else
2599 cpu_buffer->write_stamp += event->time_delta;
2600 }
2601 }
2602
rb_commit(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)2603 static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
2604 struct ring_buffer_event *event)
2605 {
2606 local_inc(&cpu_buffer->entries);
2607 rb_update_write_stamp(cpu_buffer, event);
2608 rb_end_commit(cpu_buffer);
2609 }
2610
2611 static __always_inline void
rb_wakeups(struct ring_buffer * buffer,struct ring_buffer_per_cpu * cpu_buffer)2612 rb_wakeups(struct ring_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer)
2613 {
2614 size_t nr_pages;
2615 size_t dirty;
2616 size_t full;
2617
2618 if (buffer->irq_work.waiters_pending) {
2619 buffer->irq_work.waiters_pending = false;
2620 /* irq_work_queue() supplies it's own memory barriers */
2621 irq_work_queue(&buffer->irq_work.work);
2622 }
2623
2624 if (cpu_buffer->irq_work.waiters_pending) {
2625 cpu_buffer->irq_work.waiters_pending = false;
2626 /* irq_work_queue() supplies it's own memory barriers */
2627 irq_work_queue(&cpu_buffer->irq_work.work);
2628 }
2629
2630 if (cpu_buffer->last_pages_touch == local_read(&cpu_buffer->pages_touched))
2631 return;
2632
2633 if (cpu_buffer->reader_page == cpu_buffer->commit_page)
2634 return;
2635
2636 if (!cpu_buffer->irq_work.full_waiters_pending)
2637 return;
2638
2639 cpu_buffer->last_pages_touch = local_read(&cpu_buffer->pages_touched);
2640
2641 full = cpu_buffer->shortest_full;
2642 nr_pages = cpu_buffer->nr_pages;
2643 dirty = ring_buffer_nr_dirty_pages(buffer, cpu_buffer->cpu);
2644 if (full && nr_pages && (dirty * 100) <= full * nr_pages)
2645 return;
2646
2647 cpu_buffer->irq_work.wakeup_full = true;
2648 cpu_buffer->irq_work.full_waiters_pending = false;
2649 /* irq_work_queue() supplies it's own memory barriers */
2650 irq_work_queue(&cpu_buffer->irq_work.work);
2651 }
2652
2653 /*
2654 * The lock and unlock are done within a preempt disable section.
2655 * The current_context per_cpu variable can only be modified
2656 * by the current task between lock and unlock. But it can
2657 * be modified more than once via an interrupt. To pass this
2658 * information from the lock to the unlock without having to
2659 * access the 'in_interrupt()' functions again (which do show
2660 * a bit of overhead in something as critical as function tracing,
2661 * we use a bitmask trick.
2662 *
2663 * bit 0 = NMI context
2664 * bit 1 = IRQ context
2665 * bit 2 = SoftIRQ context
2666 * bit 3 = normal context.
2667 *
2668 * This works because this is the order of contexts that can
2669 * preempt other contexts. A SoftIRQ never preempts an IRQ
2670 * context.
2671 *
2672 * When the context is determined, the corresponding bit is
2673 * checked and set (if it was set, then a recursion of that context
2674 * happened).
2675 *
2676 * On unlock, we need to clear this bit. To do so, just subtract
2677 * 1 from the current_context and AND it to itself.
2678 *
2679 * (binary)
2680 * 101 - 1 = 100
2681 * 101 & 100 = 100 (clearing bit zero)
2682 *
2683 * 1010 - 1 = 1001
2684 * 1010 & 1001 = 1000 (clearing bit 1)
2685 *
2686 * The least significant bit can be cleared this way, and it
2687 * just so happens that it is the same bit corresponding to
2688 * the current context.
2689 */
2690
2691 static __always_inline int
trace_recursive_lock(struct ring_buffer_per_cpu * cpu_buffer)2692 trace_recursive_lock(struct ring_buffer_per_cpu *cpu_buffer)
2693 {
2694 unsigned int val = cpu_buffer->current_context;
2695 unsigned long pc = preempt_count();
2696 int bit;
2697
2698 if (!(pc & (NMI_MASK | HARDIRQ_MASK | SOFTIRQ_OFFSET)))
2699 bit = RB_CTX_NORMAL;
2700 else
2701 bit = pc & NMI_MASK ? RB_CTX_NMI :
2702 pc & HARDIRQ_MASK ? RB_CTX_IRQ : RB_CTX_SOFTIRQ;
2703
2704 if (unlikely(val & (1 << (bit + cpu_buffer->nest))))
2705 return 1;
2706
2707 val |= (1 << (bit + cpu_buffer->nest));
2708 cpu_buffer->current_context = val;
2709
2710 return 0;
2711 }
2712
2713 static __always_inline void
trace_recursive_unlock(struct ring_buffer_per_cpu * cpu_buffer)2714 trace_recursive_unlock(struct ring_buffer_per_cpu *cpu_buffer)
2715 {
2716 cpu_buffer->current_context &=
2717 cpu_buffer->current_context - (1 << cpu_buffer->nest);
2718 }
2719
2720 /* The recursive locking above uses 4 bits */
2721 #define NESTED_BITS 4
2722
2723 /**
2724 * ring_buffer_nest_start - Allow to trace while nested
2725 * @buffer: The ring buffer to modify
2726 *
2727 * The ring buffer has a safety mechanism to prevent recursion.
2728 * But there may be a case where a trace needs to be done while
2729 * tracing something else. In this case, calling this function
2730 * will allow this function to nest within a currently active
2731 * ring_buffer_lock_reserve().
2732 *
2733 * Call this function before calling another ring_buffer_lock_reserve() and
2734 * call ring_buffer_nest_end() after the nested ring_buffer_unlock_commit().
2735 */
ring_buffer_nest_start(struct ring_buffer * buffer)2736 void ring_buffer_nest_start(struct ring_buffer *buffer)
2737 {
2738 struct ring_buffer_per_cpu *cpu_buffer;
2739 int cpu;
2740
2741 /* Enabled by ring_buffer_nest_end() */
2742 preempt_disable_notrace();
2743 cpu = raw_smp_processor_id();
2744 cpu_buffer = buffer->buffers[cpu];
2745 /* This is the shift value for the above recursive locking */
2746 cpu_buffer->nest += NESTED_BITS;
2747 }
2748
2749 /**
2750 * ring_buffer_nest_end - Allow to trace while nested
2751 * @buffer: The ring buffer to modify
2752 *
2753 * Must be called after ring_buffer_nest_start() and after the
2754 * ring_buffer_unlock_commit().
2755 */
ring_buffer_nest_end(struct ring_buffer * buffer)2756 void ring_buffer_nest_end(struct ring_buffer *buffer)
2757 {
2758 struct ring_buffer_per_cpu *cpu_buffer;
2759 int cpu;
2760
2761 /* disabled by ring_buffer_nest_start() */
2762 cpu = raw_smp_processor_id();
2763 cpu_buffer = buffer->buffers[cpu];
2764 /* This is the shift value for the above recursive locking */
2765 cpu_buffer->nest -= NESTED_BITS;
2766 preempt_enable_notrace();
2767 }
2768
2769 /**
2770 * ring_buffer_unlock_commit - commit a reserved
2771 * @buffer: The buffer to commit to
2772 * @event: The event pointer to commit.
2773 *
2774 * This commits the data to the ring buffer, and releases any locks held.
2775 *
2776 * Must be paired with ring_buffer_lock_reserve.
2777 */
ring_buffer_unlock_commit(struct ring_buffer * buffer,struct ring_buffer_event * event)2778 int ring_buffer_unlock_commit(struct ring_buffer *buffer,
2779 struct ring_buffer_event *event)
2780 {
2781 struct ring_buffer_per_cpu *cpu_buffer;
2782 int cpu = raw_smp_processor_id();
2783
2784 cpu_buffer = buffer->buffers[cpu];
2785
2786 rb_commit(cpu_buffer, event);
2787
2788 rb_wakeups(buffer, cpu_buffer);
2789
2790 trace_recursive_unlock(cpu_buffer);
2791
2792 preempt_enable_notrace();
2793
2794 return 0;
2795 }
2796 EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit);
2797
2798 static noinline void
rb_handle_timestamp(struct ring_buffer_per_cpu * cpu_buffer,struct rb_event_info * info)2799 rb_handle_timestamp(struct ring_buffer_per_cpu *cpu_buffer,
2800 struct rb_event_info *info)
2801 {
2802 WARN_ONCE(info->delta > (1ULL << 59),
2803 KERN_WARNING "Delta way too big! %llu ts=%llu write stamp = %llu\n%s",
2804 (unsigned long long)info->delta,
2805 (unsigned long long)info->ts,
2806 (unsigned long long)cpu_buffer->write_stamp,
2807 sched_clock_stable() ? "" :
2808 "If you just came from a suspend/resume,\n"
2809 "please switch to the trace global clock:\n"
2810 " echo global > /sys/kernel/debug/tracing/trace_clock\n"
2811 "or add trace_clock=global to the kernel command line\n");
2812 info->add_timestamp = 1;
2813 }
2814
2815 static struct ring_buffer_event *
__rb_reserve_next(struct ring_buffer_per_cpu * cpu_buffer,struct rb_event_info * info)2816 __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
2817 struct rb_event_info *info)
2818 {
2819 struct ring_buffer_event *event;
2820 struct buffer_page *tail_page;
2821 unsigned long tail, write;
2822
2823 /*
2824 * If the time delta since the last event is too big to
2825 * hold in the time field of the event, then we append a
2826 * TIME EXTEND event ahead of the data event.
2827 */
2828 if (unlikely(info->add_timestamp))
2829 info->length += RB_LEN_TIME_EXTEND;
2830
2831 /* Don't let the compiler play games with cpu_buffer->tail_page */
2832 tail_page = info->tail_page = READ_ONCE(cpu_buffer->tail_page);
2833 write = local_add_return(info->length, &tail_page->write);
2834
2835 /* set write to only the index of the write */
2836 write &= RB_WRITE_MASK;
2837 tail = write - info->length;
2838
2839 /*
2840 * If this is the first commit on the page, then it has the same
2841 * timestamp as the page itself.
2842 */
2843 if (!tail && !ring_buffer_time_stamp_abs(cpu_buffer->buffer))
2844 info->delta = 0;
2845
2846 /* See if we shot pass the end of this buffer page */
2847 if (unlikely(write > BUF_PAGE_SIZE))
2848 return rb_move_tail(cpu_buffer, tail, info);
2849
2850 /* We reserved something on the buffer */
2851
2852 event = __rb_page_index(tail_page, tail);
2853 rb_update_event(cpu_buffer, event, info);
2854
2855 local_inc(&tail_page->entries);
2856
2857 /*
2858 * If this is the first commit on the page, then update
2859 * its timestamp.
2860 */
2861 if (!tail)
2862 tail_page->page->time_stamp = info->ts;
2863
2864 /* account for these added bytes */
2865 local_add(info->length, &cpu_buffer->entries_bytes);
2866
2867 return event;
2868 }
2869
2870 static __always_inline struct ring_buffer_event *
rb_reserve_next_event(struct ring_buffer * buffer,struct ring_buffer_per_cpu * cpu_buffer,unsigned long length)2871 rb_reserve_next_event(struct ring_buffer *buffer,
2872 struct ring_buffer_per_cpu *cpu_buffer,
2873 unsigned long length)
2874 {
2875 struct ring_buffer_event *event;
2876 struct rb_event_info info;
2877 int nr_loops = 0;
2878 u64 diff;
2879
2880 rb_start_commit(cpu_buffer);
2881
2882 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
2883 /*
2884 * Due to the ability to swap a cpu buffer from a buffer
2885 * it is possible it was swapped before we committed.
2886 * (committing stops a swap). We check for it here and
2887 * if it happened, we have to fail the write.
2888 */
2889 barrier();
2890 if (unlikely(READ_ONCE(cpu_buffer->buffer) != buffer)) {
2891 local_dec(&cpu_buffer->committing);
2892 local_dec(&cpu_buffer->commits);
2893 return NULL;
2894 }
2895 #endif
2896
2897 info.length = rb_calculate_event_length(length);
2898 again:
2899 info.add_timestamp = 0;
2900 info.delta = 0;
2901
2902 /*
2903 * We allow for interrupts to reenter here and do a trace.
2904 * If one does, it will cause this original code to loop
2905 * back here. Even with heavy interrupts happening, this
2906 * should only happen a few times in a row. If this happens
2907 * 1000 times in a row, there must be either an interrupt
2908 * storm or we have something buggy.
2909 * Bail!
2910 */
2911 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
2912 goto out_fail;
2913
2914 info.ts = rb_time_stamp(cpu_buffer->buffer);
2915 diff = info.ts - cpu_buffer->write_stamp;
2916
2917 /* make sure this diff is calculated here */
2918 barrier();
2919
2920 if (ring_buffer_time_stamp_abs(buffer)) {
2921 info.delta = info.ts;
2922 rb_handle_timestamp(cpu_buffer, &info);
2923 } else /* Did the write stamp get updated already? */
2924 if (likely(info.ts >= cpu_buffer->write_stamp)) {
2925 info.delta = diff;
2926 if (unlikely(test_time_stamp(info.delta)))
2927 rb_handle_timestamp(cpu_buffer, &info);
2928 }
2929
2930 event = __rb_reserve_next(cpu_buffer, &info);
2931
2932 if (unlikely(PTR_ERR(event) == -EAGAIN)) {
2933 if (info.add_timestamp)
2934 info.length -= RB_LEN_TIME_EXTEND;
2935 goto again;
2936 }
2937
2938 if (!event)
2939 goto out_fail;
2940
2941 return event;
2942
2943 out_fail:
2944 rb_end_commit(cpu_buffer);
2945 return NULL;
2946 }
2947
2948 /**
2949 * ring_buffer_lock_reserve - reserve a part of the buffer
2950 * @buffer: the ring buffer to reserve from
2951 * @length: the length of the data to reserve (excluding event header)
2952 *
2953 * Returns a reserved event on the ring buffer to copy directly to.
2954 * The user of this interface will need to get the body to write into
2955 * and can use the ring_buffer_event_data() interface.
2956 *
2957 * The length is the length of the data needed, not the event length
2958 * which also includes the event header.
2959 *
2960 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
2961 * If NULL is returned, then nothing has been allocated or locked.
2962 */
2963 struct ring_buffer_event *
ring_buffer_lock_reserve(struct ring_buffer * buffer,unsigned long length)2964 ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length)
2965 {
2966 struct ring_buffer_per_cpu *cpu_buffer;
2967 struct ring_buffer_event *event;
2968 int cpu;
2969
2970 /* If we are tracing schedule, we don't want to recurse */
2971 preempt_disable_notrace();
2972
2973 if (unlikely(atomic_read(&buffer->record_disabled)))
2974 goto out;
2975
2976 cpu = raw_smp_processor_id();
2977
2978 if (unlikely(!cpumask_test_cpu(cpu, buffer->cpumask)))
2979 goto out;
2980
2981 cpu_buffer = buffer->buffers[cpu];
2982
2983 if (unlikely(atomic_read(&cpu_buffer->record_disabled)))
2984 goto out;
2985
2986 if (unlikely(length > BUF_MAX_DATA_SIZE))
2987 goto out;
2988
2989 if (unlikely(trace_recursive_lock(cpu_buffer)))
2990 goto out;
2991
2992 event = rb_reserve_next_event(buffer, cpu_buffer, length);
2993 if (!event)
2994 goto out_unlock;
2995
2996 return event;
2997
2998 out_unlock:
2999 trace_recursive_unlock(cpu_buffer);
3000 out:
3001 preempt_enable_notrace();
3002 return NULL;
3003 }
3004 EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve);
3005
3006 /*
3007 * Decrement the entries to the page that an event is on.
3008 * The event does not even need to exist, only the pointer
3009 * to the page it is on. This may only be called before the commit
3010 * takes place.
3011 */
3012 static inline void
rb_decrement_entry(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)3013 rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer,
3014 struct ring_buffer_event *event)
3015 {
3016 unsigned long addr = (unsigned long)event;
3017 struct buffer_page *bpage = cpu_buffer->commit_page;
3018 struct buffer_page *start;
3019
3020 addr &= PAGE_MASK;
3021
3022 /* Do the likely case first */
3023 if (likely(bpage->page == (void *)addr)) {
3024 local_dec(&bpage->entries);
3025 return;
3026 }
3027
3028 /*
3029 * Because the commit page may be on the reader page we
3030 * start with the next page and check the end loop there.
3031 */
3032 rb_inc_page(cpu_buffer, &bpage);
3033 start = bpage;
3034 do {
3035 if (bpage->page == (void *)addr) {
3036 local_dec(&bpage->entries);
3037 return;
3038 }
3039 rb_inc_page(cpu_buffer, &bpage);
3040 } while (bpage != start);
3041
3042 /* commit not part of this buffer?? */
3043 RB_WARN_ON(cpu_buffer, 1);
3044 }
3045
3046 /**
3047 * ring_buffer_commit_discard - discard an event that has not been committed
3048 * @buffer: the ring buffer
3049 * @event: non committed event to discard
3050 *
3051 * Sometimes an event that is in the ring buffer needs to be ignored.
3052 * This function lets the user discard an event in the ring buffer
3053 * and then that event will not be read later.
3054 *
3055 * This function only works if it is called before the item has been
3056 * committed. It will try to free the event from the ring buffer
3057 * if another event has not been added behind it.
3058 *
3059 * If another event has been added behind it, it will set the event
3060 * up as discarded, and perform the commit.
3061 *
3062 * If this function is called, do not call ring_buffer_unlock_commit on
3063 * the event.
3064 */
ring_buffer_discard_commit(struct ring_buffer * buffer,struct ring_buffer_event * event)3065 void ring_buffer_discard_commit(struct ring_buffer *buffer,
3066 struct ring_buffer_event *event)
3067 {
3068 struct ring_buffer_per_cpu *cpu_buffer;
3069 int cpu;
3070
3071 /* The event is discarded regardless */
3072 rb_event_discard(event);
3073
3074 cpu = smp_processor_id();
3075 cpu_buffer = buffer->buffers[cpu];
3076
3077 /*
3078 * This must only be called if the event has not been
3079 * committed yet. Thus we can assume that preemption
3080 * is still disabled.
3081 */
3082 RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing));
3083
3084 rb_decrement_entry(cpu_buffer, event);
3085 if (rb_try_to_discard(cpu_buffer, event))
3086 goto out;
3087
3088 /*
3089 * The commit is still visible by the reader, so we
3090 * must still update the timestamp.
3091 */
3092 rb_update_write_stamp(cpu_buffer, event);
3093 out:
3094 rb_end_commit(cpu_buffer);
3095
3096 trace_recursive_unlock(cpu_buffer);
3097
3098 preempt_enable_notrace();
3099
3100 }
3101 EXPORT_SYMBOL_GPL(ring_buffer_discard_commit);
3102
3103 /**
3104 * ring_buffer_write - write data to the buffer without reserving
3105 * @buffer: The ring buffer to write to.
3106 * @length: The length of the data being written (excluding the event header)
3107 * @data: The data to write to the buffer.
3108 *
3109 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
3110 * one function. If you already have the data to write to the buffer, it
3111 * may be easier to simply call this function.
3112 *
3113 * Note, like ring_buffer_lock_reserve, the length is the length of the data
3114 * and not the length of the event which would hold the header.
3115 */
ring_buffer_write(struct ring_buffer * buffer,unsigned long length,void * data)3116 int ring_buffer_write(struct ring_buffer *buffer,
3117 unsigned long length,
3118 void *data)
3119 {
3120 struct ring_buffer_per_cpu *cpu_buffer;
3121 struct ring_buffer_event *event;
3122 void *body;
3123 int ret = -EBUSY;
3124 int cpu;
3125
3126 preempt_disable_notrace();
3127
3128 if (atomic_read(&buffer->record_disabled))
3129 goto out;
3130
3131 cpu = raw_smp_processor_id();
3132
3133 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3134 goto out;
3135
3136 cpu_buffer = buffer->buffers[cpu];
3137
3138 if (atomic_read(&cpu_buffer->record_disabled))
3139 goto out;
3140
3141 if (length > BUF_MAX_DATA_SIZE)
3142 goto out;
3143
3144 if (unlikely(trace_recursive_lock(cpu_buffer)))
3145 goto out;
3146
3147 event = rb_reserve_next_event(buffer, cpu_buffer, length);
3148 if (!event)
3149 goto out_unlock;
3150
3151 body = rb_event_data(event);
3152
3153 memcpy(body, data, length);
3154
3155 rb_commit(cpu_buffer, event);
3156
3157 rb_wakeups(buffer, cpu_buffer);
3158
3159 ret = 0;
3160
3161 out_unlock:
3162 trace_recursive_unlock(cpu_buffer);
3163
3164 out:
3165 preempt_enable_notrace();
3166
3167 return ret;
3168 }
3169 EXPORT_SYMBOL_GPL(ring_buffer_write);
3170
rb_per_cpu_empty(struct ring_buffer_per_cpu * cpu_buffer)3171 static bool rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
3172 {
3173 struct buffer_page *reader = cpu_buffer->reader_page;
3174 struct buffer_page *head = rb_set_head_page(cpu_buffer);
3175 struct buffer_page *commit = cpu_buffer->commit_page;
3176
3177 /* In case of error, head will be NULL */
3178 if (unlikely(!head))
3179 return true;
3180
3181 return reader->read == rb_page_commit(reader) &&
3182 (commit == reader ||
3183 (commit == head &&
3184 head->read == rb_page_commit(commit)));
3185 }
3186
3187 /**
3188 * ring_buffer_record_disable - stop all writes into the buffer
3189 * @buffer: The ring buffer to stop writes to.
3190 *
3191 * This prevents all writes to the buffer. Any attempt to write
3192 * to the buffer after this will fail and return NULL.
3193 *
3194 * The caller should call synchronize_rcu() after this.
3195 */
ring_buffer_record_disable(struct ring_buffer * buffer)3196 void ring_buffer_record_disable(struct ring_buffer *buffer)
3197 {
3198 atomic_inc(&buffer->record_disabled);
3199 }
3200 EXPORT_SYMBOL_GPL(ring_buffer_record_disable);
3201
3202 /**
3203 * ring_buffer_record_enable - enable writes to the buffer
3204 * @buffer: The ring buffer to enable writes
3205 *
3206 * Note, multiple disables will need the same number of enables
3207 * to truly enable the writing (much like preempt_disable).
3208 */
ring_buffer_record_enable(struct ring_buffer * buffer)3209 void ring_buffer_record_enable(struct ring_buffer *buffer)
3210 {
3211 atomic_dec(&buffer->record_disabled);
3212 }
3213 EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
3214
3215 /**
3216 * ring_buffer_record_off - stop all writes into the buffer
3217 * @buffer: The ring buffer to stop writes to.
3218 *
3219 * This prevents all writes to the buffer. Any attempt to write
3220 * to the buffer after this will fail and return NULL.
3221 *
3222 * This is different than ring_buffer_record_disable() as
3223 * it works like an on/off switch, where as the disable() version
3224 * must be paired with a enable().
3225 */
ring_buffer_record_off(struct ring_buffer * buffer)3226 void ring_buffer_record_off(struct ring_buffer *buffer)
3227 {
3228 unsigned int rd;
3229 unsigned int new_rd;
3230
3231 do {
3232 rd = atomic_read(&buffer->record_disabled);
3233 new_rd = rd | RB_BUFFER_OFF;
3234 } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd);
3235 }
3236 EXPORT_SYMBOL_GPL(ring_buffer_record_off);
3237
3238 /**
3239 * ring_buffer_record_on - restart writes into the buffer
3240 * @buffer: The ring buffer to start writes to.
3241 *
3242 * This enables all writes to the buffer that was disabled by
3243 * ring_buffer_record_off().
3244 *
3245 * This is different than ring_buffer_record_enable() as
3246 * it works like an on/off switch, where as the enable() version
3247 * must be paired with a disable().
3248 */
ring_buffer_record_on(struct ring_buffer * buffer)3249 void ring_buffer_record_on(struct ring_buffer *buffer)
3250 {
3251 unsigned int rd;
3252 unsigned int new_rd;
3253
3254 do {
3255 rd = atomic_read(&buffer->record_disabled);
3256 new_rd = rd & ~RB_BUFFER_OFF;
3257 } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd);
3258 }
3259 EXPORT_SYMBOL_GPL(ring_buffer_record_on);
3260
3261 /**
3262 * ring_buffer_record_is_on - return true if the ring buffer can write
3263 * @buffer: The ring buffer to see if write is enabled
3264 *
3265 * Returns true if the ring buffer is in a state that it accepts writes.
3266 */
ring_buffer_record_is_on(struct ring_buffer * buffer)3267 bool ring_buffer_record_is_on(struct ring_buffer *buffer)
3268 {
3269 return !atomic_read(&buffer->record_disabled);
3270 }
3271
3272 /**
3273 * ring_buffer_record_is_set_on - return true if the ring buffer is set writable
3274 * @buffer: The ring buffer to see if write is set enabled
3275 *
3276 * Returns true if the ring buffer is set writable by ring_buffer_record_on().
3277 * Note that this does NOT mean it is in a writable state.
3278 *
3279 * It may return true when the ring buffer has been disabled by
3280 * ring_buffer_record_disable(), as that is a temporary disabling of
3281 * the ring buffer.
3282 */
ring_buffer_record_is_set_on(struct ring_buffer * buffer)3283 bool ring_buffer_record_is_set_on(struct ring_buffer *buffer)
3284 {
3285 return !(atomic_read(&buffer->record_disabled) & RB_BUFFER_OFF);
3286 }
3287
3288 /**
3289 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
3290 * @buffer: The ring buffer to stop writes to.
3291 * @cpu: The CPU buffer to stop
3292 *
3293 * This prevents all writes to the buffer. Any attempt to write
3294 * to the buffer after this will fail and return NULL.
3295 *
3296 * The caller should call synchronize_rcu() after this.
3297 */
ring_buffer_record_disable_cpu(struct ring_buffer * buffer,int cpu)3298 void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu)
3299 {
3300 struct ring_buffer_per_cpu *cpu_buffer;
3301
3302 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3303 return;
3304
3305 cpu_buffer = buffer->buffers[cpu];
3306 atomic_inc(&cpu_buffer->record_disabled);
3307 }
3308 EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu);
3309
3310 /**
3311 * ring_buffer_record_enable_cpu - enable writes to the buffer
3312 * @buffer: The ring buffer to enable writes
3313 * @cpu: The CPU to enable.
3314 *
3315 * Note, multiple disables will need the same number of enables
3316 * to truly enable the writing (much like preempt_disable).
3317 */
ring_buffer_record_enable_cpu(struct ring_buffer * buffer,int cpu)3318 void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu)
3319 {
3320 struct ring_buffer_per_cpu *cpu_buffer;
3321
3322 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3323 return;
3324
3325 cpu_buffer = buffer->buffers[cpu];
3326 atomic_dec(&cpu_buffer->record_disabled);
3327 }
3328 EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
3329
3330 /*
3331 * The total entries in the ring buffer is the running counter
3332 * of entries entered into the ring buffer, minus the sum of
3333 * the entries read from the ring buffer and the number of
3334 * entries that were overwritten.
3335 */
3336 static inline unsigned long
rb_num_of_entries(struct ring_buffer_per_cpu * cpu_buffer)3337 rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer)
3338 {
3339 return local_read(&cpu_buffer->entries) -
3340 (local_read(&cpu_buffer->overrun) + cpu_buffer->read);
3341 }
3342
3343 /**
3344 * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer
3345 * @buffer: The ring buffer
3346 * @cpu: The per CPU buffer to read from.
3347 */
ring_buffer_oldest_event_ts(struct ring_buffer * buffer,int cpu)3348 u64 ring_buffer_oldest_event_ts(struct ring_buffer *buffer, int cpu)
3349 {
3350 unsigned long flags;
3351 struct ring_buffer_per_cpu *cpu_buffer;
3352 struct buffer_page *bpage;
3353 u64 ret = 0;
3354
3355 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3356 return 0;
3357
3358 cpu_buffer = buffer->buffers[cpu];
3359 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
3360 /*
3361 * if the tail is on reader_page, oldest time stamp is on the reader
3362 * page
3363 */
3364 if (cpu_buffer->tail_page == cpu_buffer->reader_page)
3365 bpage = cpu_buffer->reader_page;
3366 else
3367 bpage = rb_set_head_page(cpu_buffer);
3368 if (bpage)
3369 ret = bpage->page->time_stamp;
3370 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
3371
3372 return ret;
3373 }
3374 EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts);
3375
3376 /**
3377 * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer
3378 * @buffer: The ring buffer
3379 * @cpu: The per CPU buffer to read from.
3380 */
ring_buffer_bytes_cpu(struct ring_buffer * buffer,int cpu)3381 unsigned long ring_buffer_bytes_cpu(struct ring_buffer *buffer, int cpu)
3382 {
3383 struct ring_buffer_per_cpu *cpu_buffer;
3384 unsigned long ret;
3385
3386 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3387 return 0;
3388
3389 cpu_buffer = buffer->buffers[cpu];
3390 ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes;
3391
3392 return ret;
3393 }
3394 EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu);
3395
3396 /**
3397 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
3398 * @buffer: The ring buffer
3399 * @cpu: The per CPU buffer to get the entries from.
3400 */
ring_buffer_entries_cpu(struct ring_buffer * buffer,int cpu)3401 unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
3402 {
3403 struct ring_buffer_per_cpu *cpu_buffer;
3404
3405 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3406 return 0;
3407
3408 cpu_buffer = buffer->buffers[cpu];
3409
3410 return rb_num_of_entries(cpu_buffer);
3411 }
3412 EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
3413
3414 /**
3415 * ring_buffer_overrun_cpu - get the number of overruns caused by the ring
3416 * buffer wrapping around (only if RB_FL_OVERWRITE is on).
3417 * @buffer: The ring buffer
3418 * @cpu: The per CPU buffer to get the number of overruns from
3419 */
ring_buffer_overrun_cpu(struct ring_buffer * buffer,int cpu)3420 unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu)
3421 {
3422 struct ring_buffer_per_cpu *cpu_buffer;
3423 unsigned long ret;
3424
3425 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3426 return 0;
3427
3428 cpu_buffer = buffer->buffers[cpu];
3429 ret = local_read(&cpu_buffer->overrun);
3430
3431 return ret;
3432 }
3433 EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
3434
3435 /**
3436 * ring_buffer_commit_overrun_cpu - get the number of overruns caused by
3437 * commits failing due to the buffer wrapping around while there are uncommitted
3438 * events, such as during an interrupt storm.
3439 * @buffer: The ring buffer
3440 * @cpu: The per CPU buffer to get the number of overruns from
3441 */
3442 unsigned long
ring_buffer_commit_overrun_cpu(struct ring_buffer * buffer,int cpu)3443 ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu)
3444 {
3445 struct ring_buffer_per_cpu *cpu_buffer;
3446 unsigned long ret;
3447
3448 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3449 return 0;
3450
3451 cpu_buffer = buffer->buffers[cpu];
3452 ret = local_read(&cpu_buffer->commit_overrun);
3453
3454 return ret;
3455 }
3456 EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu);
3457
3458 /**
3459 * ring_buffer_dropped_events_cpu - get the number of dropped events caused by
3460 * the ring buffer filling up (only if RB_FL_OVERWRITE is off).
3461 * @buffer: The ring buffer
3462 * @cpu: The per CPU buffer to get the number of overruns from
3463 */
3464 unsigned long
ring_buffer_dropped_events_cpu(struct ring_buffer * buffer,int cpu)3465 ring_buffer_dropped_events_cpu(struct ring_buffer *buffer, int cpu)
3466 {
3467 struct ring_buffer_per_cpu *cpu_buffer;
3468 unsigned long ret;
3469
3470 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3471 return 0;
3472
3473 cpu_buffer = buffer->buffers[cpu];
3474 ret = local_read(&cpu_buffer->dropped_events);
3475
3476 return ret;
3477 }
3478 EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu);
3479
3480 /**
3481 * ring_buffer_read_events_cpu - get the number of events successfully read
3482 * @buffer: The ring buffer
3483 * @cpu: The per CPU buffer to get the number of events read
3484 */
3485 unsigned long
ring_buffer_read_events_cpu(struct ring_buffer * buffer,int cpu)3486 ring_buffer_read_events_cpu(struct ring_buffer *buffer, int cpu)
3487 {
3488 struct ring_buffer_per_cpu *cpu_buffer;
3489
3490 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3491 return 0;
3492
3493 cpu_buffer = buffer->buffers[cpu];
3494 return cpu_buffer->read;
3495 }
3496 EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu);
3497
3498 /**
3499 * ring_buffer_entries - get the number of entries in a buffer
3500 * @buffer: The ring buffer
3501 *
3502 * Returns the total number of entries in the ring buffer
3503 * (all CPU entries)
3504 */
ring_buffer_entries(struct ring_buffer * buffer)3505 unsigned long ring_buffer_entries(struct ring_buffer *buffer)
3506 {
3507 struct ring_buffer_per_cpu *cpu_buffer;
3508 unsigned long entries = 0;
3509 int cpu;
3510
3511 /* if you care about this being correct, lock the buffer */
3512 for_each_buffer_cpu(buffer, cpu) {
3513 cpu_buffer = buffer->buffers[cpu];
3514 entries += rb_num_of_entries(cpu_buffer);
3515 }
3516
3517 return entries;
3518 }
3519 EXPORT_SYMBOL_GPL(ring_buffer_entries);
3520
3521 /**
3522 * ring_buffer_overruns - get the number of overruns in buffer
3523 * @buffer: The ring buffer
3524 *
3525 * Returns the total number of overruns in the ring buffer
3526 * (all CPU entries)
3527 */
ring_buffer_overruns(struct ring_buffer * buffer)3528 unsigned long ring_buffer_overruns(struct ring_buffer *buffer)
3529 {
3530 struct ring_buffer_per_cpu *cpu_buffer;
3531 unsigned long overruns = 0;
3532 int cpu;
3533
3534 /* if you care about this being correct, lock the buffer */
3535 for_each_buffer_cpu(buffer, cpu) {
3536 cpu_buffer = buffer->buffers[cpu];
3537 overruns += local_read(&cpu_buffer->overrun);
3538 }
3539
3540 return overruns;
3541 }
3542 EXPORT_SYMBOL_GPL(ring_buffer_overruns);
3543
rb_iter_reset(struct ring_buffer_iter * iter)3544 static void rb_iter_reset(struct ring_buffer_iter *iter)
3545 {
3546 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
3547
3548 /* Iterator usage is expected to have record disabled */
3549 iter->head_page = cpu_buffer->reader_page;
3550 iter->head = cpu_buffer->reader_page->read;
3551
3552 iter->cache_reader_page = iter->head_page;
3553 iter->cache_read = cpu_buffer->read;
3554
3555 if (iter->head)
3556 iter->read_stamp = cpu_buffer->read_stamp;
3557 else
3558 iter->read_stamp = iter->head_page->page->time_stamp;
3559 }
3560
3561 /**
3562 * ring_buffer_iter_reset - reset an iterator
3563 * @iter: The iterator to reset
3564 *
3565 * Resets the iterator, so that it will start from the beginning
3566 * again.
3567 */
ring_buffer_iter_reset(struct ring_buffer_iter * iter)3568 void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
3569 {
3570 struct ring_buffer_per_cpu *cpu_buffer;
3571 unsigned long flags;
3572
3573 if (!iter)
3574 return;
3575
3576 cpu_buffer = iter->cpu_buffer;
3577
3578 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
3579 rb_iter_reset(iter);
3580 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
3581 }
3582 EXPORT_SYMBOL_GPL(ring_buffer_iter_reset);
3583
3584 /**
3585 * ring_buffer_iter_empty - check if an iterator has no more to read
3586 * @iter: The iterator to check
3587 */
ring_buffer_iter_empty(struct ring_buffer_iter * iter)3588 int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
3589 {
3590 struct ring_buffer_per_cpu *cpu_buffer;
3591 struct buffer_page *reader;
3592 struct buffer_page *head_page;
3593 struct buffer_page *commit_page;
3594 unsigned commit;
3595
3596 cpu_buffer = iter->cpu_buffer;
3597
3598 /* Remember, trace recording is off when iterator is in use */
3599 reader = cpu_buffer->reader_page;
3600 head_page = cpu_buffer->head_page;
3601 commit_page = cpu_buffer->commit_page;
3602 commit = rb_page_commit(commit_page);
3603
3604 return ((iter->head_page == commit_page && iter->head == commit) ||
3605 (iter->head_page == reader && commit_page == head_page &&
3606 head_page->read == commit &&
3607 iter->head == rb_page_commit(cpu_buffer->reader_page)));
3608 }
3609 EXPORT_SYMBOL_GPL(ring_buffer_iter_empty);
3610
3611 static void
rb_update_read_stamp(struct ring_buffer_per_cpu * cpu_buffer,struct ring_buffer_event * event)3612 rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
3613 struct ring_buffer_event *event)
3614 {
3615 u64 delta;
3616
3617 switch (event->type_len) {
3618 case RINGBUF_TYPE_PADDING:
3619 return;
3620
3621 case RINGBUF_TYPE_TIME_EXTEND:
3622 delta = ring_buffer_event_time_stamp(event);
3623 cpu_buffer->read_stamp += delta;
3624 return;
3625
3626 case RINGBUF_TYPE_TIME_STAMP:
3627 delta = ring_buffer_event_time_stamp(event);
3628 cpu_buffer->read_stamp = delta;
3629 return;
3630
3631 case RINGBUF_TYPE_DATA:
3632 cpu_buffer->read_stamp += event->time_delta;
3633 return;
3634
3635 default:
3636 BUG();
3637 }
3638 return;
3639 }
3640
3641 static void
rb_update_iter_read_stamp(struct ring_buffer_iter * iter,struct ring_buffer_event * event)3642 rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
3643 struct ring_buffer_event *event)
3644 {
3645 u64 delta;
3646
3647 switch (event->type_len) {
3648 case RINGBUF_TYPE_PADDING:
3649 return;
3650
3651 case RINGBUF_TYPE_TIME_EXTEND:
3652 delta = ring_buffer_event_time_stamp(event);
3653 iter->read_stamp += delta;
3654 return;
3655
3656 case RINGBUF_TYPE_TIME_STAMP:
3657 delta = ring_buffer_event_time_stamp(event);
3658 iter->read_stamp = delta;
3659 return;
3660
3661 case RINGBUF_TYPE_DATA:
3662 iter->read_stamp += event->time_delta;
3663 return;
3664
3665 default:
3666 BUG();
3667 }
3668 return;
3669 }
3670
3671 static struct buffer_page *
rb_get_reader_page(struct ring_buffer_per_cpu * cpu_buffer)3672 rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
3673 {
3674 struct buffer_page *reader = NULL;
3675 unsigned long overwrite;
3676 unsigned long flags;
3677 int nr_loops = 0;
3678 int ret;
3679
3680 local_irq_save(flags);
3681 arch_spin_lock(&cpu_buffer->lock);
3682
3683 again:
3684 /*
3685 * This should normally only loop twice. But because the
3686 * start of the reader inserts an empty page, it causes
3687 * a case where we will loop three times. There should be no
3688 * reason to loop four times (that I know of).
3689 */
3690 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
3691 reader = NULL;
3692 goto out;
3693 }
3694
3695 reader = cpu_buffer->reader_page;
3696
3697 /* If there's more to read, return this page */
3698 if (cpu_buffer->reader_page->read < rb_page_size(reader))
3699 goto out;
3700
3701 /* Never should we have an index greater than the size */
3702 if (RB_WARN_ON(cpu_buffer,
3703 cpu_buffer->reader_page->read > rb_page_size(reader)))
3704 goto out;
3705
3706 /* check if we caught up to the tail */
3707 reader = NULL;
3708 if (cpu_buffer->commit_page == cpu_buffer->reader_page)
3709 goto out;
3710
3711 /* Don't bother swapping if the ring buffer is empty */
3712 if (rb_num_of_entries(cpu_buffer) == 0)
3713 goto out;
3714
3715 /*
3716 * Reset the reader page to size zero.
3717 */
3718 local_set(&cpu_buffer->reader_page->write, 0);
3719 local_set(&cpu_buffer->reader_page->entries, 0);
3720 local_set(&cpu_buffer->reader_page->page->commit, 0);
3721 cpu_buffer->reader_page->real_end = 0;
3722
3723 spin:
3724 /*
3725 * Splice the empty reader page into the list around the head.
3726 */
3727 reader = rb_set_head_page(cpu_buffer);
3728 if (!reader)
3729 goto out;
3730 cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next);
3731 cpu_buffer->reader_page->list.prev = reader->list.prev;
3732
3733 /*
3734 * cpu_buffer->pages just needs to point to the buffer, it
3735 * has no specific buffer page to point to. Lets move it out
3736 * of our way so we don't accidentally swap it.
3737 */
3738 cpu_buffer->pages = reader->list.prev;
3739
3740 /* The reader page will be pointing to the new head */
3741 rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list);
3742
3743 /*
3744 * We want to make sure we read the overruns after we set up our
3745 * pointers to the next object. The writer side does a
3746 * cmpxchg to cross pages which acts as the mb on the writer
3747 * side. Note, the reader will constantly fail the swap
3748 * while the writer is updating the pointers, so this
3749 * guarantees that the overwrite recorded here is the one we
3750 * want to compare with the last_overrun.
3751 */
3752 smp_mb();
3753 overwrite = local_read(&(cpu_buffer->overrun));
3754
3755 /*
3756 * Here's the tricky part.
3757 *
3758 * We need to move the pointer past the header page.
3759 * But we can only do that if a writer is not currently
3760 * moving it. The page before the header page has the
3761 * flag bit '1' set if it is pointing to the page we want.
3762 * but if the writer is in the process of moving it
3763 * than it will be '2' or already moved '0'.
3764 */
3765
3766 ret = rb_head_page_replace(reader, cpu_buffer->reader_page);
3767
3768 /*
3769 * If we did not convert it, then we must try again.
3770 */
3771 if (!ret)
3772 goto spin;
3773
3774 /*
3775 * Yay! We succeeded in replacing the page.
3776 *
3777 * Now make the new head point back to the reader page.
3778 */
3779 rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list;
3780 rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
3781
3782 local_inc(&cpu_buffer->pages_read);
3783
3784 /* Finally update the reader page to the new head */
3785 cpu_buffer->reader_page = reader;
3786 cpu_buffer->reader_page->read = 0;
3787
3788 if (overwrite != cpu_buffer->last_overrun) {
3789 cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun;
3790 cpu_buffer->last_overrun = overwrite;
3791 }
3792
3793 goto again;
3794
3795 out:
3796 /* Update the read_stamp on the first event */
3797 if (reader && reader->read == 0)
3798 cpu_buffer->read_stamp = reader->page->time_stamp;
3799
3800 arch_spin_unlock(&cpu_buffer->lock);
3801 local_irq_restore(flags);
3802
3803 return reader;
3804 }
3805
rb_advance_reader(struct ring_buffer_per_cpu * cpu_buffer)3806 static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
3807 {
3808 struct ring_buffer_event *event;
3809 struct buffer_page *reader;
3810 unsigned length;
3811
3812 reader = rb_get_reader_page(cpu_buffer);
3813
3814 /* This function should not be called when buffer is empty */
3815 if (RB_WARN_ON(cpu_buffer, !reader))
3816 return;
3817
3818 event = rb_reader_event(cpu_buffer);
3819
3820 if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
3821 cpu_buffer->read++;
3822
3823 rb_update_read_stamp(cpu_buffer, event);
3824
3825 length = rb_event_length(event);
3826 cpu_buffer->reader_page->read += length;
3827 }
3828
rb_advance_iter(struct ring_buffer_iter * iter)3829 static void rb_advance_iter(struct ring_buffer_iter *iter)
3830 {
3831 struct ring_buffer_per_cpu *cpu_buffer;
3832 struct ring_buffer_event *event;
3833 unsigned length;
3834
3835 cpu_buffer = iter->cpu_buffer;
3836
3837 /*
3838 * Check if we are at the end of the buffer.
3839 */
3840 if (iter->head >= rb_page_size(iter->head_page)) {
3841 /* discarded commits can make the page empty */
3842 if (iter->head_page == cpu_buffer->commit_page)
3843 return;
3844 rb_inc_iter(iter);
3845 return;
3846 }
3847
3848 event = rb_iter_head_event(iter);
3849
3850 length = rb_event_length(event);
3851
3852 /*
3853 * This should not be called to advance the header if we are
3854 * at the tail of the buffer.
3855 */
3856 if (RB_WARN_ON(cpu_buffer,
3857 (iter->head_page == cpu_buffer->commit_page) &&
3858 (iter->head + length > rb_commit_index(cpu_buffer))))
3859 return;
3860
3861 rb_update_iter_read_stamp(iter, event);
3862
3863 iter->head += length;
3864
3865 /* check for end of page padding */
3866 if ((iter->head >= rb_page_size(iter->head_page)) &&
3867 (iter->head_page != cpu_buffer->commit_page))
3868 rb_inc_iter(iter);
3869 }
3870
rb_lost_events(struct ring_buffer_per_cpu * cpu_buffer)3871 static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer)
3872 {
3873 return cpu_buffer->lost_events;
3874 }
3875
3876 static struct ring_buffer_event *
rb_buffer_peek(struct ring_buffer_per_cpu * cpu_buffer,u64 * ts,unsigned long * lost_events)3877 rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts,
3878 unsigned long *lost_events)
3879 {
3880 struct ring_buffer_event *event;
3881 struct buffer_page *reader;
3882 int nr_loops = 0;
3883
3884 if (ts)
3885 *ts = 0;
3886 again:
3887 /*
3888 * We repeat when a time extend is encountered.
3889 * Since the time extend is always attached to a data event,
3890 * we should never loop more than once.
3891 * (We never hit the following condition more than twice).
3892 */
3893 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2))
3894 return NULL;
3895
3896 reader = rb_get_reader_page(cpu_buffer);
3897 if (!reader)
3898 return NULL;
3899
3900 event = rb_reader_event(cpu_buffer);
3901
3902 switch (event->type_len) {
3903 case RINGBUF_TYPE_PADDING:
3904 if (rb_null_event(event))
3905 RB_WARN_ON(cpu_buffer, 1);
3906 /*
3907 * Because the writer could be discarding every
3908 * event it creates (which would probably be bad)
3909 * if we were to go back to "again" then we may never
3910 * catch up, and will trigger the warn on, or lock
3911 * the box. Return the padding, and we will release
3912 * the current locks, and try again.
3913 */
3914 return event;
3915
3916 case RINGBUF_TYPE_TIME_EXTEND:
3917 /* Internal data, OK to advance */
3918 rb_advance_reader(cpu_buffer);
3919 goto again;
3920
3921 case RINGBUF_TYPE_TIME_STAMP:
3922 if (ts) {
3923 *ts = ring_buffer_event_time_stamp(event);
3924 ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
3925 cpu_buffer->cpu, ts);
3926 }
3927 /* Internal data, OK to advance */
3928 rb_advance_reader(cpu_buffer);
3929 goto again;
3930
3931 case RINGBUF_TYPE_DATA:
3932 if (ts && !(*ts)) {
3933 *ts = cpu_buffer->read_stamp + event->time_delta;
3934 ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
3935 cpu_buffer->cpu, ts);
3936 }
3937 if (lost_events)
3938 *lost_events = rb_lost_events(cpu_buffer);
3939 return event;
3940
3941 default:
3942 BUG();
3943 }
3944
3945 return NULL;
3946 }
3947 EXPORT_SYMBOL_GPL(ring_buffer_peek);
3948
3949 static struct ring_buffer_event *
rb_iter_peek(struct ring_buffer_iter * iter,u64 * ts)3950 rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
3951 {
3952 struct ring_buffer *buffer;
3953 struct ring_buffer_per_cpu *cpu_buffer;
3954 struct ring_buffer_event *event;
3955 int nr_loops = 0;
3956
3957 if (ts)
3958 *ts = 0;
3959
3960 cpu_buffer = iter->cpu_buffer;
3961 buffer = cpu_buffer->buffer;
3962
3963 /*
3964 * Check if someone performed a consuming read to
3965 * the buffer. A consuming read invalidates the iterator
3966 * and we need to reset the iterator in this case.
3967 */
3968 if (unlikely(iter->cache_read != cpu_buffer->read ||
3969 iter->cache_reader_page != cpu_buffer->reader_page))
3970 rb_iter_reset(iter);
3971
3972 again:
3973 if (ring_buffer_iter_empty(iter))
3974 return NULL;
3975
3976 /*
3977 * We repeat when a time extend is encountered or we hit
3978 * the end of the page. Since the time extend is always attached
3979 * to a data event, we should never loop more than three times.
3980 * Once for going to next page, once on time extend, and
3981 * finally once to get the event.
3982 * (We never hit the following condition more than thrice).
3983 */
3984 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3))
3985 return NULL;
3986
3987 if (rb_per_cpu_empty(cpu_buffer))
3988 return NULL;
3989
3990 if (iter->head >= rb_page_size(iter->head_page)) {
3991 rb_inc_iter(iter);
3992 goto again;
3993 }
3994
3995 event = rb_iter_head_event(iter);
3996
3997 switch (event->type_len) {
3998 case RINGBUF_TYPE_PADDING:
3999 if (rb_null_event(event)) {
4000 rb_inc_iter(iter);
4001 goto again;
4002 }
4003 rb_advance_iter(iter);
4004 return event;
4005
4006 case RINGBUF_TYPE_TIME_EXTEND:
4007 /* Internal data, OK to advance */
4008 rb_advance_iter(iter);
4009 goto again;
4010
4011 case RINGBUF_TYPE_TIME_STAMP:
4012 if (ts) {
4013 *ts = ring_buffer_event_time_stamp(event);
4014 ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
4015 cpu_buffer->cpu, ts);
4016 }
4017 /* Internal data, OK to advance */
4018 rb_advance_iter(iter);
4019 goto again;
4020
4021 case RINGBUF_TYPE_DATA:
4022 if (ts && !(*ts)) {
4023 *ts = iter->read_stamp + event->time_delta;
4024 ring_buffer_normalize_time_stamp(buffer,
4025 cpu_buffer->cpu, ts);
4026 }
4027 return event;
4028
4029 default:
4030 BUG();
4031 }
4032
4033 return NULL;
4034 }
4035 EXPORT_SYMBOL_GPL(ring_buffer_iter_peek);
4036
rb_reader_lock(struct ring_buffer_per_cpu * cpu_buffer)4037 static inline bool rb_reader_lock(struct ring_buffer_per_cpu *cpu_buffer)
4038 {
4039 if (likely(!in_nmi())) {
4040 raw_spin_lock(&cpu_buffer->reader_lock);
4041 return true;
4042 }
4043
4044 /*
4045 * If an NMI die dumps out the content of the ring buffer
4046 * trylock must be used to prevent a deadlock if the NMI
4047 * preempted a task that holds the ring buffer locks. If
4048 * we get the lock then all is fine, if not, then continue
4049 * to do the read, but this can corrupt the ring buffer,
4050 * so it must be permanently disabled from future writes.
4051 * Reading from NMI is a oneshot deal.
4052 */
4053 if (raw_spin_trylock(&cpu_buffer->reader_lock))
4054 return true;
4055
4056 /* Continue without locking, but disable the ring buffer */
4057 atomic_inc(&cpu_buffer->record_disabled);
4058 return false;
4059 }
4060
4061 static inline void
rb_reader_unlock(struct ring_buffer_per_cpu * cpu_buffer,bool locked)4062 rb_reader_unlock(struct ring_buffer_per_cpu *cpu_buffer, bool locked)
4063 {
4064 if (likely(locked))
4065 raw_spin_unlock(&cpu_buffer->reader_lock);
4066 return;
4067 }
4068
4069 /**
4070 * ring_buffer_peek - peek at the next event to be read
4071 * @buffer: The ring buffer to read
4072 * @cpu: The cpu to peak at
4073 * @ts: The timestamp counter of this event.
4074 * @lost_events: a variable to store if events were lost (may be NULL)
4075 *
4076 * This will return the event that will be read next, but does
4077 * not consume the data.
4078 */
4079 struct ring_buffer_event *
ring_buffer_peek(struct ring_buffer * buffer,int cpu,u64 * ts,unsigned long * lost_events)4080 ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts,
4081 unsigned long *lost_events)
4082 {
4083 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
4084 struct ring_buffer_event *event;
4085 unsigned long flags;
4086 bool dolock;
4087
4088 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4089 return NULL;
4090
4091 again:
4092 local_irq_save(flags);
4093 dolock = rb_reader_lock(cpu_buffer);
4094 event = rb_buffer_peek(cpu_buffer, ts, lost_events);
4095 if (event && event->type_len == RINGBUF_TYPE_PADDING)
4096 rb_advance_reader(cpu_buffer);
4097 rb_reader_unlock(cpu_buffer, dolock);
4098 local_irq_restore(flags);
4099
4100 if (event && event->type_len == RINGBUF_TYPE_PADDING)
4101 goto again;
4102
4103 return event;
4104 }
4105
4106 /**
4107 * ring_buffer_iter_peek - peek at the next event to be read
4108 * @iter: The ring buffer iterator
4109 * @ts: The timestamp counter of this event.
4110 *
4111 * This will return the event that will be read next, but does
4112 * not increment the iterator.
4113 */
4114 struct ring_buffer_event *
ring_buffer_iter_peek(struct ring_buffer_iter * iter,u64 * ts)4115 ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
4116 {
4117 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
4118 struct ring_buffer_event *event;
4119 unsigned long flags;
4120
4121 again:
4122 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4123 event = rb_iter_peek(iter, ts);
4124 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4125
4126 if (event && event->type_len == RINGBUF_TYPE_PADDING)
4127 goto again;
4128
4129 return event;
4130 }
4131
4132 /**
4133 * ring_buffer_consume - return an event and consume it
4134 * @buffer: The ring buffer to get the next event from
4135 * @cpu: the cpu to read the buffer from
4136 * @ts: a variable to store the timestamp (may be NULL)
4137 * @lost_events: a variable to store if events were lost (may be NULL)
4138 *
4139 * Returns the next event in the ring buffer, and that event is consumed.
4140 * Meaning, that sequential reads will keep returning a different event,
4141 * and eventually empty the ring buffer if the producer is slower.
4142 */
4143 struct ring_buffer_event *
ring_buffer_consume(struct ring_buffer * buffer,int cpu,u64 * ts,unsigned long * lost_events)4144 ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts,
4145 unsigned long *lost_events)
4146 {
4147 struct ring_buffer_per_cpu *cpu_buffer;
4148 struct ring_buffer_event *event = NULL;
4149 unsigned long flags;
4150 bool dolock;
4151
4152 again:
4153 /* might be called in atomic */
4154 preempt_disable();
4155
4156 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4157 goto out;
4158
4159 cpu_buffer = buffer->buffers[cpu];
4160 local_irq_save(flags);
4161 dolock = rb_reader_lock(cpu_buffer);
4162
4163 event = rb_buffer_peek(cpu_buffer, ts, lost_events);
4164 if (event) {
4165 cpu_buffer->lost_events = 0;
4166 rb_advance_reader(cpu_buffer);
4167 }
4168
4169 rb_reader_unlock(cpu_buffer, dolock);
4170 local_irq_restore(flags);
4171
4172 out:
4173 preempt_enable();
4174
4175 if (event && event->type_len == RINGBUF_TYPE_PADDING)
4176 goto again;
4177
4178 return event;
4179 }
4180 EXPORT_SYMBOL_GPL(ring_buffer_consume);
4181
4182 /**
4183 * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer
4184 * @buffer: The ring buffer to read from
4185 * @cpu: The cpu buffer to iterate over
4186 * @flags: gfp flags to use for memory allocation
4187 *
4188 * This performs the initial preparations necessary to iterate
4189 * through the buffer. Memory is allocated, buffer recording
4190 * is disabled, and the iterator pointer is returned to the caller.
4191 *
4192 * Disabling buffer recording prevents the reading from being
4193 * corrupted. This is not a consuming read, so a producer is not
4194 * expected.
4195 *
4196 * After a sequence of ring_buffer_read_prepare calls, the user is
4197 * expected to make at least one call to ring_buffer_read_prepare_sync.
4198 * Afterwards, ring_buffer_read_start is invoked to get things going
4199 * for real.
4200 *
4201 * This overall must be paired with ring_buffer_read_finish.
4202 */
4203 struct ring_buffer_iter *
ring_buffer_read_prepare(struct ring_buffer * buffer,int cpu,gfp_t flags)4204 ring_buffer_read_prepare(struct ring_buffer *buffer, int cpu, gfp_t flags)
4205 {
4206 struct ring_buffer_per_cpu *cpu_buffer;
4207 struct ring_buffer_iter *iter;
4208
4209 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4210 return NULL;
4211
4212 iter = kmalloc(sizeof(*iter), flags);
4213 if (!iter)
4214 return NULL;
4215
4216 cpu_buffer = buffer->buffers[cpu];
4217
4218 iter->cpu_buffer = cpu_buffer;
4219
4220 atomic_inc(&buffer->resize_disabled);
4221 atomic_inc(&cpu_buffer->record_disabled);
4222
4223 return iter;
4224 }
4225 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare);
4226
4227 /**
4228 * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls
4229 *
4230 * All previously invoked ring_buffer_read_prepare calls to prepare
4231 * iterators will be synchronized. Afterwards, read_buffer_read_start
4232 * calls on those iterators are allowed.
4233 */
4234 void
ring_buffer_read_prepare_sync(void)4235 ring_buffer_read_prepare_sync(void)
4236 {
4237 synchronize_rcu();
4238 }
4239 EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync);
4240
4241 /**
4242 * ring_buffer_read_start - start a non consuming read of the buffer
4243 * @iter: The iterator returned by ring_buffer_read_prepare
4244 *
4245 * This finalizes the startup of an iteration through the buffer.
4246 * The iterator comes from a call to ring_buffer_read_prepare and
4247 * an intervening ring_buffer_read_prepare_sync must have been
4248 * performed.
4249 *
4250 * Must be paired with ring_buffer_read_finish.
4251 */
4252 void
ring_buffer_read_start(struct ring_buffer_iter * iter)4253 ring_buffer_read_start(struct ring_buffer_iter *iter)
4254 {
4255 struct ring_buffer_per_cpu *cpu_buffer;
4256 unsigned long flags;
4257
4258 if (!iter)
4259 return;
4260
4261 cpu_buffer = iter->cpu_buffer;
4262
4263 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4264 arch_spin_lock(&cpu_buffer->lock);
4265 rb_iter_reset(iter);
4266 arch_spin_unlock(&cpu_buffer->lock);
4267 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4268 }
4269 EXPORT_SYMBOL_GPL(ring_buffer_read_start);
4270
4271 /**
4272 * ring_buffer_read_finish - finish reading the iterator of the buffer
4273 * @iter: The iterator retrieved by ring_buffer_start
4274 *
4275 * This re-enables the recording to the buffer, and frees the
4276 * iterator.
4277 */
4278 void
ring_buffer_read_finish(struct ring_buffer_iter * iter)4279 ring_buffer_read_finish(struct ring_buffer_iter *iter)
4280 {
4281 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
4282 unsigned long flags;
4283
4284 /*
4285 * Ring buffer is disabled from recording, here's a good place
4286 * to check the integrity of the ring buffer.
4287 * Must prevent readers from trying to read, as the check
4288 * clears the HEAD page and readers require it.
4289 */
4290 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4291 rb_check_pages(cpu_buffer);
4292 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4293
4294 atomic_dec(&cpu_buffer->record_disabled);
4295 atomic_dec(&cpu_buffer->buffer->resize_disabled);
4296 kfree(iter);
4297 }
4298 EXPORT_SYMBOL_GPL(ring_buffer_read_finish);
4299
4300 /**
4301 * ring_buffer_read - read the next item in the ring buffer by the iterator
4302 * @iter: The ring buffer iterator
4303 * @ts: The time stamp of the event read.
4304 *
4305 * This reads the next event in the ring buffer and increments the iterator.
4306 */
4307 struct ring_buffer_event *
ring_buffer_read(struct ring_buffer_iter * iter,u64 * ts)4308 ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
4309 {
4310 struct ring_buffer_event *event;
4311 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
4312 unsigned long flags;
4313
4314 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4315 again:
4316 event = rb_iter_peek(iter, ts);
4317 if (!event)
4318 goto out;
4319
4320 if (event->type_len == RINGBUF_TYPE_PADDING)
4321 goto again;
4322
4323 rb_advance_iter(iter);
4324 out:
4325 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4326
4327 return event;
4328 }
4329 EXPORT_SYMBOL_GPL(ring_buffer_read);
4330
4331 /**
4332 * ring_buffer_size - return the size of the ring buffer (in bytes)
4333 * @buffer: The ring buffer.
4334 */
ring_buffer_size(struct ring_buffer * buffer,int cpu)4335 unsigned long ring_buffer_size(struct ring_buffer *buffer, int cpu)
4336 {
4337 /*
4338 * Earlier, this method returned
4339 * BUF_PAGE_SIZE * buffer->nr_pages
4340 * Since the nr_pages field is now removed, we have converted this to
4341 * return the per cpu buffer value.
4342 */
4343 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4344 return 0;
4345
4346 return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages;
4347 }
4348 EXPORT_SYMBOL_GPL(ring_buffer_size);
4349
4350 static void
rb_reset_cpu(struct ring_buffer_per_cpu * cpu_buffer)4351 rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
4352 {
4353 rb_head_page_deactivate(cpu_buffer);
4354
4355 cpu_buffer->head_page
4356 = list_entry(cpu_buffer->pages, struct buffer_page, list);
4357 local_set(&cpu_buffer->head_page->write, 0);
4358 local_set(&cpu_buffer->head_page->entries, 0);
4359 local_set(&cpu_buffer->head_page->page->commit, 0);
4360
4361 cpu_buffer->head_page->read = 0;
4362
4363 cpu_buffer->tail_page = cpu_buffer->head_page;
4364 cpu_buffer->commit_page = cpu_buffer->head_page;
4365
4366 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
4367 INIT_LIST_HEAD(&cpu_buffer->new_pages);
4368 local_set(&cpu_buffer->reader_page->write, 0);
4369 local_set(&cpu_buffer->reader_page->entries, 0);
4370 local_set(&cpu_buffer->reader_page->page->commit, 0);
4371 cpu_buffer->reader_page->read = 0;
4372
4373 local_set(&cpu_buffer->entries_bytes, 0);
4374 local_set(&cpu_buffer->overrun, 0);
4375 local_set(&cpu_buffer->commit_overrun, 0);
4376 local_set(&cpu_buffer->dropped_events, 0);
4377 local_set(&cpu_buffer->entries, 0);
4378 local_set(&cpu_buffer->committing, 0);
4379 local_set(&cpu_buffer->commits, 0);
4380 local_set(&cpu_buffer->pages_touched, 0);
4381 local_set(&cpu_buffer->pages_read, 0);
4382 cpu_buffer->last_pages_touch = 0;
4383 cpu_buffer->shortest_full = 0;
4384 cpu_buffer->read = 0;
4385 cpu_buffer->read_bytes = 0;
4386
4387 cpu_buffer->write_stamp = 0;
4388 cpu_buffer->read_stamp = 0;
4389
4390 cpu_buffer->lost_events = 0;
4391 cpu_buffer->last_overrun = 0;
4392
4393 rb_head_page_activate(cpu_buffer);
4394 }
4395
4396 /**
4397 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
4398 * @buffer: The ring buffer to reset a per cpu buffer of
4399 * @cpu: The CPU buffer to be reset
4400 */
ring_buffer_reset_cpu(struct ring_buffer * buffer,int cpu)4401 void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
4402 {
4403 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
4404 unsigned long flags;
4405
4406 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4407 return;
4408
4409 atomic_inc(&buffer->resize_disabled);
4410 atomic_inc(&cpu_buffer->record_disabled);
4411
4412 /* Make sure all commits have finished */
4413 synchronize_rcu();
4414
4415 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4416
4417 if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing)))
4418 goto out;
4419
4420 arch_spin_lock(&cpu_buffer->lock);
4421
4422 rb_reset_cpu(cpu_buffer);
4423
4424 arch_spin_unlock(&cpu_buffer->lock);
4425
4426 out:
4427 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4428
4429 atomic_dec(&cpu_buffer->record_disabled);
4430 atomic_dec(&buffer->resize_disabled);
4431 }
4432 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
4433
4434 /**
4435 * ring_buffer_reset - reset a ring buffer
4436 * @buffer: The ring buffer to reset all cpu buffers
4437 */
ring_buffer_reset(struct ring_buffer * buffer)4438 void ring_buffer_reset(struct ring_buffer *buffer)
4439 {
4440 int cpu;
4441
4442 for_each_buffer_cpu(buffer, cpu)
4443 ring_buffer_reset_cpu(buffer, cpu);
4444 }
4445 EXPORT_SYMBOL_GPL(ring_buffer_reset);
4446
4447 /**
4448 * rind_buffer_empty - is the ring buffer empty?
4449 * @buffer: The ring buffer to test
4450 */
ring_buffer_empty(struct ring_buffer * buffer)4451 bool ring_buffer_empty(struct ring_buffer *buffer)
4452 {
4453 struct ring_buffer_per_cpu *cpu_buffer;
4454 unsigned long flags;
4455 bool dolock;
4456 int cpu;
4457 int ret;
4458
4459 /* yes this is racy, but if you don't like the race, lock the buffer */
4460 for_each_buffer_cpu(buffer, cpu) {
4461 cpu_buffer = buffer->buffers[cpu];
4462 local_irq_save(flags);
4463 dolock = rb_reader_lock(cpu_buffer);
4464 ret = rb_per_cpu_empty(cpu_buffer);
4465 rb_reader_unlock(cpu_buffer, dolock);
4466 local_irq_restore(flags);
4467
4468 if (!ret)
4469 return false;
4470 }
4471
4472 return true;
4473 }
4474 EXPORT_SYMBOL_GPL(ring_buffer_empty);
4475
4476 /**
4477 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
4478 * @buffer: The ring buffer
4479 * @cpu: The CPU buffer to test
4480 */
ring_buffer_empty_cpu(struct ring_buffer * buffer,int cpu)4481 bool ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu)
4482 {
4483 struct ring_buffer_per_cpu *cpu_buffer;
4484 unsigned long flags;
4485 bool dolock;
4486 int ret;
4487
4488 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4489 return true;
4490
4491 cpu_buffer = buffer->buffers[cpu];
4492 local_irq_save(flags);
4493 dolock = rb_reader_lock(cpu_buffer);
4494 ret = rb_per_cpu_empty(cpu_buffer);
4495 rb_reader_unlock(cpu_buffer, dolock);
4496 local_irq_restore(flags);
4497
4498 return ret;
4499 }
4500 EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
4501
4502 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4503 /**
4504 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
4505 * @buffer_a: One buffer to swap with
4506 * @buffer_b: The other buffer to swap with
4507 *
4508 * This function is useful for tracers that want to take a "snapshot"
4509 * of a CPU buffer and has another back up buffer lying around.
4510 * it is expected that the tracer handles the cpu buffer not being
4511 * used at the moment.
4512 */
ring_buffer_swap_cpu(struct ring_buffer * buffer_a,struct ring_buffer * buffer_b,int cpu)4513 int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
4514 struct ring_buffer *buffer_b, int cpu)
4515 {
4516 struct ring_buffer_per_cpu *cpu_buffer_a;
4517 struct ring_buffer_per_cpu *cpu_buffer_b;
4518 int ret = -EINVAL;
4519
4520 if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
4521 !cpumask_test_cpu(cpu, buffer_b->cpumask))
4522 goto out;
4523
4524 cpu_buffer_a = buffer_a->buffers[cpu];
4525 cpu_buffer_b = buffer_b->buffers[cpu];
4526
4527 /* At least make sure the two buffers are somewhat the same */
4528 if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages)
4529 goto out;
4530
4531 ret = -EAGAIN;
4532
4533 if (atomic_read(&buffer_a->record_disabled))
4534 goto out;
4535
4536 if (atomic_read(&buffer_b->record_disabled))
4537 goto out;
4538
4539 if (atomic_read(&cpu_buffer_a->record_disabled))
4540 goto out;
4541
4542 if (atomic_read(&cpu_buffer_b->record_disabled))
4543 goto out;
4544
4545 /*
4546 * We can't do a synchronize_rcu here because this
4547 * function can be called in atomic context.
4548 * Normally this will be called from the same CPU as cpu.
4549 * If not it's up to the caller to protect this.
4550 */
4551 atomic_inc(&cpu_buffer_a->record_disabled);
4552 atomic_inc(&cpu_buffer_b->record_disabled);
4553
4554 ret = -EBUSY;
4555 if (local_read(&cpu_buffer_a->committing))
4556 goto out_dec;
4557 if (local_read(&cpu_buffer_b->committing))
4558 goto out_dec;
4559
4560 buffer_a->buffers[cpu] = cpu_buffer_b;
4561 buffer_b->buffers[cpu] = cpu_buffer_a;
4562
4563 cpu_buffer_b->buffer = buffer_a;
4564 cpu_buffer_a->buffer = buffer_b;
4565
4566 ret = 0;
4567
4568 out_dec:
4569 atomic_dec(&cpu_buffer_a->record_disabled);
4570 atomic_dec(&cpu_buffer_b->record_disabled);
4571 out:
4572 return ret;
4573 }
4574 EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
4575 #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */
4576
4577 /**
4578 * ring_buffer_alloc_read_page - allocate a page to read from buffer
4579 * @buffer: the buffer to allocate for.
4580 * @cpu: the cpu buffer to allocate.
4581 *
4582 * This function is used in conjunction with ring_buffer_read_page.
4583 * When reading a full page from the ring buffer, these functions
4584 * can be used to speed up the process. The calling function should
4585 * allocate a few pages first with this function. Then when it
4586 * needs to get pages from the ring buffer, it passes the result
4587 * of this function into ring_buffer_read_page, which will swap
4588 * the page that was allocated, with the read page of the buffer.
4589 *
4590 * Returns:
4591 * The page allocated, or ERR_PTR
4592 */
ring_buffer_alloc_read_page(struct ring_buffer * buffer,int cpu)4593 void *ring_buffer_alloc_read_page(struct ring_buffer *buffer, int cpu)
4594 {
4595 struct ring_buffer_per_cpu *cpu_buffer;
4596 struct buffer_data_page *bpage = NULL;
4597 unsigned long flags;
4598 struct page *page;
4599
4600 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4601 return ERR_PTR(-ENODEV);
4602
4603 cpu_buffer = buffer->buffers[cpu];
4604 local_irq_save(flags);
4605 arch_spin_lock(&cpu_buffer->lock);
4606
4607 if (cpu_buffer->free_page) {
4608 bpage = cpu_buffer->free_page;
4609 cpu_buffer->free_page = NULL;
4610 }
4611
4612 arch_spin_unlock(&cpu_buffer->lock);
4613 local_irq_restore(flags);
4614
4615 if (bpage)
4616 goto out;
4617
4618 page = alloc_pages_node(cpu_to_node(cpu),
4619 GFP_KERNEL | __GFP_NORETRY, 0);
4620 if (!page)
4621 return ERR_PTR(-ENOMEM);
4622
4623 bpage = page_address(page);
4624
4625 out:
4626 rb_init_page(bpage);
4627
4628 return bpage;
4629 }
4630 EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page);
4631
4632 /**
4633 * ring_buffer_free_read_page - free an allocated read page
4634 * @buffer: the buffer the page was allocate for
4635 * @cpu: the cpu buffer the page came from
4636 * @data: the page to free
4637 *
4638 * Free a page allocated from ring_buffer_alloc_read_page.
4639 */
ring_buffer_free_read_page(struct ring_buffer * buffer,int cpu,void * data)4640 void ring_buffer_free_read_page(struct ring_buffer *buffer, int cpu, void *data)
4641 {
4642 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
4643 struct buffer_data_page *bpage = data;
4644 struct page *page = virt_to_page(bpage);
4645 unsigned long flags;
4646
4647 /* If the page is still in use someplace else, we can't reuse it */
4648 if (page_ref_count(page) > 1)
4649 goto out;
4650
4651 local_irq_save(flags);
4652 arch_spin_lock(&cpu_buffer->lock);
4653
4654 if (!cpu_buffer->free_page) {
4655 cpu_buffer->free_page = bpage;
4656 bpage = NULL;
4657 }
4658
4659 arch_spin_unlock(&cpu_buffer->lock);
4660 local_irq_restore(flags);
4661
4662 out:
4663 free_page((unsigned long)bpage);
4664 }
4665 EXPORT_SYMBOL_GPL(ring_buffer_free_read_page);
4666
4667 /**
4668 * ring_buffer_read_page - extract a page from the ring buffer
4669 * @buffer: buffer to extract from
4670 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
4671 * @len: amount to extract
4672 * @cpu: the cpu of the buffer to extract
4673 * @full: should the extraction only happen when the page is full.
4674 *
4675 * This function will pull out a page from the ring buffer and consume it.
4676 * @data_page must be the address of the variable that was returned
4677 * from ring_buffer_alloc_read_page. This is because the page might be used
4678 * to swap with a page in the ring buffer.
4679 *
4680 * for example:
4681 * rpage = ring_buffer_alloc_read_page(buffer, cpu);
4682 * if (IS_ERR(rpage))
4683 * return PTR_ERR(rpage);
4684 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
4685 * if (ret >= 0)
4686 * process_page(rpage, ret);
4687 *
4688 * When @full is set, the function will not return true unless
4689 * the writer is off the reader page.
4690 *
4691 * Note: it is up to the calling functions to handle sleeps and wakeups.
4692 * The ring buffer can be used anywhere in the kernel and can not
4693 * blindly call wake_up. The layer that uses the ring buffer must be
4694 * responsible for that.
4695 *
4696 * Returns:
4697 * >=0 if data has been transferred, returns the offset of consumed data.
4698 * <0 if no data has been transferred.
4699 */
ring_buffer_read_page(struct ring_buffer * buffer,void ** data_page,size_t len,int cpu,int full)4700 int ring_buffer_read_page(struct ring_buffer *buffer,
4701 void **data_page, size_t len, int cpu, int full)
4702 {
4703 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
4704 struct ring_buffer_event *event;
4705 struct buffer_data_page *bpage;
4706 struct buffer_page *reader;
4707 unsigned long missed_events;
4708 unsigned long flags;
4709 unsigned int commit;
4710 unsigned int read;
4711 u64 save_timestamp;
4712 int ret = -1;
4713
4714 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4715 goto out;
4716
4717 /*
4718 * If len is not big enough to hold the page header, then
4719 * we can not copy anything.
4720 */
4721 if (len <= BUF_PAGE_HDR_SIZE)
4722 goto out;
4723
4724 len -= BUF_PAGE_HDR_SIZE;
4725
4726 if (!data_page)
4727 goto out;
4728
4729 bpage = *data_page;
4730 if (!bpage)
4731 goto out;
4732
4733 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4734
4735 reader = rb_get_reader_page(cpu_buffer);
4736 if (!reader)
4737 goto out_unlock;
4738
4739 event = rb_reader_event(cpu_buffer);
4740
4741 read = reader->read;
4742 commit = rb_page_commit(reader);
4743
4744 /* Check if any events were dropped */
4745 missed_events = cpu_buffer->lost_events;
4746
4747 /*
4748 * If this page has been partially read or
4749 * if len is not big enough to read the rest of the page or
4750 * a writer is still on the page, then
4751 * we must copy the data from the page to the buffer.
4752 * Otherwise, we can simply swap the page with the one passed in.
4753 */
4754 if (read || (len < (commit - read)) ||
4755 cpu_buffer->reader_page == cpu_buffer->commit_page) {
4756 struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
4757 unsigned int rpos = read;
4758 unsigned int pos = 0;
4759 unsigned int size;
4760
4761 if (full)
4762 goto out_unlock;
4763
4764 if (len > (commit - read))
4765 len = (commit - read);
4766
4767 /* Always keep the time extend and data together */
4768 size = rb_event_ts_length(event);
4769
4770 if (len < size)
4771 goto out_unlock;
4772
4773 /* save the current timestamp, since the user will need it */
4774 save_timestamp = cpu_buffer->read_stamp;
4775
4776 /* Need to copy one event at a time */
4777 do {
4778 /* We need the size of one event, because
4779 * rb_advance_reader only advances by one event,
4780 * whereas rb_event_ts_length may include the size of
4781 * one or two events.
4782 * We have already ensured there's enough space if this
4783 * is a time extend. */
4784 size = rb_event_length(event);
4785 memcpy(bpage->data + pos, rpage->data + rpos, size);
4786
4787 len -= size;
4788
4789 rb_advance_reader(cpu_buffer);
4790 rpos = reader->read;
4791 pos += size;
4792
4793 if (rpos >= commit)
4794 break;
4795
4796 event = rb_reader_event(cpu_buffer);
4797 /* Always keep the time extend and data together */
4798 size = rb_event_ts_length(event);
4799 } while (len >= size);
4800
4801 /* update bpage */
4802 local_set(&bpage->commit, pos);
4803 bpage->time_stamp = save_timestamp;
4804
4805 /* we copied everything to the beginning */
4806 read = 0;
4807 } else {
4808 /* update the entry counter */
4809 cpu_buffer->read += rb_page_entries(reader);
4810 cpu_buffer->read_bytes += BUF_PAGE_SIZE;
4811
4812 /* swap the pages */
4813 rb_init_page(bpage);
4814 bpage = reader->page;
4815 reader->page = *data_page;
4816 local_set(&reader->write, 0);
4817 local_set(&reader->entries, 0);
4818 reader->read = 0;
4819 *data_page = bpage;
4820
4821 /*
4822 * Use the real_end for the data size,
4823 * This gives us a chance to store the lost events
4824 * on the page.
4825 */
4826 if (reader->real_end)
4827 local_set(&bpage->commit, reader->real_end);
4828 }
4829 ret = read;
4830
4831 cpu_buffer->lost_events = 0;
4832
4833 commit = local_read(&bpage->commit);
4834 /*
4835 * Set a flag in the commit field if we lost events
4836 */
4837 if (missed_events) {
4838 /* If there is room at the end of the page to save the
4839 * missed events, then record it there.
4840 */
4841 if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) {
4842 memcpy(&bpage->data[commit], &missed_events,
4843 sizeof(missed_events));
4844 local_add(RB_MISSED_STORED, &bpage->commit);
4845 commit += sizeof(missed_events);
4846 }
4847 local_add(RB_MISSED_EVENTS, &bpage->commit);
4848 }
4849
4850 /*
4851 * This page may be off to user land. Zero it out here.
4852 */
4853 if (commit < BUF_PAGE_SIZE)
4854 memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit);
4855
4856 out_unlock:
4857 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4858
4859 out:
4860 return ret;
4861 }
4862 EXPORT_SYMBOL_GPL(ring_buffer_read_page);
4863
4864 /*
4865 * We only allocate new buffers, never free them if the CPU goes down.
4866 * If we were to free the buffer, then the user would lose any trace that was in
4867 * the buffer.
4868 */
trace_rb_cpu_prepare(unsigned int cpu,struct hlist_node * node)4869 int trace_rb_cpu_prepare(unsigned int cpu, struct hlist_node *node)
4870 {
4871 struct ring_buffer *buffer;
4872 long nr_pages_same;
4873 int cpu_i;
4874 unsigned long nr_pages;
4875
4876 buffer = container_of(node, struct ring_buffer, node);
4877 if (cpumask_test_cpu(cpu, buffer->cpumask))
4878 return 0;
4879
4880 nr_pages = 0;
4881 nr_pages_same = 1;
4882 /* check if all cpu sizes are same */
4883 for_each_buffer_cpu(buffer, cpu_i) {
4884 /* fill in the size from first enabled cpu */
4885 if (nr_pages == 0)
4886 nr_pages = buffer->buffers[cpu_i]->nr_pages;
4887 if (nr_pages != buffer->buffers[cpu_i]->nr_pages) {
4888 nr_pages_same = 0;
4889 break;
4890 }
4891 }
4892 /* allocate minimum pages, user can later expand it */
4893 if (!nr_pages_same)
4894 nr_pages = 2;
4895 buffer->buffers[cpu] =
4896 rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
4897 if (!buffer->buffers[cpu]) {
4898 WARN(1, "failed to allocate ring buffer on CPU %u\n",
4899 cpu);
4900 return -ENOMEM;
4901 }
4902 smp_wmb();
4903 cpumask_set_cpu(cpu, buffer->cpumask);
4904 return 0;
4905 }
4906
4907 #ifdef CONFIG_RING_BUFFER_STARTUP_TEST
4908 /*
4909 * This is a basic integrity check of the ring buffer.
4910 * Late in the boot cycle this test will run when configured in.
4911 * It will kick off a thread per CPU that will go into a loop
4912 * writing to the per cpu ring buffer various sizes of data.
4913 * Some of the data will be large items, some small.
4914 *
4915 * Another thread is created that goes into a spin, sending out
4916 * IPIs to the other CPUs to also write into the ring buffer.
4917 * this is to test the nesting ability of the buffer.
4918 *
4919 * Basic stats are recorded and reported. If something in the
4920 * ring buffer should happen that's not expected, a big warning
4921 * is displayed and all ring buffers are disabled.
4922 */
4923 static struct task_struct *rb_threads[NR_CPUS] __initdata;
4924
4925 struct rb_test_data {
4926 struct ring_buffer *buffer;
4927 unsigned long events;
4928 unsigned long bytes_written;
4929 unsigned long bytes_alloc;
4930 unsigned long bytes_dropped;
4931 unsigned long events_nested;
4932 unsigned long bytes_written_nested;
4933 unsigned long bytes_alloc_nested;
4934 unsigned long bytes_dropped_nested;
4935 int min_size_nested;
4936 int max_size_nested;
4937 int max_size;
4938 int min_size;
4939 int cpu;
4940 int cnt;
4941 };
4942
4943 static struct rb_test_data rb_data[NR_CPUS] __initdata;
4944
4945 /* 1 meg per cpu */
4946 #define RB_TEST_BUFFER_SIZE 1048576
4947
4948 static char rb_string[] __initdata =
4949 "abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\"
4950 "?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890"
4951 "!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv";
4952
4953 static bool rb_test_started __initdata;
4954
4955 struct rb_item {
4956 int size;
4957 char str[];
4958 };
4959
rb_write_something(struct rb_test_data * data,bool nested)4960 static __init int rb_write_something(struct rb_test_data *data, bool nested)
4961 {
4962 struct ring_buffer_event *event;
4963 struct rb_item *item;
4964 bool started;
4965 int event_len;
4966 int size;
4967 int len;
4968 int cnt;
4969
4970 /* Have nested writes different that what is written */
4971 cnt = data->cnt + (nested ? 27 : 0);
4972
4973 /* Multiply cnt by ~e, to make some unique increment */
4974 size = (cnt * 68 / 25) % (sizeof(rb_string) - 1);
4975
4976 len = size + sizeof(struct rb_item);
4977
4978 started = rb_test_started;
4979 /* read rb_test_started before checking buffer enabled */
4980 smp_rmb();
4981
4982 event = ring_buffer_lock_reserve(data->buffer, len);
4983 if (!event) {
4984 /* Ignore dropped events before test starts. */
4985 if (started) {
4986 if (nested)
4987 data->bytes_dropped += len;
4988 else
4989 data->bytes_dropped_nested += len;
4990 }
4991 return len;
4992 }
4993
4994 event_len = ring_buffer_event_length(event);
4995
4996 if (RB_WARN_ON(data->buffer, event_len < len))
4997 goto out;
4998
4999 item = ring_buffer_event_data(event);
5000 item->size = size;
5001 memcpy(item->str, rb_string, size);
5002
5003 if (nested) {
5004 data->bytes_alloc_nested += event_len;
5005 data->bytes_written_nested += len;
5006 data->events_nested++;
5007 if (!data->min_size_nested || len < data->min_size_nested)
5008 data->min_size_nested = len;
5009 if (len > data->max_size_nested)
5010 data->max_size_nested = len;
5011 } else {
5012 data->bytes_alloc += event_len;
5013 data->bytes_written += len;
5014 data->events++;
5015 if (!data->min_size || len < data->min_size)
5016 data->max_size = len;
5017 if (len > data->max_size)
5018 data->max_size = len;
5019 }
5020
5021 out:
5022 ring_buffer_unlock_commit(data->buffer, event);
5023
5024 return 0;
5025 }
5026
rb_test(void * arg)5027 static __init int rb_test(void *arg)
5028 {
5029 struct rb_test_data *data = arg;
5030
5031 while (!kthread_should_stop()) {
5032 rb_write_something(data, false);
5033 data->cnt++;
5034
5035 set_current_state(TASK_INTERRUPTIBLE);
5036 /* Now sleep between a min of 100-300us and a max of 1ms */
5037 usleep_range(((data->cnt % 3) + 1) * 100, 1000);
5038 }
5039
5040 return 0;
5041 }
5042
rb_ipi(void * ignore)5043 static __init void rb_ipi(void *ignore)
5044 {
5045 struct rb_test_data *data;
5046 int cpu = smp_processor_id();
5047
5048 data = &rb_data[cpu];
5049 rb_write_something(data, true);
5050 }
5051
rb_hammer_test(void * arg)5052 static __init int rb_hammer_test(void *arg)
5053 {
5054 while (!kthread_should_stop()) {
5055
5056 /* Send an IPI to all cpus to write data! */
5057 smp_call_function(rb_ipi, NULL, 1);
5058 /* No sleep, but for non preempt, let others run */
5059 schedule();
5060 }
5061
5062 return 0;
5063 }
5064
test_ringbuffer(void)5065 static __init int test_ringbuffer(void)
5066 {
5067 struct task_struct *rb_hammer;
5068 struct ring_buffer *buffer;
5069 int cpu;
5070 int ret = 0;
5071
5072 if (security_locked_down(LOCKDOWN_TRACEFS)) {
5073 pr_warning("Lockdown is enabled, skipping ring buffer tests\n");
5074 return 0;
5075 }
5076
5077 pr_info("Running ring buffer tests...\n");
5078
5079 buffer = ring_buffer_alloc(RB_TEST_BUFFER_SIZE, RB_FL_OVERWRITE);
5080 if (WARN_ON(!buffer))
5081 return 0;
5082
5083 /* Disable buffer so that threads can't write to it yet */
5084 ring_buffer_record_off(buffer);
5085
5086 for_each_online_cpu(cpu) {
5087 rb_data[cpu].buffer = buffer;
5088 rb_data[cpu].cpu = cpu;
5089 rb_data[cpu].cnt = cpu;
5090 rb_threads[cpu] = kthread_create(rb_test, &rb_data[cpu],
5091 "rbtester/%d", cpu);
5092 if (WARN_ON(IS_ERR(rb_threads[cpu]))) {
5093 pr_cont("FAILED\n");
5094 ret = PTR_ERR(rb_threads[cpu]);
5095 goto out_free;
5096 }
5097
5098 kthread_bind(rb_threads[cpu], cpu);
5099 wake_up_process(rb_threads[cpu]);
5100 }
5101
5102 /* Now create the rb hammer! */
5103 rb_hammer = kthread_run(rb_hammer_test, NULL, "rbhammer");
5104 if (WARN_ON(IS_ERR(rb_hammer))) {
5105 pr_cont("FAILED\n");
5106 ret = PTR_ERR(rb_hammer);
5107 goto out_free;
5108 }
5109
5110 ring_buffer_record_on(buffer);
5111 /*
5112 * Show buffer is enabled before setting rb_test_started.
5113 * Yes there's a small race window where events could be
5114 * dropped and the thread wont catch it. But when a ring
5115 * buffer gets enabled, there will always be some kind of
5116 * delay before other CPUs see it. Thus, we don't care about
5117 * those dropped events. We care about events dropped after
5118 * the threads see that the buffer is active.
5119 */
5120 smp_wmb();
5121 rb_test_started = true;
5122
5123 set_current_state(TASK_INTERRUPTIBLE);
5124 /* Just run for 10 seconds */;
5125 schedule_timeout(10 * HZ);
5126
5127 kthread_stop(rb_hammer);
5128
5129 out_free:
5130 for_each_online_cpu(cpu) {
5131 if (!rb_threads[cpu])
5132 break;
5133 kthread_stop(rb_threads[cpu]);
5134 }
5135 if (ret) {
5136 ring_buffer_free(buffer);
5137 return ret;
5138 }
5139
5140 /* Report! */
5141 pr_info("finished\n");
5142 for_each_online_cpu(cpu) {
5143 struct ring_buffer_event *event;
5144 struct rb_test_data *data = &rb_data[cpu];
5145 struct rb_item *item;
5146 unsigned long total_events;
5147 unsigned long total_dropped;
5148 unsigned long total_written;
5149 unsigned long total_alloc;
5150 unsigned long total_read = 0;
5151 unsigned long total_size = 0;
5152 unsigned long total_len = 0;
5153 unsigned long total_lost = 0;
5154 unsigned long lost;
5155 int big_event_size;
5156 int small_event_size;
5157
5158 ret = -1;
5159
5160 total_events = data->events + data->events_nested;
5161 total_written = data->bytes_written + data->bytes_written_nested;
5162 total_alloc = data->bytes_alloc + data->bytes_alloc_nested;
5163 total_dropped = data->bytes_dropped + data->bytes_dropped_nested;
5164
5165 big_event_size = data->max_size + data->max_size_nested;
5166 small_event_size = data->min_size + data->min_size_nested;
5167
5168 pr_info("CPU %d:\n", cpu);
5169 pr_info(" events: %ld\n", total_events);
5170 pr_info(" dropped bytes: %ld\n", total_dropped);
5171 pr_info(" alloced bytes: %ld\n", total_alloc);
5172 pr_info(" written bytes: %ld\n", total_written);
5173 pr_info(" biggest event: %d\n", big_event_size);
5174 pr_info(" smallest event: %d\n", small_event_size);
5175
5176 if (RB_WARN_ON(buffer, total_dropped))
5177 break;
5178
5179 ret = 0;
5180
5181 while ((event = ring_buffer_consume(buffer, cpu, NULL, &lost))) {
5182 total_lost += lost;
5183 item = ring_buffer_event_data(event);
5184 total_len += ring_buffer_event_length(event);
5185 total_size += item->size + sizeof(struct rb_item);
5186 if (memcmp(&item->str[0], rb_string, item->size) != 0) {
5187 pr_info("FAILED!\n");
5188 pr_info("buffer had: %.*s\n", item->size, item->str);
5189 pr_info("expected: %.*s\n", item->size, rb_string);
5190 RB_WARN_ON(buffer, 1);
5191 ret = -1;
5192 break;
5193 }
5194 total_read++;
5195 }
5196 if (ret)
5197 break;
5198
5199 ret = -1;
5200
5201 pr_info(" read events: %ld\n", total_read);
5202 pr_info(" lost events: %ld\n", total_lost);
5203 pr_info(" total events: %ld\n", total_lost + total_read);
5204 pr_info(" recorded len bytes: %ld\n", total_len);
5205 pr_info(" recorded size bytes: %ld\n", total_size);
5206 if (total_lost)
5207 pr_info(" With dropped events, record len and size may not match\n"
5208 " alloced and written from above\n");
5209 if (!total_lost) {
5210 if (RB_WARN_ON(buffer, total_len != total_alloc ||
5211 total_size != total_written))
5212 break;
5213 }
5214 if (RB_WARN_ON(buffer, total_lost + total_read != total_events))
5215 break;
5216
5217 ret = 0;
5218 }
5219 if (!ret)
5220 pr_info("Ring buffer PASSED!\n");
5221
5222 ring_buffer_free(buffer);
5223 return 0;
5224 }
5225
5226 late_initcall(test_ringbuffer);
5227 #endif /* CONFIG_RING_BUFFER_STARTUP_TEST */
5228