projects/linux·blob 1e098dec · refs

6215 lines · 164.9 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Generic ring buffer
4 *
5 * Copyright (C) 2008 Steven Rostedt <[email protected]>
6 */
7#include <linux/trace_recursion.h>
8#include <linux/trace_events.h>
9#include <linux/ring_buffer.h>
10#include <linux/trace_clock.h>
11#include <linux/sched/clock.h>
12#include <linux/trace_seq.h>
13#include <linux/spinlock.h>
14#include <linux/irq_work.h>
15#include <linux/security.h>
16#include <linux/uaccess.h>
17#include <linux/hardirq.h>
18#include <linux/kthread.h> /* for self test */
19#include <linux/module.h>
20#include <linux/percpu.h>
21#include <linux/mutex.h>
22#include <linux/delay.h>
23#include <linux/slab.h>
24#include <linux/init.h>
25#include <linux/hash.h>
26#include <linux/list.h>
27#include <linux/cpu.h>
28#include <linux/oom.h>
29
30#include <asm/local.h>
31
32/*
33 * The "absolute" timestamp in the buffer is only 59 bits.
34 * If a clock has the 5 MSBs set, it needs to be saved and
35 * reinserted.
36 */
37#define TS_MSB (0xf8ULL << 56)
38#define ABS_TS_MASK (~TS_MSB)
39
40static void update_pages_handler(struct work_struct *work);
41
42/*
43 * The ring buffer header is special. We must manually up keep it.
44 */
45int ring_buffer_print_entry_header(struct trace_seq *s)
46{
47 trace_seq_puts(s, "# compressed entry header\n");
48 trace_seq_puts(s, "\ttype_len : 5 bits\n");
49 trace_seq_puts(s, "\ttime_delta : 27 bits\n");
50 trace_seq_puts(s, "\tarray : 32 bits\n");
51 trace_seq_putc(s, '\n');
52 trace_seq_printf(s, "\tpadding : type == %d\n",
53 RINGBUF_TYPE_PADDING);
54 trace_seq_printf(s, "\ttime_extend : type == %d\n",
55 RINGBUF_TYPE_TIME_EXTEND);
56 trace_seq_printf(s, "\ttime_stamp : type == %d\n",
57 RINGBUF_TYPE_TIME_STAMP);
58 trace_seq_printf(s, "\tdata max type_len == %d\n",
59 RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
60
61 return !trace_seq_has_overflowed(s);
62}
63
64/*
65 * The ring buffer is made up of a list of pages. A separate list of pages is
66 * allocated for each CPU. A writer may only write to a buffer that is
67 * associated with the CPU it is currently executing on. A reader may read
68 * from any per cpu buffer.
69 *
70 * The reader is special. For each per cpu buffer, the reader has its own
71 * reader page. When a reader has read the entire reader page, this reader
72 * page is swapped with another page in the ring buffer.
73 *
74 * Now, as long as the writer is off the reader page, the reader can do what
75 * ever it wants with that page. The writer will never write to that page
76 * again (as long as it is out of the ring buffer).
77 *
78 * Here's some silly ASCII art.
79 *
80 * +------+
81 * |reader| RING BUFFER
82 * |page |
83 * +------+ +---+ +---+ +---+
84 * | |-->| |-->| |
85 * +---+ +---+ +---+
86 * ^ |
87 * | |
88 * +---------------+
89 *
90 *
91 * +------+
92 * |reader| RING BUFFER
93 * |page |------------------v
94 * +------+ +---+ +---+ +---+
95 * | |-->| |-->| |
96 * +---+ +---+ +---+
97 * ^ |
98 * | |
99 * +---------------+
100 *
101 *
102 * +------+
103 * |reader| RING BUFFER
104 * |page |------------------v
105 * +------+ +---+ +---+ +---+
106 * ^ | |-->| |-->| |
107 * | +---+ +---+ +---+
108 * | |
109 * | |
110 * +------------------------------+
111 *
112 *
113 * +------+
114 * |buffer| RING BUFFER
115 * |page |------------------v
116 * +------+ +---+ +---+ +---+
117 * ^ | | | |-->| |
118 * | New +---+ +---+ +---+
119 * | Reader------^ |
120 * | page |
121 * +------------------------------+
122 *
123 *
124 * After we make this swap, the reader can hand this page off to the splice
125 * code and be done with it. It can even allocate a new page if it needs to
126 * and swap that into the ring buffer.
127 *
128 * We will be using cmpxchg soon to make all this lockless.
129 *
130 */
131
132/* Used for individual buffers (after the counter) */
133#define RB_BUFFER_OFF (1 << 20)
134
135#define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
136
137#define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
138#define RB_ALIGNMENT 4U
139#define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
140#define RB_EVNT_MIN_SIZE 8U /* two 32bit words */
141
142#ifndef CONFIG_HAVE_64BIT_ALIGNED_ACCESS
143# define RB_FORCE_8BYTE_ALIGNMENT 0
144# define RB_ARCH_ALIGNMENT RB_ALIGNMENT
145#else
146# define RB_FORCE_8BYTE_ALIGNMENT 1
147# define RB_ARCH_ALIGNMENT 8U
148#endif
149
150#define RB_ALIGN_DATA __aligned(RB_ARCH_ALIGNMENT)
151
152/* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
153#define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
154
155enum {
156 RB_LEN_TIME_EXTEND = 8,
157 RB_LEN_TIME_STAMP = 8,
158};
159
160#define skip_time_extend(event) \
161 ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND))
162
163#define extended_time(event) \
164 (event->type_len >= RINGBUF_TYPE_TIME_EXTEND)
165
166static inline bool rb_null_event(struct ring_buffer_event *event)
167{
168 return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta;
169}
170
171static void rb_event_set_padding(struct ring_buffer_event *event)
172{
173 /* padding has a NULL time_delta */
174 event->type_len = RINGBUF_TYPE_PADDING;
175 event->time_delta = 0;
176}
177
178static unsigned
179rb_event_data_length(struct ring_buffer_event *event)
180{
181 unsigned length;
182
183 if (event->type_len)
184 length = event->type_len * RB_ALIGNMENT;
185 else
186 length = event->array[0];
187 return length + RB_EVNT_HDR_SIZE;
188}
189
190/*
191 * Return the length of the given event. Will return
192 * the length of the time extend if the event is a
193 * time extend.
194 */
195static inline unsigned
196rb_event_length(struct ring_buffer_event *event)
197{
198 switch (event->type_len) {
199 case RINGBUF_TYPE_PADDING:
200 if (rb_null_event(event))
201 /* undefined */
202 return -1;
203 return event->array[0] + RB_EVNT_HDR_SIZE;
204
205 case RINGBUF_TYPE_TIME_EXTEND:
206 return RB_LEN_TIME_EXTEND;
207
208 case RINGBUF_TYPE_TIME_STAMP:
209 return RB_LEN_TIME_STAMP;
210
211 case RINGBUF_TYPE_DATA:
212 return rb_event_data_length(event);
213 default:
214 WARN_ON_ONCE(1);
215 }
216 /* not hit */
217 return 0;
218}
219
220/*
221 * Return total length of time extend and data,
222 * or just the event length for all other events.
223 */
224static inline unsigned
225rb_event_ts_length(struct ring_buffer_event *event)
226{
227 unsigned len = 0;
228
229 if (extended_time(event)) {
230 /* time extends include the data event after it */
231 len = RB_LEN_TIME_EXTEND;
232 event = skip_time_extend(event);
233 }
234 return len + rb_event_length(event);
235}
236
237/**
238 * ring_buffer_event_length - return the length of the event
239 * @event: the event to get the length of
240 *
241 * Returns the size of the data load of a data event.
242 * If the event is something other than a data event, it
243 * returns the size of the event itself. With the exception
244 * of a TIME EXTEND, where it still returns the size of the
245 * data load of the data event after it.
246 */
247unsigned ring_buffer_event_length(struct ring_buffer_event *event)
248{
249 unsigned length;
250
251 if (extended_time(event))
252 event = skip_time_extend(event);
253
254 length = rb_event_length(event);
255 if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
256 return length;
257 length -= RB_EVNT_HDR_SIZE;
258 if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0]))
259 length -= sizeof(event->array[0]);
260 return length;
261}
262EXPORT_SYMBOL_GPL(ring_buffer_event_length);
263
264/* inline for ring buffer fast paths */
265static __always_inline void *
266rb_event_data(struct ring_buffer_event *event)
267{
268 if (extended_time(event))
269 event = skip_time_extend(event);
270 WARN_ON_ONCE(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
271 /* If length is in len field, then array[0] has the data */
272 if (event->type_len)
273 return (void *)&event->array[0];
274 /* Otherwise length is in array[0] and array[1] has the data */
275 return (void *)&event->array[1];
276}
277
278/**
279 * ring_buffer_event_data - return the data of the event
280 * @event: the event to get the data from
281 */
282void *ring_buffer_event_data(struct ring_buffer_event *event)
283{
284 return rb_event_data(event);
285}
286EXPORT_SYMBOL_GPL(ring_buffer_event_data);
287
288#define for_each_buffer_cpu(buffer, cpu) \
289 for_each_cpu(cpu, buffer->cpumask)
290
291#define for_each_online_buffer_cpu(buffer, cpu) \
292 for_each_cpu_and(cpu, buffer->cpumask, cpu_online_mask)
293
294#define TS_SHIFT 27
295#define TS_MASK ((1ULL << TS_SHIFT) - 1)
296#define TS_DELTA_TEST (~TS_MASK)
297
298static u64 rb_event_time_stamp(struct ring_buffer_event *event)
299{
300 u64 ts;
301
302 ts = event->array[0];
303 ts <<= TS_SHIFT;
304 ts += event->time_delta;
305
306 return ts;
307}
308
309/* Flag when events were overwritten */
310#define RB_MISSED_EVENTS (1 << 31)
311/* Missed count stored at end */
312#define RB_MISSED_STORED (1 << 30)
313
314struct buffer_data_page {
315 u64 time_stamp; /* page time stamp */
316 local_t commit; /* write committed index */
317 unsigned char data[] RB_ALIGN_DATA; /* data of buffer page */
318};
319
320/*
321 * Note, the buffer_page list must be first. The buffer pages
322 * are allocated in cache lines, which means that each buffer
323 * page will be at the beginning of a cache line, and thus
324 * the least significant bits will be zero. We use this to
325 * add flags in the list struct pointers, to make the ring buffer
326 * lockless.
327 */
328struct buffer_page {
329 struct list_head list; /* list of buffer pages */
330 local_t write; /* index for next write */
331 unsigned read; /* index for next read */
332 local_t entries; /* entries on this page */
333 unsigned long real_end; /* real end of data */
334 struct buffer_data_page *page; /* Actual data page */
335};
336
337/*
338 * The buffer page counters, write and entries, must be reset
339 * atomically when crossing page boundaries. To synchronize this
340 * update, two counters are inserted into the number. One is
341 * the actual counter for the write position or count on the page.
342 *
343 * The other is a counter of updaters. Before an update happens
344 * the update partition of the counter is incremented. This will
345 * allow the updater to update the counter atomically.
346 *
347 * The counter is 20 bits, and the state data is 12.
348 */
349#define RB_WRITE_MASK 0xfffff
350#define RB_WRITE_INTCNT (1 << 20)
351
352static void rb_init_page(struct buffer_data_page *bpage)
353{
354 local_set(&bpage->commit, 0);
355}
356
357static void free_buffer_page(struct buffer_page *bpage)
358{
359 free_page((unsigned long)bpage->page);
360 kfree(bpage);
361}
362
363/*
364 * We need to fit the time_stamp delta into 27 bits.
365 */
366static inline bool test_time_stamp(u64 delta)
367{
368 return !!(delta & TS_DELTA_TEST);
369}
370
371#define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
372
373/* Max payload is BUF_PAGE_SIZE - header (8bytes) */
374#define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
375
376int ring_buffer_print_page_header(struct trace_seq *s)
377{
378 struct buffer_data_page field;
379
380 trace_seq_printf(s, "\tfield: u64 timestamp;\t"
381 "offset:0;\tsize:%u;\tsigned:%u;\n",
382 (unsigned int)sizeof(field.time_stamp),
383 (unsigned int)is_signed_type(u64));
384
385 trace_seq_printf(s, "\tfield: local_t commit;\t"
386 "offset:%u;\tsize:%u;\tsigned:%u;\n",
387 (unsigned int)offsetof(typeof(field), commit),
388 (unsigned int)sizeof(field.commit),
389 (unsigned int)is_signed_type(long));
390
391 trace_seq_printf(s, "\tfield: int overwrite;\t"
392 "offset:%u;\tsize:%u;\tsigned:%u;\n",
393 (unsigned int)offsetof(typeof(field), commit),
394 1,
395 (unsigned int)is_signed_type(long));
396
397 trace_seq_printf(s, "\tfield: char data;\t"
398 "offset:%u;\tsize:%u;\tsigned:%u;\n",
399 (unsigned int)offsetof(typeof(field), data),
400 (unsigned int)BUF_PAGE_SIZE,
401 (unsigned int)is_signed_type(char));
402
403 return !trace_seq_has_overflowed(s);
404}
405
406struct rb_irq_work {
407 struct irq_work work;
408 wait_queue_head_t waiters;
409 wait_queue_head_t full_waiters;
410 long wait_index;
411 bool waiters_pending;
412 bool full_waiters_pending;
413 bool wakeup_full;
414};
415
416/*
417 * Structure to hold event state and handle nested events.
418 */
419struct rb_event_info {
420 u64 ts;
421 u64 delta;
422 u64 before;
423 u64 after;
424 unsigned long length;
425 struct buffer_page *tail_page;
426 int add_timestamp;
427};
428
429/*
430 * Used for the add_timestamp
431 * NONE
432 * EXTEND - wants a time extend
433 * ABSOLUTE - the buffer requests all events to have absolute time stamps
434 * FORCE - force a full time stamp.
435 */
436enum {
437 RB_ADD_STAMP_NONE = 0,
438 RB_ADD_STAMP_EXTEND = BIT(1),
439 RB_ADD_STAMP_ABSOLUTE = BIT(2),
440 RB_ADD_STAMP_FORCE = BIT(3)
441};
442/*
443 * Used for which event context the event is in.
444 * TRANSITION = 0
445 * NMI = 1
446 * IRQ = 2
447 * SOFTIRQ = 3
448 * NORMAL = 4
449 *
450 * See trace_recursive_lock() comment below for more details.
451 */
452enum {
453 RB_CTX_TRANSITION,
454 RB_CTX_NMI,
455 RB_CTX_IRQ,
456 RB_CTX_SOFTIRQ,
457 RB_CTX_NORMAL,
458 RB_CTX_MAX
459};
460
461#if BITS_PER_LONG == 32
462#define RB_TIME_32
463#endif
464
465/* To test on 64 bit machines */
466//#define RB_TIME_32
467
468#ifdef RB_TIME_32
469
470struct rb_time_struct {
471 local_t cnt;
472 local_t top;
473 local_t bottom;
474 local_t msb;
475};
476#else
477#include <asm/local64.h>
478struct rb_time_struct {
479 local64_t time;
480};
481#endif
482typedef struct rb_time_struct rb_time_t;
483
484#define MAX_NEST 5
485
486/*
487 * head_page == tail_page && head == tail then buffer is empty.
488 */
489struct ring_buffer_per_cpu {
490 int cpu;
491 atomic_t record_disabled;
492 atomic_t resize_disabled;
493 struct trace_buffer *buffer;
494 raw_spinlock_t reader_lock; /* serialize readers */
495 arch_spinlock_t lock;
496 struct lock_class_key lock_key;
497 struct buffer_data_page *free_page;
498 unsigned long nr_pages;
499 unsigned int current_context;
500 struct list_head *pages;
501 struct buffer_page *head_page; /* read from head */
502 struct buffer_page *tail_page; /* write to tail */
503 struct buffer_page *commit_page; /* committed pages */
504 struct buffer_page *reader_page;
505 unsigned long lost_events;
506 unsigned long last_overrun;
507 unsigned long nest;
508 local_t entries_bytes;
509 local_t entries;
510 local_t overrun;
511 local_t commit_overrun;
512 local_t dropped_events;
513 local_t committing;
514 local_t commits;
515 local_t pages_touched;
516 local_t pages_lost;
517 local_t pages_read;
518 long last_pages_touch;
519 size_t shortest_full;
520 unsigned long read;
521 unsigned long read_bytes;
522 rb_time_t write_stamp;
523 rb_time_t before_stamp;
524 u64 event_stamp[MAX_NEST];
525 u64 read_stamp;
526 /* ring buffer pages to update, > 0 to add, < 0 to remove */
527 long nr_pages_to_update;
528 struct list_head new_pages; /* new pages to add */
529 struct work_struct update_pages_work;
530 struct completion update_done;
531
532 struct rb_irq_work irq_work;
533};
534
535struct trace_buffer {
536 unsigned flags;
537 int cpus;
538 atomic_t record_disabled;
539 cpumask_var_t cpumask;
540
541 struct lock_class_key *reader_lock_key;
542
543 struct mutex mutex;
544
545 struct ring_buffer_per_cpu **buffers;
546
547 struct hlist_node node;
548 u64 (*clock)(void);
549
550 struct rb_irq_work irq_work;
551 bool time_stamp_abs;
552};
553
554struct ring_buffer_iter {
555 struct ring_buffer_per_cpu *cpu_buffer;
556 unsigned long head;
557 unsigned long next_event;
558 struct buffer_page *head_page;
559 struct buffer_page *cache_reader_page;
560 unsigned long cache_read;
561 u64 read_stamp;
562 u64 page_stamp;
563 struct ring_buffer_event *event;
564 int missed_events;
565};
566
567#ifdef RB_TIME_32
568
569/*
570 * On 32 bit machines, local64_t is very expensive. As the ring
571 * buffer doesn't need all the features of a true 64 bit atomic,
572 * on 32 bit, it uses these functions (64 still uses local64_t).
573 *
574 * For the ring buffer, 64 bit required operations for the time is
575 * the following:
576 *
577 * - Reads may fail if it interrupted a modification of the time stamp.
578 * It will succeed if it did not interrupt another write even if
579 * the read itself is interrupted by a write.
580 * It returns whether it was successful or not.
581 *
582 * - Writes always succeed and will overwrite other writes and writes
583 * that were done by events interrupting the current write.
584 *
585 * - A write followed by a read of the same time stamp will always succeed,
586 * but may not contain the same value.
587 *
588 * - A cmpxchg will fail if it interrupted another write or cmpxchg.
589 * Other than that, it acts like a normal cmpxchg.
590 *
591 * The 60 bit time stamp is broken up by 30 bits in a top and bottom half
592 * (bottom being the least significant 30 bits of the 60 bit time stamp).
593 *
594 * The two most significant bits of each half holds a 2 bit counter (0-3).
595 * Each update will increment this counter by one.
596 * When reading the top and bottom, if the two counter bits match then the
597 * top and bottom together make a valid 60 bit number.
598 */
599#define RB_TIME_SHIFT 30
600#define RB_TIME_VAL_MASK ((1 << RB_TIME_SHIFT) - 1)
601#define RB_TIME_MSB_SHIFT 60
602
603static inline int rb_time_cnt(unsigned long val)
604{
605 return (val >> RB_TIME_SHIFT) & 3;
606}
607
608static inline u64 rb_time_val(unsigned long top, unsigned long bottom)
609{
610 u64 val;
611
612 val = top & RB_TIME_VAL_MASK;
613 val <<= RB_TIME_SHIFT;
614 val |= bottom & RB_TIME_VAL_MASK;
615
616 return val;
617}
618
619static inline bool __rb_time_read(rb_time_t *t, u64 *ret, unsigned long *cnt)
620{
621 unsigned long top, bottom, msb;
622 unsigned long c;
623
624 /*
625 * If the read is interrupted by a write, then the cnt will
626 * be different. Loop until both top and bottom have been read
627 * without interruption.
628 */
629 do {
630 c = local_read(&t->cnt);
631 top = local_read(&t->top);
632 bottom = local_read(&t->bottom);
633 msb = local_read(&t->msb);
634 } while (c != local_read(&t->cnt));
635
636 *cnt = rb_time_cnt(top);
637
638 /* If top and bottom counts don't match, this interrupted a write */
639 if (*cnt != rb_time_cnt(bottom))
640 return false;
641
642 /* The shift to msb will lose its cnt bits */
643 *ret = rb_time_val(top, bottom) | ((u64)msb << RB_TIME_MSB_SHIFT);
644 return true;
645}
646
647static bool rb_time_read(rb_time_t *t, u64 *ret)
648{
649 unsigned long cnt;
650
651 return __rb_time_read(t, ret, &cnt);
652}
653
654static inline unsigned long rb_time_val_cnt(unsigned long val, unsigned long cnt)
655{
656 return (val & RB_TIME_VAL_MASK) | ((cnt & 3) << RB_TIME_SHIFT);
657}
658
659static inline void rb_time_split(u64 val, unsigned long *top, unsigned long *bottom,
660 unsigned long *msb)
661{
662 *top = (unsigned long)((val >> RB_TIME_SHIFT) & RB_TIME_VAL_MASK);
663 *bottom = (unsigned long)(val & RB_TIME_VAL_MASK);
664 *msb = (unsigned long)(val >> RB_TIME_MSB_SHIFT);
665}
666
667static inline void rb_time_val_set(local_t *t, unsigned long val, unsigned long cnt)
668{
669 val = rb_time_val_cnt(val, cnt);
670 local_set(t, val);
671}
672
673static void rb_time_set(rb_time_t *t, u64 val)
674{
675 unsigned long cnt, top, bottom, msb;
676
677 rb_time_split(val, &top, &bottom, &msb);
678
679 /* Writes always succeed with a valid number even if it gets interrupted. */
680 do {
681 cnt = local_inc_return(&t->cnt);
682 rb_time_val_set(&t->top, top, cnt);
683 rb_time_val_set(&t->bottom, bottom, cnt);
684 rb_time_val_set(&t->msb, val >> RB_TIME_MSB_SHIFT, cnt);
685 } while (cnt != local_read(&t->cnt));
686}
687
688static inline bool
689rb_time_read_cmpxchg(local_t *l, unsigned long expect, unsigned long set)
690{
691 unsigned long ret;
692
693 ret = local_cmpxchg(l, expect, set);
694 return ret == expect;
695}
696
697static bool rb_time_cmpxchg(rb_time_t *t, u64 expect, u64 set)
698{
699 unsigned long cnt, top, bottom, msb;
700 unsigned long cnt2, top2, bottom2, msb2;
701 u64 val;
702
703 /* The cmpxchg always fails if it interrupted an update */
704 if (!__rb_time_read(t, &val, &cnt2))
705 return false;
706
707 if (val != expect)
708 return false;
709
710 cnt = local_read(&t->cnt);
711 if ((cnt & 3) != cnt2)
712 return false;
713
714 cnt2 = cnt + 1;
715
716 rb_time_split(val, &top, &bottom, &msb);
717 top = rb_time_val_cnt(top, cnt);
718 bottom = rb_time_val_cnt(bottom, cnt);
719
720 rb_time_split(set, &top2, &bottom2, &msb2);
721 top2 = rb_time_val_cnt(top2, cnt2);
722 bottom2 = rb_time_val_cnt(bottom2, cnt2);
723
724 if (!rb_time_read_cmpxchg(&t->cnt, cnt, cnt2))
725 return false;
726 if (!rb_time_read_cmpxchg(&t->msb, msb, msb2))
727 return false;
728 if (!rb_time_read_cmpxchg(&t->top, top, top2))
729 return false;
730 if (!rb_time_read_cmpxchg(&t->bottom, bottom, bottom2))
731 return false;
732 return true;
733}
734
735#else /* 64 bits */
736
737/* local64_t always succeeds */
738
739static inline bool rb_time_read(rb_time_t *t, u64 *ret)
740{
741 *ret = local64_read(&t->time);
742 return true;
743}
744static void rb_time_set(rb_time_t *t, u64 val)
745{
746 local64_set(&t->time, val);
747}
748
749static bool rb_time_cmpxchg(rb_time_t *t, u64 expect, u64 set)
750{
751 u64 val;
752 val = local64_cmpxchg(&t->time, expect, set);
753 return val == expect;
754}
755#endif
756
757/*
758 * Enable this to make sure that the event passed to
759 * ring_buffer_event_time_stamp() is not committed and also
760 * is on the buffer that it passed in.
761 */
762//#define RB_VERIFY_EVENT
763#ifdef RB_VERIFY_EVENT
764static struct list_head *rb_list_head(struct list_head *list);
765static void verify_event(struct ring_buffer_per_cpu *cpu_buffer,
766 void *event)
767{
768 struct buffer_page *page = cpu_buffer->commit_page;
769 struct buffer_page *tail_page = READ_ONCE(cpu_buffer->tail_page);
770 struct list_head *next;
771 long commit, write;
772 unsigned long addr = (unsigned long)event;
773 bool done = false;
774 int stop = 0;
775
776 /* Make sure the event exists and is not committed yet */
777 do {
778 if (page == tail_page || WARN_ON_ONCE(stop++ > 100))
779 done = true;
780 commit = local_read(&page->page->commit);
781 write = local_read(&page->write);
782 if (addr >= (unsigned long)&page->page->data[commit] &&
783 addr < (unsigned long)&page->page->data[write])
784 return;
785
786 next = rb_list_head(page->list.next);
787 page = list_entry(next, struct buffer_page, list);
788 } while (!done);
789 WARN_ON_ONCE(1);
790}
791#else
792static inline void verify_event(struct ring_buffer_per_cpu *cpu_buffer,
793 void *event)
794{
795}
796#endif
797
798/*
799 * The absolute time stamp drops the 5 MSBs and some clocks may
800 * require them. The rb_fix_abs_ts() will take a previous full
801 * time stamp, and add the 5 MSB of that time stamp on to the
802 * saved absolute time stamp. Then they are compared in case of
803 * the unlikely event that the latest time stamp incremented
804 * the 5 MSB.
805 */
806static inline u64 rb_fix_abs_ts(u64 abs, u64 save_ts)
807{
808 if (save_ts & TS_MSB) {
809 abs |= save_ts & TS_MSB;
810 /* Check for overflow */
811 if (unlikely(abs < save_ts))
812 abs += 1ULL << 59;
813 }
814 return abs;
815}
816
817static inline u64 rb_time_stamp(struct trace_buffer *buffer);
818
819/**
820 * ring_buffer_event_time_stamp - return the event's current time stamp
821 * @buffer: The buffer that the event is on
822 * @event: the event to get the time stamp of
823 *
824 * Note, this must be called after @event is reserved, and before it is
825 * committed to the ring buffer. And must be called from the same
826 * context where the event was reserved (normal, softirq, irq, etc).
827 *
828 * Returns the time stamp associated with the current event.
829 * If the event has an extended time stamp, then that is used as
830 * the time stamp to return.
831 * In the highly unlikely case that the event was nested more than
832 * the max nesting, then the write_stamp of the buffer is returned,
833 * otherwise current time is returned, but that really neither of
834 * the last two cases should ever happen.
835 */
836u64 ring_buffer_event_time_stamp(struct trace_buffer *buffer,
837 struct ring_buffer_event *event)
838{
839 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[smp_processor_id()];
840 unsigned int nest;
841 u64 ts;
842
843 /* If the event includes an absolute time, then just use that */
844 if (event->type_len == RINGBUF_TYPE_TIME_STAMP) {
845 ts = rb_event_time_stamp(event);
846 return rb_fix_abs_ts(ts, cpu_buffer->tail_page->page->time_stamp);
847 }
848
849 nest = local_read(&cpu_buffer->committing);
850 verify_event(cpu_buffer, event);
851 if (WARN_ON_ONCE(!nest))
852 goto fail;
853
854 /* Read the current saved nesting level time stamp */
855 if (likely(--nest < MAX_NEST))
856 return cpu_buffer->event_stamp[nest];
857
858 /* Shouldn't happen, warn if it does */
859 WARN_ONCE(1, "nest (%d) greater than max", nest);
860
861 fail:
862 /* Can only fail on 32 bit */
863 if (!rb_time_read(&cpu_buffer->write_stamp, &ts))
864 /* Screw it, just read the current time */
865 ts = rb_time_stamp(cpu_buffer->buffer);
866
867 return ts;
868}
869
870/**
871 * ring_buffer_nr_pages - get the number of buffer pages in the ring buffer
872 * @buffer: The ring_buffer to get the number of pages from
873 * @cpu: The cpu of the ring_buffer to get the number of pages from
874 *
875 * Returns the number of pages used by a per_cpu buffer of the ring buffer.
876 */
877size_t ring_buffer_nr_pages(struct trace_buffer *buffer, int cpu)
878{
879 return buffer->buffers[cpu]->nr_pages;
880}
881
882/**
883 * ring_buffer_nr_dirty_pages - get the number of used pages in the ring buffer
884 * @buffer: The ring_buffer to get the number of pages from
885 * @cpu: The cpu of the ring_buffer to get the number of pages from
886 *
887 * Returns the number of pages that have content in the ring buffer.
888 */
889size_t ring_buffer_nr_dirty_pages(struct trace_buffer *buffer, int cpu)
890{
891 size_t read;
892 size_t lost;
893 size_t cnt;
894
895 read = local_read(&buffer->buffers[cpu]->pages_read);
896 lost = local_read(&buffer->buffers[cpu]->pages_lost);
897 cnt = local_read(&buffer->buffers[cpu]->pages_touched);
898
899 if (WARN_ON_ONCE(cnt < lost))
900 return 0;
901
902 cnt -= lost;
903
904 /* The reader can read an empty page, but not more than that */
905 if (cnt < read) {
906 WARN_ON_ONCE(read > cnt + 1);
907 return 0;
908 }
909
910 return cnt - read;
911}
912
913static __always_inline bool full_hit(struct trace_buffer *buffer, int cpu, int full)
914{
915 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
916 size_t nr_pages;
917 size_t dirty;
918
919 nr_pages = cpu_buffer->nr_pages;
920 if (!nr_pages || !full)
921 return true;
922
923 dirty = ring_buffer_nr_dirty_pages(buffer, cpu);
924
925 return (dirty * 100) > (full * nr_pages);
926}
927
928/*
929 * rb_wake_up_waiters - wake up tasks waiting for ring buffer input
930 *
931 * Schedules a delayed work to wake up any task that is blocked on the
932 * ring buffer waiters queue.
933 */
934static void rb_wake_up_waiters(struct irq_work *work)
935{
936 struct rb_irq_work *rbwork = container_of(work, struct rb_irq_work, work);
937
938 wake_up_all(&rbwork->waiters);
939 if (rbwork->full_waiters_pending || rbwork->wakeup_full) {
940 rbwork->wakeup_full = false;
941 rbwork->full_waiters_pending = false;
942 wake_up_all(&rbwork->full_waiters);
943 }
944}
945
946/**
947 * ring_buffer_wake_waiters - wake up any waiters on this ring buffer
948 * @buffer: The ring buffer to wake waiters on
949 *
950 * In the case of a file that represents a ring buffer is closing,
951 * it is prudent to wake up any waiters that are on this.
952 */
953void ring_buffer_wake_waiters(struct trace_buffer *buffer, int cpu)
954{
955 struct ring_buffer_per_cpu *cpu_buffer;
956 struct rb_irq_work *rbwork;
957
958 if (!buffer)
959 return;
960
961 if (cpu == RING_BUFFER_ALL_CPUS) {
962
963 /* Wake up individual ones too. One level recursion */
964 for_each_buffer_cpu(buffer, cpu)
965 ring_buffer_wake_waiters(buffer, cpu);
966
967 rbwork = &buffer->irq_work;
968 } else {
969 if (WARN_ON_ONCE(!buffer->buffers))
970 return;
971 if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
972 return;
973
974 cpu_buffer = buffer->buffers[cpu];
975 /* The CPU buffer may not have been initialized yet */
976 if (!cpu_buffer)
977 return;
978 rbwork = &cpu_buffer->irq_work;
979 }
980
981 rbwork->wait_index++;
982 /* make sure the waiters see the new index */
983 smp_wmb();
984
985 rb_wake_up_waiters(&rbwork->work);
986}
987
988/**
989 * ring_buffer_wait - wait for input to the ring buffer
990 * @buffer: buffer to wait on
991 * @cpu: the cpu buffer to wait on
992 * @full: wait until the percentage of pages are available, if @cpu != RING_BUFFER_ALL_CPUS
993 *
994 * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
995 * as data is added to any of the @buffer's cpu buffers. Otherwise
996 * it will wait for data to be added to a specific cpu buffer.
997 */
998int ring_buffer_wait(struct trace_buffer *buffer, int cpu, int full)
999{
1000 struct ring_buffer_per_cpu *cpu_buffer;
1001 DEFINE_WAIT(wait);
1002 struct rb_irq_work *work;
1003 long wait_index;
1004 int ret = 0;
1005
1006 /*
1007 * Depending on what the caller is waiting for, either any
1008 * data in any cpu buffer, or a specific buffer, put the
1009 * caller on the appropriate wait queue.
1010 */
1011 if (cpu == RING_BUFFER_ALL_CPUS) {
1012 work = &buffer->irq_work;
1013 /* Full only makes sense on per cpu reads */
1014 full = 0;
1015 } else {
1016 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1017 return -ENODEV;
1018 cpu_buffer = buffer->buffers[cpu];
1019 work = &cpu_buffer->irq_work;
1020 }
1021
1022 wait_index = READ_ONCE(work->wait_index);
1023
1024 while (true) {
1025 if (full)
1026 prepare_to_wait(&work->full_waiters, &wait, TASK_INTERRUPTIBLE);
1027 else
1028 prepare_to_wait(&work->waiters, &wait, TASK_INTERRUPTIBLE);
1029
1030 /*
1031 * The events can happen in critical sections where
1032 * checking a work queue can cause deadlocks.
1033 * After adding a task to the queue, this flag is set
1034 * only to notify events to try to wake up the queue
1035 * using irq_work.
1036 *
1037 * We don't clear it even if the buffer is no longer
1038 * empty. The flag only causes the next event to run
1039 * irq_work to do the work queue wake up. The worse
1040 * that can happen if we race with !trace_empty() is that
1041 * an event will cause an irq_work to try to wake up
1042 * an empty queue.
1043 *
1044 * There's no reason to protect this flag either, as
1045 * the work queue and irq_work logic will do the necessary
1046 * synchronization for the wake ups. The only thing
1047 * that is necessary is that the wake up happens after
1048 * a task has been queued. It's OK for spurious wake ups.
1049 */
1050 if (full)
1051 work->full_waiters_pending = true;
1052 else
1053 work->waiters_pending = true;
1054
1055 if (signal_pending(current)) {
1056 ret = -EINTR;
1057 break;
1058 }
1059
1060 if (cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer))
1061 break;
1062
1063 if (cpu != RING_BUFFER_ALL_CPUS &&
1064 !ring_buffer_empty_cpu(buffer, cpu)) {
1065 unsigned long flags;
1066 bool pagebusy;
1067 bool done;
1068
1069 if (!full)
1070 break;
1071
1072 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1073 pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page;
1074 done = !pagebusy && full_hit(buffer, cpu, full);
1075
1076 if (!cpu_buffer->shortest_full ||
1077 cpu_buffer->shortest_full > full)
1078 cpu_buffer->shortest_full = full;
1079 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1080 if (done)
1081 break;
1082 }
1083
1084 schedule();
1085
1086 /* Make sure to see the new wait index */
1087 smp_rmb();
1088 if (wait_index != work->wait_index)
1089 break;
1090 }
1091
1092 if (full)
1093 finish_wait(&work->full_waiters, &wait);
1094 else
1095 finish_wait(&work->waiters, &wait);
1096
1097 return ret;
1098}
1099
1100/**
1101 * ring_buffer_poll_wait - poll on buffer input
1102 * @buffer: buffer to wait on
1103 * @cpu: the cpu buffer to wait on
1104 * @filp: the file descriptor
1105 * @poll_table: The poll descriptor
1106 * @full: wait until the percentage of pages are available, if @cpu != RING_BUFFER_ALL_CPUS
1107 *
1108 * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
1109 * as data is added to any of the @buffer's cpu buffers. Otherwise
1110 * it will wait for data to be added to a specific cpu buffer.
1111 *
1112 * Returns EPOLLIN | EPOLLRDNORM if data exists in the buffers,
1113 * zero otherwise.
1114 */
1115__poll_t ring_buffer_poll_wait(struct trace_buffer *buffer, int cpu,
1116 struct file *filp, poll_table *poll_table, int full)
1117{
1118 struct ring_buffer_per_cpu *cpu_buffer;
1119 struct rb_irq_work *work;
1120
1121 if (cpu == RING_BUFFER_ALL_CPUS) {
1122 work = &buffer->irq_work;
1123 full = 0;
1124 } else {
1125 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1126 return -EINVAL;
1127
1128 cpu_buffer = buffer->buffers[cpu];
1129 work = &cpu_buffer->irq_work;
1130 }
1131
1132 if (full) {
1133 poll_wait(filp, &work->full_waiters, poll_table);
1134 work->full_waiters_pending = true;
1135 } else {
1136 poll_wait(filp, &work->waiters, poll_table);
1137 work->waiters_pending = true;
1138 }
1139
1140 /*
1141 * There's a tight race between setting the waiters_pending and
1142 * checking if the ring buffer is empty. Once the waiters_pending bit
1143 * is set, the next event will wake the task up, but we can get stuck
1144 * if there's only a single event in.
1145 *
1146 * FIXME: Ideally, we need a memory barrier on the writer side as well,
1147 * but adding a memory barrier to all events will cause too much of a
1148 * performance hit in the fast path. We only need a memory barrier when
1149 * the buffer goes from empty to having content. But as this race is
1150 * extremely small, and it's not a problem if another event comes in, we
1151 * will fix it later.
1152 */
1153 smp_mb();
1154
1155 if (full)
1156 return full_hit(buffer, cpu, full) ? EPOLLIN | EPOLLRDNORM : 0;
1157
1158 if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) ||
1159 (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu)))
1160 return EPOLLIN | EPOLLRDNORM;
1161 return 0;
1162}
1163
1164/* buffer may be either ring_buffer or ring_buffer_per_cpu */
1165#define RB_WARN_ON(b, cond) \
1166 ({ \
1167 int _____ret = unlikely(cond); \
1168 if (_____ret) { \
1169 if (__same_type(*(b), struct ring_buffer_per_cpu)) { \
1170 struct ring_buffer_per_cpu *__b = \
1171 (void *)b; \
1172 atomic_inc(&__b->buffer->record_disabled); \
1173 } else \
1174 atomic_inc(&b->record_disabled); \
1175 WARN_ON(1); \
1176 } \
1177 _____ret; \
1178 })
1179
1180/* Up this if you want to test the TIME_EXTENTS and normalization */
1181#define DEBUG_SHIFT 0
1182
1183static inline u64 rb_time_stamp(struct trace_buffer *buffer)
1184{
1185 u64 ts;
1186
1187 /* Skip retpolines :-( */
1188 if (IS_ENABLED(CONFIG_RETPOLINE) && likely(buffer->clock == trace_clock_local))
1189 ts = trace_clock_local();
1190 else
1191 ts = buffer->clock();
1192
1193 /* shift to debug/test normalization and TIME_EXTENTS */
1194 return ts << DEBUG_SHIFT;
1195}
1196
1197u64 ring_buffer_time_stamp(struct trace_buffer *buffer)
1198{
1199 u64 time;
1200
1201 preempt_disable_notrace();
1202 time = rb_time_stamp(buffer);
1203 preempt_enable_notrace();
1204
1205 return time;
1206}
1207EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
1208
1209void ring_buffer_normalize_time_stamp(struct trace_buffer *buffer,
1210 int cpu, u64 *ts)
1211{
1212 /* Just stupid testing the normalize function and deltas */
1213 *ts >>= DEBUG_SHIFT;
1214}
1215EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
1216
1217/*
1218 * Making the ring buffer lockless makes things tricky.
1219 * Although writes only happen on the CPU that they are on,
1220 * and they only need to worry about interrupts. Reads can
1221 * happen on any CPU.
1222 *
1223 * The reader page is always off the ring buffer, but when the
1224 * reader finishes with a page, it needs to swap its page with
1225 * a new one from the buffer. The reader needs to take from
1226 * the head (writes go to the tail). But if a writer is in overwrite
1227 * mode and wraps, it must push the head page forward.
1228 *
1229 * Here lies the problem.
1230 *
1231 * The reader must be careful to replace only the head page, and
1232 * not another one. As described at the top of the file in the
1233 * ASCII art, the reader sets its old page to point to the next
1234 * page after head. It then sets the page after head to point to
1235 * the old reader page. But if the writer moves the head page
1236 * during this operation, the reader could end up with the tail.
1237 *
1238 * We use cmpxchg to help prevent this race. We also do something
1239 * special with the page before head. We set the LSB to 1.
1240 *
1241 * When the writer must push the page forward, it will clear the
1242 * bit that points to the head page, move the head, and then set
1243 * the bit that points to the new head page.
1244 *
1245 * We also don't want an interrupt coming in and moving the head
1246 * page on another writer. Thus we use the second LSB to catch
1247 * that too. Thus:
1248 *
1249 * head->list->prev->next bit 1 bit 0
1250 * ------- -------
1251 * Normal page 0 0
1252 * Points to head page 0 1
1253 * New head page 1 0
1254 *
1255 * Note we can not trust the prev pointer of the head page, because:
1256 *
1257 * +----+ +-----+ +-----+
1258 * | |------>| T |---X--->| N |
1259 * | |<------| | | |
1260 * +----+ +-----+ +-----+
1261 * ^ ^ |
1262 * | +-----+ | |
1263 * +----------| R |----------+ |
1264 * | |<-----------+
1265 * +-----+
1266 *
1267 * Key: ---X--> HEAD flag set in pointer
1268 * T Tail page
1269 * R Reader page
1270 * N Next page
1271 *
1272 * (see __rb_reserve_next() to see where this happens)
1273 *
1274 * What the above shows is that the reader just swapped out
1275 * the reader page with a page in the buffer, but before it
1276 * could make the new header point back to the new page added
1277 * it was preempted by a writer. The writer moved forward onto
1278 * the new page added by the reader and is about to move forward
1279 * again.
1280 *
1281 * You can see, it is legitimate for the previous pointer of
1282 * the head (or any page) not to point back to itself. But only
1283 * temporarily.
1284 */
1285
1286#define RB_PAGE_NORMAL 0UL
1287#define RB_PAGE_HEAD 1UL
1288#define RB_PAGE_UPDATE 2UL
1289
1290
1291#define RB_FLAG_MASK 3UL
1292
1293/* PAGE_MOVED is not part of the mask */
1294#define RB_PAGE_MOVED 4UL
1295
1296/*
1297 * rb_list_head - remove any bit
1298 */
1299static struct list_head *rb_list_head(struct list_head *list)
1300{
1301 unsigned long val = (unsigned long)list;
1302
1303 return (struct list_head *)(val & ~RB_FLAG_MASK);
1304}
1305
1306/*
1307 * rb_is_head_page - test if the given page is the head page
1308 *
1309 * Because the reader may move the head_page pointer, we can
1310 * not trust what the head page is (it may be pointing to
1311 * the reader page). But if the next page is a header page,
1312 * its flags will be non zero.
1313 */
1314static inline int
1315rb_is_head_page(struct buffer_page *page, struct list_head *list)
1316{
1317 unsigned long val;
1318
1319 val = (unsigned long)list->next;
1320
1321 if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list)
1322 return RB_PAGE_MOVED;
1323
1324 return val & RB_FLAG_MASK;
1325}
1326
1327/*
1328 * rb_is_reader_page
1329 *
1330 * The unique thing about the reader page, is that, if the
1331 * writer is ever on it, the previous pointer never points
1332 * back to the reader page.
1333 */
1334static bool rb_is_reader_page(struct buffer_page *page)
1335{
1336 struct list_head *list = page->list.prev;
1337
1338 return rb_list_head(list->next) != &page->list;
1339}
1340
1341/*
1342 * rb_set_list_to_head - set a list_head to be pointing to head.
1343 */
1344static void rb_set_list_to_head(struct list_head *list)
1345{
1346 unsigned long *ptr;
1347
1348 ptr = (unsigned long *)&list->next;
1349 *ptr |= RB_PAGE_HEAD;
1350 *ptr &= ~RB_PAGE_UPDATE;
1351}
1352
1353/*
1354 * rb_head_page_activate - sets up head page
1355 */
1356static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer)
1357{
1358 struct buffer_page *head;
1359
1360 head = cpu_buffer->head_page;
1361 if (!head)
1362 return;
1363
1364 /*
1365 * Set the previous list pointer to have the HEAD flag.
1366 */
1367 rb_set_list_to_head(head->list.prev);
1368}
1369
1370static void rb_list_head_clear(struct list_head *list)
1371{
1372 unsigned long *ptr = (unsigned long *)&list->next;
1373
1374 *ptr &= ~RB_FLAG_MASK;
1375}
1376
1377/*
1378 * rb_head_page_deactivate - clears head page ptr (for free list)
1379 */
1380static void
1381rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer)
1382{
1383 struct list_head *hd;
1384
1385 /* Go through the whole list and clear any pointers found. */
1386 rb_list_head_clear(cpu_buffer->pages);
1387
1388 list_for_each(hd, cpu_buffer->pages)
1389 rb_list_head_clear(hd);
1390}
1391
1392static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer,
1393 struct buffer_page *head,
1394 struct buffer_page *prev,
1395 int old_flag, int new_flag)
1396{
1397 struct list_head *list;
1398 unsigned long val = (unsigned long)&head->list;
1399 unsigned long ret;
1400
1401 list = &prev->list;
1402
1403 val &= ~RB_FLAG_MASK;
1404
1405 ret = cmpxchg((unsigned long *)&list->next,
1406 val | old_flag, val | new_flag);
1407
1408 /* check if the reader took the page */
1409 if ((ret & ~RB_FLAG_MASK) != val)
1410 return RB_PAGE_MOVED;
1411
1412 return ret & RB_FLAG_MASK;
1413}
1414
1415static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer,
1416 struct buffer_page *head,
1417 struct buffer_page *prev,
1418 int old_flag)
1419{
1420 return rb_head_page_set(cpu_buffer, head, prev,
1421 old_flag, RB_PAGE_UPDATE);
1422}
1423
1424static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer,
1425 struct buffer_page *head,
1426 struct buffer_page *prev,
1427 int old_flag)
1428{
1429 return rb_head_page_set(cpu_buffer, head, prev,
1430 old_flag, RB_PAGE_HEAD);
1431}
1432
1433static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer,
1434 struct buffer_page *head,
1435 struct buffer_page *prev,
1436 int old_flag)
1437{
1438 return rb_head_page_set(cpu_buffer, head, prev,
1439 old_flag, RB_PAGE_NORMAL);
1440}
1441
1442static inline void rb_inc_page(struct buffer_page **bpage)
1443{
1444 struct list_head *p = rb_list_head((*bpage)->list.next);
1445
1446 *bpage = list_entry(p, struct buffer_page, list);
1447}
1448
1449static struct buffer_page *
1450rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer)
1451{
1452 struct buffer_page *head;
1453 struct buffer_page *page;
1454 struct list_head *list;
1455 int i;
1456
1457 if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page))
1458 return NULL;
1459
1460 /* sanity check */
1461 list = cpu_buffer->pages;
1462 if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list))
1463 return NULL;
1464
1465 page = head = cpu_buffer->head_page;
1466 /*
1467 * It is possible that the writer moves the header behind
1468 * where we started, and we miss in one loop.
1469 * A second loop should grab the header, but we'll do
1470 * three loops just because I'm paranoid.
1471 */
1472 for (i = 0; i < 3; i++) {
1473 do {
1474 if (rb_is_head_page(page, page->list.prev)) {
1475 cpu_buffer->head_page = page;
1476 return page;
1477 }
1478 rb_inc_page(&page);
1479 } while (page != head);
1480 }
1481
1482 RB_WARN_ON(cpu_buffer, 1);
1483
1484 return NULL;
1485}
1486
1487static bool rb_head_page_replace(struct buffer_page *old,
1488 struct buffer_page *new)
1489{
1490 unsigned long *ptr = (unsigned long *)&old->list.prev->next;
1491 unsigned long val;
1492 unsigned long ret;
1493
1494 val = *ptr & ~RB_FLAG_MASK;
1495 val |= RB_PAGE_HEAD;
1496
1497 ret = cmpxchg(ptr, val, (unsigned long)&new->list);
1498
1499 return ret == val;
1500}
1501
1502/*
1503 * rb_tail_page_update - move the tail page forward
1504 */
1505static void rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer,
1506 struct buffer_page *tail_page,
1507 struct buffer_page *next_page)
1508{
1509 unsigned long old_entries;
1510 unsigned long old_write;
1511
1512 /*
1513 * The tail page now needs to be moved forward.
1514 *
1515 * We need to reset the tail page, but without messing
1516 * with possible erasing of data brought in by interrupts
1517 * that have moved the tail page and are currently on it.
1518 *
1519 * We add a counter to the write field to denote this.
1520 */
1521 old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write);
1522 old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries);
1523
1524 local_inc(&cpu_buffer->pages_touched);
1525 /*
1526 * Just make sure we have seen our old_write and synchronize
1527 * with any interrupts that come in.
1528 */
1529 barrier();
1530
1531 /*
1532 * If the tail page is still the same as what we think
1533 * it is, then it is up to us to update the tail
1534 * pointer.
1535 */
1536 if (tail_page == READ_ONCE(cpu_buffer->tail_page)) {
1537 /* Zero the write counter */
1538 unsigned long val = old_write & ~RB_WRITE_MASK;
1539 unsigned long eval = old_entries & ~RB_WRITE_MASK;
1540
1541 /*
1542 * This will only succeed if an interrupt did
1543 * not come in and change it. In which case, we
1544 * do not want to modify it.
1545 *
1546 * We add (void) to let the compiler know that we do not care
1547 * about the return value of these functions. We use the
1548 * cmpxchg to only update if an interrupt did not already
1549 * do it for us. If the cmpxchg fails, we don't care.
1550 */
1551 (void)local_cmpxchg(&next_page->write, old_write, val);
1552 (void)local_cmpxchg(&next_page->entries, old_entries, eval);
1553
1554 /*
1555 * No need to worry about races with clearing out the commit.
1556 * it only can increment when a commit takes place. But that
1557 * only happens in the outer most nested commit.
1558 */
1559 local_set(&next_page->page->commit, 0);
1560
1561 /* Again, either we update tail_page or an interrupt does */
1562 (void)cmpxchg(&cpu_buffer->tail_page, tail_page, next_page);
1563 }
1564}
1565
1566static void rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer,
1567 struct buffer_page *bpage)
1568{
1569 unsigned long val = (unsigned long)bpage;
1570
1571 RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK);
1572}
1573
1574/**
1575 * rb_check_pages - integrity check of buffer pages
1576 * @cpu_buffer: CPU buffer with pages to test
1577 *
1578 * As a safety measure we check to make sure the data pages have not
1579 * been corrupted.
1580 */
1581static void rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
1582{
1583 struct list_head *head = rb_list_head(cpu_buffer->pages);
1584 struct list_head *tmp;
1585
1586 if (RB_WARN_ON(cpu_buffer,
1587 rb_list_head(rb_list_head(head->next)->prev) != head))
1588 return;
1589
1590 if (RB_WARN_ON(cpu_buffer,
1591 rb_list_head(rb_list_head(head->prev)->next) != head))
1592 return;
1593
1594 for (tmp = rb_list_head(head->next); tmp != head; tmp = rb_list_head(tmp->next)) {
1595 if (RB_WARN_ON(cpu_buffer,
1596 rb_list_head(rb_list_head(tmp->next)->prev) != tmp))
1597 return;
1598
1599 if (RB_WARN_ON(cpu_buffer,
1600 rb_list_head(rb_list_head(tmp->prev)->next) != tmp))
1601 return;
1602 }
1603}
1604
1605static int __rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
1606 long nr_pages, struct list_head *pages)
1607{
1608 struct buffer_page *bpage, *tmp;
1609 bool user_thread = current->mm != NULL;
1610 gfp_t mflags;
1611 long i;
1612
1613 /*
1614 * Check if the available memory is there first.
1615 * Note, si_mem_available() only gives us a rough estimate of available
1616 * memory. It may not be accurate. But we don't care, we just want
1617 * to prevent doing any allocation when it is obvious that it is
1618 * not going to succeed.
1619 */
1620 i = si_mem_available();
1621 if (i < nr_pages)
1622 return -ENOMEM;
1623
1624 /*
1625 * __GFP_RETRY_MAYFAIL flag makes sure that the allocation fails
1626 * gracefully without invoking oom-killer and the system is not
1627 * destabilized.
1628 */
1629 mflags = GFP_KERNEL | __GFP_RETRY_MAYFAIL;
1630
1631 /*
1632 * If a user thread allocates too much, and si_mem_available()
1633 * reports there's enough memory, even though there is not.
1634 * Make sure the OOM killer kills this thread. This can happen
1635 * even with RETRY_MAYFAIL because another task may be doing
1636 * an allocation after this task has taken all memory.
1637 * This is the task the OOM killer needs to take out during this
1638 * loop, even if it was triggered by an allocation somewhere else.
1639 */
1640 if (user_thread)
1641 set_current_oom_origin();
1642 for (i = 0; i < nr_pages; i++) {
1643 struct page *page;
1644
1645 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
1646 mflags, cpu_to_node(cpu_buffer->cpu));
1647 if (!bpage)
1648 goto free_pages;
1649
1650 rb_check_bpage(cpu_buffer, bpage);
1651
1652 list_add(&bpage->list, pages);
1653
1654 page = alloc_pages_node(cpu_to_node(cpu_buffer->cpu), mflags, 0);
1655 if (!page)
1656 goto free_pages;
1657 bpage->page = page_address(page);
1658 rb_init_page(bpage->page);
1659
1660 if (user_thread && fatal_signal_pending(current))
1661 goto free_pages;
1662 }
1663 if (user_thread)
1664 clear_current_oom_origin();
1665
1666 return 0;
1667
1668free_pages:
1669 list_for_each_entry_safe(bpage, tmp, pages, list) {
1670 list_del_init(&bpage->list);
1671 free_buffer_page(bpage);
1672 }
1673 if (user_thread)
1674 clear_current_oom_origin();
1675
1676 return -ENOMEM;
1677}
1678
1679static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
1680 unsigned long nr_pages)
1681{
1682 LIST_HEAD(pages);
1683
1684 WARN_ON(!nr_pages);
1685
1686 if (__rb_allocate_pages(cpu_buffer, nr_pages, &pages))
1687 return -ENOMEM;
1688
1689 /*
1690 * The ring buffer page list is a circular list that does not
1691 * start and end with a list head. All page list items point to
1692 * other pages.
1693 */
1694 cpu_buffer->pages = pages.next;
1695 list_del(&pages);
1696
1697 cpu_buffer->nr_pages = nr_pages;
1698
1699 rb_check_pages(cpu_buffer);
1700
1701 return 0;
1702}
1703
1704static struct ring_buffer_per_cpu *
1705rb_allocate_cpu_buffer(struct trace_buffer *buffer, long nr_pages, int cpu)
1706{
1707 struct ring_buffer_per_cpu *cpu_buffer;
1708 struct buffer_page *bpage;
1709 struct page *page;
1710 int ret;
1711
1712 cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
1713 GFP_KERNEL, cpu_to_node(cpu));
1714 if (!cpu_buffer)
1715 return NULL;
1716
1717 cpu_buffer->cpu = cpu;
1718 cpu_buffer->buffer = buffer;
1719 raw_spin_lock_init(&cpu_buffer->reader_lock);
1720 lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key);
1721 cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1722 INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler);
1723 init_completion(&cpu_buffer->update_done);
1724 init_irq_work(&cpu_buffer->irq_work.work, rb_wake_up_waiters);
1725 init_waitqueue_head(&cpu_buffer->irq_work.waiters);
1726 init_waitqueue_head(&cpu_buffer->irq_work.full_waiters);
1727
1728 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
1729 GFP_KERNEL, cpu_to_node(cpu));
1730 if (!bpage)
1731 goto fail_free_buffer;
1732
1733 rb_check_bpage(cpu_buffer, bpage);
1734
1735 cpu_buffer->reader_page = bpage;
1736 page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0);
1737 if (!page)
1738 goto fail_free_reader;
1739 bpage->page = page_address(page);
1740 rb_init_page(bpage->page);
1741
1742 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
1743 INIT_LIST_HEAD(&cpu_buffer->new_pages);
1744
1745 ret = rb_allocate_pages(cpu_buffer, nr_pages);
1746 if (ret < 0)
1747 goto fail_free_reader;
1748
1749 cpu_buffer->head_page
1750 = list_entry(cpu_buffer->pages, struct buffer_page, list);
1751 cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
1752
1753 rb_head_page_activate(cpu_buffer);
1754
1755 return cpu_buffer;
1756
1757 fail_free_reader:
1758 free_buffer_page(cpu_buffer->reader_page);
1759
1760 fail_free_buffer:
1761 kfree(cpu_buffer);
1762 return NULL;
1763}
1764
1765static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
1766{
1767 struct list_head *head = cpu_buffer->pages;
1768 struct buffer_page *bpage, *tmp;
1769
1770 irq_work_sync(&cpu_buffer->irq_work.work);
1771
1772 free_buffer_page(cpu_buffer->reader_page);
1773
1774 if (head) {
1775 rb_head_page_deactivate(cpu_buffer);
1776
1777 list_for_each_entry_safe(bpage, tmp, head, list) {
1778 list_del_init(&bpage->list);
1779 free_buffer_page(bpage);
1780 }
1781 bpage = list_entry(head, struct buffer_page, list);
1782 free_buffer_page(bpage);
1783 }
1784
1785 kfree(cpu_buffer);
1786}
1787
1788/**
1789 * __ring_buffer_alloc - allocate a new ring_buffer
1790 * @size: the size in bytes per cpu that is needed.
1791 * @flags: attributes to set for the ring buffer.
1792 * @key: ring buffer reader_lock_key.
1793 *
1794 * Currently the only flag that is available is the RB_FL_OVERWRITE
1795 * flag. This flag means that the buffer will overwrite old data
1796 * when the buffer wraps. If this flag is not set, the buffer will
1797 * drop data when the tail hits the head.
1798 */
1799struct trace_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags,
1800 struct lock_class_key *key)
1801{
1802 struct trace_buffer *buffer;
1803 long nr_pages;
1804 int bsize;
1805 int cpu;
1806 int ret;
1807
1808 /* keep it in its own cache line */
1809 buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
1810 GFP_KERNEL);
1811 if (!buffer)
1812 return NULL;
1813
1814 if (!zalloc_cpumask_var(&buffer->cpumask, GFP_KERNEL))
1815 goto fail_free_buffer;
1816
1817 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
1818 buffer->flags = flags;
1819 buffer->clock = trace_clock_local;
1820 buffer->reader_lock_key = key;
1821
1822 init_irq_work(&buffer->irq_work.work, rb_wake_up_waiters);
1823 init_waitqueue_head(&buffer->irq_work.waiters);
1824
1825 /* need at least two pages */
1826 if (nr_pages < 2)
1827 nr_pages = 2;
1828
1829 buffer->cpus = nr_cpu_ids;
1830
1831 bsize = sizeof(void *) * nr_cpu_ids;
1832 buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
1833 GFP_KERNEL);
1834 if (!buffer->buffers)
1835 goto fail_free_cpumask;
1836
1837 cpu = raw_smp_processor_id();
1838 cpumask_set_cpu(cpu, buffer->cpumask);
1839 buffer->buffers[cpu] = rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
1840 if (!buffer->buffers[cpu])
1841 goto fail_free_buffers;
1842
1843 ret = cpuhp_state_add_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node);
1844 if (ret < 0)
1845 goto fail_free_buffers;
1846
1847 mutex_init(&buffer->mutex);
1848
1849 return buffer;
1850
1851 fail_free_buffers:
1852 for_each_buffer_cpu(buffer, cpu) {
1853 if (buffer->buffers[cpu])
1854 rb_free_cpu_buffer(buffer->buffers[cpu]);
1855 }
1856 kfree(buffer->buffers);
1857
1858 fail_free_cpumask:
1859 free_cpumask_var(buffer->cpumask);
1860
1861 fail_free_buffer:
1862 kfree(buffer);
1863 return NULL;
1864}
1865EXPORT_SYMBOL_GPL(__ring_buffer_alloc);
1866
1867/**
1868 * ring_buffer_free - free a ring buffer.
1869 * @buffer: the buffer to free.
1870 */
1871void
1872ring_buffer_free(struct trace_buffer *buffer)
1873{
1874 int cpu;
1875
1876 cpuhp_state_remove_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node);
1877
1878 irq_work_sync(&buffer->irq_work.work);
1879
1880 for_each_buffer_cpu(buffer, cpu)
1881 rb_free_cpu_buffer(buffer->buffers[cpu]);
1882
1883 kfree(buffer->buffers);
1884 free_cpumask_var(buffer->cpumask);
1885
1886 kfree(buffer);
1887}
1888EXPORT_SYMBOL_GPL(ring_buffer_free);
1889
1890void ring_buffer_set_clock(struct trace_buffer *buffer,
1891 u64 (*clock)(void))
1892{
1893 buffer->clock = clock;
1894}
1895
1896void ring_buffer_set_time_stamp_abs(struct trace_buffer *buffer, bool abs)
1897{
1898 buffer->time_stamp_abs = abs;
1899}
1900
1901bool ring_buffer_time_stamp_abs(struct trace_buffer *buffer)
1902{
1903 return buffer->time_stamp_abs;
1904}
1905
1906static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
1907
1908static inline unsigned long rb_page_entries(struct buffer_page *bpage)
1909{
1910 return local_read(&bpage->entries) & RB_WRITE_MASK;
1911}
1912
1913static inline unsigned long rb_page_write(struct buffer_page *bpage)
1914{
1915 return local_read(&bpage->write) & RB_WRITE_MASK;
1916}
1917
1918static bool
1919rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned long nr_pages)
1920{
1921 struct list_head *tail_page, *to_remove, *next_page;
1922 struct buffer_page *to_remove_page, *tmp_iter_page;
1923 struct buffer_page *last_page, *first_page;
1924 unsigned long nr_removed;
1925 unsigned long head_bit;
1926 int page_entries;
1927
1928 head_bit = 0;
1929
1930 raw_spin_lock_irq(&cpu_buffer->reader_lock);
1931 atomic_inc(&cpu_buffer->record_disabled);
1932 /*
1933 * We don't race with the readers since we have acquired the reader
1934 * lock. We also don't race with writers after disabling recording.
1935 * This makes it easy to figure out the first and the last page to be
1936 * removed from the list. We unlink all the pages in between including
1937 * the first and last pages. This is done in a busy loop so that we
1938 * lose the least number of traces.
1939 * The pages are freed after we restart recording and unlock readers.
1940 */
1941 tail_page = &cpu_buffer->tail_page->list;
1942
1943 /*
1944 * tail page might be on reader page, we remove the next page
1945 * from the ring buffer
1946 */
1947 if (cpu_buffer->tail_page == cpu_buffer->reader_page)
1948 tail_page = rb_list_head(tail_page->next);
1949 to_remove = tail_page;
1950
1951 /* start of pages to remove */
1952 first_page = list_entry(rb_list_head(to_remove->next),
1953 struct buffer_page, list);
1954
1955 for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) {
1956 to_remove = rb_list_head(to_remove)->next;
1957 head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD;
1958 }
1959
1960 next_page = rb_list_head(to_remove)->next;
1961
1962 /*
1963 * Now we remove all pages between tail_page and next_page.
1964 * Make sure that we have head_bit value preserved for the
1965 * next page
1966 */
1967 tail_page->next = (struct list_head *)((unsigned long)next_page |
1968 head_bit);
1969 next_page = rb_list_head(next_page);
1970 next_page->prev = tail_page;
1971
1972 /* make sure pages points to a valid page in the ring buffer */
1973 cpu_buffer->pages = next_page;
1974
1975 /* update head page */
1976 if (head_bit)
1977 cpu_buffer->head_page = list_entry(next_page,
1978 struct buffer_page, list);
1979
1980 /*
1981 * change read pointer to make sure any read iterators reset
1982 * themselves
1983 */
1984 cpu_buffer->read = 0;
1985
1986 /* pages are removed, resume tracing and then free the pages */
1987 atomic_dec(&cpu_buffer->record_disabled);
1988 raw_spin_unlock_irq(&cpu_buffer->reader_lock);
1989
1990 RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages));
1991
1992 /* last buffer page to remove */
1993 last_page = list_entry(rb_list_head(to_remove), struct buffer_page,
1994 list);
1995 tmp_iter_page = first_page;
1996
1997 do {
1998 cond_resched();
1999
2000 to_remove_page = tmp_iter_page;
2001 rb_inc_page(&tmp_iter_page);
2002
2003 /* update the counters */
2004 page_entries = rb_page_entries(to_remove_page);
2005 if (page_entries) {
2006 /*
2007 * If something was added to this page, it was full
2008 * since it is not the tail page. So we deduct the
2009 * bytes consumed in ring buffer from here.
2010 * Increment overrun to account for the lost events.
2011 */
2012 local_add(page_entries, &cpu_buffer->overrun);
2013 local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
2014 local_inc(&cpu_buffer->pages_lost);
2015 }
2016
2017 /*
2018 * We have already removed references to this list item, just
2019 * free up the buffer_page and its page
2020 */
2021 free_buffer_page(to_remove_page);
2022 nr_removed--;
2023
2024 } while (to_remove_page != last_page);
2025
2026 RB_WARN_ON(cpu_buffer, nr_removed);
2027
2028 return nr_removed == 0;
2029}
2030
2031static bool
2032rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer)
2033{
2034 struct list_head *pages = &cpu_buffer->new_pages;
2035 unsigned long flags;
2036 bool success;
2037 int retries;
2038
2039 /* Can be called at early boot up, where interrupts must not been enabled */
2040 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2041 /*
2042 * We are holding the reader lock, so the reader page won't be swapped
2043 * in the ring buffer. Now we are racing with the writer trying to
2044 * move head page and the tail page.
2045 * We are going to adapt the reader page update process where:
2046 * 1. We first splice the start and end of list of new pages between
2047 * the head page and its previous page.
2048 * 2. We cmpxchg the prev_page->next to point from head page to the
2049 * start of new pages list.
2050 * 3. Finally, we update the head->prev to the end of new list.
2051 *
2052 * We will try this process 10 times, to make sure that we don't keep
2053 * spinning.
2054 */
2055 retries = 10;
2056 success = false;
2057 while (retries--) {
2058 struct list_head *head_page, *prev_page, *r;
2059 struct list_head *last_page, *first_page;
2060 struct list_head *head_page_with_bit;
2061 struct buffer_page *hpage = rb_set_head_page(cpu_buffer);
2062
2063 if (!hpage)
2064 break;
2065 head_page = &hpage->list;
2066 prev_page = head_page->prev;
2067
2068 first_page = pages->next;
2069 last_page = pages->prev;
2070
2071 head_page_with_bit = (struct list_head *)
2072 ((unsigned long)head_page | RB_PAGE_HEAD);
2073
2074 last_page->next = head_page_with_bit;
2075 first_page->prev = prev_page;
2076
2077 r = cmpxchg(&prev_page->next, head_page_with_bit, first_page);
2078
2079 if (r == head_page_with_bit) {
2080 /*
2081 * yay, we replaced the page pointer to our new list,
2082 * now, we just have to update to head page's prev
2083 * pointer to point to end of list
2084 */
2085 head_page->prev = last_page;
2086 success = true;
2087 break;
2088 }
2089 }
2090
2091 if (success)
2092 INIT_LIST_HEAD(pages);
2093 /*
2094 * If we weren't successful in adding in new pages, warn and stop
2095 * tracing
2096 */
2097 RB_WARN_ON(cpu_buffer, !success);
2098 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2099
2100 /* free pages if they weren't inserted */
2101 if (!success) {
2102 struct buffer_page *bpage, *tmp;
2103 list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
2104 list) {
2105 list_del_init(&bpage->list);
2106 free_buffer_page(bpage);
2107 }
2108 }
2109 return success;
2110}
2111
2112static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer)
2113{
2114 bool success;
2115
2116 if (cpu_buffer->nr_pages_to_update > 0)
2117 success = rb_insert_pages(cpu_buffer);
2118 else
2119 success = rb_remove_pages(cpu_buffer,
2120 -cpu_buffer->nr_pages_to_update);
2121
2122 if (success)
2123 cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update;
2124}
2125
2126static void update_pages_handler(struct work_struct *work)
2127{
2128 struct ring_buffer_per_cpu *cpu_buffer = container_of(work,
2129 struct ring_buffer_per_cpu, update_pages_work);
2130 rb_update_pages(cpu_buffer);
2131 complete(&cpu_buffer->update_done);
2132}
2133
2134/**
2135 * ring_buffer_resize - resize the ring buffer
2136 * @buffer: the buffer to resize.
2137 * @size: the new size.
2138 * @cpu_id: the cpu buffer to resize
2139 *
2140 * Minimum size is 2 * BUF_PAGE_SIZE.
2141 *
2142 * Returns 0 on success and < 0 on failure.
2143 */
2144int ring_buffer_resize(struct trace_buffer *buffer, unsigned long size,
2145 int cpu_id)
2146{
2147 struct ring_buffer_per_cpu *cpu_buffer;
2148 unsigned long nr_pages;
2149 int cpu, err;
2150
2151 /*
2152 * Always succeed at resizing a non-existent buffer:
2153 */
2154 if (!buffer)
2155 return 0;
2156
2157 /* Make sure the requested buffer exists */
2158 if (cpu_id != RING_BUFFER_ALL_CPUS &&
2159 !cpumask_test_cpu(cpu_id, buffer->cpumask))
2160 return 0;
2161
2162 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
2163
2164 /* we need a minimum of two pages */
2165 if (nr_pages < 2)
2166 nr_pages = 2;
2167
2168 /* prevent another thread from changing buffer sizes */
2169 mutex_lock(&buffer->mutex);
2170
2171
2172 if (cpu_id == RING_BUFFER_ALL_CPUS) {
2173 /*
2174 * Don't succeed if resizing is disabled, as a reader might be
2175 * manipulating the ring buffer and is expecting a sane state while
2176 * this is true.
2177 */
2178 for_each_buffer_cpu(buffer, cpu) {
2179 cpu_buffer = buffer->buffers[cpu];
2180 if (atomic_read(&cpu_buffer->resize_disabled)) {
2181 err = -EBUSY;
2182 goto out_err_unlock;
2183 }
2184 }
2185
2186 /* calculate the pages to update */
2187 for_each_buffer_cpu(buffer, cpu) {
2188 cpu_buffer = buffer->buffers[cpu];
2189
2190 cpu_buffer->nr_pages_to_update = nr_pages -
2191 cpu_buffer->nr_pages;
2192 /*
2193 * nothing more to do for removing pages or no update
2194 */
2195 if (cpu_buffer->nr_pages_to_update <= 0)
2196 continue;
2197 /*
2198 * to add pages, make sure all new pages can be
2199 * allocated without receiving ENOMEM
2200 */
2201 INIT_LIST_HEAD(&cpu_buffer->new_pages);
2202 if (__rb_allocate_pages(cpu_buffer, cpu_buffer->nr_pages_to_update,
2203 &cpu_buffer->new_pages)) {
2204 /* not enough memory for new pages */
2205 err = -ENOMEM;
2206 goto out_err;
2207 }
2208 }
2209
2210 cpus_read_lock();
2211 /*
2212 * Fire off all the required work handlers
2213 * We can't schedule on offline CPUs, but it's not necessary
2214 * since we can change their buffer sizes without any race.
2215 */
2216 for_each_buffer_cpu(buffer, cpu) {
2217 cpu_buffer = buffer->buffers[cpu];
2218 if (!cpu_buffer->nr_pages_to_update)
2219 continue;
2220
2221 /* Can't run something on an offline CPU. */
2222 if (!cpu_online(cpu)) {
2223 rb_update_pages(cpu_buffer);
2224 cpu_buffer->nr_pages_to_update = 0;
2225 } else {
2226 /* Run directly if possible. */
2227 migrate_disable();
2228 if (cpu != smp_processor_id()) {
2229 migrate_enable();
2230 schedule_work_on(cpu,
2231 &cpu_buffer->update_pages_work);
2232 } else {
2233 update_pages_handler(&cpu_buffer->update_pages_work);
2234 migrate_enable();
2235 }
2236 }
2237 }
2238
2239 /* wait for all the updates to complete */
2240 for_each_buffer_cpu(buffer, cpu) {
2241 cpu_buffer = buffer->buffers[cpu];
2242 if (!cpu_buffer->nr_pages_to_update)
2243 continue;
2244
2245 if (cpu_online(cpu))
2246 wait_for_completion(&cpu_buffer->update_done);
2247 cpu_buffer->nr_pages_to_update = 0;
2248 }
2249
2250 cpus_read_unlock();
2251 } else {
2252 cpu_buffer = buffer->buffers[cpu_id];
2253
2254 if (nr_pages == cpu_buffer->nr_pages)
2255 goto out;
2256
2257 /*
2258 * Don't succeed if resizing is disabled, as a reader might be
2259 * manipulating the ring buffer and is expecting a sane state while
2260 * this is true.
2261 */
2262 if (atomic_read(&cpu_buffer->resize_disabled)) {
2263 err = -EBUSY;
2264 goto out_err_unlock;
2265 }
2266
2267 cpu_buffer->nr_pages_to_update = nr_pages -
2268 cpu_buffer->nr_pages;
2269
2270 INIT_LIST_HEAD(&cpu_buffer->new_pages);
2271 if (cpu_buffer->nr_pages_to_update > 0 &&
2272 __rb_allocate_pages(cpu_buffer, cpu_buffer->nr_pages_to_update,
2273 &cpu_buffer->new_pages)) {
2274 err = -ENOMEM;
2275 goto out_err;
2276 }
2277
2278 cpus_read_lock();
2279
2280 /* Can't run something on an offline CPU. */
2281 if (!cpu_online(cpu_id))
2282 rb_update_pages(cpu_buffer);
2283 else {
2284 /* Run directly if possible. */
2285 migrate_disable();
2286 if (cpu_id == smp_processor_id()) {
2287 rb_update_pages(cpu_buffer);
2288 migrate_enable();
2289 } else {
2290 migrate_enable();
2291 schedule_work_on(cpu_id,
2292 &cpu_buffer->update_pages_work);
2293 wait_for_completion(&cpu_buffer->update_done);
2294 }
2295 }
2296
2297 cpu_buffer->nr_pages_to_update = 0;
2298 cpus_read_unlock();
2299 }
2300
2301 out:
2302 /*
2303 * The ring buffer resize can happen with the ring buffer
2304 * enabled, so that the update disturbs the tracing as little
2305 * as possible. But if the buffer is disabled, we do not need
2306 * to worry about that, and we can take the time to verify
2307 * that the buffer is not corrupt.
2308 */
2309 if (atomic_read(&buffer->record_disabled)) {
2310 atomic_inc(&buffer->record_disabled);
2311 /*
2312 * Even though the buffer was disabled, we must make sure
2313 * that it is truly disabled before calling rb_check_pages.
2314 * There could have been a race between checking
2315 * record_disable and incrementing it.
2316 */
2317 synchronize_rcu();
2318 for_each_buffer_cpu(buffer, cpu) {
2319 cpu_buffer = buffer->buffers[cpu];
2320 rb_check_pages(cpu_buffer);
2321 }
2322 atomic_dec(&buffer->record_disabled);
2323 }
2324
2325 mutex_unlock(&buffer->mutex);
2326 return 0;
2327
2328 out_err:
2329 for_each_buffer_cpu(buffer, cpu) {
2330 struct buffer_page *bpage, *tmp;
2331
2332 cpu_buffer = buffer->buffers[cpu];
2333 cpu_buffer->nr_pages_to_update = 0;
2334
2335 if (list_empty(&cpu_buffer->new_pages))
2336 continue;
2337
2338 list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
2339 list) {
2340 list_del_init(&bpage->list);
2341 free_buffer_page(bpage);
2342 }
2343 }
2344 out_err_unlock:
2345 mutex_unlock(&buffer->mutex);
2346 return err;
2347}
2348EXPORT_SYMBOL_GPL(ring_buffer_resize);
2349
2350void ring_buffer_change_overwrite(struct trace_buffer *buffer, int val)
2351{
2352 mutex_lock(&buffer->mutex);
2353 if (val)
2354 buffer->flags |= RB_FL_OVERWRITE;
2355 else
2356 buffer->flags &= ~RB_FL_OVERWRITE;
2357 mutex_unlock(&buffer->mutex);
2358}
2359EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite);
2360
2361static __always_inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
2362{
2363 return bpage->page->data + index;
2364}
2365
2366static __always_inline struct ring_buffer_event *
2367rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
2368{
2369 return __rb_page_index(cpu_buffer->reader_page,
2370 cpu_buffer->reader_page->read);
2371}
2372
2373static __always_inline unsigned rb_page_commit(struct buffer_page *bpage)
2374{
2375 return local_read(&bpage->page->commit);
2376}
2377
2378static struct ring_buffer_event *
2379rb_iter_head_event(struct ring_buffer_iter *iter)
2380{
2381 struct ring_buffer_event *event;
2382 struct buffer_page *iter_head_page = iter->head_page;
2383 unsigned long commit;
2384 unsigned length;
2385
2386 if (iter->head != iter->next_event)
2387 return iter->event;
2388
2389 /*
2390 * When the writer goes across pages, it issues a cmpxchg which
2391 * is a mb(), which will synchronize with the rmb here.
2392 * (see rb_tail_page_update() and __rb_reserve_next())
2393 */
2394 commit = rb_page_commit(iter_head_page);
2395 smp_rmb();
2396 event = __rb_page_index(iter_head_page, iter->head);
2397 length = rb_event_length(event);
2398
2399 /*
2400 * READ_ONCE() doesn't work on functions and we don't want the
2401 * compiler doing any crazy optimizations with length.
2402 */
2403 barrier();
2404
2405 if ((iter->head + length) > commit || length > BUF_MAX_DATA_SIZE)
2406 /* Writer corrupted the read? */
2407 goto reset;
2408
2409 memcpy(iter->event, event, length);
2410 /*
2411 * If the page stamp is still the same after this rmb() then the
2412 * event was safely copied without the writer entering the page.
2413 */
2414 smp_rmb();
2415
2416 /* Make sure the page didn't change since we read this */
2417 if (iter->page_stamp != iter_head_page->page->time_stamp ||
2418 commit > rb_page_commit(iter_head_page))
2419 goto reset;
2420
2421 iter->next_event = iter->head + length;
2422 return iter->event;
2423 reset:
2424 /* Reset to the beginning */
2425 iter->page_stamp = iter->read_stamp = iter->head_page->page->time_stamp;
2426 iter->head = 0;
2427 iter->next_event = 0;
2428 iter->missed_events = 1;
2429 return NULL;
2430}
2431
2432/* Size is determined by what has been committed */
2433static __always_inline unsigned rb_page_size(struct buffer_page *bpage)
2434{
2435 return rb_page_commit(bpage);
2436}
2437
2438static __always_inline unsigned
2439rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
2440{
2441 return rb_page_commit(cpu_buffer->commit_page);
2442}
2443
2444static __always_inline unsigned
2445rb_event_index(struct ring_buffer_event *event)
2446{
2447 unsigned long addr = (unsigned long)event;
2448
2449 return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE;
2450}
2451
2452static void rb_inc_iter(struct ring_buffer_iter *iter)
2453{
2454 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2455
2456 /*
2457 * The iterator could be on the reader page (it starts there).
2458 * But the head could have moved, since the reader was
2459 * found. Check for this case and assign the iterator
2460 * to the head page instead of next.
2461 */
2462 if (iter->head_page == cpu_buffer->reader_page)
2463 iter->head_page = rb_set_head_page(cpu_buffer);
2464 else
2465 rb_inc_page(&iter->head_page);
2466
2467 iter->page_stamp = iter->read_stamp = iter->head_page->page->time_stamp;
2468 iter->head = 0;
2469 iter->next_event = 0;
2470}
2471
2472/*
2473 * rb_handle_head_page - writer hit the head page
2474 *
2475 * Returns: +1 to retry page
2476 * 0 to continue
2477 * -1 on error
2478 */
2479static int
2480rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer,
2481 struct buffer_page *tail_page,
2482 struct buffer_page *next_page)
2483{
2484 struct buffer_page *new_head;
2485 int entries;
2486 int type;
2487 int ret;
2488
2489 entries = rb_page_entries(next_page);
2490
2491 /*
2492 * The hard part is here. We need to move the head
2493 * forward, and protect against both readers on
2494 * other CPUs and writers coming in via interrupts.
2495 */
2496 type = rb_head_page_set_update(cpu_buffer, next_page, tail_page,
2497 RB_PAGE_HEAD);
2498
2499 /*
2500 * type can be one of four:
2501 * NORMAL - an interrupt already moved it for us
2502 * HEAD - we are the first to get here.
2503 * UPDATE - we are the interrupt interrupting
2504 * a current move.
2505 * MOVED - a reader on another CPU moved the next
2506 * pointer to its reader page. Give up
2507 * and try again.
2508 */
2509
2510 switch (type) {
2511 case RB_PAGE_HEAD:
2512 /*
2513 * We changed the head to UPDATE, thus
2514 * it is our responsibility to update
2515 * the counters.
2516 */
2517 local_add(entries, &cpu_buffer->overrun);
2518 local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
2519 local_inc(&cpu_buffer->pages_lost);
2520
2521 /*
2522 * The entries will be zeroed out when we move the
2523 * tail page.
2524 */
2525
2526 /* still more to do */
2527 break;
2528
2529 case RB_PAGE_UPDATE:
2530 /*
2531 * This is an interrupt that interrupt the
2532 * previous update. Still more to do.
2533 */
2534 break;
2535 case RB_PAGE_NORMAL:
2536 /*
2537 * An interrupt came in before the update
2538 * and processed this for us.
2539 * Nothing left to do.
2540 */
2541 return 1;
2542 case RB_PAGE_MOVED:
2543 /*
2544 * The reader is on another CPU and just did
2545 * a swap with our next_page.
2546 * Try again.
2547 */
2548 return 1;
2549 default:
2550 RB_WARN_ON(cpu_buffer, 1); /* WTF??? */
2551 return -1;
2552 }
2553
2554 /*
2555 * Now that we are here, the old head pointer is
2556 * set to UPDATE. This will keep the reader from
2557 * swapping the head page with the reader page.
2558 * The reader (on another CPU) will spin till
2559 * we are finished.
2560 *
2561 * We just need to protect against interrupts
2562 * doing the job. We will set the next pointer
2563 * to HEAD. After that, we set the old pointer
2564 * to NORMAL, but only if it was HEAD before.
2565 * otherwise we are an interrupt, and only
2566 * want the outer most commit to reset it.
2567 */
2568 new_head = next_page;
2569 rb_inc_page(&new_head);
2570
2571 ret = rb_head_page_set_head(cpu_buffer, new_head, next_page,
2572 RB_PAGE_NORMAL);
2573
2574 /*
2575 * Valid returns are:
2576 * HEAD - an interrupt came in and already set it.
2577 * NORMAL - One of two things:
2578 * 1) We really set it.
2579 * 2) A bunch of interrupts came in and moved
2580 * the page forward again.
2581 */
2582 switch (ret) {
2583 case RB_PAGE_HEAD:
2584 case RB_PAGE_NORMAL:
2585 /* OK */
2586 break;
2587 default:
2588 RB_WARN_ON(cpu_buffer, 1);
2589 return -1;
2590 }
2591
2592 /*
2593 * It is possible that an interrupt came in,
2594 * set the head up, then more interrupts came in
2595 * and moved it again. When we get back here,
2596 * the page would have been set to NORMAL but we
2597 * just set it back to HEAD.
2598 *
2599 * How do you detect this? Well, if that happened
2600 * the tail page would have moved.
2601 */
2602 if (ret == RB_PAGE_NORMAL) {
2603 struct buffer_page *buffer_tail_page;
2604
2605 buffer_tail_page = READ_ONCE(cpu_buffer->tail_page);
2606 /*
2607 * If the tail had moved passed next, then we need
2608 * to reset the pointer.
2609 */
2610 if (buffer_tail_page != tail_page &&
2611 buffer_tail_page != next_page)
2612 rb_head_page_set_normal(cpu_buffer, new_head,
2613 next_page,
2614 RB_PAGE_HEAD);
2615 }
2616
2617 /*
2618 * If this was the outer most commit (the one that
2619 * changed the original pointer from HEAD to UPDATE),
2620 * then it is up to us to reset it to NORMAL.
2621 */
2622 if (type == RB_PAGE_HEAD) {
2623 ret = rb_head_page_set_normal(cpu_buffer, next_page,
2624 tail_page,
2625 RB_PAGE_UPDATE);
2626 if (RB_WARN_ON(cpu_buffer,
2627 ret != RB_PAGE_UPDATE))
2628 return -1;
2629 }
2630
2631 return 0;
2632}
2633
2634static inline void
2635rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer,
2636 unsigned long tail, struct rb_event_info *info)
2637{
2638 struct buffer_page *tail_page = info->tail_page;
2639 struct ring_buffer_event *event;
2640 unsigned long length = info->length;
2641
2642 /*
2643 * Only the event that crossed the page boundary
2644 * must fill the old tail_page with padding.
2645 */
2646 if (tail >= BUF_PAGE_SIZE) {
2647 /*
2648 * If the page was filled, then we still need
2649 * to update the real_end. Reset it to zero
2650 * and the reader will ignore it.
2651 */
2652 if (tail == BUF_PAGE_SIZE)
2653 tail_page->real_end = 0;
2654
2655 local_sub(length, &tail_page->write);
2656 return;
2657 }
2658
2659 event = __rb_page_index(tail_page, tail);
2660
2661 /* account for padding bytes */
2662 local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes);
2663
2664 /*
2665 * Save the original length to the meta data.
2666 * This will be used by the reader to add lost event
2667 * counter.
2668 */
2669 tail_page->real_end = tail;
2670
2671 /*
2672 * If this event is bigger than the minimum size, then
2673 * we need to be careful that we don't subtract the
2674 * write counter enough to allow another writer to slip
2675 * in on this page.
2676 * We put in a discarded commit instead, to make sure
2677 * that this space is not used again.
2678 *
2679 * If we are less than the minimum size, we don't need to
2680 * worry about it.
2681 */
2682 if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) {
2683 /* No room for any events */
2684
2685 /* Mark the rest of the page with padding */
2686 rb_event_set_padding(event);
2687
2688 /* Make sure the padding is visible before the write update */
2689 smp_wmb();
2690
2691 /* Set the write back to the previous setting */
2692 local_sub(length, &tail_page->write);
2693 return;
2694 }
2695
2696 /* Put in a discarded event */
2697 event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE;
2698 event->type_len = RINGBUF_TYPE_PADDING;
2699 /* time delta must be non zero */
2700 event->time_delta = 1;
2701
2702 /* Make sure the padding is visible before the tail_page->write update */
2703 smp_wmb();
2704
2705 /* Set write to end of buffer */
2706 length = (tail + length) - BUF_PAGE_SIZE;
2707 local_sub(length, &tail_page->write);
2708}
2709
2710static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer);
2711
2712/*
2713 * This is the slow path, force gcc not to inline it.
2714 */
2715static noinline struct ring_buffer_event *
2716rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer,
2717 unsigned long tail, struct rb_event_info *info)
2718{
2719 struct buffer_page *tail_page = info->tail_page;
2720 struct buffer_page *commit_page = cpu_buffer->commit_page;
2721 struct trace_buffer *buffer = cpu_buffer->buffer;
2722 struct buffer_page *next_page;
2723 int ret;
2724
2725 next_page = tail_page;
2726
2727 rb_inc_page(&next_page);
2728
2729 /*
2730 * If for some reason, we had an interrupt storm that made
2731 * it all the way around the buffer, bail, and warn
2732 * about it.
2733 */
2734 if (unlikely(next_page == commit_page)) {
2735 local_inc(&cpu_buffer->commit_overrun);
2736 goto out_reset;
2737 }
2738
2739 /*
2740 * This is where the fun begins!
2741 *
2742 * We are fighting against races between a reader that
2743 * could be on another CPU trying to swap its reader
2744 * page with the buffer head.
2745 *
2746 * We are also fighting against interrupts coming in and
2747 * moving the head or tail on us as well.
2748 *
2749 * If the next page is the head page then we have filled
2750 * the buffer, unless the commit page is still on the
2751 * reader page.
2752 */
2753 if (rb_is_head_page(next_page, &tail_page->list)) {
2754
2755 /*
2756 * If the commit is not on the reader page, then
2757 * move the header page.
2758 */
2759 if (!rb_is_reader_page(cpu_buffer->commit_page)) {
2760 /*
2761 * If we are not in overwrite mode,
2762 * this is easy, just stop here.
2763 */
2764 if (!(buffer->flags & RB_FL_OVERWRITE)) {
2765 local_inc(&cpu_buffer->dropped_events);
2766 goto out_reset;
2767 }
2768
2769 ret = rb_handle_head_page(cpu_buffer,
2770 tail_page,
2771 next_page);
2772 if (ret < 0)
2773 goto out_reset;
2774 if (ret)
2775 goto out_again;
2776 } else {
2777 /*
2778 * We need to be careful here too. The
2779 * commit page could still be on the reader
2780 * page. We could have a small buffer, and
2781 * have filled up the buffer with events
2782 * from interrupts and such, and wrapped.
2783 *
2784 * Note, if the tail page is also on the
2785 * reader_page, we let it move out.
2786 */
2787 if (unlikely((cpu_buffer->commit_page !=
2788 cpu_buffer->tail_page) &&
2789 (cpu_buffer->commit_page ==
2790 cpu_buffer->reader_page))) {
2791 local_inc(&cpu_buffer->commit_overrun);
2792 goto out_reset;
2793 }
2794 }
2795 }
2796
2797 rb_tail_page_update(cpu_buffer, tail_page, next_page);
2798
2799 out_again:
2800
2801 rb_reset_tail(cpu_buffer, tail, info);
2802
2803 /* Commit what we have for now. */
2804 rb_end_commit(cpu_buffer);
2805 /* rb_end_commit() decs committing */
2806 local_inc(&cpu_buffer->committing);
2807
2808 /* fail and let the caller try again */
2809 return ERR_PTR(-EAGAIN);
2810
2811 out_reset:
2812 /* reset write */
2813 rb_reset_tail(cpu_buffer, tail, info);
2814
2815 return NULL;
2816}
2817
2818/* Slow path */
2819static struct ring_buffer_event *
2820rb_add_time_stamp(struct ring_buffer_event *event, u64 delta, bool abs)
2821{
2822 if (abs)
2823 event->type_len = RINGBUF_TYPE_TIME_STAMP;
2824 else
2825 event->type_len = RINGBUF_TYPE_TIME_EXTEND;
2826
2827 /* Not the first event on the page, or not delta? */
2828 if (abs || rb_event_index(event)) {
2829 event->time_delta = delta & TS_MASK;
2830 event->array[0] = delta >> TS_SHIFT;
2831 } else {
2832 /* nope, just zero it */
2833 event->time_delta = 0;
2834 event->array[0] = 0;
2835 }
2836
2837 return skip_time_extend(event);
2838}
2839
2840#ifndef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
2841static inline bool sched_clock_stable(void)
2842{
2843 return true;
2844}
2845#endif
2846
2847static void
2848rb_check_timestamp(struct ring_buffer_per_cpu *cpu_buffer,
2849 struct rb_event_info *info)
2850{
2851 u64 write_stamp;
2852
2853 WARN_ONCE(1, "Delta way too big! %llu ts=%llu before=%llu after=%llu write stamp=%llu\n%s",
2854 (unsigned long long)info->delta,
2855 (unsigned long long)info->ts,
2856 (unsigned long long)info->before,
2857 (unsigned long long)info->after,
2858 (unsigned long long)(rb_time_read(&cpu_buffer->write_stamp, &write_stamp) ? write_stamp : 0),
2859 sched_clock_stable() ? "" :
2860 "If you just came from a suspend/resume,\n"
2861 "please switch to the trace global clock:\n"
2862 " echo global > /sys/kernel/tracing/trace_clock\n"
2863 "or add trace_clock=global to the kernel command line\n");
2864}
2865
2866static void rb_add_timestamp(struct ring_buffer_per_cpu *cpu_buffer,
2867 struct ring_buffer_event **event,
2868 struct rb_event_info *info,
2869 u64 *delta,
2870 unsigned int *length)
2871{
2872 bool abs = info->add_timestamp &
2873 (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE);
2874
2875 if (unlikely(info->delta > (1ULL << 59))) {
2876 /*
2877 * Some timers can use more than 59 bits, and when a timestamp
2878 * is added to the buffer, it will lose those bits.
2879 */
2880 if (abs && (info->ts & TS_MSB)) {
2881 info->delta &= ABS_TS_MASK;
2882
2883 /* did the clock go backwards */
2884 } else if (info->before == info->after && info->before > info->ts) {
2885 /* not interrupted */
2886 static int once;
2887
2888 /*
2889 * This is possible with a recalibrating of the TSC.
2890 * Do not produce a call stack, but just report it.
2891 */
2892 if (!once) {
2893 once++;
2894 pr_warn("Ring buffer clock went backwards: %llu -> %llu\n",
2895 info->before, info->ts);
2896 }
2897 } else
2898 rb_check_timestamp(cpu_buffer, info);
2899 if (!abs)
2900 info->delta = 0;
2901 }
2902 *event = rb_add_time_stamp(*event, info->delta, abs);
2903 *length -= RB_LEN_TIME_EXTEND;
2904 *delta = 0;
2905}
2906
2907/**
2908 * rb_update_event - update event type and data
2909 * @cpu_buffer: The per cpu buffer of the @event
2910 * @event: the event to update
2911 * @info: The info to update the @event with (contains length and delta)
2912 *
2913 * Update the type and data fields of the @event. The length
2914 * is the actual size that is written to the ring buffer,
2915 * and with this, we can determine what to place into the
2916 * data field.
2917 */
2918static void
2919rb_update_event(struct ring_buffer_per_cpu *cpu_buffer,
2920 struct ring_buffer_event *event,
2921 struct rb_event_info *info)
2922{
2923 unsigned length = info->length;
2924 u64 delta = info->delta;
2925 unsigned int nest = local_read(&cpu_buffer->committing) - 1;
2926
2927 if (!WARN_ON_ONCE(nest >= MAX_NEST))
2928 cpu_buffer->event_stamp[nest] = info->ts;
2929
2930 /*
2931 * If we need to add a timestamp, then we
2932 * add it to the start of the reserved space.
2933 */
2934 if (unlikely(info->add_timestamp))
2935 rb_add_timestamp(cpu_buffer, &event, info, &delta, &length);
2936
2937 event->time_delta = delta;
2938 length -= RB_EVNT_HDR_SIZE;
2939 if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) {
2940 event->type_len = 0;
2941 event->array[0] = length;
2942 } else
2943 event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT);
2944}
2945
2946static unsigned rb_calculate_event_length(unsigned length)
2947{
2948 struct ring_buffer_event event; /* Used only for sizeof array */
2949
2950 /* zero length can cause confusions */
2951 if (!length)
2952 length++;
2953
2954 if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT)
2955 length += sizeof(event.array[0]);
2956
2957 length += RB_EVNT_HDR_SIZE;
2958 length = ALIGN(length, RB_ARCH_ALIGNMENT);
2959
2960 /*
2961 * In case the time delta is larger than the 27 bits for it
2962 * in the header, we need to add a timestamp. If another
2963 * event comes in when trying to discard this one to increase
2964 * the length, then the timestamp will be added in the allocated
2965 * space of this event. If length is bigger than the size needed
2966 * for the TIME_EXTEND, then padding has to be used. The events
2967 * length must be either RB_LEN_TIME_EXTEND, or greater than or equal
2968 * to RB_LEN_TIME_EXTEND + 8, as 8 is the minimum size for padding.
2969 * As length is a multiple of 4, we only need to worry if it
2970 * is 12 (RB_LEN_TIME_EXTEND + 4).
2971 */
2972 if (length == RB_LEN_TIME_EXTEND + RB_ALIGNMENT)
2973 length += RB_ALIGNMENT;
2974
2975 return length;
2976}
2977
2978static u64 rb_time_delta(struct ring_buffer_event *event)
2979{
2980 switch (event->type_len) {
2981 case RINGBUF_TYPE_PADDING:
2982 return 0;
2983
2984 case RINGBUF_TYPE_TIME_EXTEND:
2985 return rb_event_time_stamp(event);
2986
2987 case RINGBUF_TYPE_TIME_STAMP:
2988 return 0;
2989
2990 case RINGBUF_TYPE_DATA:
2991 return event->time_delta;
2992 default:
2993 return 0;
2994 }
2995}
2996
2997static inline bool
2998rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer,
2999 struct ring_buffer_event *event)
3000{
3001 unsigned long new_index, old_index;
3002 struct buffer_page *bpage;
3003 unsigned long index;
3004 unsigned long addr;
3005 u64 write_stamp;
3006 u64 delta;
3007
3008 new_index = rb_event_index(event);
3009 old_index = new_index + rb_event_ts_length(event);
3010 addr = (unsigned long)event;
3011 addr &= PAGE_MASK;
3012
3013 bpage = READ_ONCE(cpu_buffer->tail_page);
3014
3015 delta = rb_time_delta(event);
3016
3017 if (!rb_time_read(&cpu_buffer->write_stamp, &write_stamp))
3018 return false;
3019
3020 /* Make sure the write stamp is read before testing the location */
3021 barrier();
3022
3023 if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) {
3024 unsigned long write_mask =
3025 local_read(&bpage->write) & ~RB_WRITE_MASK;
3026 unsigned long event_length = rb_event_length(event);
3027
3028 /* Something came in, can't discard */
3029 if (!rb_time_cmpxchg(&cpu_buffer->write_stamp,
3030 write_stamp, write_stamp - delta))
3031 return false;
3032
3033 /*
3034 * It's possible that the event time delta is zero
3035 * (has the same time stamp as the previous event)
3036 * in which case write_stamp and before_stamp could
3037 * be the same. In such a case, force before_stamp
3038 * to be different than write_stamp. It doesn't
3039 * matter what it is, as long as its different.
3040 */
3041 if (!delta)
3042 rb_time_set(&cpu_buffer->before_stamp, 0);
3043
3044 /*
3045 * If an event were to come in now, it would see that the
3046 * write_stamp and the before_stamp are different, and assume
3047 * that this event just added itself before updating
3048 * the write stamp. The interrupting event will fix the
3049 * write stamp for us, and use the before stamp as its delta.
3050 */
3051
3052 /*
3053 * This is on the tail page. It is possible that
3054 * a write could come in and move the tail page
3055 * and write to the next page. That is fine
3056 * because we just shorten what is on this page.
3057 */
3058 old_index += write_mask;
3059 new_index += write_mask;
3060 index = local_cmpxchg(&bpage->write, old_index, new_index);
3061 if (index == old_index) {
3062 /* update counters */
3063 local_sub(event_length, &cpu_buffer->entries_bytes);
3064 return true;
3065 }
3066 }
3067
3068 /* could not discard */
3069 return false;
3070}
3071
3072static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer)
3073{
3074 local_inc(&cpu_buffer->committing);
3075 local_inc(&cpu_buffer->commits);
3076}
3077
3078static __always_inline void
3079rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
3080{
3081 unsigned long max_count;
3082
3083 /*
3084 * We only race with interrupts and NMIs on this CPU.
3085 * If we own the commit event, then we can commit
3086 * all others that interrupted us, since the interruptions
3087 * are in stack format (they finish before they come
3088 * back to us). This allows us to do a simple loop to
3089 * assign the commit to the tail.
3090 */
3091 again:
3092 max_count = cpu_buffer->nr_pages * 100;
3093
3094 while (cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)) {
3095 if (RB_WARN_ON(cpu_buffer, !(--max_count)))
3096 return;
3097 if (RB_WARN_ON(cpu_buffer,
3098 rb_is_reader_page(cpu_buffer->tail_page)))
3099 return;
3100 /*
3101 * No need for a memory barrier here, as the update
3102 * of the tail_page did it for this page.
3103 */
3104 local_set(&cpu_buffer->commit_page->page->commit,
3105 rb_page_write(cpu_buffer->commit_page));
3106 rb_inc_page(&cpu_buffer->commit_page);
3107 /* add barrier to keep gcc from optimizing too much */
3108 barrier();
3109 }
3110 while (rb_commit_index(cpu_buffer) !=
3111 rb_page_write(cpu_buffer->commit_page)) {
3112
3113 /* Make sure the readers see the content of what is committed. */
3114 smp_wmb();
3115 local_set(&cpu_buffer->commit_page->page->commit,
3116 rb_page_write(cpu_buffer->commit_page));
3117 RB_WARN_ON(cpu_buffer,
3118 local_read(&cpu_buffer->commit_page->page->commit) &
3119 ~RB_WRITE_MASK);
3120 barrier();
3121 }
3122
3123 /* again, keep gcc from optimizing */
3124 barrier();
3125
3126 /*
3127 * If an interrupt came in just after the first while loop
3128 * and pushed the tail page forward, we will be left with
3129 * a dangling commit that will never go forward.
3130 */
3131 if (unlikely(cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)))
3132 goto again;
3133}
3134
3135static __always_inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer)
3136{
3137 unsigned long commits;
3138
3139 if (RB_WARN_ON(cpu_buffer,
3140 !local_read(&cpu_buffer->committing)))
3141 return;
3142
3143 again:
3144 commits = local_read(&cpu_buffer->commits);
3145 /* synchronize with interrupts */
3146 barrier();
3147 if (local_read(&cpu_buffer->committing) == 1)
3148 rb_set_commit_to_write(cpu_buffer);
3149
3150 local_dec(&cpu_buffer->committing);
3151
3152 /* synchronize with interrupts */
3153 barrier();
3154
3155 /*
3156 * Need to account for interrupts coming in between the
3157 * updating of the commit page and the clearing of the
3158 * committing counter.
3159 */
3160 if (unlikely(local_read(&cpu_buffer->commits) != commits) &&
3161 !local_read(&cpu_buffer->committing)) {
3162 local_inc(&cpu_buffer->committing);
3163 goto again;
3164 }
3165}
3166
3167static inline void rb_event_discard(struct ring_buffer_event *event)
3168{
3169 if (extended_time(event))
3170 event = skip_time_extend(event);
3171
3172 /* array[0] holds the actual length for the discarded event */
3173 event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE;
3174 event->type_len = RINGBUF_TYPE_PADDING;
3175 /* time delta must be non zero */
3176 if (!event->time_delta)
3177 event->time_delta = 1;
3178}
3179
3180static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer)
3181{
3182 local_inc(&cpu_buffer->entries);
3183 rb_end_commit(cpu_buffer);
3184}
3185
3186static __always_inline void
3187rb_wakeups(struct trace_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer)
3188{
3189 if (buffer->irq_work.waiters_pending) {
3190 buffer->irq_work.waiters_pending = false;
3191 /* irq_work_queue() supplies it's own memory barriers */
3192 irq_work_queue(&buffer->irq_work.work);
3193 }
3194
3195 if (cpu_buffer->irq_work.waiters_pending) {
3196 cpu_buffer->irq_work.waiters_pending = false;
3197 /* irq_work_queue() supplies it's own memory barriers */
3198 irq_work_queue(&cpu_buffer->irq_work.work);
3199 }
3200
3201 if (cpu_buffer->last_pages_touch == local_read(&cpu_buffer->pages_touched))
3202 return;
3203
3204 if (cpu_buffer->reader_page == cpu_buffer->commit_page)
3205 return;
3206
3207 if (!cpu_buffer->irq_work.full_waiters_pending)
3208 return;
3209
3210 cpu_buffer->last_pages_touch = local_read(&cpu_buffer->pages_touched);
3211
3212 if (!full_hit(buffer, cpu_buffer->cpu, cpu_buffer->shortest_full))
3213 return;
3214
3215 cpu_buffer->irq_work.wakeup_full = true;
3216 cpu_buffer->irq_work.full_waiters_pending = false;
3217 /* irq_work_queue() supplies it's own memory barriers */
3218 irq_work_queue(&cpu_buffer->irq_work.work);
3219}
3220
3221#ifdef CONFIG_RING_BUFFER_RECORD_RECURSION
3222# define do_ring_buffer_record_recursion() \
3223 do_ftrace_record_recursion(_THIS_IP_, _RET_IP_)
3224#else
3225# define do_ring_buffer_record_recursion() do { } while (0)
3226#endif
3227
3228/*
3229 * The lock and unlock are done within a preempt disable section.
3230 * The current_context per_cpu variable can only be modified
3231 * by the current task between lock and unlock. But it can
3232 * be modified more than once via an interrupt. To pass this
3233 * information from the lock to the unlock without having to
3234 * access the 'in_interrupt()' functions again (which do show
3235 * a bit of overhead in something as critical as function tracing,
3236 * we use a bitmask trick.
3237 *
3238 * bit 1 = NMI context
3239 * bit 2 = IRQ context
3240 * bit 3 = SoftIRQ context
3241 * bit 4 = normal context.
3242 *
3243 * This works because this is the order of contexts that can
3244 * preempt other contexts. A SoftIRQ never preempts an IRQ
3245 * context.
3246 *
3247 * When the context is determined, the corresponding bit is
3248 * checked and set (if it was set, then a recursion of that context
3249 * happened).
3250 *
3251 * On unlock, we need to clear this bit. To do so, just subtract
3252 * 1 from the current_context and AND it to itself.
3253 *
3254 * (binary)
3255 * 101 - 1 = 100
3256 * 101 & 100 = 100 (clearing bit zero)
3257 *
3258 * 1010 - 1 = 1001
3259 * 1010 & 1001 = 1000 (clearing bit 1)
3260 *
3261 * The least significant bit can be cleared this way, and it
3262 * just so happens that it is the same bit corresponding to
3263 * the current context.
3264 *
3265 * Now the TRANSITION bit breaks the above slightly. The TRANSITION bit
3266 * is set when a recursion is detected at the current context, and if
3267 * the TRANSITION bit is already set, it will fail the recursion.
3268 * This is needed because there's a lag between the changing of
3269 * interrupt context and updating the preempt count. In this case,
3270 * a false positive will be found. To handle this, one extra recursion
3271 * is allowed, and this is done by the TRANSITION bit. If the TRANSITION
3272 * bit is already set, then it is considered a recursion and the function
3273 * ends. Otherwise, the TRANSITION bit is set, and that bit is returned.
3274 *
3275 * On the trace_recursive_unlock(), the TRANSITION bit will be the first
3276 * to be cleared. Even if it wasn't the context that set it. That is,
3277 * if an interrupt comes in while NORMAL bit is set and the ring buffer
3278 * is called before preempt_count() is updated, since the check will
3279 * be on the NORMAL bit, the TRANSITION bit will then be set. If an
3280 * NMI then comes in, it will set the NMI bit, but when the NMI code
3281 * does the trace_recursive_unlock() it will clear the TRANSITION bit
3282 * and leave the NMI bit set. But this is fine, because the interrupt
3283 * code that set the TRANSITION bit will then clear the NMI bit when it
3284 * calls trace_recursive_unlock(). If another NMI comes in, it will
3285 * set the TRANSITION bit and continue.
3286 *
3287 * Note: The TRANSITION bit only handles a single transition between context.
3288 */
3289
3290static __always_inline bool
3291trace_recursive_lock(struct ring_buffer_per_cpu *cpu_buffer)
3292{
3293 unsigned int val = cpu_buffer->current_context;
3294 int bit = interrupt_context_level();
3295
3296 bit = RB_CTX_NORMAL - bit;
3297
3298 if (unlikely(val & (1 << (bit + cpu_buffer->nest)))) {
3299 /*
3300 * It is possible that this was called by transitioning
3301 * between interrupt context, and preempt_count() has not
3302 * been updated yet. In this case, use the TRANSITION bit.
3303 */
3304 bit = RB_CTX_TRANSITION;
3305 if (val & (1 << (bit + cpu_buffer->nest))) {
3306 do_ring_buffer_record_recursion();
3307 return true;
3308 }
3309 }
3310
3311 val |= (1 << (bit + cpu_buffer->nest));
3312 cpu_buffer->current_context = val;
3313
3314 return false;
3315}
3316
3317static __always_inline void
3318trace_recursive_unlock(struct ring_buffer_per_cpu *cpu_buffer)
3319{
3320 cpu_buffer->current_context &=
3321 cpu_buffer->current_context - (1 << cpu_buffer->nest);
3322}
3323
3324/* The recursive locking above uses 5 bits */
3325#define NESTED_BITS 5
3326
3327/**
3328 * ring_buffer_nest_start - Allow to trace while nested
3329 * @buffer: The ring buffer to modify
3330 *
3331 * The ring buffer has a safety mechanism to prevent recursion.
3332 * But there may be a case where a trace needs to be done while
3333 * tracing something else. In this case, calling this function
3334 * will allow this function to nest within a currently active
3335 * ring_buffer_lock_reserve().
3336 *
3337 * Call this function before calling another ring_buffer_lock_reserve() and
3338 * call ring_buffer_nest_end() after the nested ring_buffer_unlock_commit().
3339 */
3340void ring_buffer_nest_start(struct trace_buffer *buffer)
3341{
3342 struct ring_buffer_per_cpu *cpu_buffer;
3343 int cpu;
3344
3345 /* Enabled by ring_buffer_nest_end() */
3346 preempt_disable_notrace();
3347 cpu = raw_smp_processor_id();
3348 cpu_buffer = buffer->buffers[cpu];
3349 /* This is the shift value for the above recursive locking */
3350 cpu_buffer->nest += NESTED_BITS;
3351}
3352
3353/**
3354 * ring_buffer_nest_end - Allow to trace while nested
3355 * @buffer: The ring buffer to modify
3356 *
3357 * Must be called after ring_buffer_nest_start() and after the
3358 * ring_buffer_unlock_commit().
3359 */
3360void ring_buffer_nest_end(struct trace_buffer *buffer)
3361{
3362 struct ring_buffer_per_cpu *cpu_buffer;
3363 int cpu;
3364
3365 /* disabled by ring_buffer_nest_start() */
3366 cpu = raw_smp_processor_id();
3367 cpu_buffer = buffer->buffers[cpu];
3368 /* This is the shift value for the above recursive locking */
3369 cpu_buffer->nest -= NESTED_BITS;
3370 preempt_enable_notrace();
3371}
3372
3373/**
3374 * ring_buffer_unlock_commit - commit a reserved
3375 * @buffer: The buffer to commit to
3376 * @event: The event pointer to commit.
3377 *
3378 * This commits the data to the ring buffer, and releases any locks held.
3379 *
3380 * Must be paired with ring_buffer_lock_reserve.
3381 */
3382int ring_buffer_unlock_commit(struct trace_buffer *buffer)
3383{
3384 struct ring_buffer_per_cpu *cpu_buffer;
3385 int cpu = raw_smp_processor_id();
3386
3387 cpu_buffer = buffer->buffers[cpu];
3388
3389 rb_commit(cpu_buffer);
3390
3391 rb_wakeups(buffer, cpu_buffer);
3392
3393 trace_recursive_unlock(cpu_buffer);
3394
3395 preempt_enable_notrace();
3396
3397 return 0;
3398}
3399EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit);
3400
3401/* Special value to validate all deltas on a page. */
3402#define CHECK_FULL_PAGE 1L
3403
3404#ifdef CONFIG_RING_BUFFER_VALIDATE_TIME_DELTAS
3405static void dump_buffer_page(struct buffer_data_page *bpage,
3406 struct rb_event_info *info,
3407 unsigned long tail)
3408{
3409 struct ring_buffer_event *event;
3410 u64 ts, delta;
3411 int e;
3412
3413 ts = bpage->time_stamp;
3414 pr_warn(" [%lld] PAGE TIME STAMP\n", ts);
3415
3416 for (e = 0; e < tail; e += rb_event_length(event)) {
3417
3418 event = (struct ring_buffer_event *)(bpage->data + e);
3419
3420 switch (event->type_len) {
3421
3422 case RINGBUF_TYPE_TIME_EXTEND:
3423 delta = rb_event_time_stamp(event);
3424 ts += delta;
3425 pr_warn(" [%lld] delta:%lld TIME EXTEND\n", ts, delta);
3426 break;
3427
3428 case RINGBUF_TYPE_TIME_STAMP:
3429 delta = rb_event_time_stamp(event);
3430 ts = rb_fix_abs_ts(delta, ts);
3431 pr_warn(" [%lld] absolute:%lld TIME STAMP\n", ts, delta);
3432 break;
3433
3434 case RINGBUF_TYPE_PADDING:
3435 ts += event->time_delta;
3436 pr_warn(" [%lld] delta:%d PADDING\n", ts, event->time_delta);
3437 break;
3438
3439 case RINGBUF_TYPE_DATA:
3440 ts += event->time_delta;
3441 pr_warn(" [%lld] delta:%d\n", ts, event->time_delta);
3442 break;
3443
3444 default:
3445 break;
3446 }
3447 }
3448}
3449
3450static DEFINE_PER_CPU(atomic_t, checking);
3451static atomic_t ts_dump;
3452
3453/*
3454 * Check if the current event time stamp matches the deltas on
3455 * the buffer page.
3456 */
3457static void check_buffer(struct ring_buffer_per_cpu *cpu_buffer,
3458 struct rb_event_info *info,
3459 unsigned long tail)
3460{
3461 struct ring_buffer_event *event;
3462 struct buffer_data_page *bpage;
3463 u64 ts, delta;
3464 bool full = false;
3465 int e;
3466
3467 bpage = info->tail_page->page;
3468
3469 if (tail == CHECK_FULL_PAGE) {
3470 full = true;
3471 tail = local_read(&bpage->commit);
3472 } else if (info->add_timestamp &
3473 (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE)) {
3474 /* Ignore events with absolute time stamps */
3475 return;
3476 }
3477
3478 /*
3479 * Do not check the first event (skip possible extends too).
3480 * Also do not check if previous events have not been committed.
3481 */
3482 if (tail <= 8 || tail > local_read(&bpage->commit))
3483 return;
3484
3485 /*
3486 * If this interrupted another event,
3487 */
3488 if (atomic_inc_return(this_cpu_ptr(&checking)) != 1)
3489 goto out;
3490
3491 ts = bpage->time_stamp;
3492
3493 for (e = 0; e < tail; e += rb_event_length(event)) {
3494
3495 event = (struct ring_buffer_event *)(bpage->data + e);
3496
3497 switch (event->type_len) {
3498
3499 case RINGBUF_TYPE_TIME_EXTEND:
3500 delta = rb_event_time_stamp(event);
3501 ts += delta;
3502 break;
3503
3504 case RINGBUF_TYPE_TIME_STAMP:
3505 delta = rb_event_time_stamp(event);
3506 ts = rb_fix_abs_ts(delta, ts);
3507 break;
3508
3509 case RINGBUF_TYPE_PADDING:
3510 if (event->time_delta == 1)
3511 break;
3512 fallthrough;
3513 case RINGBUF_TYPE_DATA:
3514 ts += event->time_delta;
3515 break;
3516
3517 default:
3518 RB_WARN_ON(cpu_buffer, 1);
3519 }
3520 }
3521 if ((full && ts > info->ts) ||
3522 (!full && ts + info->delta != info->ts)) {
3523 /* If another report is happening, ignore this one */
3524 if (atomic_inc_return(&ts_dump) != 1) {
3525 atomic_dec(&ts_dump);
3526 goto out;
3527 }
3528 atomic_inc(&cpu_buffer->record_disabled);
3529 /* There's some cases in boot up that this can happen */
3530 WARN_ON_ONCE(system_state != SYSTEM_BOOTING);
3531 pr_warn("[CPU: %d]TIME DOES NOT MATCH expected:%lld actual:%lld delta:%lld before:%lld after:%lld%s\n",
3532 cpu_buffer->cpu,
3533 ts + info->delta, info->ts, info->delta,
3534 info->before, info->after,
3535 full ? " (full)" : "");
3536 dump_buffer_page(bpage, info, tail);
3537 atomic_dec(&ts_dump);
3538 /* Do not re-enable checking */
3539 return;
3540 }
3541out:
3542 atomic_dec(this_cpu_ptr(&checking));
3543}
3544#else
3545static inline void check_buffer(struct ring_buffer_per_cpu *cpu_buffer,
3546 struct rb_event_info *info,
3547 unsigned long tail)
3548{
3549}
3550#endif /* CONFIG_RING_BUFFER_VALIDATE_TIME_DELTAS */
3551
3552static struct ring_buffer_event *
3553__rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
3554 struct rb_event_info *info)
3555{
3556 struct ring_buffer_event *event;
3557 struct buffer_page *tail_page;
3558 unsigned long tail, write, w;
3559 bool a_ok;
3560 bool b_ok;
3561
3562 /* Don't let the compiler play games with cpu_buffer->tail_page */
3563 tail_page = info->tail_page = READ_ONCE(cpu_buffer->tail_page);
3564
3565 /*A*/ w = local_read(&tail_page->write) & RB_WRITE_MASK;
3566 barrier();
3567 b_ok = rb_time_read(&cpu_buffer->before_stamp, &info->before);
3568 a_ok = rb_time_read(&cpu_buffer->write_stamp, &info->after);
3569 barrier();
3570 info->ts = rb_time_stamp(cpu_buffer->buffer);
3571
3572 if ((info->add_timestamp & RB_ADD_STAMP_ABSOLUTE)) {
3573 info->delta = info->ts;
3574 } else {
3575 /*
3576 * If interrupting an event time update, we may need an
3577 * absolute timestamp.
3578 * Don't bother if this is the start of a new page (w == 0).
3579 */
3580 if (unlikely(!a_ok || !b_ok || (info->before != info->after && w))) {
3581 info->add_timestamp |= RB_ADD_STAMP_FORCE | RB_ADD_STAMP_EXTEND;
3582 info->length += RB_LEN_TIME_EXTEND;
3583 } else {
3584 info->delta = info->ts - info->after;
3585 if (unlikely(test_time_stamp(info->delta))) {
3586 info->add_timestamp |= RB_ADD_STAMP_EXTEND;
3587 info->length += RB_LEN_TIME_EXTEND;
3588 }
3589 }
3590 }
3591
3592 /*B*/ rb_time_set(&cpu_buffer->before_stamp, info->ts);
3593
3594 /*C*/ write = local_add_return(info->length, &tail_page->write);
3595
3596 /* set write to only the index of the write */
3597 write &= RB_WRITE_MASK;
3598
3599 tail = write - info->length;
3600
3601 /* See if we shot pass the end of this buffer page */
3602 if (unlikely(write > BUF_PAGE_SIZE)) {
3603 /* before and after may now different, fix it up*/
3604 b_ok = rb_time_read(&cpu_buffer->before_stamp, &info->before);
3605 a_ok = rb_time_read(&cpu_buffer->write_stamp, &info->after);
3606 if (a_ok && b_ok && info->before != info->after)
3607 (void)rb_time_cmpxchg(&cpu_buffer->before_stamp,
3608 info->before, info->after);
3609 if (a_ok && b_ok)
3610 check_buffer(cpu_buffer, info, CHECK_FULL_PAGE);
3611 return rb_move_tail(cpu_buffer, tail, info);
3612 }
3613
3614 if (likely(tail == w)) {
3615 u64 save_before;
3616 bool s_ok;
3617
3618 /* Nothing interrupted us between A and C */
3619 /*D*/ rb_time_set(&cpu_buffer->write_stamp, info->ts);
3620 barrier();
3621 /*E*/ s_ok = rb_time_read(&cpu_buffer->before_stamp, &save_before);
3622 RB_WARN_ON(cpu_buffer, !s_ok);
3623 if (likely(!(info->add_timestamp &
3624 (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE))))
3625 /* This did not interrupt any time update */
3626 info->delta = info->ts - info->after;
3627 else
3628 /* Just use full timestamp for interrupting event */
3629 info->delta = info->ts;
3630 barrier();
3631 check_buffer(cpu_buffer, info, tail);
3632 if (unlikely(info->ts != save_before)) {
3633 /* SLOW PATH - Interrupted between C and E */
3634
3635 a_ok = rb_time_read(&cpu_buffer->write_stamp, &info->after);
3636 RB_WARN_ON(cpu_buffer, !a_ok);
3637
3638 /* Write stamp must only go forward */
3639 if (save_before > info->after) {
3640 /*
3641 * We do not care about the result, only that
3642 * it gets updated atomically.
3643 */
3644 (void)rb_time_cmpxchg(&cpu_buffer->write_stamp,
3645 info->after, save_before);
3646 }
3647 }
3648 } else {
3649 u64 ts;
3650 /* SLOW PATH - Interrupted between A and C */
3651 a_ok = rb_time_read(&cpu_buffer->write_stamp, &info->after);
3652 /* Was interrupted before here, write_stamp must be valid */
3653 RB_WARN_ON(cpu_buffer, !a_ok);
3654 ts = rb_time_stamp(cpu_buffer->buffer);
3655 barrier();
3656 /*E*/ if (write == (local_read(&tail_page->write) & RB_WRITE_MASK) &&
3657 info->after < ts &&
3658 rb_time_cmpxchg(&cpu_buffer->write_stamp,
3659 info->after, ts)) {
3660 /* Nothing came after this event between C and E */
3661 info->delta = ts - info->after;
3662 } else {
3663 /*
3664 * Interrupted between C and E:
3665 * Lost the previous events time stamp. Just set the
3666 * delta to zero, and this will be the same time as
3667 * the event this event interrupted. And the events that
3668 * came after this will still be correct (as they would
3669 * have built their delta on the previous event.
3670 */
3671 info->delta = 0;
3672 }
3673 info->ts = ts;
3674 info->add_timestamp &= ~RB_ADD_STAMP_FORCE;
3675 }
3676
3677 /*
3678 * If this is the first commit on the page, then it has the same
3679 * timestamp as the page itself.
3680 */
3681 if (unlikely(!tail && !(info->add_timestamp &
3682 (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE))))
3683 info->delta = 0;
3684
3685 /* We reserved something on the buffer */
3686
3687 event = __rb_page_index(tail_page, tail);
3688 rb_update_event(cpu_buffer, event, info);
3689
3690 local_inc(&tail_page->entries);
3691
3692 /*
3693 * If this is the first commit on the page, then update
3694 * its timestamp.
3695 */
3696 if (unlikely(!tail))
3697 tail_page->page->time_stamp = info->ts;
3698
3699 /* account for these added bytes */
3700 local_add(info->length, &cpu_buffer->entries_bytes);
3701
3702 return event;
3703}
3704
3705static __always_inline struct ring_buffer_event *
3706rb_reserve_next_event(struct trace_buffer *buffer,
3707 struct ring_buffer_per_cpu *cpu_buffer,
3708 unsigned long length)
3709{
3710 struct ring_buffer_event *event;
3711 struct rb_event_info info;
3712 int nr_loops = 0;
3713 int add_ts_default;
3714
3715 rb_start_commit(cpu_buffer);
3716 /* The commit page can not change after this */
3717
3718#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
3719 /*
3720 * Due to the ability to swap a cpu buffer from a buffer
3721 * it is possible it was swapped before we committed.
3722 * (committing stops a swap). We check for it here and
3723 * if it happened, we have to fail the write.
3724 */
3725 barrier();
3726 if (unlikely(READ_ONCE(cpu_buffer->buffer) != buffer)) {
3727 local_dec(&cpu_buffer->committing);
3728 local_dec(&cpu_buffer->commits);
3729 return NULL;
3730 }
3731#endif
3732
3733 info.length = rb_calculate_event_length(length);
3734
3735 if (ring_buffer_time_stamp_abs(cpu_buffer->buffer)) {
3736 add_ts_default = RB_ADD_STAMP_ABSOLUTE;
3737 info.length += RB_LEN_TIME_EXTEND;
3738 } else {
3739 add_ts_default = RB_ADD_STAMP_NONE;
3740 }
3741
3742 again:
3743 info.add_timestamp = add_ts_default;
3744 info.delta = 0;
3745
3746 /*
3747 * We allow for interrupts to reenter here and do a trace.
3748 * If one does, it will cause this original code to loop
3749 * back here. Even with heavy interrupts happening, this
3750 * should only happen a few times in a row. If this happens
3751 * 1000 times in a row, there must be either an interrupt
3752 * storm or we have something buggy.
3753 * Bail!
3754 */
3755 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
3756 goto out_fail;
3757
3758 event = __rb_reserve_next(cpu_buffer, &info);
3759
3760 if (unlikely(PTR_ERR(event) == -EAGAIN)) {
3761 if (info.add_timestamp & (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_EXTEND))
3762 info.length -= RB_LEN_TIME_EXTEND;
3763 goto again;
3764 }
3765
3766 if (likely(event))
3767 return event;
3768 out_fail:
3769 rb_end_commit(cpu_buffer);
3770 return NULL;
3771}
3772
3773/**
3774 * ring_buffer_lock_reserve - reserve a part of the buffer
3775 * @buffer: the ring buffer to reserve from
3776 * @length: the length of the data to reserve (excluding event header)
3777 *
3778 * Returns a reserved event on the ring buffer to copy directly to.
3779 * The user of this interface will need to get the body to write into
3780 * and can use the ring_buffer_event_data() interface.
3781 *
3782 * The length is the length of the data needed, not the event length
3783 * which also includes the event header.
3784 *
3785 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
3786 * If NULL is returned, then nothing has been allocated or locked.
3787 */
3788struct ring_buffer_event *
3789ring_buffer_lock_reserve(struct trace_buffer *buffer, unsigned long length)
3790{
3791 struct ring_buffer_per_cpu *cpu_buffer;
3792 struct ring_buffer_event *event;
3793 int cpu;
3794
3795 /* If we are tracing schedule, we don't want to recurse */
3796 preempt_disable_notrace();
3797
3798 if (unlikely(atomic_read(&buffer->record_disabled)))
3799 goto out;
3800
3801 cpu = raw_smp_processor_id();
3802
3803 if (unlikely(!cpumask_test_cpu(cpu, buffer->cpumask)))
3804 goto out;
3805
3806 cpu_buffer = buffer->buffers[cpu];
3807
3808 if (unlikely(atomic_read(&cpu_buffer->record_disabled)))
3809 goto out;
3810
3811 if (unlikely(length > BUF_MAX_DATA_SIZE))
3812 goto out;
3813
3814 if (unlikely(trace_recursive_lock(cpu_buffer)))
3815 goto out;
3816
3817 event = rb_reserve_next_event(buffer, cpu_buffer, length);
3818 if (!event)
3819 goto out_unlock;
3820
3821 return event;
3822
3823 out_unlock:
3824 trace_recursive_unlock(cpu_buffer);
3825 out:
3826 preempt_enable_notrace();
3827 return NULL;
3828}
3829EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve);
3830
3831/*
3832 * Decrement the entries to the page that an event is on.
3833 * The event does not even need to exist, only the pointer
3834 * to the page it is on. This may only be called before the commit
3835 * takes place.
3836 */
3837static inline void
3838rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer,
3839 struct ring_buffer_event *event)
3840{
3841 unsigned long addr = (unsigned long)event;
3842 struct buffer_page *bpage = cpu_buffer->commit_page;
3843 struct buffer_page *start;
3844
3845 addr &= PAGE_MASK;
3846
3847 /* Do the likely case first */
3848 if (likely(bpage->page == (void *)addr)) {
3849 local_dec(&bpage->entries);
3850 return;
3851 }
3852
3853 /*
3854 * Because the commit page may be on the reader page we
3855 * start with the next page and check the end loop there.
3856 */
3857 rb_inc_page(&bpage);
3858 start = bpage;
3859 do {
3860 if (bpage->page == (void *)addr) {
3861 local_dec(&bpage->entries);
3862 return;
3863 }
3864 rb_inc_page(&bpage);
3865 } while (bpage != start);
3866
3867 /* commit not part of this buffer?? */
3868 RB_WARN_ON(cpu_buffer, 1);
3869}
3870
3871/**
3872 * ring_buffer_discard_commit - discard an event that has not been committed
3873 * @buffer: the ring buffer
3874 * @event: non committed event to discard
3875 *
3876 * Sometimes an event that is in the ring buffer needs to be ignored.
3877 * This function lets the user discard an event in the ring buffer
3878 * and then that event will not be read later.
3879 *
3880 * This function only works if it is called before the item has been
3881 * committed. It will try to free the event from the ring buffer
3882 * if another event has not been added behind it.
3883 *
3884 * If another event has been added behind it, it will set the event
3885 * up as discarded, and perform the commit.
3886 *
3887 * If this function is called, do not call ring_buffer_unlock_commit on
3888 * the event.
3889 */
3890void ring_buffer_discard_commit(struct trace_buffer *buffer,
3891 struct ring_buffer_event *event)
3892{
3893 struct ring_buffer_per_cpu *cpu_buffer;
3894 int cpu;
3895
3896 /* The event is discarded regardless */
3897 rb_event_discard(event);
3898
3899 cpu = smp_processor_id();
3900 cpu_buffer = buffer->buffers[cpu];
3901
3902 /*
3903 * This must only be called if the event has not been
3904 * committed yet. Thus we can assume that preemption
3905 * is still disabled.
3906 */
3907 RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing));
3908
3909 rb_decrement_entry(cpu_buffer, event);
3910 if (rb_try_to_discard(cpu_buffer, event))
3911 goto out;
3912
3913 out:
3914 rb_end_commit(cpu_buffer);
3915
3916 trace_recursive_unlock(cpu_buffer);
3917
3918 preempt_enable_notrace();
3919
3920}
3921EXPORT_SYMBOL_GPL(ring_buffer_discard_commit);
3922
3923/**
3924 * ring_buffer_write - write data to the buffer without reserving
3925 * @buffer: The ring buffer to write to.
3926 * @length: The length of the data being written (excluding the event header)
3927 * @data: The data to write to the buffer.
3928 *
3929 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
3930 * one function. If you already have the data to write to the buffer, it
3931 * may be easier to simply call this function.
3932 *
3933 * Note, like ring_buffer_lock_reserve, the length is the length of the data
3934 * and not the length of the event which would hold the header.
3935 */
3936int ring_buffer_write(struct trace_buffer *buffer,
3937 unsigned long length,
3938 void *data)
3939{
3940 struct ring_buffer_per_cpu *cpu_buffer;
3941 struct ring_buffer_event *event;
3942 void *body;
3943 int ret = -EBUSY;
3944 int cpu;
3945
3946 preempt_disable_notrace();
3947
3948 if (atomic_read(&buffer->record_disabled))
3949 goto out;
3950
3951 cpu = raw_smp_processor_id();
3952
3953 if (!cpumask_test_cpu(cpu, buffer->cpumask))
3954 goto out;
3955
3956 cpu_buffer = buffer->buffers[cpu];
3957
3958 if (atomic_read(&cpu_buffer->record_disabled))
3959 goto out;
3960
3961 if (length > BUF_MAX_DATA_SIZE)
3962 goto out;
3963
3964 if (unlikely(trace_recursive_lock(cpu_buffer)))
3965 goto out;
3966
3967 event = rb_reserve_next_event(buffer, cpu_buffer, length);
3968 if (!event)
3969 goto out_unlock;
3970
3971 body = rb_event_data(event);
3972
3973 memcpy(body, data, length);
3974
3975 rb_commit(cpu_buffer);
3976
3977 rb_wakeups(buffer, cpu_buffer);
3978
3979 ret = 0;
3980
3981 out_unlock:
3982 trace_recursive_unlock(cpu_buffer);
3983
3984 out:
3985 preempt_enable_notrace();
3986
3987 return ret;
3988}
3989EXPORT_SYMBOL_GPL(ring_buffer_write);
3990
3991static bool rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
3992{
3993 struct buffer_page *reader = cpu_buffer->reader_page;
3994 struct buffer_page *head = rb_set_head_page(cpu_buffer);
3995 struct buffer_page *commit = cpu_buffer->commit_page;
3996
3997 /* In case of error, head will be NULL */
3998 if (unlikely(!head))
3999 return true;
4000
4001 /* Reader should exhaust content in reader page */
4002 if (reader->read != rb_page_commit(reader))
4003 return false;
4004
4005 /*
4006 * If writers are committing on the reader page, knowing all
4007 * committed content has been read, the ring buffer is empty.
4008 */
4009 if (commit == reader)
4010 return true;
4011
4012 /*
4013 * If writers are committing on a page other than reader page
4014 * and head page, there should always be content to read.
4015 */
4016 if (commit != head)
4017 return false;
4018
4019 /*
4020 * Writers are committing on the head page, we just need
4021 * to care about there're committed data, and the reader will
4022 * swap reader page with head page when it is to read data.
4023 */
4024 return rb_page_commit(commit) == 0;
4025}
4026
4027/**
4028 * ring_buffer_record_disable - stop all writes into the buffer
4029 * @buffer: The ring buffer to stop writes to.
4030 *
4031 * This prevents all writes to the buffer. Any attempt to write
4032 * to the buffer after this will fail and return NULL.
4033 *
4034 * The caller should call synchronize_rcu() after this.
4035 */
4036void ring_buffer_record_disable(struct trace_buffer *buffer)
4037{
4038 atomic_inc(&buffer->record_disabled);
4039}
4040EXPORT_SYMBOL_GPL(ring_buffer_record_disable);
4041
4042/**
4043 * ring_buffer_record_enable - enable writes to the buffer
4044 * @buffer: The ring buffer to enable writes
4045 *
4046 * Note, multiple disables will need the same number of enables
4047 * to truly enable the writing (much like preempt_disable).
4048 */
4049void ring_buffer_record_enable(struct trace_buffer *buffer)
4050{
4051 atomic_dec(&buffer->record_disabled);
4052}
4053EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
4054
4055/**
4056 * ring_buffer_record_off - stop all writes into the buffer
4057 * @buffer: The ring buffer to stop writes to.
4058 *
4059 * This prevents all writes to the buffer. Any attempt to write
4060 * to the buffer after this will fail and return NULL.
4061 *
4062 * This is different than ring_buffer_record_disable() as
4063 * it works like an on/off switch, where as the disable() version
4064 * must be paired with a enable().
4065 */
4066void ring_buffer_record_off(struct trace_buffer *buffer)
4067{
4068 unsigned int rd;
4069 unsigned int new_rd;
4070
4071 rd = atomic_read(&buffer->record_disabled);
4072 do {
4073 new_rd = rd | RB_BUFFER_OFF;
4074 } while (!atomic_try_cmpxchg(&buffer->record_disabled, &rd, new_rd));
4075}
4076EXPORT_SYMBOL_GPL(ring_buffer_record_off);
4077
4078/**
4079 * ring_buffer_record_on - restart writes into the buffer
4080 * @buffer: The ring buffer to start writes to.
4081 *
4082 * This enables all writes to the buffer that was disabled by
4083 * ring_buffer_record_off().
4084 *
4085 * This is different than ring_buffer_record_enable() as
4086 * it works like an on/off switch, where as the enable() version
4087 * must be paired with a disable().
4088 */
4089void ring_buffer_record_on(struct trace_buffer *buffer)
4090{
4091 unsigned int rd;
4092 unsigned int new_rd;
4093
4094 rd = atomic_read(&buffer->record_disabled);
4095 do {
4096 new_rd = rd & ~RB_BUFFER_OFF;
4097 } while (!atomic_try_cmpxchg(&buffer->record_disabled, &rd, new_rd));
4098}
4099EXPORT_SYMBOL_GPL(ring_buffer_record_on);
4100
4101/**
4102 * ring_buffer_record_is_on - return true if the ring buffer can write
4103 * @buffer: The ring buffer to see if write is enabled
4104 *
4105 * Returns true if the ring buffer is in a state that it accepts writes.
4106 */
4107bool ring_buffer_record_is_on(struct trace_buffer *buffer)
4108{
4109 return !atomic_read(&buffer->record_disabled);
4110}
4111
4112/**
4113 * ring_buffer_record_is_set_on - return true if the ring buffer is set writable
4114 * @buffer: The ring buffer to see if write is set enabled
4115 *
4116 * Returns true if the ring buffer is set writable by ring_buffer_record_on().
4117 * Note that this does NOT mean it is in a writable state.
4118 *
4119 * It may return true when the ring buffer has been disabled by
4120 * ring_buffer_record_disable(), as that is a temporary disabling of
4121 * the ring buffer.
4122 */
4123bool ring_buffer_record_is_set_on(struct trace_buffer *buffer)
4124{
4125 return !(atomic_read(&buffer->record_disabled) & RB_BUFFER_OFF);
4126}
4127
4128/**
4129 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
4130 * @buffer: The ring buffer to stop writes to.
4131 * @cpu: The CPU buffer to stop
4132 *
4133 * This prevents all writes to the buffer. Any attempt to write
4134 * to the buffer after this will fail and return NULL.
4135 *
4136 * The caller should call synchronize_rcu() after this.
4137 */
4138void ring_buffer_record_disable_cpu(struct trace_buffer *buffer, int cpu)
4139{
4140 struct ring_buffer_per_cpu *cpu_buffer;
4141
4142 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4143 return;
4144
4145 cpu_buffer = buffer->buffers[cpu];
4146 atomic_inc(&cpu_buffer->record_disabled);
4147}
4148EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu);
4149
4150/**
4151 * ring_buffer_record_enable_cpu - enable writes to the buffer
4152 * @buffer: The ring buffer to enable writes
4153 * @cpu: The CPU to enable.
4154 *
4155 * Note, multiple disables will need the same number of enables
4156 * to truly enable the writing (much like preempt_disable).
4157 */
4158void ring_buffer_record_enable_cpu(struct trace_buffer *buffer, int cpu)
4159{
4160 struct ring_buffer_per_cpu *cpu_buffer;
4161
4162 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4163 return;
4164
4165 cpu_buffer = buffer->buffers[cpu];
4166 atomic_dec(&cpu_buffer->record_disabled);
4167}
4168EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
4169
4170/*
4171 * The total entries in the ring buffer is the running counter
4172 * of entries entered into the ring buffer, minus the sum of
4173 * the entries read from the ring buffer and the number of
4174 * entries that were overwritten.
4175 */
4176static inline unsigned long
4177rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer)
4178{
4179 return local_read(&cpu_buffer->entries) -
4180 (local_read(&cpu_buffer->overrun) + cpu_buffer->read);
4181}
4182
4183/**
4184 * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer
4185 * @buffer: The ring buffer
4186 * @cpu: The per CPU buffer to read from.
4187 */
4188u64 ring_buffer_oldest_event_ts(struct trace_buffer *buffer, int cpu)
4189{
4190 unsigned long flags;
4191 struct ring_buffer_per_cpu *cpu_buffer;
4192 struct buffer_page *bpage;
4193 u64 ret = 0;
4194
4195 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4196 return 0;
4197
4198 cpu_buffer = buffer->buffers[cpu];
4199 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4200 /*
4201 * if the tail is on reader_page, oldest time stamp is on the reader
4202 * page
4203 */
4204 if (cpu_buffer->tail_page == cpu_buffer->reader_page)
4205 bpage = cpu_buffer->reader_page;
4206 else
4207 bpage = rb_set_head_page(cpu_buffer);
4208 if (bpage)
4209 ret = bpage->page->time_stamp;
4210 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4211
4212 return ret;
4213}
4214EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts);
4215
4216/**
4217 * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer
4218 * @buffer: The ring buffer
4219 * @cpu: The per CPU buffer to read from.
4220 */
4221unsigned long ring_buffer_bytes_cpu(struct trace_buffer *buffer, int cpu)
4222{
4223 struct ring_buffer_per_cpu *cpu_buffer;
4224 unsigned long ret;
4225
4226 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4227 return 0;
4228
4229 cpu_buffer = buffer->buffers[cpu];
4230 ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes;
4231
4232 return ret;
4233}
4234EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu);
4235
4236/**
4237 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
4238 * @buffer: The ring buffer
4239 * @cpu: The per CPU buffer to get the entries from.
4240 */
4241unsigned long ring_buffer_entries_cpu(struct trace_buffer *buffer, int cpu)
4242{
4243 struct ring_buffer_per_cpu *cpu_buffer;
4244
4245 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4246 return 0;
4247
4248 cpu_buffer = buffer->buffers[cpu];
4249
4250 return rb_num_of_entries(cpu_buffer);
4251}
4252EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
4253
4254/**
4255 * ring_buffer_overrun_cpu - get the number of overruns caused by the ring
4256 * buffer wrapping around (only if RB_FL_OVERWRITE is on).
4257 * @buffer: The ring buffer
4258 * @cpu: The per CPU buffer to get the number of overruns from
4259 */
4260unsigned long ring_buffer_overrun_cpu(struct trace_buffer *buffer, int cpu)
4261{
4262 struct ring_buffer_per_cpu *cpu_buffer;
4263 unsigned long ret;
4264
4265 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4266 return 0;
4267
4268 cpu_buffer = buffer->buffers[cpu];
4269 ret = local_read(&cpu_buffer->overrun);
4270
4271 return ret;
4272}
4273EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
4274
4275/**
4276 * ring_buffer_commit_overrun_cpu - get the number of overruns caused by
4277 * commits failing due to the buffer wrapping around while there are uncommitted
4278 * events, such as during an interrupt storm.
4279 * @buffer: The ring buffer
4280 * @cpu: The per CPU buffer to get the number of overruns from
4281 */
4282unsigned long
4283ring_buffer_commit_overrun_cpu(struct trace_buffer *buffer, int cpu)
4284{
4285 struct ring_buffer_per_cpu *cpu_buffer;
4286 unsigned long ret;
4287
4288 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4289 return 0;
4290
4291 cpu_buffer = buffer->buffers[cpu];
4292 ret = local_read(&cpu_buffer->commit_overrun);
4293
4294 return ret;
4295}
4296EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu);
4297
4298/**
4299 * ring_buffer_dropped_events_cpu - get the number of dropped events caused by
4300 * the ring buffer filling up (only if RB_FL_OVERWRITE is off).
4301 * @buffer: The ring buffer
4302 * @cpu: The per CPU buffer to get the number of overruns from
4303 */
4304unsigned long
4305ring_buffer_dropped_events_cpu(struct trace_buffer *buffer, int cpu)
4306{
4307 struct ring_buffer_per_cpu *cpu_buffer;
4308 unsigned long ret;
4309
4310 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4311 return 0;
4312
4313 cpu_buffer = buffer->buffers[cpu];
4314 ret = local_read(&cpu_buffer->dropped_events);
4315
4316 return ret;
4317}
4318EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu);
4319
4320/**
4321 * ring_buffer_read_events_cpu - get the number of events successfully read
4322 * @buffer: The ring buffer
4323 * @cpu: The per CPU buffer to get the number of events read
4324 */
4325unsigned long
4326ring_buffer_read_events_cpu(struct trace_buffer *buffer, int cpu)
4327{
4328 struct ring_buffer_per_cpu *cpu_buffer;
4329
4330 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4331 return 0;
4332
4333 cpu_buffer = buffer->buffers[cpu];
4334 return cpu_buffer->read;
4335}
4336EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu);
4337
4338/**
4339 * ring_buffer_entries - get the number of entries in a buffer
4340 * @buffer: The ring buffer
4341 *
4342 * Returns the total number of entries in the ring buffer
4343 * (all CPU entries)
4344 */
4345unsigned long ring_buffer_entries(struct trace_buffer *buffer)
4346{
4347 struct ring_buffer_per_cpu *cpu_buffer;
4348 unsigned long entries = 0;
4349 int cpu;
4350
4351 /* if you care about this being correct, lock the buffer */
4352 for_each_buffer_cpu(buffer, cpu) {
4353 cpu_buffer = buffer->buffers[cpu];
4354 entries += rb_num_of_entries(cpu_buffer);
4355 }
4356
4357 return entries;
4358}
4359EXPORT_SYMBOL_GPL(ring_buffer_entries);
4360
4361/**
4362 * ring_buffer_overruns - get the number of overruns in buffer
4363 * @buffer: The ring buffer
4364 *
4365 * Returns the total number of overruns in the ring buffer
4366 * (all CPU entries)
4367 */
4368unsigned long ring_buffer_overruns(struct trace_buffer *buffer)
4369{
4370 struct ring_buffer_per_cpu *cpu_buffer;
4371 unsigned long overruns = 0;
4372 int cpu;
4373
4374 /* if you care about this being correct, lock the buffer */
4375 for_each_buffer_cpu(buffer, cpu) {
4376 cpu_buffer = buffer->buffers[cpu];
4377 overruns += local_read(&cpu_buffer->overrun);
4378 }
4379
4380 return overruns;
4381}
4382EXPORT_SYMBOL_GPL(ring_buffer_overruns);
4383
4384static void rb_iter_reset(struct ring_buffer_iter *iter)
4385{
4386 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
4387
4388 /* Iterator usage is expected to have record disabled */
4389 iter->head_page = cpu_buffer->reader_page;
4390 iter->head = cpu_buffer->reader_page->read;
4391 iter->next_event = iter->head;
4392
4393 iter->cache_reader_page = iter->head_page;
4394 iter->cache_read = cpu_buffer->read;
4395
4396 if (iter->head) {
4397 iter->read_stamp = cpu_buffer->read_stamp;
4398 iter->page_stamp = cpu_buffer->reader_page->page->time_stamp;
4399 } else {
4400 iter->read_stamp = iter->head_page->page->time_stamp;
4401 iter->page_stamp = iter->read_stamp;
4402 }
4403}
4404
4405/**
4406 * ring_buffer_iter_reset - reset an iterator
4407 * @iter: The iterator to reset
4408 *
4409 * Resets the iterator, so that it will start from the beginning
4410 * again.
4411 */
4412void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
4413{
4414 struct ring_buffer_per_cpu *cpu_buffer;
4415 unsigned long flags;
4416
4417 if (!iter)
4418 return;
4419
4420 cpu_buffer = iter->cpu_buffer;
4421
4422 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
4423 rb_iter_reset(iter);
4424 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
4425}
4426EXPORT_SYMBOL_GPL(ring_buffer_iter_reset);
4427
4428/**
4429 * ring_buffer_iter_empty - check if an iterator has no more to read
4430 * @iter: The iterator to check
4431 */
4432int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
4433{
4434 struct ring_buffer_per_cpu *cpu_buffer;
4435 struct buffer_page *reader;
4436 struct buffer_page *head_page;
4437 struct buffer_page *commit_page;
4438 struct buffer_page *curr_commit_page;
4439 unsigned commit;
4440 u64 curr_commit_ts;
4441 u64 commit_ts;
4442
4443 cpu_buffer = iter->cpu_buffer;
4444 reader = cpu_buffer->reader_page;
4445 head_page = cpu_buffer->head_page;
4446 commit_page = cpu_buffer->commit_page;
4447 commit_ts = commit_page->page->time_stamp;
4448
4449 /*
4450 * When the writer goes across pages, it issues a cmpxchg which
4451 * is a mb(), which will synchronize with the rmb here.
4452 * (see rb_tail_page_update())
4453 */
4454 smp_rmb();
4455 commit = rb_page_commit(commit_page);
4456 /* We want to make sure that the commit page doesn't change */
4457 smp_rmb();
4458
4459 /* Make sure commit page didn't change */
4460 curr_commit_page = READ_ONCE(cpu_buffer->commit_page);
4461 curr_commit_ts = READ_ONCE(curr_commit_page->page->time_stamp);
4462
4463 /* If the commit page changed, then there's more data */
4464 if (curr_commit_page != commit_page ||
4465 curr_commit_ts != commit_ts)
4466 return 0;
4467
4468 /* Still racy, as it may return a false positive, but that's OK */
4469 return ((iter->head_page == commit_page && iter->head >= commit) ||
4470 (iter->head_page == reader && commit_page == head_page &&
4471 head_page->read == commit &&
4472 iter->head == rb_page_commit(cpu_buffer->reader_page)));
4473}
4474EXPORT_SYMBOL_GPL(ring_buffer_iter_empty);
4475
4476static void
4477rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
4478 struct ring_buffer_event *event)
4479{
4480 u64 delta;
4481
4482 switch (event->type_len) {
4483 case RINGBUF_TYPE_PADDING:
4484 return;
4485
4486 case RINGBUF_TYPE_TIME_EXTEND:
4487 delta = rb_event_time_stamp(event);
4488 cpu_buffer->read_stamp += delta;
4489 return;
4490
4491 case RINGBUF_TYPE_TIME_STAMP:
4492 delta = rb_event_time_stamp(event);
4493 delta = rb_fix_abs_ts(delta, cpu_buffer->read_stamp);
4494 cpu_buffer->read_stamp = delta;
4495 return;
4496
4497 case RINGBUF_TYPE_DATA:
4498 cpu_buffer->read_stamp += event->time_delta;
4499 return;
4500
4501 default:
4502 RB_WARN_ON(cpu_buffer, 1);
4503 }
4504}
4505
4506static void
4507rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
4508 struct ring_buffer_event *event)
4509{
4510 u64 delta;
4511
4512 switch (event->type_len) {
4513 case RINGBUF_TYPE_PADDING:
4514 return;
4515
4516 case RINGBUF_TYPE_TIME_EXTEND:
4517 delta = rb_event_time_stamp(event);
4518 iter->read_stamp += delta;
4519 return;
4520
4521 case RINGBUF_TYPE_TIME_STAMP:
4522 delta = rb_event_time_stamp(event);
4523 delta = rb_fix_abs_ts(delta, iter->read_stamp);
4524 iter->read_stamp = delta;
4525 return;
4526
4527 case RINGBUF_TYPE_DATA:
4528 iter->read_stamp += event->time_delta;
4529 return;
4530
4531 default:
4532 RB_WARN_ON(iter->cpu_buffer, 1);
4533 }
4534}
4535
4536static struct buffer_page *
4537rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
4538{
4539 struct buffer_page *reader = NULL;
4540 unsigned long overwrite;
4541 unsigned long flags;
4542 int nr_loops = 0;
4543 bool ret;
4544
4545 local_irq_save(flags);
4546 arch_spin_lock(&cpu_buffer->lock);
4547
4548 again:
4549 /*
4550 * This should normally only loop twice. But because the
4551 * start of the reader inserts an empty page, it causes
4552 * a case where we will loop three times. There should be no
4553 * reason to loop four times (that I know of).
4554 */
4555 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
4556 reader = NULL;
4557 goto out;
4558 }
4559
4560 reader = cpu_buffer->reader_page;
4561
4562 /* If there's more to read, return this page */
4563 if (cpu_buffer->reader_page->read < rb_page_size(reader))
4564 goto out;
4565
4566 /* Never should we have an index greater than the size */
4567 if (RB_WARN_ON(cpu_buffer,
4568 cpu_buffer->reader_page->read > rb_page_size(reader)))
4569 goto out;
4570
4571 /* check if we caught up to the tail */
4572 reader = NULL;
4573 if (cpu_buffer->commit_page == cpu_buffer->reader_page)
4574 goto out;
4575
4576 /* Don't bother swapping if the ring buffer is empty */
4577 if (rb_num_of_entries(cpu_buffer) == 0)
4578 goto out;
4579
4580 /*
4581 * Reset the reader page to size zero.
4582 */
4583 local_set(&cpu_buffer->reader_page->write, 0);
4584 local_set(&cpu_buffer->reader_page->entries, 0);
4585 local_set(&cpu_buffer->reader_page->page->commit, 0);
4586 cpu_buffer->reader_page->real_end = 0;
4587
4588 spin:
4589 /*
4590 * Splice the empty reader page into the list around the head.
4591 */
4592 reader = rb_set_head_page(cpu_buffer);
4593 if (!reader)
4594 goto out;
4595 cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next);
4596 cpu_buffer->reader_page->list.prev = reader->list.prev;
4597
4598 /*
4599 * cpu_buffer->pages just needs to point to the buffer, it
4600 * has no specific buffer page to point to. Lets move it out
4601 * of our way so we don't accidentally swap it.
4602 */
4603 cpu_buffer->pages = reader->list.prev;
4604
4605 /* The reader page will be pointing to the new head */
4606 rb_set_list_to_head(&cpu_buffer->reader_page->list);
4607
4608 /*
4609 * We want to make sure we read the overruns after we set up our
4610 * pointers to the next object. The writer side does a
4611 * cmpxchg to cross pages which acts as the mb on the writer
4612 * side. Note, the reader will constantly fail the swap
4613 * while the writer is updating the pointers, so this
4614 * guarantees that the overwrite recorded here is the one we
4615 * want to compare with the last_overrun.
4616 */
4617 smp_mb();
4618 overwrite = local_read(&(cpu_buffer->overrun));
4619
4620 /*
4621 * Here's the tricky part.
4622 *
4623 * We need to move the pointer past the header page.
4624 * But we can only do that if a writer is not currently
4625 * moving it. The page before the header page has the
4626 * flag bit '1' set if it is pointing to the page we want.
4627 * but if the writer is in the process of moving it
4628 * than it will be '2' or already moved '0'.
4629 */
4630
4631 ret = rb_head_page_replace(reader, cpu_buffer->reader_page);
4632
4633 /*
4634 * If we did not convert it, then we must try again.
4635 */
4636 if (!ret)
4637 goto spin;
4638
4639 /*
4640 * Yay! We succeeded in replacing the page.
4641 *
4642 * Now make the new head point back to the reader page.
4643 */
4644 rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list;
4645 rb_inc_page(&cpu_buffer->head_page);
4646
4647 local_inc(&cpu_buffer->pages_read);
4648
4649 /* Finally update the reader page to the new head */
4650 cpu_buffer->reader_page = reader;
4651 cpu_buffer->reader_page->read = 0;
4652
4653 if (overwrite != cpu_buffer->last_overrun) {
4654 cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun;
4655 cpu_buffer->last_overrun = overwrite;
4656 }
4657
4658 goto again;
4659
4660 out:
4661 /* Update the read_stamp on the first event */
4662 if (reader && reader->read == 0)
4663 cpu_buffer->read_stamp = reader->page->time_stamp;
4664
4665 arch_spin_unlock(&cpu_buffer->lock);
4666 local_irq_restore(flags);
4667
4668 /*
4669 * The writer has preempt disable, wait for it. But not forever
4670 * Although, 1 second is pretty much "forever"
4671 */
4672#define USECS_WAIT 1000000
4673 for (nr_loops = 0; nr_loops < USECS_WAIT; nr_loops++) {
4674 /* If the write is past the end of page, a writer is still updating it */
4675 if (likely(!reader || rb_page_write(reader) <= BUF_PAGE_SIZE))
4676 break;
4677
4678 udelay(1);
4679
4680 /* Get the latest version of the reader write value */
4681 smp_rmb();
4682 }
4683
4684 /* The writer is not moving forward? Something is wrong */
4685 if (RB_WARN_ON(cpu_buffer, nr_loops == USECS_WAIT))
4686 reader = NULL;
4687
4688 /*
4689 * Make sure we see any padding after the write update
4690 * (see rb_reset_tail()).
4691 *
4692 * In addition, a writer may be writing on the reader page
4693 * if the page has not been fully filled, so the read barrier
4694 * is also needed to make sure we see the content of what is
4695 * committed by the writer (see rb_set_commit_to_write()).
4696 */
4697 smp_rmb();
4698
4699
4700 return reader;
4701}
4702
4703static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
4704{
4705 struct ring_buffer_event *event;
4706 struct buffer_page *reader;
4707 unsigned length;
4708
4709 reader = rb_get_reader_page(cpu_buffer);
4710
4711 /* This function should not be called when buffer is empty */
4712 if (RB_WARN_ON(cpu_buffer, !reader))
4713 return;
4714
4715 event = rb_reader_event(cpu_buffer);
4716
4717 if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
4718 cpu_buffer->read++;
4719
4720 rb_update_read_stamp(cpu_buffer, event);
4721
4722 length = rb_event_length(event);
4723 cpu_buffer->reader_page->read += length;
4724}
4725
4726static void rb_advance_iter(struct ring_buffer_iter *iter)
4727{
4728 struct ring_buffer_per_cpu *cpu_buffer;
4729
4730 cpu_buffer = iter->cpu_buffer;
4731
4732 /* If head == next_event then we need to jump to the next event */
4733 if (iter->head == iter->next_event) {
4734 /* If the event gets overwritten again, there's nothing to do */
4735 if (rb_iter_head_event(iter) == NULL)
4736 return;
4737 }
4738
4739 iter->head = iter->next_event;
4740
4741 /*
4742 * Check if we are at the end of the buffer.
4743 */
4744 if (iter->next_event >= rb_page_size(iter->head_page)) {
4745 /* discarded commits can make the page empty */
4746 if (iter->head_page == cpu_buffer->commit_page)
4747 return;
4748 rb_inc_iter(iter);
4749 return;
4750 }
4751
4752 rb_update_iter_read_stamp(iter, iter->event);
4753}
4754
4755static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer)
4756{
4757 return cpu_buffer->lost_events;
4758}
4759
4760static struct ring_buffer_event *
4761rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts,
4762 unsigned long *lost_events)
4763{
4764 struct ring_buffer_event *event;
4765 struct buffer_page *reader;
4766 int nr_loops = 0;
4767
4768 if (ts)
4769 *ts = 0;
4770 again:
4771 /*
4772 * We repeat when a time extend is encountered.
4773 * Since the time extend is always attached to a data event,
4774 * we should never loop more than once.
4775 * (We never hit the following condition more than twice).
4776 */
4777 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2))
4778 return NULL;
4779
4780 reader = rb_get_reader_page(cpu_buffer);
4781 if (!reader)
4782 return NULL;
4783
4784 event = rb_reader_event(cpu_buffer);
4785
4786 switch (event->type_len) {
4787 case RINGBUF_TYPE_PADDING:
4788 if (rb_null_event(event))
4789 RB_WARN_ON(cpu_buffer, 1);
4790 /*
4791 * Because the writer could be discarding every
4792 * event it creates (which would probably be bad)
4793 * if we were to go back to "again" then we may never
4794 * catch up, and will trigger the warn on, or lock
4795 * the box. Return the padding, and we will release
4796 * the current locks, and try again.
4797 */
4798 return event;
4799
4800 case RINGBUF_TYPE_TIME_EXTEND:
4801 /* Internal data, OK to advance */
4802 rb_advance_reader(cpu_buffer);
4803 goto again;
4804
4805 case RINGBUF_TYPE_TIME_STAMP:
4806 if (ts) {
4807 *ts = rb_event_time_stamp(event);
4808 *ts = rb_fix_abs_ts(*ts, reader->page->time_stamp);
4809 ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
4810 cpu_buffer->cpu, ts);
4811 }
4812 /* Internal data, OK to advance */
4813 rb_advance_reader(cpu_buffer);
4814 goto again;
4815
4816 case RINGBUF_TYPE_DATA:
4817 if (ts && !(*ts)) {
4818 *ts = cpu_buffer->read_stamp + event->time_delta;
4819 ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
4820 cpu_buffer->cpu, ts);
4821 }
4822 if (lost_events)
4823 *lost_events = rb_lost_events(cpu_buffer);
4824 return event;
4825
4826 default:
4827 RB_WARN_ON(cpu_buffer, 1);
4828 }
4829
4830 return NULL;
4831}
4832EXPORT_SYMBOL_GPL(ring_buffer_peek);
4833
4834static struct ring_buffer_event *
4835rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
4836{
4837 struct trace_buffer *buffer;
4838 struct ring_buffer_per_cpu *cpu_buffer;
4839 struct ring_buffer_event *event;
4840 int nr_loops = 0;
4841
4842 if (ts)
4843 *ts = 0;
4844
4845 cpu_buffer = iter->cpu_buffer;
4846 buffer = cpu_buffer->buffer;
4847
4848 /*
4849 * Check if someone performed a consuming read to
4850 * the buffer. A consuming read invalidates the iterator
4851 * and we need to reset the iterator in this case.
4852 */
4853 if (unlikely(iter->cache_read != cpu_buffer->read ||
4854 iter->cache_reader_page != cpu_buffer->reader_page))
4855 rb_iter_reset(iter);
4856
4857 again:
4858 if (ring_buffer_iter_empty(iter))
4859 return NULL;
4860
4861 /*
4862 * As the writer can mess with what the iterator is trying
4863 * to read, just give up if we fail to get an event after
4864 * three tries. The iterator is not as reliable when reading
4865 * the ring buffer with an active write as the consumer is.
4866 * Do not warn if the three failures is reached.
4867 */
4868 if (++nr_loops > 3)
4869 return NULL;
4870
4871 if (rb_per_cpu_empty(cpu_buffer))
4872 return NULL;
4873
4874 if (iter->head >= rb_page_size(iter->head_page)) {
4875 rb_inc_iter(iter);
4876 goto again;
4877 }
4878
4879 event = rb_iter_head_event(iter);
4880 if (!event)
4881 goto again;
4882
4883 switch (event->type_len) {
4884 case RINGBUF_TYPE_PADDING:
4885 if (rb_null_event(event)) {
4886 rb_inc_iter(iter);
4887 goto again;
4888 }
4889 rb_advance_iter(iter);
4890 return event;
4891
4892 case RINGBUF_TYPE_TIME_EXTEND:
4893 /* Internal data, OK to advance */
4894 rb_advance_iter(iter);
4895 goto again;
4896
4897 case RINGBUF_TYPE_TIME_STAMP:
4898 if (ts) {
4899 *ts = rb_event_time_stamp(event);
4900 *ts = rb_fix_abs_ts(*ts, iter->head_page->page->time_stamp);
4901 ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
4902 cpu_buffer->cpu, ts);
4903 }
4904 /* Internal data, OK to advance */
4905 rb_advance_iter(iter);
4906 goto again;
4907
4908 case RINGBUF_TYPE_DATA:
4909 if (ts && !(*ts)) {
4910 *ts = iter->read_stamp + event->time_delta;
4911 ring_buffer_normalize_time_stamp(buffer,
4912 cpu_buffer->cpu, ts);
4913 }
4914 return event;
4915
4916 default:
4917 RB_WARN_ON(cpu_buffer, 1);
4918 }
4919
4920 return NULL;
4921}
4922EXPORT_SYMBOL_GPL(ring_buffer_iter_peek);
4923
4924static inline bool rb_reader_lock(struct ring_buffer_per_cpu *cpu_buffer)
4925{
4926 if (likely(!in_nmi())) {
4927 raw_spin_lock(&cpu_buffer->reader_lock);
4928 return true;
4929 }
4930
4931 /*
4932 * If an NMI die dumps out the content of the ring buffer
4933 * trylock must be used to prevent a deadlock if the NMI
4934 * preempted a task that holds the ring buffer locks. If
4935 * we get the lock then all is fine, if not, then continue
4936 * to do the read, but this can corrupt the ring buffer,
4937 * so it must be permanently disabled from future writes.
4938 * Reading from NMI is a oneshot deal.
4939 */
4940 if (raw_spin_trylock(&cpu_buffer->reader_lock))
4941 return true;
4942
4943 /* Continue without locking, but disable the ring buffer */
4944 atomic_inc(&cpu_buffer->record_disabled);
4945 return false;
4946}
4947
4948static inline void
4949rb_reader_unlock(struct ring_buffer_per_cpu *cpu_buffer, bool locked)
4950{
4951 if (likely(locked))
4952 raw_spin_unlock(&cpu_buffer->reader_lock);
4953}
4954
4955/**
4956 * ring_buffer_peek - peek at the next event to be read
4957 * @buffer: The ring buffer to read
4958 * @cpu: The cpu to peak at
4959 * @ts: The timestamp counter of this event.
4960 * @lost_events: a variable to store if events were lost (may be NULL)
4961 *
4962 * This will return the event that will be read next, but does
4963 * not consume the data.
4964 */
4965struct ring_buffer_event *
4966ring_buffer_peek(struct trace_buffer *buffer, int cpu, u64 *ts,
4967 unsigned long *lost_events)
4968{
4969 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
4970 struct ring_buffer_event *event;
4971 unsigned long flags;
4972 bool dolock;
4973
4974 if (!cpumask_test_cpu(cpu, buffer->cpumask))
4975 return NULL;
4976
4977 again:
4978 local_irq_save(flags);
4979 dolock = rb_reader_lock(cpu_buffer);
4980 event = rb_buffer_peek(cpu_buffer, ts, lost_events);
4981 if (event && event->type_len == RINGBUF_TYPE_PADDING)
4982 rb_advance_reader(cpu_buffer);
4983 rb_reader_unlock(cpu_buffer, dolock);
4984 local_irq_restore(flags);
4985
4986 if (event && event->type_len == RINGBUF_TYPE_PADDING)
4987 goto again;
4988
4989 return event;
4990}
4991
4992/** ring_buffer_iter_dropped - report if there are dropped events
4993 * @iter: The ring buffer iterator
4994 *
4995 * Returns true if there was dropped events since the last peek.
4996 */
4997bool ring_buffer_iter_dropped(struct ring_buffer_iter *iter)
4998{
4999 bool ret = iter->missed_events != 0;
5000
5001 iter->missed_events = 0;
5002 return ret;
5003}
5004EXPORT_SYMBOL_GPL(ring_buffer_iter_dropped);
5005
5006/**
5007 * ring_buffer_iter_peek - peek at the next event to be read
5008 * @iter: The ring buffer iterator
5009 * @ts: The timestamp counter of this event.
5010 *
5011 * This will return the event that will be read next, but does
5012 * not increment the iterator.
5013 */
5014struct ring_buffer_event *
5015ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
5016{
5017 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
5018 struct ring_buffer_event *event;
5019 unsigned long flags;
5020
5021 again:
5022 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5023 event = rb_iter_peek(iter, ts);
5024 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5025
5026 if (event && event->type_len == RINGBUF_TYPE_PADDING)
5027 goto again;
5028
5029 return event;
5030}
5031
5032/**
5033 * ring_buffer_consume - return an event and consume it
5034 * @buffer: The ring buffer to get the next event from
5035 * @cpu: the cpu to read the buffer from
5036 * @ts: a variable to store the timestamp (may be NULL)
5037 * @lost_events: a variable to store if events were lost (may be NULL)
5038 *
5039 * Returns the next event in the ring buffer, and that event is consumed.
5040 * Meaning, that sequential reads will keep returning a different event,
5041 * and eventually empty the ring buffer if the producer is slower.
5042 */
5043struct ring_buffer_event *
5044ring_buffer_consume(struct trace_buffer *buffer, int cpu, u64 *ts,
5045 unsigned long *lost_events)
5046{
5047 struct ring_buffer_per_cpu *cpu_buffer;
5048 struct ring_buffer_event *event = NULL;
5049 unsigned long flags;
5050 bool dolock;
5051
5052 again:
5053 /* might be called in atomic */
5054 preempt_disable();
5055
5056 if (!cpumask_test_cpu(cpu, buffer->cpumask))
5057 goto out;
5058
5059 cpu_buffer = buffer->buffers[cpu];
5060 local_irq_save(flags);
5061 dolock = rb_reader_lock(cpu_buffer);
5062
5063 event = rb_buffer_peek(cpu_buffer, ts, lost_events);
5064 if (event) {
5065 cpu_buffer->lost_events = 0;
5066 rb_advance_reader(cpu_buffer);
5067 }
5068
5069 rb_reader_unlock(cpu_buffer, dolock);
5070 local_irq_restore(flags);
5071
5072 out:
5073 preempt_enable();
5074
5075 if (event && event->type_len == RINGBUF_TYPE_PADDING)
5076 goto again;
5077
5078 return event;
5079}
5080EXPORT_SYMBOL_GPL(ring_buffer_consume);
5081
5082/**
5083 * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer
5084 * @buffer: The ring buffer to read from
5085 * @cpu: The cpu buffer to iterate over
5086 * @flags: gfp flags to use for memory allocation
5087 *
5088 * This performs the initial preparations necessary to iterate
5089 * through the buffer. Memory is allocated, buffer recording
5090 * is disabled, and the iterator pointer is returned to the caller.
5091 *
5092 * Disabling buffer recording prevents the reading from being
5093 * corrupted. This is not a consuming read, so a producer is not
5094 * expected.
5095 *
5096 * After a sequence of ring_buffer_read_prepare calls, the user is
5097 * expected to make at least one call to ring_buffer_read_prepare_sync.
5098 * Afterwards, ring_buffer_read_start is invoked to get things going
5099 * for real.
5100 *
5101 * This overall must be paired with ring_buffer_read_finish.
5102 */
5103struct ring_buffer_iter *
5104ring_buffer_read_prepare(struct trace_buffer *buffer, int cpu, gfp_t flags)
5105{
5106 struct ring_buffer_per_cpu *cpu_buffer;
5107 struct ring_buffer_iter *iter;
5108
5109 if (!cpumask_test_cpu(cpu, buffer->cpumask))
5110 return NULL;
5111
5112 iter = kzalloc(sizeof(*iter), flags);
5113 if (!iter)
5114 return NULL;
5115
5116 iter->event = kmalloc(BUF_MAX_DATA_SIZE, flags);
5117 if (!iter->event) {
5118 kfree(iter);
5119 return NULL;
5120 }
5121
5122 cpu_buffer = buffer->buffers[cpu];
5123
5124 iter->cpu_buffer = cpu_buffer;
5125
5126 atomic_inc(&cpu_buffer->resize_disabled);
5127
5128 return iter;
5129}
5130EXPORT_SYMBOL_GPL(ring_buffer_read_prepare);
5131
5132/**
5133 * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls
5134 *
5135 * All previously invoked ring_buffer_read_prepare calls to prepare
5136 * iterators will be synchronized. Afterwards, read_buffer_read_start
5137 * calls on those iterators are allowed.
5138 */
5139void
5140ring_buffer_read_prepare_sync(void)
5141{
5142 synchronize_rcu();
5143}
5144EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync);
5145
5146/**
5147 * ring_buffer_read_start - start a non consuming read of the buffer
5148 * @iter: The iterator returned by ring_buffer_read_prepare
5149 *
5150 * This finalizes the startup of an iteration through the buffer.
5151 * The iterator comes from a call to ring_buffer_read_prepare and
5152 * an intervening ring_buffer_read_prepare_sync must have been
5153 * performed.
5154 *
5155 * Must be paired with ring_buffer_read_finish.
5156 */
5157void
5158ring_buffer_read_start(struct ring_buffer_iter *iter)
5159{
5160 struct ring_buffer_per_cpu *cpu_buffer;
5161 unsigned long flags;
5162
5163 if (!iter)
5164 return;
5165
5166 cpu_buffer = iter->cpu_buffer;
5167
5168 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5169 arch_spin_lock(&cpu_buffer->lock);
5170 rb_iter_reset(iter);
5171 arch_spin_unlock(&cpu_buffer->lock);
5172 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5173}
5174EXPORT_SYMBOL_GPL(ring_buffer_read_start);
5175
5176/**
5177 * ring_buffer_read_finish - finish reading the iterator of the buffer
5178 * @iter: The iterator retrieved by ring_buffer_start
5179 *
5180 * This re-enables the recording to the buffer, and frees the
5181 * iterator.
5182 */
5183void
5184ring_buffer_read_finish(struct ring_buffer_iter *iter)
5185{
5186 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
5187 unsigned long flags;
5188
5189 /*
5190 * Ring buffer is disabled from recording, here's a good place
5191 * to check the integrity of the ring buffer.
5192 * Must prevent readers from trying to read, as the check
5193 * clears the HEAD page and readers require it.
5194 */
5195 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5196 rb_check_pages(cpu_buffer);
5197 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5198
5199 atomic_dec(&cpu_buffer->resize_disabled);
5200 kfree(iter->event);
5201 kfree(iter);
5202}
5203EXPORT_SYMBOL_GPL(ring_buffer_read_finish);
5204
5205/**
5206 * ring_buffer_iter_advance - advance the iterator to the next location
5207 * @iter: The ring buffer iterator
5208 *
5209 * Move the location of the iterator such that the next read will
5210 * be the next location of the iterator.
5211 */
5212void ring_buffer_iter_advance(struct ring_buffer_iter *iter)
5213{
5214 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
5215 unsigned long flags;
5216
5217 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5218
5219 rb_advance_iter(iter);
5220
5221 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5222}
5223EXPORT_SYMBOL_GPL(ring_buffer_iter_advance);
5224
5225/**
5226 * ring_buffer_size - return the size of the ring buffer (in bytes)
5227 * @buffer: The ring buffer.
5228 * @cpu: The CPU to get ring buffer size from.
5229 */
5230unsigned long ring_buffer_size(struct trace_buffer *buffer, int cpu)
5231{
5232 /*
5233 * Earlier, this method returned
5234 * BUF_PAGE_SIZE * buffer->nr_pages
5235 * Since the nr_pages field is now removed, we have converted this to
5236 * return the per cpu buffer value.
5237 */
5238 if (!cpumask_test_cpu(cpu, buffer->cpumask))
5239 return 0;
5240
5241 return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages;
5242}
5243EXPORT_SYMBOL_GPL(ring_buffer_size);
5244
5245static void
5246rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
5247{
5248 rb_head_page_deactivate(cpu_buffer);
5249
5250 cpu_buffer->head_page
5251 = list_entry(cpu_buffer->pages, struct buffer_page, list);
5252 local_set(&cpu_buffer->head_page->write, 0);
5253 local_set(&cpu_buffer->head_page->entries, 0);
5254 local_set(&cpu_buffer->head_page->page->commit, 0);
5255
5256 cpu_buffer->head_page->read = 0;
5257
5258 cpu_buffer->tail_page = cpu_buffer->head_page;
5259 cpu_buffer->commit_page = cpu_buffer->head_page;
5260
5261 INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
5262 INIT_LIST_HEAD(&cpu_buffer->new_pages);
5263 local_set(&cpu_buffer->reader_page->write, 0);
5264 local_set(&cpu_buffer->reader_page->entries, 0);
5265 local_set(&cpu_buffer->reader_page->page->commit, 0);
5266 cpu_buffer->reader_page->read = 0;
5267
5268 local_set(&cpu_buffer->entries_bytes, 0);
5269 local_set(&cpu_buffer->overrun, 0);
5270 local_set(&cpu_buffer->commit_overrun, 0);
5271 local_set(&cpu_buffer->dropped_events, 0);
5272 local_set(&cpu_buffer->entries, 0);
5273 local_set(&cpu_buffer->committing, 0);
5274 local_set(&cpu_buffer->commits, 0);
5275 local_set(&cpu_buffer->pages_touched, 0);
5276 local_set(&cpu_buffer->pages_lost, 0);
5277 local_set(&cpu_buffer->pages_read, 0);
5278 cpu_buffer->last_pages_touch = 0;
5279 cpu_buffer->shortest_full = 0;
5280 cpu_buffer->read = 0;
5281 cpu_buffer->read_bytes = 0;
5282
5283 rb_time_set(&cpu_buffer->write_stamp, 0);
5284 rb_time_set(&cpu_buffer->before_stamp, 0);
5285
5286 memset(cpu_buffer->event_stamp, 0, sizeof(cpu_buffer->event_stamp));
5287
5288 cpu_buffer->lost_events = 0;
5289 cpu_buffer->last_overrun = 0;
5290
5291 rb_head_page_activate(cpu_buffer);
5292}
5293
5294/* Must have disabled the cpu buffer then done a synchronize_rcu */
5295static void reset_disabled_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
5296{
5297 unsigned long flags;
5298
5299 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5300
5301 if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing)))
5302 goto out;
5303
5304 arch_spin_lock(&cpu_buffer->lock);
5305
5306 rb_reset_cpu(cpu_buffer);
5307
5308 arch_spin_unlock(&cpu_buffer->lock);
5309
5310 out:
5311 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5312}
5313
5314/**
5315 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
5316 * @buffer: The ring buffer to reset a per cpu buffer of
5317 * @cpu: The CPU buffer to be reset
5318 */
5319void ring_buffer_reset_cpu(struct trace_buffer *buffer, int cpu)
5320{
5321 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
5322
5323 if (!cpumask_test_cpu(cpu, buffer->cpumask))
5324 return;
5325
5326 /* prevent another thread from changing buffer sizes */
5327 mutex_lock(&buffer->mutex);
5328
5329 atomic_inc(&cpu_buffer->resize_disabled);
5330 atomic_inc(&cpu_buffer->record_disabled);
5331
5332 /* Make sure all commits have finished */
5333 synchronize_rcu();
5334
5335 reset_disabled_cpu_buffer(cpu_buffer);
5336
5337 atomic_dec(&cpu_buffer->record_disabled);
5338 atomic_dec(&cpu_buffer->resize_disabled);
5339
5340 mutex_unlock(&buffer->mutex);
5341}
5342EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
5343
5344/* Flag to ensure proper resetting of atomic variables */
5345#define RESET_BIT (1 << 30)
5346
5347/**
5348 * ring_buffer_reset_online_cpus - reset a ring buffer per CPU buffer
5349 * @buffer: The ring buffer to reset a per cpu buffer of
5350 * @cpu: The CPU buffer to be reset
5351 */
5352void ring_buffer_reset_online_cpus(struct trace_buffer *buffer)
5353{
5354 struct ring_buffer_per_cpu *cpu_buffer;
5355 int cpu;
5356
5357 /* prevent another thread from changing buffer sizes */
5358 mutex_lock(&buffer->mutex);
5359
5360 for_each_online_buffer_cpu(buffer, cpu) {
5361 cpu_buffer = buffer->buffers[cpu];
5362
5363 atomic_add(RESET_BIT, &cpu_buffer->resize_disabled);
5364 atomic_inc(&cpu_buffer->record_disabled);
5365 }
5366
5367 /* Make sure all commits have finished */
5368 synchronize_rcu();
5369
5370 for_each_buffer_cpu(buffer, cpu) {
5371 cpu_buffer = buffer->buffers[cpu];
5372
5373 /*
5374 * If a CPU came online during the synchronize_rcu(), then
5375 * ignore it.
5376 */
5377 if (!(atomic_read(&cpu_buffer->resize_disabled) & RESET_BIT))
5378 continue;
5379
5380 reset_disabled_cpu_buffer(cpu_buffer);
5381
5382 atomic_dec(&cpu_buffer->record_disabled);
5383 atomic_sub(RESET_BIT, &cpu_buffer->resize_disabled);
5384 }
5385
5386 mutex_unlock(&buffer->mutex);
5387}
5388
5389/**
5390 * ring_buffer_reset - reset a ring buffer
5391 * @buffer: The ring buffer to reset all cpu buffers
5392 */
5393void ring_buffer_reset(struct trace_buffer *buffer)
5394{
5395 struct ring_buffer_per_cpu *cpu_buffer;
5396 int cpu;
5397
5398 /* prevent another thread from changing buffer sizes */
5399 mutex_lock(&buffer->mutex);
5400
5401 for_each_buffer_cpu(buffer, cpu) {
5402 cpu_buffer = buffer->buffers[cpu];
5403
5404 atomic_inc(&cpu_buffer->resize_disabled);
5405 atomic_inc(&cpu_buffer->record_disabled);
5406 }
5407
5408 /* Make sure all commits have finished */
5409 synchronize_rcu();
5410
5411 for_each_buffer_cpu(buffer, cpu) {
5412 cpu_buffer = buffer->buffers[cpu];
5413
5414 reset_disabled_cpu_buffer(cpu_buffer);
5415
5416 atomic_dec(&cpu_buffer->record_disabled);
5417 atomic_dec(&cpu_buffer->resize_disabled);
5418 }
5419
5420 mutex_unlock(&buffer->mutex);
5421}
5422EXPORT_SYMBOL_GPL(ring_buffer_reset);
5423
5424/**
5425 * ring_buffer_empty - is the ring buffer empty?
5426 * @buffer: The ring buffer to test
5427 */
5428bool ring_buffer_empty(struct trace_buffer *buffer)
5429{
5430 struct ring_buffer_per_cpu *cpu_buffer;
5431 unsigned long flags;
5432 bool dolock;
5433 bool ret;
5434 int cpu;
5435
5436 /* yes this is racy, but if you don't like the race, lock the buffer */
5437 for_each_buffer_cpu(buffer, cpu) {
5438 cpu_buffer = buffer->buffers[cpu];
5439 local_irq_save(flags);
5440 dolock = rb_reader_lock(cpu_buffer);
5441 ret = rb_per_cpu_empty(cpu_buffer);
5442 rb_reader_unlock(cpu_buffer, dolock);
5443 local_irq_restore(flags);
5444
5445 if (!ret)
5446 return false;
5447 }
5448
5449 return true;
5450}
5451EXPORT_SYMBOL_GPL(ring_buffer_empty);
5452
5453/**
5454 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
5455 * @buffer: The ring buffer
5456 * @cpu: The CPU buffer to test
5457 */
5458bool ring_buffer_empty_cpu(struct trace_buffer *buffer, int cpu)
5459{
5460 struct ring_buffer_per_cpu *cpu_buffer;
5461 unsigned long flags;
5462 bool dolock;
5463 bool ret;
5464
5465 if (!cpumask_test_cpu(cpu, buffer->cpumask))
5466 return true;
5467
5468 cpu_buffer = buffer->buffers[cpu];
5469 local_irq_save(flags);
5470 dolock = rb_reader_lock(cpu_buffer);
5471 ret = rb_per_cpu_empty(cpu_buffer);
5472 rb_reader_unlock(cpu_buffer, dolock);
5473 local_irq_restore(flags);
5474
5475 return ret;
5476}
5477EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
5478
5479#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
5480/**
5481 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
5482 * @buffer_a: One buffer to swap with
5483 * @buffer_b: The other buffer to swap with
5484 * @cpu: the CPU of the buffers to swap
5485 *
5486 * This function is useful for tracers that want to take a "snapshot"
5487 * of a CPU buffer and has another back up buffer lying around.
5488 * it is expected that the tracer handles the cpu buffer not being
5489 * used at the moment.
5490 */
5491int ring_buffer_swap_cpu(struct trace_buffer *buffer_a,
5492 struct trace_buffer *buffer_b, int cpu)
5493{
5494 struct ring_buffer_per_cpu *cpu_buffer_a;
5495 struct ring_buffer_per_cpu *cpu_buffer_b;
5496 int ret = -EINVAL;
5497
5498 if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
5499 !cpumask_test_cpu(cpu, buffer_b->cpumask))
5500 goto out;
5501
5502 cpu_buffer_a = buffer_a->buffers[cpu];
5503 cpu_buffer_b = buffer_b->buffers[cpu];
5504
5505 /* At least make sure the two buffers are somewhat the same */
5506 if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages)
5507 goto out;
5508
5509 ret = -EAGAIN;
5510
5511 if (atomic_read(&buffer_a->record_disabled))
5512 goto out;
5513
5514 if (atomic_read(&buffer_b->record_disabled))
5515 goto out;
5516
5517 if (atomic_read(&cpu_buffer_a->record_disabled))
5518 goto out;
5519
5520 if (atomic_read(&cpu_buffer_b->record_disabled))
5521 goto out;
5522
5523 /*
5524 * We can't do a synchronize_rcu here because this
5525 * function can be called in atomic context.
5526 * Normally this will be called from the same CPU as cpu.
5527 * If not it's up to the caller to protect this.
5528 */
5529 atomic_inc(&cpu_buffer_a->record_disabled);
5530 atomic_inc(&cpu_buffer_b->record_disabled);
5531
5532 ret = -EBUSY;
5533 if (local_read(&cpu_buffer_a->committing))
5534 goto out_dec;
5535 if (local_read(&cpu_buffer_b->committing))
5536 goto out_dec;
5537
5538 buffer_a->buffers[cpu] = cpu_buffer_b;
5539 buffer_b->buffers[cpu] = cpu_buffer_a;
5540
5541 cpu_buffer_b->buffer = buffer_a;
5542 cpu_buffer_a->buffer = buffer_b;
5543
5544 ret = 0;
5545
5546out_dec:
5547 atomic_dec(&cpu_buffer_a->record_disabled);
5548 atomic_dec(&cpu_buffer_b->record_disabled);
5549out:
5550 return ret;
5551}
5552EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
5553#endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */
5554
5555/**
5556 * ring_buffer_alloc_read_page - allocate a page to read from buffer
5557 * @buffer: the buffer to allocate for.
5558 * @cpu: the cpu buffer to allocate.
5559 *
5560 * This function is used in conjunction with ring_buffer_read_page.
5561 * When reading a full page from the ring buffer, these functions
5562 * can be used to speed up the process. The calling function should
5563 * allocate a few pages first with this function. Then when it
5564 * needs to get pages from the ring buffer, it passes the result
5565 * of this function into ring_buffer_read_page, which will swap
5566 * the page that was allocated, with the read page of the buffer.
5567 *
5568 * Returns:
5569 * The page allocated, or ERR_PTR
5570 */
5571void *ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
5572{
5573 struct ring_buffer_per_cpu *cpu_buffer;
5574 struct buffer_data_page *bpage = NULL;
5575 unsigned long flags;
5576 struct page *page;
5577
5578 if (!cpumask_test_cpu(cpu, buffer->cpumask))
5579 return ERR_PTR(-ENODEV);
5580
5581 cpu_buffer = buffer->buffers[cpu];
5582 local_irq_save(flags);
5583 arch_spin_lock(&cpu_buffer->lock);
5584
5585 if (cpu_buffer->free_page) {
5586 bpage = cpu_buffer->free_page;
5587 cpu_buffer->free_page = NULL;
5588 }
5589
5590 arch_spin_unlock(&cpu_buffer->lock);
5591 local_irq_restore(flags);
5592
5593 if (bpage)
5594 goto out;
5595
5596 page = alloc_pages_node(cpu_to_node(cpu),
5597 GFP_KERNEL | __GFP_NORETRY, 0);
5598 if (!page)
5599 return ERR_PTR(-ENOMEM);
5600
5601 bpage = page_address(page);
5602
5603 out:
5604 rb_init_page(bpage);
5605
5606 return bpage;
5607}
5608EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page);
5609
5610/**
5611 * ring_buffer_free_read_page - free an allocated read page
5612 * @buffer: the buffer the page was allocate for
5613 * @cpu: the cpu buffer the page came from
5614 * @data: the page to free
5615 *
5616 * Free a page allocated from ring_buffer_alloc_read_page.
5617 */
5618void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu, void *data)
5619{
5620 struct ring_buffer_per_cpu *cpu_buffer;
5621 struct buffer_data_page *bpage = data;
5622 struct page *page = virt_to_page(bpage);
5623 unsigned long flags;
5624
5625 if (!buffer || !buffer->buffers || !buffer->buffers[cpu])
5626 return;
5627
5628 cpu_buffer = buffer->buffers[cpu];
5629
5630 /* If the page is still in use someplace else, we can't reuse it */
5631 if (page_ref_count(page) > 1)
5632 goto out;
5633
5634 local_irq_save(flags);
5635 arch_spin_lock(&cpu_buffer->lock);
5636
5637 if (!cpu_buffer->free_page) {
5638 cpu_buffer->free_page = bpage;
5639 bpage = NULL;
5640 }
5641
5642 arch_spin_unlock(&cpu_buffer->lock);
5643 local_irq_restore(flags);
5644
5645 out:
5646 free_page((unsigned long)bpage);
5647}
5648EXPORT_SYMBOL_GPL(ring_buffer_free_read_page);
5649
5650/**
5651 * ring_buffer_read_page - extract a page from the ring buffer
5652 * @buffer: buffer to extract from
5653 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
5654 * @len: amount to extract
5655 * @cpu: the cpu of the buffer to extract
5656 * @full: should the extraction only happen when the page is full.
5657 *
5658 * This function will pull out a page from the ring buffer and consume it.
5659 * @data_page must be the address of the variable that was returned
5660 * from ring_buffer_alloc_read_page. This is because the page might be used
5661 * to swap with a page in the ring buffer.
5662 *
5663 * for example:
5664 * rpage = ring_buffer_alloc_read_page(buffer, cpu);
5665 * if (IS_ERR(rpage))
5666 * return PTR_ERR(rpage);
5667 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
5668 * if (ret >= 0)
5669 * process_page(rpage, ret);
5670 *
5671 * When @full is set, the function will not return true unless
5672 * the writer is off the reader page.
5673 *
5674 * Note: it is up to the calling functions to handle sleeps and wakeups.
5675 * The ring buffer can be used anywhere in the kernel and can not
5676 * blindly call wake_up. The layer that uses the ring buffer must be
5677 * responsible for that.
5678 *
5679 * Returns:
5680 * >=0 if data has been transferred, returns the offset of consumed data.
5681 * <0 if no data has been transferred.
5682 */
5683int ring_buffer_read_page(struct trace_buffer *buffer,
5684 void **data_page, size_t len, int cpu, int full)
5685{
5686 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
5687 struct ring_buffer_event *event;
5688 struct buffer_data_page *bpage;
5689 struct buffer_page *reader;
5690 unsigned long missed_events;
5691 unsigned long flags;
5692 unsigned int commit;
5693 unsigned int read;
5694 u64 save_timestamp;
5695 int ret = -1;
5696
5697 if (!cpumask_test_cpu(cpu, buffer->cpumask))
5698 goto out;
5699
5700 /*
5701 * If len is not big enough to hold the page header, then
5702 * we can not copy anything.
5703 */
5704 if (len <= BUF_PAGE_HDR_SIZE)
5705 goto out;
5706
5707 len -= BUF_PAGE_HDR_SIZE;
5708
5709 if (!data_page)
5710 goto out;
5711
5712 bpage = *data_page;
5713 if (!bpage)
5714 goto out;
5715
5716 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5717
5718 reader = rb_get_reader_page(cpu_buffer);
5719 if (!reader)
5720 goto out_unlock;
5721
5722 event = rb_reader_event(cpu_buffer);
5723
5724 read = reader->read;
5725 commit = rb_page_commit(reader);
5726
5727 /* Check if any events were dropped */
5728 missed_events = cpu_buffer->lost_events;
5729
5730 /*
5731 * If this page has been partially read or
5732 * if len is not big enough to read the rest of the page or
5733 * a writer is still on the page, then
5734 * we must copy the data from the page to the buffer.
5735 * Otherwise, we can simply swap the page with the one passed in.
5736 */
5737 if (read || (len < (commit - read)) ||
5738 cpu_buffer->reader_page == cpu_buffer->commit_page) {
5739 struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
5740 unsigned int rpos = read;
5741 unsigned int pos = 0;
5742 unsigned int size;
5743
5744 /*
5745 * If a full page is expected, this can still be returned
5746 * if there's been a previous partial read and the
5747 * rest of the page can be read and the commit page is off
5748 * the reader page.
5749 */
5750 if (full &&
5751 (!read || (len < (commit - read)) ||
5752 cpu_buffer->reader_page == cpu_buffer->commit_page))
5753 goto out_unlock;
5754
5755 if (len > (commit - read))
5756 len = (commit - read);
5757
5758 /* Always keep the time extend and data together */
5759 size = rb_event_ts_length(event);
5760
5761 if (len < size)
5762 goto out_unlock;
5763
5764 /* save the current timestamp, since the user will need it */
5765 save_timestamp = cpu_buffer->read_stamp;
5766
5767 /* Need to copy one event at a time */
5768 do {
5769 /* We need the size of one event, because
5770 * rb_advance_reader only advances by one event,
5771 * whereas rb_event_ts_length may include the size of
5772 * one or two events.
5773 * We have already ensured there's enough space if this
5774 * is a time extend. */
5775 size = rb_event_length(event);
5776 memcpy(bpage->data + pos, rpage->data + rpos, size);
5777
5778 len -= size;
5779
5780 rb_advance_reader(cpu_buffer);
5781 rpos = reader->read;
5782 pos += size;
5783
5784 if (rpos >= commit)
5785 break;
5786
5787 event = rb_reader_event(cpu_buffer);
5788 /* Always keep the time extend and data together */
5789 size = rb_event_ts_length(event);
5790 } while (len >= size);
5791
5792 /* update bpage */
5793 local_set(&bpage->commit, pos);
5794 bpage->time_stamp = save_timestamp;
5795
5796 /* we copied everything to the beginning */
5797 read = 0;
5798 } else {
5799 /* update the entry counter */
5800 cpu_buffer->read += rb_page_entries(reader);
5801 cpu_buffer->read_bytes += BUF_PAGE_SIZE;
5802
5803 /* swap the pages */
5804 rb_init_page(bpage);
5805 bpage = reader->page;
5806 reader->page = *data_page;
5807 local_set(&reader->write, 0);
5808 local_set(&reader->entries, 0);
5809 reader->read = 0;
5810 *data_page = bpage;
5811
5812 /*
5813 * Use the real_end for the data size,
5814 * This gives us a chance to store the lost events
5815 * on the page.
5816 */
5817 if (reader->real_end)
5818 local_set(&bpage->commit, reader->real_end);
5819 }
5820 ret = read;
5821
5822 cpu_buffer->lost_events = 0;
5823
5824 commit = local_read(&bpage->commit);
5825 /*
5826 * Set a flag in the commit field if we lost events
5827 */
5828 if (missed_events) {
5829 /* If there is room at the end of the page to save the
5830 * missed events, then record it there.
5831 */
5832 if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) {
5833 memcpy(&bpage->data[commit], &missed_events,
5834 sizeof(missed_events));
5835 local_add(RB_MISSED_STORED, &bpage->commit);
5836 commit += sizeof(missed_events);
5837 }
5838 local_add(RB_MISSED_EVENTS, &bpage->commit);
5839 }
5840
5841 /*
5842 * This page may be off to user land. Zero it out here.
5843 */
5844 if (commit < BUF_PAGE_SIZE)
5845 memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit);
5846
5847 out_unlock:
5848 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5849
5850 out:
5851 return ret;
5852}
5853EXPORT_SYMBOL_GPL(ring_buffer_read_page);
5854
5855/*
5856 * We only allocate new buffers, never free them if the CPU goes down.
5857 * If we were to free the buffer, then the user would lose any trace that was in
5858 * the buffer.
5859 */
5860int trace_rb_cpu_prepare(unsigned int cpu, struct hlist_node *node)
5861{
5862 struct trace_buffer *buffer;
5863 long nr_pages_same;
5864 int cpu_i;
5865 unsigned long nr_pages;
5866
5867 buffer = container_of(node, struct trace_buffer, node);
5868 if (cpumask_test_cpu(cpu, buffer->cpumask))
5869 return 0;
5870
5871 nr_pages = 0;
5872 nr_pages_same = 1;
5873 /* check if all cpu sizes are same */
5874 for_each_buffer_cpu(buffer, cpu_i) {
5875 /* fill in the size from first enabled cpu */
5876 if (nr_pages == 0)
5877 nr_pages = buffer->buffers[cpu_i]->nr_pages;
5878 if (nr_pages != buffer->buffers[cpu_i]->nr_pages) {
5879 nr_pages_same = 0;
5880 break;
5881 }
5882 }
5883 /* allocate minimum pages, user can later expand it */
5884 if (!nr_pages_same)
5885 nr_pages = 2;
5886 buffer->buffers[cpu] =
5887 rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
5888 if (!buffer->buffers[cpu]) {
5889 WARN(1, "failed to allocate ring buffer on CPU %u\n",
5890 cpu);
5891 return -ENOMEM;
5892 }
5893 smp_wmb();
5894 cpumask_set_cpu(cpu, buffer->cpumask);
5895 return 0;
5896}
5897
5898#ifdef CONFIG_RING_BUFFER_STARTUP_TEST
5899/*
5900 * This is a basic integrity check of the ring buffer.
5901 * Late in the boot cycle this test will run when configured in.
5902 * It will kick off a thread per CPU that will go into a loop
5903 * writing to the per cpu ring buffer various sizes of data.
5904 * Some of the data will be large items, some small.
5905 *
5906 * Another thread is created that goes into a spin, sending out
5907 * IPIs to the other CPUs to also write into the ring buffer.
5908 * this is to test the nesting ability of the buffer.
5909 *
5910 * Basic stats are recorded and reported. If something in the
5911 * ring buffer should happen that's not expected, a big warning
5912 * is displayed and all ring buffers are disabled.
5913 */
5914static struct task_struct *rb_threads[NR_CPUS] __initdata;
5915
5916struct rb_test_data {
5917 struct trace_buffer *buffer;
5918 unsigned long events;
5919 unsigned long bytes_written;
5920 unsigned long bytes_alloc;
5921 unsigned long bytes_dropped;
5922 unsigned long events_nested;
5923 unsigned long bytes_written_nested;
5924 unsigned long bytes_alloc_nested;
5925 unsigned long bytes_dropped_nested;
5926 int min_size_nested;
5927 int max_size_nested;
5928 int max_size;
5929 int min_size;
5930 int cpu;
5931 int cnt;
5932};
5933
5934static struct rb_test_data rb_data[NR_CPUS] __initdata;
5935
5936/* 1 meg per cpu */
5937#define RB_TEST_BUFFER_SIZE 1048576
5938
5939static char rb_string[] __initdata =
5940 "abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\"
5941 "?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890"
5942 "!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv";
5943
5944static bool rb_test_started __initdata;
5945
5946struct rb_item {
5947 int size;
5948 char str[];
5949};
5950
5951static __init int rb_write_something(struct rb_test_data *data, bool nested)
5952{
5953 struct ring_buffer_event *event;
5954 struct rb_item *item;
5955 bool started;
5956 int event_len;
5957 int size;
5958 int len;
5959 int cnt;
5960
5961 /* Have nested writes different that what is written */
5962 cnt = data->cnt + (nested ? 27 : 0);
5963
5964 /* Multiply cnt by ~e, to make some unique increment */
5965 size = (cnt * 68 / 25) % (sizeof(rb_string) - 1);
5966
5967 len = size + sizeof(struct rb_item);
5968
5969 started = rb_test_started;
5970 /* read rb_test_started before checking buffer enabled */
5971 smp_rmb();
5972
5973 event = ring_buffer_lock_reserve(data->buffer, len);
5974 if (!event) {
5975 /* Ignore dropped events before test starts. */
5976 if (started) {
5977 if (nested)
5978 data->bytes_dropped += len;
5979 else
5980 data->bytes_dropped_nested += len;
5981 }
5982 return len;
5983 }
5984
5985 event_len = ring_buffer_event_length(event);
5986
5987 if (RB_WARN_ON(data->buffer, event_len < len))
5988 goto out;
5989
5990 item = ring_buffer_event_data(event);
5991 item->size = size;
5992 memcpy(item->str, rb_string, size);
5993
5994 if (nested) {
5995 data->bytes_alloc_nested += event_len;
5996 data->bytes_written_nested += len;
5997 data->events_nested++;
5998 if (!data->min_size_nested || len < data->min_size_nested)
5999 data->min_size_nested = len;
6000 if (len > data->max_size_nested)
6001 data->max_size_nested = len;
6002 } else {
6003 data->bytes_alloc += event_len;
6004 data->bytes_written += len;
6005 data->events++;
6006 if (!data->min_size || len < data->min_size)
6007 data->max_size = len;
6008 if (len > data->max_size)
6009 data->max_size = len;
6010 }
6011
6012 out:
6013 ring_buffer_unlock_commit(data->buffer);
6014
6015 return 0;
6016}
6017
6018static __init int rb_test(void *arg)
6019{
6020 struct rb_test_data *data = arg;
6021
6022 while (!kthread_should_stop()) {
6023 rb_write_something(data, false);
6024 data->cnt++;
6025
6026 set_current_state(TASK_INTERRUPTIBLE);
6027 /* Now sleep between a min of 100-300us and a max of 1ms */
6028 usleep_range(((data->cnt % 3) + 1) * 100, 1000);
6029 }
6030
6031 return 0;
6032}
6033
6034static __init void rb_ipi(void *ignore)
6035{
6036 struct rb_test_data *data;
6037 int cpu = smp_processor_id();
6038
6039 data = &rb_data[cpu];
6040 rb_write_something(data, true);
6041}
6042
6043static __init int rb_hammer_test(void *arg)
6044{
6045 while (!kthread_should_stop()) {
6046
6047 /* Send an IPI to all cpus to write data! */
6048 smp_call_function(rb_ipi, NULL, 1);
6049 /* No sleep, but for non preempt, let others run */
6050 schedule();
6051 }
6052
6053 return 0;
6054}
6055
6056static __init int test_ringbuffer(void)
6057{
6058 struct task_struct *rb_hammer;
6059 struct trace_buffer *buffer;
6060 int cpu;
6061 int ret = 0;
6062
6063 if (security_locked_down(LOCKDOWN_TRACEFS)) {
6064 pr_warn("Lockdown is enabled, skipping ring buffer tests\n");
6065 return 0;
6066 }
6067
6068 pr_info("Running ring buffer tests...\n");
6069
6070 buffer = ring_buffer_alloc(RB_TEST_BUFFER_SIZE, RB_FL_OVERWRITE);
6071 if (WARN_ON(!buffer))
6072 return 0;
6073
6074 /* Disable buffer so that threads can't write to it yet */
6075 ring_buffer_record_off(buffer);
6076
6077 for_each_online_cpu(cpu) {
6078 rb_data[cpu].buffer = buffer;
6079 rb_data[cpu].cpu = cpu;
6080 rb_data[cpu].cnt = cpu;
6081 rb_threads[cpu] = kthread_run_on_cpu(rb_test, &rb_data[cpu],
6082 cpu, "rbtester/%u");
6083 if (WARN_ON(IS_ERR(rb_threads[cpu]))) {
6084 pr_cont("FAILED\n");
6085 ret = PTR_ERR(rb_threads[cpu]);
6086 goto out_free;
6087 }
6088 }
6089
6090 /* Now create the rb hammer! */
6091 rb_hammer = kthread_run(rb_hammer_test, NULL, "rbhammer");
6092 if (WARN_ON(IS_ERR(rb_hammer))) {
6093 pr_cont("FAILED\n");
6094 ret = PTR_ERR(rb_hammer);
6095 goto out_free;
6096 }
6097
6098 ring_buffer_record_on(buffer);
6099 /*
6100 * Show buffer is enabled before setting rb_test_started.
6101 * Yes there's a small race window where events could be
6102 * dropped and the thread wont catch it. But when a ring
6103 * buffer gets enabled, there will always be some kind of
6104 * delay before other CPUs see it. Thus, we don't care about
6105 * those dropped events. We care about events dropped after
6106 * the threads see that the buffer is active.
6107 */
6108 smp_wmb();
6109 rb_test_started = true;
6110
6111 set_current_state(TASK_INTERRUPTIBLE);
6112 /* Just run for 10 seconds */;
6113 schedule_timeout(10 * HZ);
6114
6115 kthread_stop(rb_hammer);
6116
6117 out_free:
6118 for_each_online_cpu(cpu) {
6119 if (!rb_threads[cpu])
6120 break;
6121 kthread_stop(rb_threads[cpu]);
6122 }
6123 if (ret) {
6124 ring_buffer_free(buffer);
6125 return ret;
6126 }
6127
6128 /* Report! */
6129 pr_info("finished\n");
6130 for_each_online_cpu(cpu) {
6131 struct ring_buffer_event *event;
6132 struct rb_test_data *data = &rb_data[cpu];
6133 struct rb_item *item;
6134 unsigned long total_events;
6135 unsigned long total_dropped;
6136 unsigned long total_written;
6137 unsigned long total_alloc;
6138 unsigned long total_read = 0;
6139 unsigned long total_size = 0;
6140 unsigned long total_len = 0;
6141 unsigned long total_lost = 0;
6142 unsigned long lost;
6143 int big_event_size;
6144 int small_event_size;
6145
6146 ret = -1;
6147
6148 total_events = data->events + data->events_nested;
6149 total_written = data->bytes_written + data->bytes_written_nested;
6150 total_alloc = data->bytes_alloc + data->bytes_alloc_nested;
6151 total_dropped = data->bytes_dropped + data->bytes_dropped_nested;
6152
6153 big_event_size = data->max_size + data->max_size_nested;
6154 small_event_size = data->min_size + data->min_size_nested;
6155
6156 pr_info("CPU %d:\n", cpu);
6157 pr_info(" events: %ld\n", total_events);
6158 pr_info(" dropped bytes: %ld\n", total_dropped);
6159 pr_info(" alloced bytes: %ld\n", total_alloc);
6160 pr_info(" written bytes: %ld\n", total_written);
6161 pr_info(" biggest event: %d\n", big_event_size);
6162 pr_info(" smallest event: %d\n", small_event_size);
6163
6164 if (RB_WARN_ON(buffer, total_dropped))
6165 break;
6166
6167 ret = 0;
6168
6169 while ((event = ring_buffer_consume(buffer, cpu, NULL, &lost))) {
6170 total_lost += lost;
6171 item = ring_buffer_event_data(event);
6172 total_len += ring_buffer_event_length(event);
6173 total_size += item->size + sizeof(struct rb_item);
6174 if (memcmp(&item->str[0], rb_string, item->size) != 0) {
6175 pr_info("FAILED!\n");
6176 pr_info("buffer had: %.*s\n", item->size, item->str);
6177 pr_info("expected: %.*s\n", item->size, rb_string);
6178 RB_WARN_ON(buffer, 1);
6179 ret = -1;
6180 break;
6181 }
6182 total_read++;
6183 }
6184 if (ret)
6185 break;
6186
6187 ret = -1;
6188
6189 pr_info(" read events: %ld\n", total_read);
6190 pr_info(" lost events: %ld\n", total_lost);
6191 pr_info(" total events: %ld\n", total_lost + total_read);
6192 pr_info(" recorded len bytes: %ld\n", total_len);
6193 pr_info(" recorded size bytes: %ld\n", total_size);
6194 if (total_lost) {
6195 pr_info(" With dropped events, record len and size may not match\n"
6196 " alloced and written from above\n");
6197 } else {
6198 if (RB_WARN_ON(buffer, total_len != total_alloc ||
6199 total_size != total_written))
6200 break;
6201 }
6202 if (RB_WARN_ON(buffer, total_lost + total_read != total_events))
6203 break;
6204
6205 ret = 0;
6206 }
6207 if (!ret)
6208 pr_info("Ring buffer PASSED!\n");
6209
6210 ring_buffer_free(buffer);
6211 return 0;
6212}
6213
6214late_initcall(test_ringbuffer);
6215#endif /* CONFIG_RING_BUFFER_STARTUP_TEST */