1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Generic ring buffer
4 *
5 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
6 */
7#include <linux/trace_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/cacheflush.h>
13#include <linux/trace_seq.h>
14#include <linux/spinlock.h>
15#include <linux/irq_work.h>
16#include <linux/security.h>
17#include <linux/uaccess.h>
18#include <linux/hardirq.h>
19#include <linux/kthread.h> /* for self test */
20#include <linux/module.h>
21#include <linux/percpu.h>
22#include <linux/mutex.h>
23#include <linux/delay.h>
24#include <linux/slab.h>
25#include <linux/init.h>
26#include <linux/hash.h>
27#include <linux/list.h>
28#include <linux/cpu.h>
29#include <linux/oom.h>
30#include <linux/mm.h>
31
32#include <asm/local64.h>
33#include <asm/local.h>
34#include <asm/setup.h>
35
36#include "trace.h"
37
38/*
39 * The "absolute" timestamp in the buffer is only 59 bits.
40 * If a clock has the 5 MSBs set, it needs to be saved and
41 * reinserted.
42 */
43#define TS_MSB (0xf8ULL << 56)
44#define ABS_TS_MASK (~TS_MSB)
45
46static void update_pages_handler(struct work_struct *work);
47
48#define RING_BUFFER_META_MAGIC 0xBADFEED
49
50struct ring_buffer_meta {
51 int magic;
52 int struct_sizes;
53 unsigned long total_size;
54 unsigned long buffers_offset;
55};
56
57struct ring_buffer_cpu_meta {
58 unsigned long first_buffer;
59 unsigned long head_buffer;
60 unsigned long commit_buffer;
61 __u32 subbuf_size;
62 __u32 nr_subbufs;
63 int buffers[];
64};
65
66/*
67 * The ring buffer header is special. We must manually up keep it.
68 */
69int ring_buffer_print_entry_header(struct trace_seq *s)
70{
71 trace_seq_puts(s, str: "# compressed entry header\n");
72 trace_seq_puts(s, str: "\ttype_len : 5 bits\n");
73 trace_seq_puts(s, str: "\ttime_delta : 27 bits\n");
74 trace_seq_puts(s, str: "\tarray : 32 bits\n");
75 trace_seq_putc(s, c: '\n');
76 trace_seq_printf(s, fmt: "\tpadding : type == %d\n",
77 RINGBUF_TYPE_PADDING);
78 trace_seq_printf(s, fmt: "\ttime_extend : type == %d\n",
79 RINGBUF_TYPE_TIME_EXTEND);
80 trace_seq_printf(s, fmt: "\ttime_stamp : type == %d\n",
81 RINGBUF_TYPE_TIME_STAMP);
82 trace_seq_printf(s, fmt: "\tdata max type_len == %d\n",
83 RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
84
85 return !trace_seq_has_overflowed(s);
86}
87
88/*
89 * The ring buffer is made up of a list of pages. A separate list of pages is
90 * allocated for each CPU. A writer may only write to a buffer that is
91 * associated with the CPU it is currently executing on. A reader may read
92 * from any per cpu buffer.
93 *
94 * The reader is special. For each per cpu buffer, the reader has its own
95 * reader page. When a reader has read the entire reader page, this reader
96 * page is swapped with another page in the ring buffer.
97 *
98 * Now, as long as the writer is off the reader page, the reader can do what
99 * ever it wants with that page. The writer will never write to that page
100 * again (as long as it is out of the ring buffer).
101 *
102 * Here's some silly ASCII art.
103 *
104 * +------+
105 * |reader| RING BUFFER
106 * |page |
107 * +------+ +---+ +---+ +---+
108 * | |-->| |-->| |
109 * +---+ +---+ +---+
110 * ^ |
111 * | |
112 * +---------------+
113 *
114 *
115 * +------+
116 * |reader| RING BUFFER
117 * |page |------------------v
118 * +------+ +---+ +---+ +---+
119 * | |-->| |-->| |
120 * +---+ +---+ +---+
121 * ^ |
122 * | |
123 * +---------------+
124 *
125 *
126 * +------+
127 * |reader| RING BUFFER
128 * |page |------------------v
129 * +------+ +---+ +---+ +---+
130 * ^ | |-->| |-->| |
131 * | +---+ +---+ +---+
132 * | |
133 * | |
134 * +------------------------------+
135 *
136 *
137 * +------+
138 * |buffer| RING BUFFER
139 * |page |------------------v
140 * +------+ +---+ +---+ +---+
141 * ^ | | | |-->| |
142 * | New +---+ +---+ +---+
143 * | Reader------^ |
144 * | page |
145 * +------------------------------+
146 *
147 *
148 * After we make this swap, the reader can hand this page off to the splice
149 * code and be done with it. It can even allocate a new page if it needs to
150 * and swap that into the ring buffer.
151 *
152 * We will be using cmpxchg soon to make all this lockless.
153 *
154 */
155
156/* Used for individual buffers (after the counter) */
157#define RB_BUFFER_OFF (1 << 20)
158
159#define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
160
161#define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
162#define RB_ALIGNMENT 4U
163#define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
164#define RB_EVNT_MIN_SIZE 8U /* two 32bit words */
165
166#ifndef CONFIG_HAVE_64BIT_ALIGNED_ACCESS
167# define RB_FORCE_8BYTE_ALIGNMENT 0
168# define RB_ARCH_ALIGNMENT RB_ALIGNMENT
169#else
170# define RB_FORCE_8BYTE_ALIGNMENT 1
171# define RB_ARCH_ALIGNMENT 8U
172#endif
173
174#define RB_ALIGN_DATA __aligned(RB_ARCH_ALIGNMENT)
175
176/* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
177#define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
178
179enum {
180 RB_LEN_TIME_EXTEND = 8,
181 RB_LEN_TIME_STAMP = 8,
182};
183
184#define skip_time_extend(event) \
185 ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND))
186
187#define extended_time(event) \
188 (event->type_len >= RINGBUF_TYPE_TIME_EXTEND)
189
190static inline bool rb_null_event(struct ring_buffer_event *event)
191{
192 return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta;
193}
194
195static void rb_event_set_padding(struct ring_buffer_event *event)
196{
197 /* padding has a NULL time_delta */
198 event->type_len = RINGBUF_TYPE_PADDING;
199 event->time_delta = 0;
200}
201
202static unsigned
203rb_event_data_length(struct ring_buffer_event *event)
204{
205 unsigned length;
206
207 if (event->type_len)
208 length = event->type_len * RB_ALIGNMENT;
209 else
210 length = event->array[0];
211 return length + RB_EVNT_HDR_SIZE;
212}
213
214/*
215 * Return the length of the given event. Will return
216 * the length of the time extend if the event is a
217 * time extend.
218 */
219static inline unsigned
220rb_event_length(struct ring_buffer_event *event)
221{
222 switch (event->type_len) {
223 case RINGBUF_TYPE_PADDING:
224 if (rb_null_event(event))
225 /* undefined */
226 return -1;
227 return event->array[0] + RB_EVNT_HDR_SIZE;
228
229 case RINGBUF_TYPE_TIME_EXTEND:
230 return RB_LEN_TIME_EXTEND;
231
232 case RINGBUF_TYPE_TIME_STAMP:
233 return RB_LEN_TIME_STAMP;
234
235 case RINGBUF_TYPE_DATA:
236 return rb_event_data_length(event);
237 default:
238 WARN_ON_ONCE(1);
239 }
240 /* not hit */
241 return 0;
242}
243
244/*
245 * Return total length of time extend and data,
246 * or just the event length for all other events.
247 */
248static inline unsigned
249rb_event_ts_length(struct ring_buffer_event *event)
250{
251 unsigned len = 0;
252
253 if (extended_time(event)) {
254 /* time extends include the data event after it */
255 len = RB_LEN_TIME_EXTEND;
256 event = skip_time_extend(event);
257 }
258 return len + rb_event_length(event);
259}
260
261/**
262 * ring_buffer_event_length - return the length of the event
263 * @event: the event to get the length of
264 *
265 * Returns the size of the data load of a data event.
266 * If the event is something other than a data event, it
267 * returns the size of the event itself. With the exception
268 * of a TIME EXTEND, where it still returns the size of the
269 * data load of the data event after it.
270 */
271unsigned ring_buffer_event_length(struct ring_buffer_event *event)
272{
273 unsigned length;
274
275 if (extended_time(event))
276 event = skip_time_extend(event);
277
278 length = rb_event_length(event);
279 if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
280 return length;
281 length -= RB_EVNT_HDR_SIZE;
282 if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0]))
283 length -= sizeof(event->array[0]);
284 return length;
285}
286EXPORT_SYMBOL_GPL(ring_buffer_event_length);
287
288/* inline for ring buffer fast paths */
289static __always_inline void *
290rb_event_data(struct ring_buffer_event *event)
291{
292 if (extended_time(event))
293 event = skip_time_extend(event);
294 WARN_ON_ONCE(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX);
295 /* If length is in len field, then array[0] has the data */
296 if (event->type_len)
297 return (void *)&event->array[0];
298 /* Otherwise length is in array[0] and array[1] has the data */
299 return (void *)&event->array[1];
300}
301
302/**
303 * ring_buffer_event_data - return the data of the event
304 * @event: the event to get the data from
305 */
306void *ring_buffer_event_data(struct ring_buffer_event *event)
307{
308 return rb_event_data(event);
309}
310EXPORT_SYMBOL_GPL(ring_buffer_event_data);
311
312#define for_each_buffer_cpu(buffer, cpu) \
313 for_each_cpu(cpu, buffer->cpumask)
314
315#define for_each_online_buffer_cpu(buffer, cpu) \
316 for_each_cpu_and(cpu, buffer->cpumask, cpu_online_mask)
317
318#define TS_SHIFT 27
319#define TS_MASK ((1ULL << TS_SHIFT) - 1)
320#define TS_DELTA_TEST (~TS_MASK)
321
322static u64 rb_event_time_stamp(struct ring_buffer_event *event)
323{
324 u64 ts;
325
326 ts = event->array[0];
327 ts <<= TS_SHIFT;
328 ts += event->time_delta;
329
330 return ts;
331}
332
333/* Flag when events were overwritten */
334#define RB_MISSED_EVENTS (1 << 31)
335/* Missed count stored at end */
336#define RB_MISSED_STORED (1 << 30)
337
338#define RB_MISSED_MASK (3 << 30)
339
340struct buffer_data_page {
341 u64 time_stamp; /* page time stamp */
342 local_t commit; /* write committed index */
343 unsigned char data[] RB_ALIGN_DATA; /* data of buffer page */
344};
345
346struct buffer_data_read_page {
347 unsigned order; /* order of the page */
348 struct buffer_data_page *data; /* actual data, stored in this page */
349};
350
351/*
352 * Note, the buffer_page list must be first. The buffer pages
353 * are allocated in cache lines, which means that each buffer
354 * page will be at the beginning of a cache line, and thus
355 * the least significant bits will be zero. We use this to
356 * add flags in the list struct pointers, to make the ring buffer
357 * lockless.
358 */
359struct buffer_page {
360 struct list_head list; /* list of buffer pages */
361 local_t write; /* index for next write */
362 unsigned read; /* index for next read */
363 local_t entries; /* entries on this page */
364 unsigned long real_end; /* real end of data */
365 unsigned order; /* order of the page */
366 u32 id:30; /* ID for external mapping */
367 u32 range:1; /* Mapped via a range */
368 struct buffer_data_page *page; /* Actual data page */
369};
370
371/*
372 * The buffer page counters, write and entries, must be reset
373 * atomically when crossing page boundaries. To synchronize this
374 * update, two counters are inserted into the number. One is
375 * the actual counter for the write position or count on the page.
376 *
377 * The other is a counter of updaters. Before an update happens
378 * the update partition of the counter is incremented. This will
379 * allow the updater to update the counter atomically.
380 *
381 * The counter is 20 bits, and the state data is 12.
382 */
383#define RB_WRITE_MASK 0xfffff
384#define RB_WRITE_INTCNT (1 << 20)
385
386static void rb_init_page(struct buffer_data_page *bpage)
387{
388 local_set(&bpage->commit, 0);
389}
390
391static __always_inline unsigned int rb_page_commit(struct buffer_page *bpage)
392{
393 return local_read(&bpage->page->commit);
394}
395
396static void free_buffer_page(struct buffer_page *bpage)
397{
398 /* Range pages are not to be freed */
399 if (!bpage->range)
400 free_pages(addr: (unsigned long)bpage->page, order: bpage->order);
401 kfree(objp: bpage);
402}
403
404/*
405 * We need to fit the time_stamp delta into 27 bits.
406 */
407static inline bool test_time_stamp(u64 delta)
408{
409 return !!(delta & TS_DELTA_TEST);
410}
411
412struct rb_irq_work {
413 struct irq_work work;
414 wait_queue_head_t waiters;
415 wait_queue_head_t full_waiters;
416 atomic_t seq;
417 bool waiters_pending;
418 bool full_waiters_pending;
419 bool wakeup_full;
420};
421
422/*
423 * Structure to hold event state and handle nested events.
424 */
425struct rb_event_info {
426 u64 ts;
427 u64 delta;
428 u64 before;
429 u64 after;
430 unsigned long length;
431 struct buffer_page *tail_page;
432 int add_timestamp;
433};
434
435/*
436 * Used for the add_timestamp
437 * NONE
438 * EXTEND - wants a time extend
439 * ABSOLUTE - the buffer requests all events to have absolute time stamps
440 * FORCE - force a full time stamp.
441 */
442enum {
443 RB_ADD_STAMP_NONE = 0,
444 RB_ADD_STAMP_EXTEND = BIT(1),
445 RB_ADD_STAMP_ABSOLUTE = BIT(2),
446 RB_ADD_STAMP_FORCE = BIT(3)
447};
448/*
449 * Used for which event context the event is in.
450 * TRANSITION = 0
451 * NMI = 1
452 * IRQ = 2
453 * SOFTIRQ = 3
454 * NORMAL = 4
455 *
456 * See trace_recursive_lock() comment below for more details.
457 */
458enum {
459 RB_CTX_TRANSITION,
460 RB_CTX_NMI,
461 RB_CTX_IRQ,
462 RB_CTX_SOFTIRQ,
463 RB_CTX_NORMAL,
464 RB_CTX_MAX
465};
466
467struct rb_time_struct {
468 local64_t time;
469};
470typedef struct rb_time_struct rb_time_t;
471
472#define MAX_NEST 5
473
474/*
475 * head_page == tail_page && head == tail then buffer is empty.
476 */
477struct ring_buffer_per_cpu {
478 int cpu;
479 atomic_t record_disabled;
480 atomic_t resize_disabled;
481 struct trace_buffer *buffer;
482 raw_spinlock_t reader_lock; /* serialize readers */
483 arch_spinlock_t lock;
484 struct lock_class_key lock_key;
485 struct buffer_data_page *free_page;
486 unsigned long nr_pages;
487 unsigned int current_context;
488 struct list_head *pages;
489 /* pages generation counter, incremented when the list changes */
490 unsigned long cnt;
491 struct buffer_page *head_page; /* read from head */
492 struct buffer_page *tail_page; /* write to tail */
493 struct buffer_page *commit_page; /* committed pages */
494 struct buffer_page *reader_page;
495 unsigned long lost_events;
496 unsigned long last_overrun;
497 unsigned long nest;
498 local_t entries_bytes;
499 local_t entries;
500 local_t overrun;
501 local_t commit_overrun;
502 local_t dropped_events;
503 local_t committing;
504 local_t commits;
505 local_t pages_touched;
506 local_t pages_lost;
507 local_t pages_read;
508 long last_pages_touch;
509 size_t shortest_full;
510 unsigned long read;
511 unsigned long read_bytes;
512 rb_time_t write_stamp;
513 rb_time_t before_stamp;
514 u64 event_stamp[MAX_NEST];
515 u64 read_stamp;
516 /* pages removed since last reset */
517 unsigned long pages_removed;
518
519 unsigned int mapped;
520 unsigned int user_mapped; /* user space mapping */
521 struct mutex mapping_lock;
522 unsigned long *subbuf_ids; /* ID to subbuf VA */
523 struct trace_buffer_meta *meta_page;
524 struct ring_buffer_cpu_meta *ring_meta;
525
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 atomic_t resizing;
540 cpumask_var_t cpumask;
541
542 struct lock_class_key *reader_lock_key;
543
544 struct mutex mutex;
545
546 struct ring_buffer_per_cpu **buffers;
547
548 struct hlist_node node;
549 u64 (*clock)(void);
550
551 struct rb_irq_work irq_work;
552 bool time_stamp_abs;
553
554 unsigned long range_addr_start;
555 unsigned long range_addr_end;
556
557 struct ring_buffer_meta *meta;
558
559 unsigned int subbuf_size;
560 unsigned int subbuf_order;
561 unsigned int max_data_size;
562};
563
564struct ring_buffer_iter {
565 struct ring_buffer_per_cpu *cpu_buffer;
566 unsigned long head;
567 unsigned long next_event;
568 struct buffer_page *head_page;
569 struct buffer_page *cache_reader_page;
570 unsigned long cache_read;
571 unsigned long cache_pages_removed;
572 u64 read_stamp;
573 u64 page_stamp;
574 struct ring_buffer_event *event;
575 size_t event_size;
576 int missed_events;
577};
578
579int ring_buffer_print_page_header(struct trace_buffer *buffer, struct trace_seq *s)
580{
581 struct buffer_data_page field;
582
583 trace_seq_printf(s, fmt: "\tfield: u64 timestamp;\t"
584 "offset:0;\tsize:%u;\tsigned:%u;\n",
585 (unsigned int)sizeof(field.time_stamp),
586 (unsigned int)is_signed_type(u64));
587
588 trace_seq_printf(s, fmt: "\tfield: local_t commit;\t"
589 "offset:%u;\tsize:%u;\tsigned:%u;\n",
590 (unsigned int)offsetof(typeof(field), commit),
591 (unsigned int)sizeof(field.commit),
592 (unsigned int)is_signed_type(long));
593
594 trace_seq_printf(s, fmt: "\tfield: int overwrite;\t"
595 "offset:%u;\tsize:%u;\tsigned:%u;\n",
596 (unsigned int)offsetof(typeof(field), commit),
597 1,
598 (unsigned int)is_signed_type(long));
599
600 trace_seq_printf(s, fmt: "\tfield: char data;\t"
601 "offset:%u;\tsize:%u;\tsigned:%u;\n",
602 (unsigned int)offsetof(typeof(field), data),
603 (unsigned int)buffer->subbuf_size,
604 (unsigned int)is_signed_type(char));
605
606 return !trace_seq_has_overflowed(s);
607}
608
609static inline void rb_time_read(rb_time_t *t, u64 *ret)
610{
611 *ret = local64_read(&t->time);
612}
613static void rb_time_set(rb_time_t *t, u64 val)
614{
615 local64_set(&t->time, val);
616}
617
618/*
619 * Enable this to make sure that the event passed to
620 * ring_buffer_event_time_stamp() is not committed and also
621 * is on the buffer that it passed in.
622 */
623//#define RB_VERIFY_EVENT
624#ifdef RB_VERIFY_EVENT
625static struct list_head *rb_list_head(struct list_head *list);
626static void verify_event(struct ring_buffer_per_cpu *cpu_buffer,
627 void *event)
628{
629 struct buffer_page *page = cpu_buffer->commit_page;
630 struct buffer_page *tail_page = READ_ONCE(cpu_buffer->tail_page);
631 struct list_head *next;
632 long commit, write;
633 unsigned long addr = (unsigned long)event;
634 bool done = false;
635 int stop = 0;
636
637 /* Make sure the event exists and is not committed yet */
638 do {
639 if (page == tail_page || WARN_ON_ONCE(stop++ > 100))
640 done = true;
641 commit = local_read(&page->page->commit);
642 write = local_read(&page->write);
643 if (addr >= (unsigned long)&page->page->data[commit] &&
644 addr < (unsigned long)&page->page->data[write])
645 return;
646
647 next = rb_list_head(page->list.next);
648 page = list_entry(next, struct buffer_page, list);
649 } while (!done);
650 WARN_ON_ONCE(1);
651}
652#else
653static inline void verify_event(struct ring_buffer_per_cpu *cpu_buffer,
654 void *event)
655{
656}
657#endif
658
659/*
660 * The absolute time stamp drops the 5 MSBs and some clocks may
661 * require them. The rb_fix_abs_ts() will take a previous full
662 * time stamp, and add the 5 MSB of that time stamp on to the
663 * saved absolute time stamp. Then they are compared in case of
664 * the unlikely event that the latest time stamp incremented
665 * the 5 MSB.
666 */
667static inline u64 rb_fix_abs_ts(u64 abs, u64 save_ts)
668{
669 if (save_ts & TS_MSB) {
670 abs |= save_ts & TS_MSB;
671 /* Check for overflow */
672 if (unlikely(abs < save_ts))
673 abs += 1ULL << 59;
674 }
675 return abs;
676}
677
678static inline u64 rb_time_stamp(struct trace_buffer *buffer);
679
680/**
681 * ring_buffer_event_time_stamp - return the event's current time stamp
682 * @buffer: The buffer that the event is on
683 * @event: the event to get the time stamp of
684 *
685 * Note, this must be called after @event is reserved, and before it is
686 * committed to the ring buffer. And must be called from the same
687 * context where the event was reserved (normal, softirq, irq, etc).
688 *
689 * Returns the time stamp associated with the current event.
690 * If the event has an extended time stamp, then that is used as
691 * the time stamp to return.
692 * In the highly unlikely case that the event was nested more than
693 * the max nesting, then the write_stamp of the buffer is returned,
694 * otherwise current time is returned, but that really neither of
695 * the last two cases should ever happen.
696 */
697u64 ring_buffer_event_time_stamp(struct trace_buffer *buffer,
698 struct ring_buffer_event *event)
699{
700 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[smp_processor_id()];
701 unsigned int nest;
702 u64 ts;
703
704 /* If the event includes an absolute time, then just use that */
705 if (event->type_len == RINGBUF_TYPE_TIME_STAMP) {
706 ts = rb_event_time_stamp(event);
707 return rb_fix_abs_ts(abs: ts, save_ts: cpu_buffer->tail_page->page->time_stamp);
708 }
709
710 nest = local_read(&cpu_buffer->committing);
711 verify_event(cpu_buffer, event);
712 if (WARN_ON_ONCE(!nest))
713 goto fail;
714
715 /* Read the current saved nesting level time stamp */
716 if (likely(--nest < MAX_NEST))
717 return cpu_buffer->event_stamp[nest];
718
719 /* Shouldn't happen, warn if it does */
720 WARN_ONCE(1, "nest (%d) greater than max", nest);
721
722 fail:
723 rb_time_read(t: &cpu_buffer->write_stamp, ret: &ts);
724
725 return ts;
726}
727
728/**
729 * ring_buffer_nr_dirty_pages - get the number of used pages in the ring buffer
730 * @buffer: The ring_buffer to get the number of pages from
731 * @cpu: The cpu of the ring_buffer to get the number of pages from
732 *
733 * Returns the number of pages that have content in the ring buffer.
734 */
735size_t ring_buffer_nr_dirty_pages(struct trace_buffer *buffer, int cpu)
736{
737 size_t read;
738 size_t lost;
739 size_t cnt;
740
741 read = local_read(&buffer->buffers[cpu]->pages_read);
742 lost = local_read(&buffer->buffers[cpu]->pages_lost);
743 cnt = local_read(&buffer->buffers[cpu]->pages_touched);
744
745 if (WARN_ON_ONCE(cnt < lost))
746 return 0;
747
748 cnt -= lost;
749
750 /* The reader can read an empty page, but not more than that */
751 if (cnt < read) {
752 WARN_ON_ONCE(read > cnt + 1);
753 return 0;
754 }
755
756 return cnt - read;
757}
758
759static __always_inline bool full_hit(struct trace_buffer *buffer, int cpu, int full)
760{
761 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
762 size_t nr_pages;
763 size_t dirty;
764
765 nr_pages = cpu_buffer->nr_pages;
766 if (!nr_pages || !full)
767 return true;
768
769 /*
770 * Add one as dirty will never equal nr_pages, as the sub-buffer
771 * that the writer is on is not counted as dirty.
772 * This is needed if "buffer_percent" is set to 100.
773 */
774 dirty = ring_buffer_nr_dirty_pages(buffer, cpu) + 1;
775
776 return (dirty * 100) >= (full * nr_pages);
777}
778
779/*
780 * rb_wake_up_waiters - wake up tasks waiting for ring buffer input
781 *
782 * Schedules a delayed work to wake up any task that is blocked on the
783 * ring buffer waiters queue.
784 */
785static void rb_wake_up_waiters(struct irq_work *work)
786{
787 struct rb_irq_work *rbwork = container_of(work, struct rb_irq_work, work);
788
789 /* For waiters waiting for the first wake up */
790 (void)atomic_fetch_inc_release(v: &rbwork->seq);
791
792 wake_up_all(&rbwork->waiters);
793 if (rbwork->full_waiters_pending || rbwork->wakeup_full) {
794 /* Only cpu_buffer sets the above flags */
795 struct ring_buffer_per_cpu *cpu_buffer =
796 container_of(rbwork, struct ring_buffer_per_cpu, irq_work);
797
798 /* Called from interrupt context */
799 raw_spin_lock(&cpu_buffer->reader_lock);
800 rbwork->wakeup_full = false;
801 rbwork->full_waiters_pending = false;
802
803 /* Waking up all waiters, they will reset the shortest full */
804 cpu_buffer->shortest_full = 0;
805 raw_spin_unlock(&cpu_buffer->reader_lock);
806
807 wake_up_all(&rbwork->full_waiters);
808 }
809}
810
811/**
812 * ring_buffer_wake_waiters - wake up any waiters on this ring buffer
813 * @buffer: The ring buffer to wake waiters on
814 * @cpu: The CPU buffer to wake waiters on
815 *
816 * In the case of a file that represents a ring buffer is closing,
817 * it is prudent to wake up any waiters that are on this.
818 */
819void ring_buffer_wake_waiters(struct trace_buffer *buffer, int cpu)
820{
821 struct ring_buffer_per_cpu *cpu_buffer;
822 struct rb_irq_work *rbwork;
823
824 if (!buffer)
825 return;
826
827 if (cpu == RING_BUFFER_ALL_CPUS) {
828
829 /* Wake up individual ones too. One level recursion */
830 for_each_buffer_cpu(buffer, cpu)
831 ring_buffer_wake_waiters(buffer, cpu);
832
833 rbwork = &buffer->irq_work;
834 } else {
835 if (WARN_ON_ONCE(!buffer->buffers))
836 return;
837 if (WARN_ON_ONCE(cpu >= nr_cpu_ids))
838 return;
839
840 cpu_buffer = buffer->buffers[cpu];
841 /* The CPU buffer may not have been initialized yet */
842 if (!cpu_buffer)
843 return;
844 rbwork = &cpu_buffer->irq_work;
845 }
846
847 /* This can be called in any context */
848 irq_work_queue(work: &rbwork->work);
849}
850
851static bool rb_watermark_hit(struct trace_buffer *buffer, int cpu, int full)
852{
853 struct ring_buffer_per_cpu *cpu_buffer;
854 bool ret = false;
855
856 /* Reads of all CPUs always waits for any data */
857 if (cpu == RING_BUFFER_ALL_CPUS)
858 return !ring_buffer_empty(buffer);
859
860 cpu_buffer = buffer->buffers[cpu];
861
862 if (!ring_buffer_empty_cpu(buffer, cpu)) {
863 unsigned long flags;
864 bool pagebusy;
865
866 if (!full)
867 return true;
868
869 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
870 pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page;
871 ret = !pagebusy && full_hit(buffer, cpu, full);
872
873 if (!ret && (!cpu_buffer->shortest_full ||
874 cpu_buffer->shortest_full > full)) {
875 cpu_buffer->shortest_full = full;
876 }
877 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
878 }
879 return ret;
880}
881
882static inline bool
883rb_wait_cond(struct rb_irq_work *rbwork, struct trace_buffer *buffer,
884 int cpu, int full, ring_buffer_cond_fn cond, void *data)
885{
886 if (rb_watermark_hit(buffer, cpu, full))
887 return true;
888
889 if (cond(data))
890 return true;
891
892 /*
893 * The events can happen in critical sections where
894 * checking a work queue can cause deadlocks.
895 * After adding a task to the queue, this flag is set
896 * only to notify events to try to wake up the queue
897 * using irq_work.
898 *
899 * We don't clear it even if the buffer is no longer
900 * empty. The flag only causes the next event to run
901 * irq_work to do the work queue wake up. The worse
902 * that can happen if we race with !trace_empty() is that
903 * an event will cause an irq_work to try to wake up
904 * an empty queue.
905 *
906 * There's no reason to protect this flag either, as
907 * the work queue and irq_work logic will do the necessary
908 * synchronization for the wake ups. The only thing
909 * that is necessary is that the wake up happens after
910 * a task has been queued. It's OK for spurious wake ups.
911 */
912 if (full)
913 rbwork->full_waiters_pending = true;
914 else
915 rbwork->waiters_pending = true;
916
917 return false;
918}
919
920struct rb_wait_data {
921 struct rb_irq_work *irq_work;
922 int seq;
923};
924
925/*
926 * The default wait condition for ring_buffer_wait() is to just to exit the
927 * wait loop the first time it is woken up.
928 */
929static bool rb_wait_once(void *data)
930{
931 struct rb_wait_data *rdata = data;
932 struct rb_irq_work *rbwork = rdata->irq_work;
933
934 return atomic_read_acquire(v: &rbwork->seq) != rdata->seq;
935}
936
937/**
938 * ring_buffer_wait - wait for input to the ring buffer
939 * @buffer: buffer to wait on
940 * @cpu: the cpu buffer to wait on
941 * @full: wait until the percentage of pages are available, if @cpu != RING_BUFFER_ALL_CPUS
942 * @cond: condition function to break out of wait (NULL to run once)
943 * @data: the data to pass to @cond.
944 *
945 * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
946 * as data is added to any of the @buffer's cpu buffers. Otherwise
947 * it will wait for data to be added to a specific cpu buffer.
948 */
949int ring_buffer_wait(struct trace_buffer *buffer, int cpu, int full,
950 ring_buffer_cond_fn cond, void *data)
951{
952 struct ring_buffer_per_cpu *cpu_buffer;
953 struct wait_queue_head *waitq;
954 struct rb_irq_work *rbwork;
955 struct rb_wait_data rdata;
956 int ret = 0;
957
958 /*
959 * Depending on what the caller is waiting for, either any
960 * data in any cpu buffer, or a specific buffer, put the
961 * caller on the appropriate wait queue.
962 */
963 if (cpu == RING_BUFFER_ALL_CPUS) {
964 rbwork = &buffer->irq_work;
965 /* Full only makes sense on per cpu reads */
966 full = 0;
967 } else {
968 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
969 return -ENODEV;
970 cpu_buffer = buffer->buffers[cpu];
971 rbwork = &cpu_buffer->irq_work;
972 }
973
974 if (full)
975 waitq = &rbwork->full_waiters;
976 else
977 waitq = &rbwork->waiters;
978
979 /* Set up to exit loop as soon as it is woken */
980 if (!cond) {
981 cond = rb_wait_once;
982 rdata.irq_work = rbwork;
983 rdata.seq = atomic_read_acquire(v: &rbwork->seq);
984 data = &rdata;
985 }
986
987 ret = wait_event_interruptible((*waitq),
988 rb_wait_cond(rbwork, buffer, cpu, full, cond, data));
989
990 return ret;
991}
992
993/**
994 * ring_buffer_poll_wait - poll on buffer input
995 * @buffer: buffer to wait on
996 * @cpu: the cpu buffer to wait on
997 * @filp: the file descriptor
998 * @poll_table: The poll descriptor
999 * @full: wait until the percentage of pages are available, if @cpu != RING_BUFFER_ALL_CPUS
1000 *
1001 * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon
1002 * as data is added to any of the @buffer's cpu buffers. Otherwise
1003 * it will wait for data to be added to a specific cpu buffer.
1004 *
1005 * Returns EPOLLIN | EPOLLRDNORM if data exists in the buffers,
1006 * zero otherwise.
1007 */
1008__poll_t ring_buffer_poll_wait(struct trace_buffer *buffer, int cpu,
1009 struct file *filp, poll_table *poll_table, int full)
1010{
1011 struct ring_buffer_per_cpu *cpu_buffer;
1012 struct rb_irq_work *rbwork;
1013
1014 if (cpu == RING_BUFFER_ALL_CPUS) {
1015 rbwork = &buffer->irq_work;
1016 full = 0;
1017 } else {
1018 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
1019 return EPOLLERR;
1020
1021 cpu_buffer = buffer->buffers[cpu];
1022 rbwork = &cpu_buffer->irq_work;
1023 }
1024
1025 if (full) {
1026 poll_wait(filp, wait_address: &rbwork->full_waiters, p: poll_table);
1027
1028 if (rb_watermark_hit(buffer, cpu, full))
1029 return EPOLLIN | EPOLLRDNORM;
1030 /*
1031 * Only allow full_waiters_pending update to be seen after
1032 * the shortest_full is set (in rb_watermark_hit). If the
1033 * writer sees the full_waiters_pending flag set, it will
1034 * compare the amount in the ring buffer to shortest_full.
1035 * If the amount in the ring buffer is greater than the
1036 * shortest_full percent, it will call the irq_work handler
1037 * to wake up this list. The irq_handler will reset shortest_full
1038 * back to zero. That's done under the reader_lock, but
1039 * the below smp_mb() makes sure that the update to
1040 * full_waiters_pending doesn't leak up into the above.
1041 */
1042 smp_mb();
1043 rbwork->full_waiters_pending = true;
1044 return 0;
1045 }
1046
1047 poll_wait(filp, wait_address: &rbwork->waiters, p: poll_table);
1048 rbwork->waiters_pending = true;
1049
1050 /*
1051 * There's a tight race between setting the waiters_pending and
1052 * checking if the ring buffer is empty. Once the waiters_pending bit
1053 * is set, the next event will wake the task up, but we can get stuck
1054 * if there's only a single event in.
1055 *
1056 * FIXME: Ideally, we need a memory barrier on the writer side as well,
1057 * but adding a memory barrier to all events will cause too much of a
1058 * performance hit in the fast path. We only need a memory barrier when
1059 * the buffer goes from empty to having content. But as this race is
1060 * extremely small, and it's not a problem if another event comes in, we
1061 * will fix it later.
1062 */
1063 smp_mb();
1064
1065 if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) ||
1066 (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu)))
1067 return EPOLLIN | EPOLLRDNORM;
1068 return 0;
1069}
1070
1071/* buffer may be either ring_buffer or ring_buffer_per_cpu */
1072#define RB_WARN_ON(b, cond) \
1073 ({ \
1074 int _____ret = unlikely(cond); \
1075 if (_____ret) { \
1076 if (__same_type(*(b), struct ring_buffer_per_cpu)) { \
1077 struct ring_buffer_per_cpu *__b = \
1078 (void *)b; \
1079 atomic_inc(&__b->buffer->record_disabled); \
1080 } else \
1081 atomic_inc(&b->record_disabled); \
1082 WARN_ON(1); \
1083 } \
1084 _____ret; \
1085 })
1086
1087/* Up this if you want to test the TIME_EXTENTS and normalization */
1088#define DEBUG_SHIFT 0
1089
1090static inline u64 rb_time_stamp(struct trace_buffer *buffer)
1091{
1092 u64 ts;
1093
1094 /* Skip retpolines :-( */
1095 if (IS_ENABLED(CONFIG_MITIGATION_RETPOLINE) && likely(buffer->clock == trace_clock_local))
1096 ts = trace_clock_local();
1097 else
1098 ts = buffer->clock();
1099
1100 /* shift to debug/test normalization and TIME_EXTENTS */
1101 return ts << DEBUG_SHIFT;
1102}
1103
1104u64 ring_buffer_time_stamp(struct trace_buffer *buffer)
1105{
1106 u64 time;
1107
1108 preempt_disable_notrace();
1109 time = rb_time_stamp(buffer);
1110 preempt_enable_notrace();
1111
1112 return time;
1113}
1114EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
1115
1116void ring_buffer_normalize_time_stamp(struct trace_buffer *buffer,
1117 int cpu, u64 *ts)
1118{
1119 /* Just stupid testing the normalize function and deltas */
1120 *ts >>= DEBUG_SHIFT;
1121}
1122EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
1123
1124/*
1125 * Making the ring buffer lockless makes things tricky.
1126 * Although writes only happen on the CPU that they are on,
1127 * and they only need to worry about interrupts. Reads can
1128 * happen on any CPU.
1129 *
1130 * The reader page is always off the ring buffer, but when the
1131 * reader finishes with a page, it needs to swap its page with
1132 * a new one from the buffer. The reader needs to take from
1133 * the head (writes go to the tail). But if a writer is in overwrite
1134 * mode and wraps, it must push the head page forward.
1135 *
1136 * Here lies the problem.
1137 *
1138 * The reader must be careful to replace only the head page, and
1139 * not another one. As described at the top of the file in the
1140 * ASCII art, the reader sets its old page to point to the next
1141 * page after head. It then sets the page after head to point to
1142 * the old reader page. But if the writer moves the head page
1143 * during this operation, the reader could end up with the tail.
1144 *
1145 * We use cmpxchg to help prevent this race. We also do something
1146 * special with the page before head. We set the LSB to 1.
1147 *
1148 * When the writer must push the page forward, it will clear the
1149 * bit that points to the head page, move the head, and then set
1150 * the bit that points to the new head page.
1151 *
1152 * We also don't want an interrupt coming in and moving the head
1153 * page on another writer. Thus we use the second LSB to catch
1154 * that too. Thus:
1155 *
1156 * head->list->prev->next bit 1 bit 0
1157 * ------- -------
1158 * Normal page 0 0
1159 * Points to head page 0 1
1160 * New head page 1 0
1161 *
1162 * Note we can not trust the prev pointer of the head page, because:
1163 *
1164 * +----+ +-----+ +-----+
1165 * | |------>| T |---X--->| N |
1166 * | |<------| | | |
1167 * +----+ +-----+ +-----+
1168 * ^ ^ |
1169 * | +-----+ | |
1170 * +----------| R |----------+ |
1171 * | |<-----------+
1172 * +-----+
1173 *
1174 * Key: ---X--> HEAD flag set in pointer
1175 * T Tail page
1176 * R Reader page
1177 * N Next page
1178 *
1179 * (see __rb_reserve_next() to see where this happens)
1180 *
1181 * What the above shows is that the reader just swapped out
1182 * the reader page with a page in the buffer, but before it
1183 * could make the new header point back to the new page added
1184 * it was preempted by a writer. The writer moved forward onto
1185 * the new page added by the reader and is about to move forward
1186 * again.
1187 *
1188 * You can see, it is legitimate for the previous pointer of
1189 * the head (or any page) not to point back to itself. But only
1190 * temporarily.
1191 */
1192
1193#define RB_PAGE_NORMAL 0UL
1194#define RB_PAGE_HEAD 1UL
1195#define RB_PAGE_UPDATE 2UL
1196
1197
1198#define RB_FLAG_MASK 3UL
1199
1200/* PAGE_MOVED is not part of the mask */
1201#define RB_PAGE_MOVED 4UL
1202
1203/*
1204 * rb_list_head - remove any bit
1205 */
1206static struct list_head *rb_list_head(struct list_head *list)
1207{
1208 unsigned long val = (unsigned long)list;
1209
1210 return (struct list_head *)(val & ~RB_FLAG_MASK);
1211}
1212
1213/*
1214 * rb_is_head_page - test if the given page is the head page
1215 *
1216 * Because the reader may move the head_page pointer, we can
1217 * not trust what the head page is (it may be pointing to
1218 * the reader page). But if the next page is a header page,
1219 * its flags will be non zero.
1220 */
1221static inline int
1222rb_is_head_page(struct buffer_page *page, struct list_head *list)
1223{
1224 unsigned long val;
1225
1226 val = (unsigned long)list->next;
1227
1228 if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list)
1229 return RB_PAGE_MOVED;
1230
1231 return val & RB_FLAG_MASK;
1232}
1233
1234/*
1235 * rb_is_reader_page
1236 *
1237 * The unique thing about the reader page, is that, if the
1238 * writer is ever on it, the previous pointer never points
1239 * back to the reader page.
1240 */
1241static bool rb_is_reader_page(struct buffer_page *page)
1242{
1243 struct list_head *list = page->list.prev;
1244
1245 return rb_list_head(list: list->next) != &page->list;
1246}
1247
1248/*
1249 * rb_set_list_to_head - set a list_head to be pointing to head.
1250 */
1251static void rb_set_list_to_head(struct list_head *list)
1252{
1253 unsigned long *ptr;
1254
1255 ptr = (unsigned long *)&list->next;
1256 *ptr |= RB_PAGE_HEAD;
1257 *ptr &= ~RB_PAGE_UPDATE;
1258}
1259
1260/*
1261 * rb_head_page_activate - sets up head page
1262 */
1263static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer)
1264{
1265 struct buffer_page *head;
1266
1267 head = cpu_buffer->head_page;
1268 if (!head)
1269 return;
1270
1271 /*
1272 * Set the previous list pointer to have the HEAD flag.
1273 */
1274 rb_set_list_to_head(list: head->list.prev);
1275
1276 if (cpu_buffer->ring_meta) {
1277 struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
1278 meta->head_buffer = (unsigned long)head->page;
1279 }
1280}
1281
1282static void rb_list_head_clear(struct list_head *list)
1283{
1284 unsigned long *ptr = (unsigned long *)&list->next;
1285
1286 *ptr &= ~RB_FLAG_MASK;
1287}
1288
1289/*
1290 * rb_head_page_deactivate - clears head page ptr (for free list)
1291 */
1292static void
1293rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer)
1294{
1295 struct list_head *hd;
1296
1297 /* Go through the whole list and clear any pointers found. */
1298 rb_list_head_clear(list: cpu_buffer->pages);
1299
1300 list_for_each(hd, cpu_buffer->pages)
1301 rb_list_head_clear(list: hd);
1302}
1303
1304static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer,
1305 struct buffer_page *head,
1306 struct buffer_page *prev,
1307 int old_flag, int new_flag)
1308{
1309 struct list_head *list;
1310 unsigned long val = (unsigned long)&head->list;
1311 unsigned long ret;
1312
1313 list = &prev->list;
1314
1315 val &= ~RB_FLAG_MASK;
1316
1317 ret = cmpxchg((unsigned long *)&list->next,
1318 val | old_flag, val | new_flag);
1319
1320 /* check if the reader took the page */
1321 if ((ret & ~RB_FLAG_MASK) != val)
1322 return RB_PAGE_MOVED;
1323
1324 return ret & RB_FLAG_MASK;
1325}
1326
1327static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer,
1328 struct buffer_page *head,
1329 struct buffer_page *prev,
1330 int old_flag)
1331{
1332 return rb_head_page_set(cpu_buffer, head, prev,
1333 old_flag, RB_PAGE_UPDATE);
1334}
1335
1336static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer,
1337 struct buffer_page *head,
1338 struct buffer_page *prev,
1339 int old_flag)
1340{
1341 return rb_head_page_set(cpu_buffer, head, prev,
1342 old_flag, RB_PAGE_HEAD);
1343}
1344
1345static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer,
1346 struct buffer_page *head,
1347 struct buffer_page *prev,
1348 int old_flag)
1349{
1350 return rb_head_page_set(cpu_buffer, head, prev,
1351 old_flag, RB_PAGE_NORMAL);
1352}
1353
1354static inline void rb_inc_page(struct buffer_page **bpage)
1355{
1356 struct list_head *p = rb_list_head(list: (*bpage)->list.next);
1357
1358 *bpage = list_entry(p, struct buffer_page, list);
1359}
1360
1361static inline void rb_dec_page(struct buffer_page **bpage)
1362{
1363 struct list_head *p = rb_list_head(list: (*bpage)->list.prev);
1364
1365 *bpage = list_entry(p, struct buffer_page, list);
1366}
1367
1368static struct buffer_page *
1369rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer)
1370{
1371 struct buffer_page *head;
1372 struct buffer_page *page;
1373 struct list_head *list;
1374 int i;
1375
1376 if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page))
1377 return NULL;
1378
1379 /* sanity check */
1380 list = cpu_buffer->pages;
1381 if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list))
1382 return NULL;
1383
1384 page = head = cpu_buffer->head_page;
1385 /*
1386 * It is possible that the writer moves the header behind
1387 * where we started, and we miss in one loop.
1388 * A second loop should grab the header, but we'll do
1389 * three loops just because I'm paranoid.
1390 */
1391 for (i = 0; i < 3; i++) {
1392 do {
1393 if (rb_is_head_page(page, list: page->list.prev)) {
1394 cpu_buffer->head_page = page;
1395 return page;
1396 }
1397 rb_inc_page(bpage: &page);
1398 } while (page != head);
1399 }
1400
1401 RB_WARN_ON(cpu_buffer, 1);
1402
1403 return NULL;
1404}
1405
1406static bool rb_head_page_replace(struct buffer_page *old,
1407 struct buffer_page *new)
1408{
1409 unsigned long *ptr = (unsigned long *)&old->list.prev->next;
1410 unsigned long val;
1411
1412 val = *ptr & ~RB_FLAG_MASK;
1413 val |= RB_PAGE_HEAD;
1414
1415 return try_cmpxchg(ptr, &val, (unsigned long)&new->list);
1416}
1417
1418/*
1419 * rb_tail_page_update - move the tail page forward
1420 */
1421static void rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer,
1422 struct buffer_page *tail_page,
1423 struct buffer_page *next_page)
1424{
1425 unsigned long old_entries;
1426 unsigned long old_write;
1427
1428 /*
1429 * The tail page now needs to be moved forward.
1430 *
1431 * We need to reset the tail page, but without messing
1432 * with possible erasing of data brought in by interrupts
1433 * that have moved the tail page and are currently on it.
1434 *
1435 * We add a counter to the write field to denote this.
1436 */
1437 old_write = local_add_return(RB_WRITE_INTCNT, l: &next_page->write);
1438 old_entries = local_add_return(RB_WRITE_INTCNT, l: &next_page->entries);
1439
1440 /*
1441 * Just make sure we have seen our old_write and synchronize
1442 * with any interrupts that come in.
1443 */
1444 barrier();
1445
1446 /*
1447 * If the tail page is still the same as what we think
1448 * it is, then it is up to us to update the tail
1449 * pointer.
1450 */
1451 if (tail_page == READ_ONCE(cpu_buffer->tail_page)) {
1452 /* Zero the write counter */
1453 unsigned long val = old_write & ~RB_WRITE_MASK;
1454 unsigned long eval = old_entries & ~RB_WRITE_MASK;
1455
1456 /*
1457 * This will only succeed if an interrupt did
1458 * not come in and change it. In which case, we
1459 * do not want to modify it.
1460 *
1461 * We add (void) to let the compiler know that we do not care
1462 * about the return value of these functions. We use the
1463 * cmpxchg to only update if an interrupt did not already
1464 * do it for us. If the cmpxchg fails, we don't care.
1465 */
1466 (void)local_cmpxchg(l: &next_page->write, old: old_write, new: val);
1467 (void)local_cmpxchg(l: &next_page->entries, old: old_entries, new: eval);
1468
1469 /*
1470 * No need to worry about races with clearing out the commit.
1471 * it only can increment when a commit takes place. But that
1472 * only happens in the outer most nested commit.
1473 */
1474 local_set(&next_page->page->commit, 0);
1475
1476 /* Either we update tail_page or an interrupt does */
1477 if (try_cmpxchg(&cpu_buffer->tail_page, &tail_page, next_page))
1478 local_inc(l: &cpu_buffer->pages_touched);
1479 }
1480}
1481
1482static void rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer,
1483 struct buffer_page *bpage)
1484{
1485 unsigned long val = (unsigned long)bpage;
1486
1487 RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK);
1488}
1489
1490static bool rb_check_links(struct ring_buffer_per_cpu *cpu_buffer,
1491 struct list_head *list)
1492{
1493 if (RB_WARN_ON(cpu_buffer,
1494 rb_list_head(rb_list_head(list->next)->prev) != list))
1495 return false;
1496
1497 if (RB_WARN_ON(cpu_buffer,
1498 rb_list_head(rb_list_head(list->prev)->next) != list))
1499 return false;
1500
1501 return true;
1502}
1503
1504/**
1505 * rb_check_pages - integrity check of buffer pages
1506 * @cpu_buffer: CPU buffer with pages to test
1507 *
1508 * As a safety measure we check to make sure the data pages have not
1509 * been corrupted.
1510 */
1511static void rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
1512{
1513 struct list_head *head, *tmp;
1514 unsigned long buffer_cnt;
1515 unsigned long flags;
1516 int nr_loops = 0;
1517
1518 /*
1519 * Walk the linked list underpinning the ring buffer and validate all
1520 * its next and prev links.
1521 *
1522 * The check acquires the reader_lock to avoid concurrent processing
1523 * with code that could be modifying the list. However, the lock cannot
1524 * be held for the entire duration of the walk, as this would make the
1525 * time when interrupts are disabled non-deterministic, dependent on the
1526 * ring buffer size. Therefore, the code releases and re-acquires the
1527 * lock after checking each page. The ring_buffer_per_cpu.cnt variable
1528 * is then used to detect if the list was modified while the lock was
1529 * not held, in which case the check needs to be restarted.
1530 *
1531 * The code attempts to perform the check at most three times before
1532 * giving up. This is acceptable because this is only a self-validation
1533 * to detect problems early on. In practice, the list modification
1534 * operations are fairly spaced, and so this check typically succeeds at
1535 * most on the second try.
1536 */
1537again:
1538 if (++nr_loops > 3)
1539 return;
1540
1541 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1542 head = rb_list_head(list: cpu_buffer->pages);
1543 if (!rb_check_links(cpu_buffer, list: head))
1544 goto out_locked;
1545 buffer_cnt = cpu_buffer->cnt;
1546 tmp = head;
1547 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1548
1549 while (true) {
1550 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1551
1552 if (buffer_cnt != cpu_buffer->cnt) {
1553 /* The list was updated, try again. */
1554 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1555 goto again;
1556 }
1557
1558 tmp = rb_list_head(list: tmp->next);
1559 if (tmp == head)
1560 /* The iteration circled back, all is done. */
1561 goto out_locked;
1562
1563 if (!rb_check_links(cpu_buffer, list: tmp))
1564 goto out_locked;
1565
1566 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1567 }
1568
1569out_locked:
1570 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
1571}
1572
1573/*
1574 * Take an address, add the meta data size as well as the array of
1575 * array subbuffer indexes, then align it to a subbuffer size.
1576 *
1577 * This is used to help find the next per cpu subbuffer within a mapped range.
1578 */
1579static unsigned long
1580rb_range_align_subbuf(unsigned long addr, int subbuf_size, int nr_subbufs)
1581{
1582 addr += sizeof(struct ring_buffer_cpu_meta) +
1583 sizeof(int) * nr_subbufs;
1584 return ALIGN(addr, subbuf_size);
1585}
1586
1587/*
1588 * Return the ring_buffer_meta for a given @cpu.
1589 */
1590static void *rb_range_meta(struct trace_buffer *buffer, int nr_pages, int cpu)
1591{
1592 int subbuf_size = buffer->subbuf_size + BUF_PAGE_HDR_SIZE;
1593 struct ring_buffer_cpu_meta *meta;
1594 struct ring_buffer_meta *bmeta;
1595 unsigned long ptr;
1596 int nr_subbufs;
1597
1598 bmeta = buffer->meta;
1599 if (!bmeta)
1600 return NULL;
1601
1602 ptr = (unsigned long)bmeta + bmeta->buffers_offset;
1603 meta = (struct ring_buffer_cpu_meta *)ptr;
1604
1605 /* When nr_pages passed in is zero, the first meta has already been initialized */
1606 if (!nr_pages) {
1607 nr_subbufs = meta->nr_subbufs;
1608 } else {
1609 /* Include the reader page */
1610 nr_subbufs = nr_pages + 1;
1611 }
1612
1613 /*
1614 * The first chunk may not be subbuffer aligned, where as
1615 * the rest of the chunks are.
1616 */
1617 if (cpu) {
1618 ptr = rb_range_align_subbuf(addr: ptr, subbuf_size, nr_subbufs);
1619 ptr += subbuf_size * nr_subbufs;
1620
1621 /* We can use multiplication to find chunks greater than 1 */
1622 if (cpu > 1) {
1623 unsigned long size;
1624 unsigned long p;
1625
1626 /* Save the beginning of this CPU chunk */
1627 p = ptr;
1628 ptr = rb_range_align_subbuf(addr: ptr, subbuf_size, nr_subbufs);
1629 ptr += subbuf_size * nr_subbufs;
1630
1631 /* Now all chunks after this are the same size */
1632 size = ptr - p;
1633 ptr += size * (cpu - 2);
1634 }
1635 }
1636 return (void *)ptr;
1637}
1638
1639/* Return the start of subbufs given the meta pointer */
1640static void *rb_subbufs_from_meta(struct ring_buffer_cpu_meta *meta)
1641{
1642 int subbuf_size = meta->subbuf_size;
1643 unsigned long ptr;
1644
1645 ptr = (unsigned long)meta;
1646 ptr = rb_range_align_subbuf(addr: ptr, subbuf_size, nr_subbufs: meta->nr_subbufs);
1647
1648 return (void *)ptr;
1649}
1650
1651/*
1652 * Return a specific sub-buffer for a given @cpu defined by @idx.
1653 */
1654static void *rb_range_buffer(struct ring_buffer_per_cpu *cpu_buffer, int idx)
1655{
1656 struct ring_buffer_cpu_meta *meta;
1657 unsigned long ptr;
1658 int subbuf_size;
1659
1660 meta = rb_range_meta(buffer: cpu_buffer->buffer, nr_pages: 0, cpu: cpu_buffer->cpu);
1661 if (!meta)
1662 return NULL;
1663
1664 if (WARN_ON_ONCE(idx >= meta->nr_subbufs))
1665 return NULL;
1666
1667 subbuf_size = meta->subbuf_size;
1668
1669 /* Map this buffer to the order that's in meta->buffers[] */
1670 idx = meta->buffers[idx];
1671
1672 ptr = (unsigned long)rb_subbufs_from_meta(meta);
1673
1674 ptr += subbuf_size * idx;
1675 if (ptr + subbuf_size > cpu_buffer->buffer->range_addr_end)
1676 return NULL;
1677
1678 return (void *)ptr;
1679}
1680
1681/*
1682 * See if the existing memory contains a valid meta section.
1683 * if so, use that, otherwise initialize it.
1684 */
1685static bool rb_meta_init(struct trace_buffer *buffer, int scratch_size)
1686{
1687 unsigned long ptr = buffer->range_addr_start;
1688 struct ring_buffer_meta *bmeta;
1689 unsigned long total_size;
1690 int struct_sizes;
1691
1692 bmeta = (struct ring_buffer_meta *)ptr;
1693 buffer->meta = bmeta;
1694
1695 total_size = buffer->range_addr_end - buffer->range_addr_start;
1696
1697 struct_sizes = sizeof(struct ring_buffer_cpu_meta);
1698 struct_sizes |= sizeof(*bmeta) << 16;
1699
1700 /* The first buffer will start word size after the meta page */
1701 ptr += sizeof(*bmeta);
1702 ptr = ALIGN(ptr, sizeof(long));
1703 ptr += scratch_size;
1704
1705 if (bmeta->magic != RING_BUFFER_META_MAGIC) {
1706 pr_info("Ring buffer boot meta mismatch of magic\n");
1707 goto init;
1708 }
1709
1710 if (bmeta->struct_sizes != struct_sizes) {
1711 pr_info("Ring buffer boot meta mismatch of struct size\n");
1712 goto init;
1713 }
1714
1715 if (bmeta->total_size != total_size) {
1716 pr_info("Ring buffer boot meta mismatch of total size\n");
1717 goto init;
1718 }
1719
1720 if (bmeta->buffers_offset > bmeta->total_size) {
1721 pr_info("Ring buffer boot meta mismatch of offset outside of total size\n");
1722 goto init;
1723 }
1724
1725 if (bmeta->buffers_offset != (void *)ptr - (void *)bmeta) {
1726 pr_info("Ring buffer boot meta mismatch of first buffer offset\n");
1727 goto init;
1728 }
1729
1730 return true;
1731
1732 init:
1733 bmeta->magic = RING_BUFFER_META_MAGIC;
1734 bmeta->struct_sizes = struct_sizes;
1735 bmeta->total_size = total_size;
1736 bmeta->buffers_offset = (void *)ptr - (void *)bmeta;
1737
1738 /* Zero out the scatch pad */
1739 memset(s: (void *)bmeta + sizeof(*bmeta), c: 0, n: bmeta->buffers_offset - sizeof(*bmeta));
1740
1741 return false;
1742}
1743
1744/*
1745 * See if the existing memory contains valid ring buffer data.
1746 * As the previous kernel must be the same as this kernel, all
1747 * the calculations (size of buffers and number of buffers)
1748 * must be the same.
1749 */
1750static bool rb_cpu_meta_valid(struct ring_buffer_cpu_meta *meta, int cpu,
1751 struct trace_buffer *buffer, int nr_pages,
1752 unsigned long *subbuf_mask)
1753{
1754 int subbuf_size = PAGE_SIZE;
1755 struct buffer_data_page *subbuf;
1756 unsigned long buffers_start;
1757 unsigned long buffers_end;
1758 int i;
1759
1760 if (!subbuf_mask)
1761 return false;
1762
1763 buffers_start = meta->first_buffer;
1764 buffers_end = meta->first_buffer + (subbuf_size * meta->nr_subbufs);
1765
1766 /* Is the head and commit buffers within the range of buffers? */
1767 if (meta->head_buffer < buffers_start ||
1768 meta->head_buffer >= buffers_end) {
1769 pr_info("Ring buffer boot meta [%d] head buffer out of range\n", cpu);
1770 return false;
1771 }
1772
1773 if (meta->commit_buffer < buffers_start ||
1774 meta->commit_buffer >= buffers_end) {
1775 pr_info("Ring buffer boot meta [%d] commit buffer out of range\n", cpu);
1776 return false;
1777 }
1778
1779 subbuf = rb_subbufs_from_meta(meta);
1780
1781 bitmap_clear(map: subbuf_mask, start: 0, nbits: meta->nr_subbufs);
1782
1783 /* Is the meta buffers and the subbufs themselves have correct data? */
1784 for (i = 0; i < meta->nr_subbufs; i++) {
1785 if (meta->buffers[i] < 0 ||
1786 meta->buffers[i] >= meta->nr_subbufs) {
1787 pr_info("Ring buffer boot meta [%d] array out of range\n", cpu);
1788 return false;
1789 }
1790
1791 if ((unsigned)local_read(&subbuf->commit) > subbuf_size) {
1792 pr_info("Ring buffer boot meta [%d] buffer invalid commit\n", cpu);
1793 return false;
1794 }
1795
1796 if (test_bit(meta->buffers[i], subbuf_mask)) {
1797 pr_info("Ring buffer boot meta [%d] array has duplicates\n", cpu);
1798 return false;
1799 }
1800
1801 set_bit(nr: meta->buffers[i], addr: subbuf_mask);
1802 subbuf = (void *)subbuf + subbuf_size;
1803 }
1804
1805 return true;
1806}
1807
1808static int rb_meta_subbuf_idx(struct ring_buffer_cpu_meta *meta, void *subbuf);
1809
1810static int rb_read_data_buffer(struct buffer_data_page *dpage, int tail, int cpu,
1811 unsigned long long *timestamp, u64 *delta_ptr)
1812{
1813 struct ring_buffer_event *event;
1814 u64 ts, delta;
1815 int events = 0;
1816 int e;
1817
1818 *delta_ptr = 0;
1819 *timestamp = 0;
1820
1821 ts = dpage->time_stamp;
1822
1823 for (e = 0; e < tail; e += rb_event_length(event)) {
1824
1825 event = (struct ring_buffer_event *)(dpage->data + e);
1826
1827 switch (event->type_len) {
1828
1829 case RINGBUF_TYPE_TIME_EXTEND:
1830 delta = rb_event_time_stamp(event);
1831 ts += delta;
1832 break;
1833
1834 case RINGBUF_TYPE_TIME_STAMP:
1835 delta = rb_event_time_stamp(event);
1836 delta = rb_fix_abs_ts(abs: delta, save_ts: ts);
1837 if (delta < ts) {
1838 *delta_ptr = delta;
1839 *timestamp = ts;
1840 return -1;
1841 }
1842 ts = delta;
1843 break;
1844
1845 case RINGBUF_TYPE_PADDING:
1846 if (event->time_delta == 1)
1847 break;
1848 fallthrough;
1849 case RINGBUF_TYPE_DATA:
1850 events++;
1851 ts += event->time_delta;
1852 break;
1853
1854 default:
1855 return -1;
1856 }
1857 }
1858 *timestamp = ts;
1859 return events;
1860}
1861
1862static int rb_validate_buffer(struct buffer_data_page *dpage, int cpu)
1863{
1864 unsigned long long ts;
1865 u64 delta;
1866 int tail;
1867
1868 tail = local_read(&dpage->commit);
1869 return rb_read_data_buffer(dpage, tail, cpu, timestamp: &ts, delta_ptr: &delta);
1870}
1871
1872/* If the meta data has been validated, now validate the events */
1873static void rb_meta_validate_events(struct ring_buffer_per_cpu *cpu_buffer)
1874{
1875 struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
1876 struct buffer_page *head_page, *orig_head;
1877 unsigned long entry_bytes = 0;
1878 unsigned long entries = 0;
1879 int ret;
1880 u64 ts;
1881 int i;
1882
1883 if (!meta || !meta->head_buffer)
1884 return;
1885
1886 /* Do the reader page first */
1887 ret = rb_validate_buffer(dpage: cpu_buffer->reader_page->page, cpu: cpu_buffer->cpu);
1888 if (ret < 0) {
1889 pr_info("Ring buffer reader page is invalid\n");
1890 goto invalid;
1891 }
1892 entries += ret;
1893 entry_bytes += local_read(&cpu_buffer->reader_page->page->commit);
1894 local_set(&cpu_buffer->reader_page->entries, ret);
1895
1896 orig_head = head_page = cpu_buffer->head_page;
1897 ts = head_page->page->time_stamp;
1898
1899 /*
1900 * Try to rewind the head so that we can read the pages which already
1901 * read in the previous boot.
1902 */
1903 if (head_page == cpu_buffer->tail_page)
1904 goto skip_rewind;
1905
1906 rb_dec_page(bpage: &head_page);
1907 for (i = 0; i < meta->nr_subbufs + 1; i++, rb_dec_page(bpage: &head_page)) {
1908
1909 /* Rewind until tail (writer) page. */
1910 if (head_page == cpu_buffer->tail_page)
1911 break;
1912
1913 /* Ensure the page has older data than head. */
1914 if (ts < head_page->page->time_stamp)
1915 break;
1916
1917 ts = head_page->page->time_stamp;
1918 /* Ensure the page has correct timestamp and some data. */
1919 if (!ts || rb_page_commit(bpage: head_page) == 0)
1920 break;
1921
1922 /* Stop rewind if the page is invalid. */
1923 ret = rb_validate_buffer(dpage: head_page->page, cpu: cpu_buffer->cpu);
1924 if (ret < 0)
1925 break;
1926
1927 /* Recover the number of entries and update stats. */
1928 local_set(&head_page->entries, ret);
1929 if (ret)
1930 local_inc(l: &cpu_buffer->pages_touched);
1931 entries += ret;
1932 entry_bytes += rb_page_commit(bpage: head_page);
1933 }
1934 if (i)
1935 pr_info("Ring buffer [%d] rewound %d pages\n", cpu_buffer->cpu, i);
1936
1937 /* The last rewound page must be skipped. */
1938 if (head_page != orig_head)
1939 rb_inc_page(bpage: &head_page);
1940
1941 /*
1942 * If the ring buffer was rewound, then inject the reader page
1943 * into the location just before the original head page.
1944 */
1945 if (head_page != orig_head) {
1946 struct buffer_page *bpage = orig_head;
1947
1948 rb_dec_page(bpage: &bpage);
1949 /*
1950 * Insert the reader_page before the original head page.
1951 * Since the list encode RB_PAGE flags, general list
1952 * operations should be avoided.
1953 */
1954 cpu_buffer->reader_page->list.next = &orig_head->list;
1955 cpu_buffer->reader_page->list.prev = orig_head->list.prev;
1956 orig_head->list.prev = &cpu_buffer->reader_page->list;
1957 bpage->list.next = &cpu_buffer->reader_page->list;
1958
1959 /* Make the head_page the reader page */
1960 cpu_buffer->reader_page = head_page;
1961 bpage = head_page;
1962 rb_inc_page(bpage: &head_page);
1963 head_page->list.prev = bpage->list.prev;
1964 rb_dec_page(bpage: &bpage);
1965 bpage->list.next = &head_page->list;
1966 rb_set_list_to_head(list: &bpage->list);
1967 cpu_buffer->pages = &head_page->list;
1968
1969 cpu_buffer->head_page = head_page;
1970 meta->head_buffer = (unsigned long)head_page->page;
1971
1972 /* Reset all the indexes */
1973 bpage = cpu_buffer->reader_page;
1974 meta->buffers[0] = rb_meta_subbuf_idx(meta, subbuf: bpage->page);
1975 bpage->id = 0;
1976
1977 for (i = 1, bpage = head_page; i < meta->nr_subbufs;
1978 i++, rb_inc_page(bpage: &bpage)) {
1979 meta->buffers[i] = rb_meta_subbuf_idx(meta, subbuf: bpage->page);
1980 bpage->id = i;
1981 }
1982
1983 /* We'll restart verifying from orig_head */
1984 head_page = orig_head;
1985 }
1986
1987 skip_rewind:
1988 /* If the commit_buffer is the reader page, update the commit page */
1989 if (meta->commit_buffer == (unsigned long)cpu_buffer->reader_page->page) {
1990 cpu_buffer->commit_page = cpu_buffer->reader_page;
1991 /* Nothing more to do, the only page is the reader page */
1992 goto done;
1993 }
1994
1995 /* Iterate until finding the commit page */
1996 for (i = 0; i < meta->nr_subbufs + 1; i++, rb_inc_page(bpage: &head_page)) {
1997
1998 /* Reader page has already been done */
1999 if (head_page == cpu_buffer->reader_page)
2000 continue;
2001
2002 ret = rb_validate_buffer(dpage: head_page->page, cpu: cpu_buffer->cpu);
2003 if (ret < 0) {
2004 pr_info("Ring buffer meta [%d] invalid buffer page\n",
2005 cpu_buffer->cpu);
2006 goto invalid;
2007 }
2008
2009 /* If the buffer has content, update pages_touched */
2010 if (ret)
2011 local_inc(l: &cpu_buffer->pages_touched);
2012
2013 entries += ret;
2014 entry_bytes += local_read(&head_page->page->commit);
2015 local_set(&cpu_buffer->head_page->entries, ret);
2016
2017 if (head_page == cpu_buffer->commit_page)
2018 break;
2019 }
2020
2021 if (head_page != cpu_buffer->commit_page) {
2022 pr_info("Ring buffer meta [%d] commit page not found\n",
2023 cpu_buffer->cpu);
2024 goto invalid;
2025 }
2026 done:
2027 local_set(&cpu_buffer->entries, entries);
2028 local_set(&cpu_buffer->entries_bytes, entry_bytes);
2029
2030 pr_info("Ring buffer meta [%d] is from previous boot!\n", cpu_buffer->cpu);
2031 return;
2032
2033 invalid:
2034 /* The content of the buffers are invalid, reset the meta data */
2035 meta->head_buffer = 0;
2036 meta->commit_buffer = 0;
2037
2038 /* Reset the reader page */
2039 local_set(&cpu_buffer->reader_page->entries, 0);
2040 local_set(&cpu_buffer->reader_page->page->commit, 0);
2041
2042 /* Reset all the subbuffers */
2043 for (i = 0; i < meta->nr_subbufs - 1; i++, rb_inc_page(bpage: &head_page)) {
2044 local_set(&head_page->entries, 0);
2045 local_set(&head_page->page->commit, 0);
2046 }
2047}
2048
2049static void rb_range_meta_init(struct trace_buffer *buffer, int nr_pages, int scratch_size)
2050{
2051 struct ring_buffer_cpu_meta *meta;
2052 unsigned long *subbuf_mask;
2053 unsigned long delta;
2054 void *subbuf;
2055 bool valid = false;
2056 int cpu;
2057 int i;
2058
2059 /* Create a mask to test the subbuf array */
2060 subbuf_mask = bitmap_alloc(nbits: nr_pages + 1, GFP_KERNEL);
2061 /* If subbuf_mask fails to allocate, then rb_meta_valid() will return false */
2062
2063 if (rb_meta_init(buffer, scratch_size))
2064 valid = true;
2065
2066 for (cpu = 0; cpu < nr_cpu_ids; cpu++) {
2067 void *next_meta;
2068
2069 meta = rb_range_meta(buffer, nr_pages, cpu);
2070
2071 if (valid && rb_cpu_meta_valid(meta, cpu, buffer, nr_pages, subbuf_mask)) {
2072 /* Make the mappings match the current address */
2073 subbuf = rb_subbufs_from_meta(meta);
2074 delta = (unsigned long)subbuf - meta->first_buffer;
2075 meta->first_buffer += delta;
2076 meta->head_buffer += delta;
2077 meta->commit_buffer += delta;
2078 continue;
2079 }
2080
2081 if (cpu < nr_cpu_ids - 1)
2082 next_meta = rb_range_meta(buffer, nr_pages, cpu: cpu + 1);
2083 else
2084 next_meta = (void *)buffer->range_addr_end;
2085
2086 memset(s: meta, c: 0, n: next_meta - (void *)meta);
2087
2088 meta->nr_subbufs = nr_pages + 1;
2089 meta->subbuf_size = PAGE_SIZE;
2090
2091 subbuf = rb_subbufs_from_meta(meta);
2092
2093 meta->first_buffer = (unsigned long)subbuf;
2094
2095 /*
2096 * The buffers[] array holds the order of the sub-buffers
2097 * that are after the meta data. The sub-buffers may
2098 * be swapped out when read and inserted into a different
2099 * location of the ring buffer. Although their addresses
2100 * remain the same, the buffers[] array contains the
2101 * index into the sub-buffers holding their actual order.
2102 */
2103 for (i = 0; i < meta->nr_subbufs; i++) {
2104 meta->buffers[i] = i;
2105 rb_init_page(bpage: subbuf);
2106 subbuf += meta->subbuf_size;
2107 }
2108 }
2109 bitmap_free(bitmap: subbuf_mask);
2110}
2111
2112static void *rbm_start(struct seq_file *m, loff_t *pos)
2113{
2114 struct ring_buffer_per_cpu *cpu_buffer = m->private;
2115 struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
2116 unsigned long val;
2117
2118 if (!meta)
2119 return NULL;
2120
2121 if (*pos > meta->nr_subbufs)
2122 return NULL;
2123
2124 val = *pos;
2125 val++;
2126
2127 return (void *)val;
2128}
2129
2130static void *rbm_next(struct seq_file *m, void *v, loff_t *pos)
2131{
2132 (*pos)++;
2133
2134 return rbm_start(m, pos);
2135}
2136
2137static int rbm_show(struct seq_file *m, void *v)
2138{
2139 struct ring_buffer_per_cpu *cpu_buffer = m->private;
2140 struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
2141 unsigned long val = (unsigned long)v;
2142
2143 if (val == 1) {
2144 seq_printf(m, fmt: "head_buffer: %d\n",
2145 rb_meta_subbuf_idx(meta, subbuf: (void *)meta->head_buffer));
2146 seq_printf(m, fmt: "commit_buffer: %d\n",
2147 rb_meta_subbuf_idx(meta, subbuf: (void *)meta->commit_buffer));
2148 seq_printf(m, fmt: "subbuf_size: %d\n", meta->subbuf_size);
2149 seq_printf(m, fmt: "nr_subbufs: %d\n", meta->nr_subbufs);
2150 return 0;
2151 }
2152
2153 val -= 2;
2154 seq_printf(m, fmt: "buffer[%ld]: %d\n", val, meta->buffers[val]);
2155
2156 return 0;
2157}
2158
2159static void rbm_stop(struct seq_file *m, void *p)
2160{
2161}
2162
2163static const struct seq_operations rb_meta_seq_ops = {
2164 .start = rbm_start,
2165 .next = rbm_next,
2166 .show = rbm_show,
2167 .stop = rbm_stop,
2168};
2169
2170int ring_buffer_meta_seq_init(struct file *file, struct trace_buffer *buffer, int cpu)
2171{
2172 struct seq_file *m;
2173 int ret;
2174
2175 ret = seq_open(file, &rb_meta_seq_ops);
2176 if (ret)
2177 return ret;
2178
2179 m = file->private_data;
2180 m->private = buffer->buffers[cpu];
2181
2182 return 0;
2183}
2184
2185/* Map the buffer_pages to the previous head and commit pages */
2186static void rb_meta_buffer_update(struct ring_buffer_per_cpu *cpu_buffer,
2187 struct buffer_page *bpage)
2188{
2189 struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
2190
2191 if (meta->head_buffer == (unsigned long)bpage->page)
2192 cpu_buffer->head_page = bpage;
2193
2194 if (meta->commit_buffer == (unsigned long)bpage->page) {
2195 cpu_buffer->commit_page = bpage;
2196 cpu_buffer->tail_page = bpage;
2197 }
2198}
2199
2200static int __rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
2201 long nr_pages, struct list_head *pages)
2202{
2203 struct trace_buffer *buffer = cpu_buffer->buffer;
2204 struct ring_buffer_cpu_meta *meta = NULL;
2205 struct buffer_page *bpage, *tmp;
2206 bool user_thread = current->mm != NULL;
2207 gfp_t mflags;
2208 long i;
2209
2210 /*
2211 * Check if the available memory is there first.
2212 * Note, si_mem_available() only gives us a rough estimate of available
2213 * memory. It may not be accurate. But we don't care, we just want
2214 * to prevent doing any allocation when it is obvious that it is
2215 * not going to succeed.
2216 */
2217 i = si_mem_available();
2218 if (i < nr_pages)
2219 return -ENOMEM;
2220
2221 /*
2222 * __GFP_RETRY_MAYFAIL flag makes sure that the allocation fails
2223 * gracefully without invoking oom-killer and the system is not
2224 * destabilized.
2225 */
2226 mflags = GFP_KERNEL | __GFP_RETRY_MAYFAIL;
2227
2228 /*
2229 * If a user thread allocates too much, and si_mem_available()
2230 * reports there's enough memory, even though there is not.
2231 * Make sure the OOM killer kills this thread. This can happen
2232 * even with RETRY_MAYFAIL because another task may be doing
2233 * an allocation after this task has taken all memory.
2234 * This is the task the OOM killer needs to take out during this
2235 * loop, even if it was triggered by an allocation somewhere else.
2236 */
2237 if (user_thread)
2238 set_current_oom_origin();
2239
2240 if (buffer->range_addr_start)
2241 meta = rb_range_meta(buffer, nr_pages, cpu: cpu_buffer->cpu);
2242
2243 for (i = 0; i < nr_pages; i++) {
2244 struct page *page;
2245
2246 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
2247 mflags, cpu_to_node(cpu_buffer->cpu));
2248 if (!bpage)
2249 goto free_pages;
2250
2251 rb_check_bpage(cpu_buffer, bpage);
2252
2253 /*
2254 * Append the pages as for mapped buffers we want to keep
2255 * the order
2256 */
2257 list_add_tail(new: &bpage->list, head: pages);
2258
2259 if (meta) {
2260 /* A range was given. Use that for the buffer page */
2261 bpage->page = rb_range_buffer(cpu_buffer, idx: i + 1);
2262 if (!bpage->page)
2263 goto free_pages;
2264 /* If this is valid from a previous boot */
2265 if (meta->head_buffer)
2266 rb_meta_buffer_update(cpu_buffer, bpage);
2267 bpage->range = 1;
2268 bpage->id = i + 1;
2269 } else {
2270 page = alloc_pages_node(cpu_to_node(cpu_buffer->cpu),
2271 mflags | __GFP_COMP | __GFP_ZERO,
2272 cpu_buffer->buffer->subbuf_order);
2273 if (!page)
2274 goto free_pages;
2275 bpage->page = page_address(page);
2276 rb_init_page(bpage: bpage->page);
2277 }
2278 bpage->order = cpu_buffer->buffer->subbuf_order;
2279
2280 if (user_thread && fatal_signal_pending(current))
2281 goto free_pages;
2282 }
2283 if (user_thread)
2284 clear_current_oom_origin();
2285
2286 return 0;
2287
2288free_pages:
2289 list_for_each_entry_safe(bpage, tmp, pages, list) {
2290 list_del_init(entry: &bpage->list);
2291 free_buffer_page(bpage);
2292 }
2293 if (user_thread)
2294 clear_current_oom_origin();
2295
2296 return -ENOMEM;
2297}
2298
2299static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
2300 unsigned long nr_pages)
2301{
2302 LIST_HEAD(pages);
2303
2304 WARN_ON(!nr_pages);
2305
2306 if (__rb_allocate_pages(cpu_buffer, nr_pages, pages: &pages))
2307 return -ENOMEM;
2308
2309 /*
2310 * The ring buffer page list is a circular list that does not
2311 * start and end with a list head. All page list items point to
2312 * other pages.
2313 */
2314 cpu_buffer->pages = pages.next;
2315 list_del(entry: &pages);
2316
2317 cpu_buffer->nr_pages = nr_pages;
2318
2319 rb_check_pages(cpu_buffer);
2320
2321 return 0;
2322}
2323
2324static struct ring_buffer_per_cpu *
2325rb_allocate_cpu_buffer(struct trace_buffer *buffer, long nr_pages, int cpu)
2326{
2327 struct ring_buffer_per_cpu *cpu_buffer __free(kfree) = NULL;
2328 struct ring_buffer_cpu_meta *meta;
2329 struct buffer_page *bpage;
2330 struct page *page;
2331 int ret;
2332
2333 cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
2334 GFP_KERNEL, cpu_to_node(cpu));
2335 if (!cpu_buffer)
2336 return NULL;
2337
2338 cpu_buffer->cpu = cpu;
2339 cpu_buffer->buffer = buffer;
2340 raw_spin_lock_init(&cpu_buffer->reader_lock);
2341 lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key);
2342 cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
2343 INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler);
2344 init_completion(x: &cpu_buffer->update_done);
2345 init_irq_work(work: &cpu_buffer->irq_work.work, func: rb_wake_up_waiters);
2346 init_waitqueue_head(&cpu_buffer->irq_work.waiters);
2347 init_waitqueue_head(&cpu_buffer->irq_work.full_waiters);
2348 mutex_init(&cpu_buffer->mapping_lock);
2349
2350 bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
2351 GFP_KERNEL, cpu_to_node(cpu));
2352 if (!bpage)
2353 return NULL;
2354
2355 rb_check_bpage(cpu_buffer, bpage);
2356
2357 cpu_buffer->reader_page = bpage;
2358
2359 if (buffer->range_addr_start) {
2360 /*
2361 * Range mapped buffers have the same restrictions as memory
2362 * mapped ones do.
2363 */
2364 cpu_buffer->mapped = 1;
2365 cpu_buffer->ring_meta = rb_range_meta(buffer, nr_pages, cpu);
2366 bpage->page = rb_range_buffer(cpu_buffer, idx: 0);
2367 if (!bpage->page)
2368 goto fail_free_reader;
2369 if (cpu_buffer->ring_meta->head_buffer)
2370 rb_meta_buffer_update(cpu_buffer, bpage);
2371 bpage->range = 1;
2372 } else {
2373 page = alloc_pages_node(cpu_to_node(cpu),
2374 GFP_KERNEL | __GFP_COMP | __GFP_ZERO,
2375 cpu_buffer->buffer->subbuf_order);
2376 if (!page)
2377 goto fail_free_reader;
2378 bpage->page = page_address(page);
2379 rb_init_page(bpage: bpage->page);
2380 }
2381
2382 INIT_LIST_HEAD(list: &cpu_buffer->reader_page->list);
2383 INIT_LIST_HEAD(list: &cpu_buffer->new_pages);
2384
2385 ret = rb_allocate_pages(cpu_buffer, nr_pages);
2386 if (ret < 0)
2387 goto fail_free_reader;
2388
2389 rb_meta_validate_events(cpu_buffer);
2390
2391 /* If the boot meta was valid then this has already been updated */
2392 meta = cpu_buffer->ring_meta;
2393 if (!meta || !meta->head_buffer ||
2394 !cpu_buffer->head_page || !cpu_buffer->commit_page || !cpu_buffer->tail_page) {
2395 if (meta && meta->head_buffer &&
2396 (cpu_buffer->head_page || cpu_buffer->commit_page || cpu_buffer->tail_page)) {
2397 pr_warn("Ring buffer meta buffers not all mapped\n");
2398 if (!cpu_buffer->head_page)
2399 pr_warn(" Missing head_page\n");
2400 if (!cpu_buffer->commit_page)
2401 pr_warn(" Missing commit_page\n");
2402 if (!cpu_buffer->tail_page)
2403 pr_warn(" Missing tail_page\n");
2404 }
2405
2406 cpu_buffer->head_page
2407 = list_entry(cpu_buffer->pages, struct buffer_page, list);
2408 cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
2409
2410 rb_head_page_activate(cpu_buffer);
2411
2412 if (cpu_buffer->ring_meta)
2413 meta->commit_buffer = meta->head_buffer;
2414 } else {
2415 /* The valid meta buffer still needs to activate the head page */
2416 rb_head_page_activate(cpu_buffer);
2417 }
2418
2419 return_ptr(cpu_buffer);
2420
2421 fail_free_reader:
2422 free_buffer_page(bpage: cpu_buffer->reader_page);
2423
2424 return NULL;
2425}
2426
2427static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
2428{
2429 struct list_head *head = cpu_buffer->pages;
2430 struct buffer_page *bpage, *tmp;
2431
2432 irq_work_sync(work: &cpu_buffer->irq_work.work);
2433
2434 free_buffer_page(bpage: cpu_buffer->reader_page);
2435
2436 if (head) {
2437 rb_head_page_deactivate(cpu_buffer);
2438
2439 list_for_each_entry_safe(bpage, tmp, head, list) {
2440 list_del_init(entry: &bpage->list);
2441 free_buffer_page(bpage);
2442 }
2443 bpage = list_entry(head, struct buffer_page, list);
2444 free_buffer_page(bpage);
2445 }
2446
2447 free_page((unsigned long)cpu_buffer->free_page);
2448
2449 kfree(objp: cpu_buffer);
2450}
2451
2452static struct trace_buffer *alloc_buffer(unsigned long size, unsigned flags,
2453 int order, unsigned long start,
2454 unsigned long end,
2455 unsigned long scratch_size,
2456 struct lock_class_key *key)
2457{
2458 struct trace_buffer *buffer __free(kfree) = NULL;
2459 long nr_pages;
2460 int subbuf_size;
2461 int bsize;
2462 int cpu;
2463 int ret;
2464
2465 /* keep it in its own cache line */
2466 buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
2467 GFP_KERNEL);
2468 if (!buffer)
2469 return NULL;
2470
2471 if (!zalloc_cpumask_var(mask: &buffer->cpumask, GFP_KERNEL))
2472 return NULL;
2473
2474 buffer->subbuf_order = order;
2475 subbuf_size = (PAGE_SIZE << order);
2476 buffer->subbuf_size = subbuf_size - BUF_PAGE_HDR_SIZE;
2477
2478 /* Max payload is buffer page size - header (8bytes) */
2479 buffer->max_data_size = buffer->subbuf_size - (sizeof(u32) * 2);
2480
2481 buffer->flags = flags;
2482 buffer->clock = trace_clock_local;
2483 buffer->reader_lock_key = key;
2484
2485 init_irq_work(work: &buffer->irq_work.work, func: rb_wake_up_waiters);
2486 init_waitqueue_head(&buffer->irq_work.waiters);
2487
2488 buffer->cpus = nr_cpu_ids;
2489
2490 bsize = sizeof(void *) * nr_cpu_ids;
2491 buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()),
2492 GFP_KERNEL);
2493 if (!buffer->buffers)
2494 goto fail_free_cpumask;
2495
2496 /* If start/end are specified, then that overrides size */
2497 if (start && end) {
2498 unsigned long buffers_start;
2499 unsigned long ptr;
2500 int n;
2501
2502 /* Make sure that start is word aligned */
2503 start = ALIGN(start, sizeof(long));
2504
2505 /* scratch_size needs to be aligned too */
2506 scratch_size = ALIGN(scratch_size, sizeof(long));
2507
2508 /* Subtract the buffer meta data and word aligned */
2509 buffers_start = start + sizeof(struct ring_buffer_cpu_meta);
2510 buffers_start = ALIGN(buffers_start, sizeof(long));
2511 buffers_start += scratch_size;
2512
2513 /* Calculate the size for the per CPU data */
2514 size = end - buffers_start;
2515 size = size / nr_cpu_ids;
2516
2517 /*
2518 * The number of sub-buffers (nr_pages) is determined by the
2519 * total size allocated minus the meta data size.
2520 * Then that is divided by the number of per CPU buffers
2521 * needed, plus account for the integer array index that
2522 * will be appended to the meta data.
2523 */
2524 nr_pages = (size - sizeof(struct ring_buffer_cpu_meta)) /
2525 (subbuf_size + sizeof(int));
2526 /* Need at least two pages plus the reader page */
2527 if (nr_pages < 3)
2528 goto fail_free_buffers;
2529
2530 again:
2531 /* Make sure that the size fits aligned */
2532 for (n = 0, ptr = buffers_start; n < nr_cpu_ids; n++) {
2533 ptr += sizeof(struct ring_buffer_cpu_meta) +
2534 sizeof(int) * nr_pages;
2535 ptr = ALIGN(ptr, subbuf_size);
2536 ptr += subbuf_size * nr_pages;
2537 }
2538 if (ptr > end) {
2539 if (nr_pages <= 3)
2540 goto fail_free_buffers;
2541 nr_pages--;
2542 goto again;
2543 }
2544
2545 /* nr_pages should not count the reader page */
2546 nr_pages--;
2547 buffer->range_addr_start = start;
2548 buffer->range_addr_end = end;
2549
2550 rb_range_meta_init(buffer, nr_pages, scratch_size);
2551 } else {
2552
2553 /* need at least two pages */
2554 nr_pages = DIV_ROUND_UP(size, buffer->subbuf_size);
2555 if (nr_pages < 2)
2556 nr_pages = 2;
2557 }
2558
2559 cpu = raw_smp_processor_id();
2560 cpumask_set_cpu(cpu, dstp: buffer->cpumask);
2561 buffer->buffers[cpu] = rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
2562 if (!buffer->buffers[cpu])
2563 goto fail_free_buffers;
2564
2565 ret = cpuhp_state_add_instance(state: CPUHP_TRACE_RB_PREPARE, node: &buffer->node);
2566 if (ret < 0)
2567 goto fail_free_buffers;
2568
2569 mutex_init(&buffer->mutex);
2570
2571 return_ptr(buffer);
2572
2573 fail_free_buffers:
2574 for_each_buffer_cpu(buffer, cpu) {
2575 if (buffer->buffers[cpu])
2576 rb_free_cpu_buffer(cpu_buffer: buffer->buffers[cpu]);
2577 }
2578 kfree(objp: buffer->buffers);
2579
2580 fail_free_cpumask:
2581 free_cpumask_var(mask: buffer->cpumask);
2582
2583 return NULL;
2584}
2585
2586/**
2587 * __ring_buffer_alloc - allocate a new ring_buffer
2588 * @size: the size in bytes per cpu that is needed.
2589 * @flags: attributes to set for the ring buffer.
2590 * @key: ring buffer reader_lock_key.
2591 *
2592 * Currently the only flag that is available is the RB_FL_OVERWRITE
2593 * flag. This flag means that the buffer will overwrite old data
2594 * when the buffer wraps. If this flag is not set, the buffer will
2595 * drop data when the tail hits the head.
2596 */
2597struct trace_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags,
2598 struct lock_class_key *key)
2599{
2600 /* Default buffer page size - one system page */
2601 return alloc_buffer(size, flags, order: 0, start: 0, end: 0, scratch_size: 0, key);
2602
2603}
2604EXPORT_SYMBOL_GPL(__ring_buffer_alloc);
2605
2606/**
2607 * __ring_buffer_alloc_range - allocate a new ring_buffer from existing memory
2608 * @size: the size in bytes per cpu that is needed.
2609 * @flags: attributes to set for the ring buffer.
2610 * @order: sub-buffer order
2611 * @start: start of allocated range
2612 * @range_size: size of allocated range
2613 * @scratch_size: size of scratch area (for preallocated memory buffers)
2614 * @key: ring buffer reader_lock_key.
2615 *
2616 * Currently the only flag that is available is the RB_FL_OVERWRITE
2617 * flag. This flag means that the buffer will overwrite old data
2618 * when the buffer wraps. If this flag is not set, the buffer will
2619 * drop data when the tail hits the head.
2620 */
2621struct trace_buffer *__ring_buffer_alloc_range(unsigned long size, unsigned flags,
2622 int order, unsigned long start,
2623 unsigned long range_size,
2624 unsigned long scratch_size,
2625 struct lock_class_key *key)
2626{
2627 return alloc_buffer(size, flags, order, start, end: start + range_size,
2628 scratch_size, key);
2629}
2630
2631void *ring_buffer_meta_scratch(struct trace_buffer *buffer, unsigned int *size)
2632{
2633 struct ring_buffer_meta *meta;
2634 void *ptr;
2635
2636 if (!buffer || !buffer->meta)
2637 return NULL;
2638
2639 meta = buffer->meta;
2640
2641 ptr = (void *)ALIGN((unsigned long)meta + sizeof(*meta), sizeof(long));
2642
2643 if (size)
2644 *size = (void *)meta + meta->buffers_offset - ptr;
2645
2646 return ptr;
2647}
2648
2649/**
2650 * ring_buffer_free - free a ring buffer.
2651 * @buffer: the buffer to free.
2652 */
2653void
2654ring_buffer_free(struct trace_buffer *buffer)
2655{
2656 int cpu;
2657
2658 cpuhp_state_remove_instance(state: CPUHP_TRACE_RB_PREPARE, node: &buffer->node);
2659
2660 irq_work_sync(work: &buffer->irq_work.work);
2661
2662 for_each_buffer_cpu(buffer, cpu)
2663 rb_free_cpu_buffer(cpu_buffer: buffer->buffers[cpu]);
2664
2665 kfree(objp: buffer->buffers);
2666 free_cpumask_var(mask: buffer->cpumask);
2667
2668 kfree(objp: buffer);
2669}
2670EXPORT_SYMBOL_GPL(ring_buffer_free);
2671
2672void ring_buffer_set_clock(struct trace_buffer *buffer,
2673 u64 (*clock)(void))
2674{
2675 buffer->clock = clock;
2676}
2677
2678void ring_buffer_set_time_stamp_abs(struct trace_buffer *buffer, bool abs)
2679{
2680 buffer->time_stamp_abs = abs;
2681}
2682
2683bool ring_buffer_time_stamp_abs(struct trace_buffer *buffer)
2684{
2685 return buffer->time_stamp_abs;
2686}
2687
2688static inline unsigned long rb_page_entries(struct buffer_page *bpage)
2689{
2690 return local_read(&bpage->entries) & RB_WRITE_MASK;
2691}
2692
2693static inline unsigned long rb_page_write(struct buffer_page *bpage)
2694{
2695 return local_read(&bpage->write) & RB_WRITE_MASK;
2696}
2697
2698static bool
2699rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned long nr_pages)
2700{
2701 struct list_head *tail_page, *to_remove, *next_page;
2702 struct buffer_page *to_remove_page, *tmp_iter_page;
2703 struct buffer_page *last_page, *first_page;
2704 unsigned long nr_removed;
2705 unsigned long head_bit;
2706 int page_entries;
2707
2708 head_bit = 0;
2709
2710 raw_spin_lock_irq(&cpu_buffer->reader_lock);
2711 atomic_inc(v: &cpu_buffer->record_disabled);
2712 /*
2713 * We don't race with the readers since we have acquired the reader
2714 * lock. We also don't race with writers after disabling recording.
2715 * This makes it easy to figure out the first and the last page to be
2716 * removed from the list. We unlink all the pages in between including
2717 * the first and last pages. This is done in a busy loop so that we
2718 * lose the least number of traces.
2719 * The pages are freed after we restart recording and unlock readers.
2720 */
2721 tail_page = &cpu_buffer->tail_page->list;
2722
2723 /*
2724 * tail page might be on reader page, we remove the next page
2725 * from the ring buffer
2726 */
2727 if (cpu_buffer->tail_page == cpu_buffer->reader_page)
2728 tail_page = rb_list_head(list: tail_page->next);
2729 to_remove = tail_page;
2730
2731 /* start of pages to remove */
2732 first_page = list_entry(rb_list_head(to_remove->next),
2733 struct buffer_page, list);
2734
2735 for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) {
2736 to_remove = rb_list_head(list: to_remove)->next;
2737 head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD;
2738 }
2739 /* Read iterators need to reset themselves when some pages removed */
2740 cpu_buffer->pages_removed += nr_removed;
2741
2742 next_page = rb_list_head(list: to_remove)->next;
2743
2744 /*
2745 * Now we remove all pages between tail_page and next_page.
2746 * Make sure that we have head_bit value preserved for the
2747 * next page
2748 */
2749 tail_page->next = (struct list_head *)((unsigned long)next_page |
2750 head_bit);
2751 next_page = rb_list_head(list: next_page);
2752 next_page->prev = tail_page;
2753
2754 /* make sure pages points to a valid page in the ring buffer */
2755 cpu_buffer->pages = next_page;
2756 cpu_buffer->cnt++;
2757
2758 /* update head page */
2759 if (head_bit)
2760 cpu_buffer->head_page = list_entry(next_page,
2761 struct buffer_page, list);
2762
2763 /* pages are removed, resume tracing and then free the pages */
2764 atomic_dec(v: &cpu_buffer->record_disabled);
2765 raw_spin_unlock_irq(&cpu_buffer->reader_lock);
2766
2767 RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages));
2768
2769 /* last buffer page to remove */
2770 last_page = list_entry(rb_list_head(to_remove), struct buffer_page,
2771 list);
2772 tmp_iter_page = first_page;
2773
2774 do {
2775 cond_resched();
2776
2777 to_remove_page = tmp_iter_page;
2778 rb_inc_page(bpage: &tmp_iter_page);
2779
2780 /* update the counters */
2781 page_entries = rb_page_entries(bpage: to_remove_page);
2782 if (page_entries) {
2783 /*
2784 * If something was added to this page, it was full
2785 * since it is not the tail page. So we deduct the
2786 * bytes consumed in ring buffer from here.
2787 * Increment overrun to account for the lost events.
2788 */
2789 local_add(i: page_entries, l: &cpu_buffer->overrun);
2790 local_sub(i: rb_page_commit(bpage: to_remove_page), l: &cpu_buffer->entries_bytes);
2791 local_inc(l: &cpu_buffer->pages_lost);
2792 }
2793
2794 /*
2795 * We have already removed references to this list item, just
2796 * free up the buffer_page and its page
2797 */
2798 free_buffer_page(bpage: to_remove_page);
2799 nr_removed--;
2800
2801 } while (to_remove_page != last_page);
2802
2803 RB_WARN_ON(cpu_buffer, nr_removed);
2804
2805 return nr_removed == 0;
2806}
2807
2808static bool
2809rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer)
2810{
2811 struct list_head *pages = &cpu_buffer->new_pages;
2812 unsigned long flags;
2813 bool success;
2814 int retries;
2815
2816 /* Can be called at early boot up, where interrupts must not been enabled */
2817 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2818 /*
2819 * We are holding the reader lock, so the reader page won't be swapped
2820 * in the ring buffer. Now we are racing with the writer trying to
2821 * move head page and the tail page.
2822 * We are going to adapt the reader page update process where:
2823 * 1. We first splice the start and end of list of new pages between
2824 * the head page and its previous page.
2825 * 2. We cmpxchg the prev_page->next to point from head page to the
2826 * start of new pages list.
2827 * 3. Finally, we update the head->prev to the end of new list.
2828 *
2829 * We will try this process 10 times, to make sure that we don't keep
2830 * spinning.
2831 */
2832 retries = 10;
2833 success = false;
2834 while (retries--) {
2835 struct list_head *head_page, *prev_page;
2836 struct list_head *last_page, *first_page;
2837 struct list_head *head_page_with_bit;
2838 struct buffer_page *hpage = rb_set_head_page(cpu_buffer);
2839
2840 if (!hpage)
2841 break;
2842 head_page = &hpage->list;
2843 prev_page = head_page->prev;
2844
2845 first_page = pages->next;
2846 last_page = pages->prev;
2847
2848 head_page_with_bit = (struct list_head *)
2849 ((unsigned long)head_page | RB_PAGE_HEAD);
2850
2851 last_page->next = head_page_with_bit;
2852 first_page->prev = prev_page;
2853
2854 /* caution: head_page_with_bit gets updated on cmpxchg failure */
2855 if (try_cmpxchg(&prev_page->next,
2856 &head_page_with_bit, first_page)) {
2857 /*
2858 * yay, we replaced the page pointer to our new list,
2859 * now, we just have to update to head page's prev
2860 * pointer to point to end of list
2861 */
2862 head_page->prev = last_page;
2863 cpu_buffer->cnt++;
2864 success = true;
2865 break;
2866 }
2867 }
2868
2869 if (success)
2870 INIT_LIST_HEAD(list: pages);
2871 /*
2872 * If we weren't successful in adding in new pages, warn and stop
2873 * tracing
2874 */
2875 RB_WARN_ON(cpu_buffer, !success);
2876 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2877
2878 /* free pages if they weren't inserted */
2879 if (!success) {
2880 struct buffer_page *bpage, *tmp;
2881 list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
2882 list) {
2883 list_del_init(entry: &bpage->list);
2884 free_buffer_page(bpage);
2885 }
2886 }
2887 return success;
2888}
2889
2890static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer)
2891{
2892 bool success;
2893
2894 if (cpu_buffer->nr_pages_to_update > 0)
2895 success = rb_insert_pages(cpu_buffer);
2896 else
2897 success = rb_remove_pages(cpu_buffer,
2898 nr_pages: -cpu_buffer->nr_pages_to_update);
2899
2900 if (success)
2901 cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update;
2902}
2903
2904static void update_pages_handler(struct work_struct *work)
2905{
2906 struct ring_buffer_per_cpu *cpu_buffer = container_of(work,
2907 struct ring_buffer_per_cpu, update_pages_work);
2908 rb_update_pages(cpu_buffer);
2909 complete(&cpu_buffer->update_done);
2910}
2911
2912/**
2913 * ring_buffer_resize - resize the ring buffer
2914 * @buffer: the buffer to resize.
2915 * @size: the new size.
2916 * @cpu_id: the cpu buffer to resize
2917 *
2918 * Minimum size is 2 * buffer->subbuf_size.
2919 *
2920 * Returns 0 on success and < 0 on failure.
2921 */
2922int ring_buffer_resize(struct trace_buffer *buffer, unsigned long size,
2923 int cpu_id)
2924{
2925 struct ring_buffer_per_cpu *cpu_buffer;
2926 unsigned long nr_pages;
2927 int cpu, err;
2928
2929 /*
2930 * Always succeed at resizing a non-existent buffer:
2931 */
2932 if (!buffer)
2933 return 0;
2934
2935 /* Make sure the requested buffer exists */
2936 if (cpu_id != RING_BUFFER_ALL_CPUS &&
2937 !cpumask_test_cpu(cpu: cpu_id, cpumask: buffer->cpumask))
2938 return 0;
2939
2940 nr_pages = DIV_ROUND_UP(size, buffer->subbuf_size);
2941
2942 /* we need a minimum of two pages */
2943 if (nr_pages < 2)
2944 nr_pages = 2;
2945
2946 /*
2947 * Keep CPUs from coming online while resizing to synchronize
2948 * with new per CPU buffers being created.
2949 */
2950 guard(cpus_read_lock)();
2951
2952 /* prevent another thread from changing buffer sizes */
2953 mutex_lock(lock: &buffer->mutex);
2954 atomic_inc(v: &buffer->resizing);
2955
2956 if (cpu_id == RING_BUFFER_ALL_CPUS) {
2957 /*
2958 * Don't succeed if resizing is disabled, as a reader might be
2959 * manipulating the ring buffer and is expecting a sane state while
2960 * this is true.
2961 */
2962 for_each_buffer_cpu(buffer, cpu) {
2963 cpu_buffer = buffer->buffers[cpu];
2964 if (atomic_read(v: &cpu_buffer->resize_disabled)) {
2965 err = -EBUSY;
2966 goto out_err_unlock;
2967 }
2968 }
2969
2970 /* calculate the pages to update */
2971 for_each_buffer_cpu(buffer, cpu) {
2972 cpu_buffer = buffer->buffers[cpu];
2973
2974 cpu_buffer->nr_pages_to_update = nr_pages -
2975 cpu_buffer->nr_pages;
2976 /*
2977 * nothing more to do for removing pages or no update
2978 */
2979 if (cpu_buffer->nr_pages_to_update <= 0)
2980 continue;
2981 /*
2982 * to add pages, make sure all new pages can be
2983 * allocated without receiving ENOMEM
2984 */
2985 INIT_LIST_HEAD(list: &cpu_buffer->new_pages);
2986 if (__rb_allocate_pages(cpu_buffer, nr_pages: cpu_buffer->nr_pages_to_update,
2987 pages: &cpu_buffer->new_pages)) {
2988 /* not enough memory for new pages */
2989 err = -ENOMEM;
2990 goto out_err;
2991 }
2992
2993 cond_resched();
2994 }
2995
2996 /*
2997 * Fire off all the required work handlers
2998 * We can't schedule on offline CPUs, but it's not necessary
2999 * since we can change their buffer sizes without any race.
3000 */
3001 for_each_buffer_cpu(buffer, cpu) {
3002 cpu_buffer = buffer->buffers[cpu];
3003 if (!cpu_buffer->nr_pages_to_update)
3004 continue;
3005
3006 /* Can't run something on an offline CPU. */
3007 if (!cpu_online(cpu)) {
3008 rb_update_pages(cpu_buffer);
3009 cpu_buffer->nr_pages_to_update = 0;
3010 } else {
3011 /* Run directly if possible. */
3012 migrate_disable();
3013 if (cpu != smp_processor_id()) {
3014 migrate_enable();
3015 schedule_work_on(cpu,
3016 work: &cpu_buffer->update_pages_work);
3017 } else {
3018 update_pages_handler(work: &cpu_buffer->update_pages_work);
3019 migrate_enable();
3020 }
3021 }
3022 }
3023
3024 /* wait for all the updates to complete */
3025 for_each_buffer_cpu(buffer, cpu) {
3026 cpu_buffer = buffer->buffers[cpu];
3027 if (!cpu_buffer->nr_pages_to_update)
3028 continue;
3029
3030 if (cpu_online(cpu))
3031 wait_for_completion(&cpu_buffer->update_done);
3032 cpu_buffer->nr_pages_to_update = 0;
3033 }
3034
3035 } else {
3036 cpu_buffer = buffer->buffers[cpu_id];
3037
3038 if (nr_pages == cpu_buffer->nr_pages)
3039 goto out;
3040
3041 /*
3042 * Don't succeed if resizing is disabled, as a reader might be
3043 * manipulating the ring buffer and is expecting a sane state while
3044 * this is true.
3045 */
3046 if (atomic_read(v: &cpu_buffer->resize_disabled)) {
3047 err = -EBUSY;
3048 goto out_err_unlock;
3049 }
3050
3051 cpu_buffer->nr_pages_to_update = nr_pages -
3052 cpu_buffer->nr_pages;
3053
3054 INIT_LIST_HEAD(list: &cpu_buffer->new_pages);
3055 if (cpu_buffer->nr_pages_to_update > 0 &&
3056 __rb_allocate_pages(cpu_buffer, nr_pages: cpu_buffer->nr_pages_to_update,
3057 pages: &cpu_buffer->new_pages)) {
3058 err = -ENOMEM;
3059 goto out_err;
3060 }
3061
3062 /* Can't run something on an offline CPU. */
3063 if (!cpu_online(cpu: cpu_id))
3064 rb_update_pages(cpu_buffer);
3065 else {
3066 /* Run directly if possible. */
3067 migrate_disable();
3068 if (cpu_id == smp_processor_id()) {
3069 rb_update_pages(cpu_buffer);
3070 migrate_enable();
3071 } else {
3072 migrate_enable();
3073 schedule_work_on(cpu: cpu_id,
3074 work: &cpu_buffer->update_pages_work);
3075 wait_for_completion(&cpu_buffer->update_done);
3076 }
3077 }
3078
3079 cpu_buffer->nr_pages_to_update = 0;
3080 }
3081
3082 out:
3083 /*
3084 * The ring buffer resize can happen with the ring buffer
3085 * enabled, so that the update disturbs the tracing as little
3086 * as possible. But if the buffer is disabled, we do not need
3087 * to worry about that, and we can take the time to verify
3088 * that the buffer is not corrupt.
3089 */
3090 if (atomic_read(v: &buffer->record_disabled)) {
3091 atomic_inc(v: &buffer->record_disabled);
3092 /*
3093 * Even though the buffer was disabled, we must make sure
3094 * that it is truly disabled before calling rb_check_pages.
3095 * There could have been a race between checking
3096 * record_disable and incrementing it.
3097 */
3098 synchronize_rcu();
3099 for_each_buffer_cpu(buffer, cpu) {
3100 cpu_buffer = buffer->buffers[cpu];
3101 rb_check_pages(cpu_buffer);
3102 }
3103 atomic_dec(v: &buffer->record_disabled);
3104 }
3105
3106 atomic_dec(v: &buffer->resizing);
3107 mutex_unlock(lock: &buffer->mutex);
3108 return 0;
3109
3110 out_err:
3111 for_each_buffer_cpu(buffer, cpu) {
3112 struct buffer_page *bpage, *tmp;
3113
3114 cpu_buffer = buffer->buffers[cpu];
3115 cpu_buffer->nr_pages_to_update = 0;
3116
3117 if (list_empty(head: &cpu_buffer->new_pages))
3118 continue;
3119
3120 list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages,
3121 list) {
3122 list_del_init(entry: &bpage->list);
3123 free_buffer_page(bpage);
3124 }
3125 }
3126 out_err_unlock:
3127 atomic_dec(v: &buffer->resizing);
3128 mutex_unlock(lock: &buffer->mutex);
3129 return err;
3130}
3131EXPORT_SYMBOL_GPL(ring_buffer_resize);
3132
3133void ring_buffer_change_overwrite(struct trace_buffer *buffer, int val)
3134{
3135 mutex_lock(lock: &buffer->mutex);
3136 if (val)
3137 buffer->flags |= RB_FL_OVERWRITE;
3138 else
3139 buffer->flags &= ~RB_FL_OVERWRITE;
3140 mutex_unlock(lock: &buffer->mutex);
3141}
3142EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite);
3143
3144static __always_inline void *__rb_page_index(struct buffer_page *bpage, unsigned index)
3145{
3146 return bpage->page->data + index;
3147}
3148
3149static __always_inline struct ring_buffer_event *
3150rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
3151{
3152 return __rb_page_index(bpage: cpu_buffer->reader_page,
3153 index: cpu_buffer->reader_page->read);
3154}
3155
3156static struct ring_buffer_event *
3157rb_iter_head_event(struct ring_buffer_iter *iter)
3158{
3159 struct ring_buffer_event *event;
3160 struct buffer_page *iter_head_page = iter->head_page;
3161 unsigned long commit;
3162 unsigned length;
3163
3164 if (iter->head != iter->next_event)
3165 return iter->event;
3166
3167 /*
3168 * When the writer goes across pages, it issues a cmpxchg which
3169 * is a mb(), which will synchronize with the rmb here.
3170 * (see rb_tail_page_update() and __rb_reserve_next())
3171 */
3172 commit = rb_page_commit(bpage: iter_head_page);
3173 smp_rmb();
3174
3175 /* An event needs to be at least 8 bytes in size */
3176 if (iter->head > commit - 8)
3177 goto reset;
3178
3179 event = __rb_page_index(bpage: iter_head_page, index: iter->head);
3180 length = rb_event_length(event);
3181
3182 /*
3183 * READ_ONCE() doesn't work on functions and we don't want the
3184 * compiler doing any crazy optimizations with length.
3185 */
3186 barrier();
3187
3188 if ((iter->head + length) > commit || length > iter->event_size)
3189 /* Writer corrupted the read? */
3190 goto reset;
3191
3192 memcpy(to: iter->event, from: event, len: length);
3193 /*
3194 * If the page stamp is still the same after this rmb() then the
3195 * event was safely copied without the writer entering the page.
3196 */
3197 smp_rmb();
3198
3199 /* Make sure the page didn't change since we read this */
3200 if (iter->page_stamp != iter_head_page->page->time_stamp ||
3201 commit > rb_page_commit(bpage: iter_head_page))
3202 goto reset;
3203
3204 iter->next_event = iter->head + length;
3205 return iter->event;
3206 reset:
3207 /* Reset to the beginning */
3208 iter->page_stamp = iter->read_stamp = iter->head_page->page->time_stamp;
3209 iter->head = 0;
3210 iter->next_event = 0;
3211 iter->missed_events = 1;
3212 return NULL;
3213}
3214
3215/* Size is determined by what has been committed */
3216static __always_inline unsigned rb_page_size(struct buffer_page *bpage)
3217{
3218 return rb_page_commit(bpage) & ~RB_MISSED_MASK;
3219}
3220
3221static __always_inline unsigned
3222rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
3223{
3224 return rb_page_commit(bpage: cpu_buffer->commit_page);
3225}
3226
3227static __always_inline unsigned
3228rb_event_index(struct ring_buffer_per_cpu *cpu_buffer, struct ring_buffer_event *event)
3229{
3230 unsigned long addr = (unsigned long)event;
3231
3232 addr &= (PAGE_SIZE << cpu_buffer->buffer->subbuf_order) - 1;
3233
3234 return addr - BUF_PAGE_HDR_SIZE;
3235}
3236
3237static void rb_inc_iter(struct ring_buffer_iter *iter)
3238{
3239 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
3240
3241 /*
3242 * The iterator could be on the reader page (it starts there).
3243 * But the head could have moved, since the reader was
3244 * found. Check for this case and assign the iterator
3245 * to the head page instead of next.
3246 */
3247 if (iter->head_page == cpu_buffer->reader_page)
3248 iter->head_page = rb_set_head_page(cpu_buffer);
3249 else
3250 rb_inc_page(bpage: &iter->head_page);
3251
3252 iter->page_stamp = iter->read_stamp = iter->head_page->page->time_stamp;
3253 iter->head = 0;
3254 iter->next_event = 0;
3255}
3256
3257/* Return the index into the sub-buffers for a given sub-buffer */
3258static int rb_meta_subbuf_idx(struct ring_buffer_cpu_meta *meta, void *subbuf)
3259{
3260 void *subbuf_array;
3261
3262 subbuf_array = (void *)meta + sizeof(int) * meta->nr_subbufs;
3263 subbuf_array = (void *)ALIGN((unsigned long)subbuf_array, meta->subbuf_size);
3264 return (subbuf - subbuf_array) / meta->subbuf_size;
3265}
3266
3267static void rb_update_meta_head(struct ring_buffer_per_cpu *cpu_buffer,
3268 struct buffer_page *next_page)
3269{
3270 struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
3271 unsigned long old_head = (unsigned long)next_page->page;
3272 unsigned long new_head;
3273
3274 rb_inc_page(bpage: &next_page);
3275 new_head = (unsigned long)next_page->page;
3276
3277 /*
3278 * Only move it forward once, if something else came in and
3279 * moved it forward, then we don't want to touch it.
3280 */
3281 (void)cmpxchg(&meta->head_buffer, old_head, new_head);
3282}
3283
3284static void rb_update_meta_reader(struct ring_buffer_per_cpu *cpu_buffer,
3285 struct buffer_page *reader)
3286{
3287 struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
3288 void *old_reader = cpu_buffer->reader_page->page;
3289 void *new_reader = reader->page;
3290 int id;
3291
3292 id = reader->id;
3293 cpu_buffer->reader_page->id = id;
3294 reader->id = 0;
3295
3296 meta->buffers[0] = rb_meta_subbuf_idx(meta, subbuf: new_reader);
3297 meta->buffers[id] = rb_meta_subbuf_idx(meta, subbuf: old_reader);
3298
3299 /* The head pointer is the one after the reader */
3300 rb_update_meta_head(cpu_buffer, next_page: reader);
3301}
3302
3303/*
3304 * rb_handle_head_page - writer hit the head page
3305 *
3306 * Returns: +1 to retry page
3307 * 0 to continue
3308 * -1 on error
3309 */
3310static int
3311rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer,
3312 struct buffer_page *tail_page,
3313 struct buffer_page *next_page)
3314{
3315 struct buffer_page *new_head;
3316 int entries;
3317 int type;
3318 int ret;
3319
3320 entries = rb_page_entries(bpage: next_page);
3321
3322 /*
3323 * The hard part is here. We need to move the head
3324 * forward, and protect against both readers on
3325 * other CPUs and writers coming in via interrupts.
3326 */
3327 type = rb_head_page_set_update(cpu_buffer, head: next_page, prev: tail_page,
3328 RB_PAGE_HEAD);
3329
3330 /*
3331 * type can be one of four:
3332 * NORMAL - an interrupt already moved it for us
3333 * HEAD - we are the first to get here.
3334 * UPDATE - we are the interrupt interrupting
3335 * a current move.
3336 * MOVED - a reader on another CPU moved the next
3337 * pointer to its reader page. Give up
3338 * and try again.
3339 */
3340
3341 switch (type) {
3342 case RB_PAGE_HEAD:
3343 /*
3344 * We changed the head to UPDATE, thus
3345 * it is our responsibility to update
3346 * the counters.
3347 */
3348 local_add(i: entries, l: &cpu_buffer->overrun);
3349 local_sub(i: rb_page_commit(bpage: next_page), l: &cpu_buffer->entries_bytes);
3350 local_inc(l: &cpu_buffer->pages_lost);
3351
3352 if (cpu_buffer->ring_meta)
3353 rb_update_meta_head(cpu_buffer, next_page);
3354 /*
3355 * The entries will be zeroed out when we move the
3356 * tail page.
3357 */
3358
3359 /* still more to do */
3360 break;
3361
3362 case RB_PAGE_UPDATE:
3363 /*
3364 * This is an interrupt that interrupt the
3365 * previous update. Still more to do.
3366 */
3367 break;
3368 case RB_PAGE_NORMAL:
3369 /*
3370 * An interrupt came in before the update
3371 * and processed this for us.
3372 * Nothing left to do.
3373 */
3374 return 1;
3375 case RB_PAGE_MOVED:
3376 /*
3377 * The reader is on another CPU and just did
3378 * a swap with our next_page.
3379 * Try again.
3380 */
3381 return 1;
3382 default:
3383 RB_WARN_ON(cpu_buffer, 1); /* WTF??? */
3384 return -1;
3385 }
3386
3387 /*
3388 * Now that we are here, the old head pointer is
3389 * set to UPDATE. This will keep the reader from
3390 * swapping the head page with the reader page.
3391 * The reader (on another CPU) will spin till
3392 * we are finished.
3393 *
3394 * We just need to protect against interrupts
3395 * doing the job. We will set the next pointer
3396 * to HEAD. After that, we set the old pointer
3397 * to NORMAL, but only if it was HEAD before.
3398 * otherwise we are an interrupt, and only
3399 * want the outer most commit to reset it.
3400 */
3401 new_head = next_page;
3402 rb_inc_page(bpage: &new_head);
3403
3404 ret = rb_head_page_set_head(cpu_buffer, head: new_head, prev: next_page,
3405 RB_PAGE_NORMAL);
3406
3407 /*
3408 * Valid returns are:
3409 * HEAD - an interrupt came in and already set it.
3410 * NORMAL - One of two things:
3411 * 1) We really set it.
3412 * 2) A bunch of interrupts came in and moved
3413 * the page forward again.
3414 */
3415 switch (ret) {
3416 case RB_PAGE_HEAD:
3417 case RB_PAGE_NORMAL:
3418 /* OK */
3419 break;
3420 default:
3421 RB_WARN_ON(cpu_buffer, 1);
3422 return -1;
3423 }
3424
3425 /*
3426 * It is possible that an interrupt came in,
3427 * set the head up, then more interrupts came in
3428 * and moved it again. When we get back here,
3429 * the page would have been set to NORMAL but we
3430 * just set it back to HEAD.
3431 *
3432 * How do you detect this? Well, if that happened
3433 * the tail page would have moved.
3434 */
3435 if (ret == RB_PAGE_NORMAL) {
3436 struct buffer_page *buffer_tail_page;
3437
3438 buffer_tail_page = READ_ONCE(cpu_buffer->tail_page);
3439 /*
3440 * If the tail had moved passed next, then we need
3441 * to reset the pointer.
3442 */
3443 if (buffer_tail_page != tail_page &&
3444 buffer_tail_page != next_page)
3445 rb_head_page_set_normal(cpu_buffer, head: new_head,
3446 prev: next_page,
3447 RB_PAGE_HEAD);
3448 }
3449
3450 /*
3451 * If this was the outer most commit (the one that
3452 * changed the original pointer from HEAD to UPDATE),
3453 * then it is up to us to reset it to NORMAL.
3454 */
3455 if (type == RB_PAGE_HEAD) {
3456 ret = rb_head_page_set_normal(cpu_buffer, head: next_page,
3457 prev: tail_page,
3458 RB_PAGE_UPDATE);
3459 if (RB_WARN_ON(cpu_buffer,
3460 ret != RB_PAGE_UPDATE))
3461 return -1;
3462 }
3463
3464 return 0;
3465}
3466
3467static inline void
3468rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer,
3469 unsigned long tail, struct rb_event_info *info)
3470{
3471 unsigned long bsize = READ_ONCE(cpu_buffer->buffer->subbuf_size);
3472 struct buffer_page *tail_page = info->tail_page;
3473 struct ring_buffer_event *event;
3474 unsigned long length = info->length;
3475
3476 /*
3477 * Only the event that crossed the page boundary
3478 * must fill the old tail_page with padding.
3479 */
3480 if (tail >= bsize) {
3481 /*
3482 * If the page was filled, then we still need
3483 * to update the real_end. Reset it to zero
3484 * and the reader will ignore it.
3485 */
3486 if (tail == bsize)
3487 tail_page->real_end = 0;
3488
3489 local_sub(i: length, l: &tail_page->write);
3490 return;
3491 }
3492
3493 event = __rb_page_index(bpage: tail_page, index: tail);
3494
3495 /*
3496 * Save the original length to the meta data.
3497 * This will be used by the reader to add lost event
3498 * counter.
3499 */
3500 tail_page->real_end = tail;
3501
3502 /*
3503 * If this event is bigger than the minimum size, then
3504 * we need to be careful that we don't subtract the
3505 * write counter enough to allow another writer to slip
3506 * in on this page.
3507 * We put in a discarded commit instead, to make sure
3508 * that this space is not used again, and this space will
3509 * not be accounted into 'entries_bytes'.
3510 *
3511 * If we are less than the minimum size, we don't need to
3512 * worry about it.
3513 */
3514 if (tail > (bsize - RB_EVNT_MIN_SIZE)) {
3515 /* No room for any events */
3516
3517 /* Mark the rest of the page with padding */
3518 rb_event_set_padding(event);
3519
3520 /* Make sure the padding is visible before the write update */
3521 smp_wmb();
3522
3523 /* Set the write back to the previous setting */
3524 local_sub(i: length, l: &tail_page->write);
3525 return;
3526 }
3527
3528 /* Put in a discarded event */
3529 event->array[0] = (bsize - tail) - RB_EVNT_HDR_SIZE;
3530 event->type_len = RINGBUF_TYPE_PADDING;
3531 /* time delta must be non zero */
3532 event->time_delta = 1;
3533
3534 /* account for padding bytes */
3535 local_add(i: bsize - tail, l: &cpu_buffer->entries_bytes);
3536
3537 /* Make sure the padding is visible before the tail_page->write update */
3538 smp_wmb();
3539
3540 /* Set write to end of buffer */
3541 length = (tail + length) - bsize;
3542 local_sub(i: length, l: &tail_page->write);
3543}
3544
3545static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer);
3546
3547/*
3548 * This is the slow path, force gcc not to inline it.
3549 */
3550static noinline struct ring_buffer_event *
3551rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer,
3552 unsigned long tail, struct rb_event_info *info)
3553{
3554 struct buffer_page *tail_page = info->tail_page;
3555 struct buffer_page *commit_page = cpu_buffer->commit_page;
3556 struct trace_buffer *buffer = cpu_buffer->buffer;
3557 struct buffer_page *next_page;
3558 int ret;
3559
3560 next_page = tail_page;
3561
3562 rb_inc_page(bpage: &next_page);
3563
3564 /*
3565 * If for some reason, we had an interrupt storm that made
3566 * it all the way around the buffer, bail, and warn
3567 * about it.
3568 */
3569 if (unlikely(next_page == commit_page)) {
3570 local_inc(l: &cpu_buffer->commit_overrun);
3571 goto out_reset;
3572 }
3573
3574 /*
3575 * This is where the fun begins!
3576 *
3577 * We are fighting against races between a reader that
3578 * could be on another CPU trying to swap its reader
3579 * page with the buffer head.
3580 *
3581 * We are also fighting against interrupts coming in and
3582 * moving the head or tail on us as well.
3583 *
3584 * If the next page is the head page then we have filled
3585 * the buffer, unless the commit page is still on the
3586 * reader page.
3587 */
3588 if (rb_is_head_page(page: next_page, list: &tail_page->list)) {
3589
3590 /*
3591 * If the commit is not on the reader page, then
3592 * move the header page.
3593 */
3594 if (!rb_is_reader_page(page: cpu_buffer->commit_page)) {
3595 /*
3596 * If we are not in overwrite mode,
3597 * this is easy, just stop here.
3598 */
3599 if (!(buffer->flags & RB_FL_OVERWRITE)) {
3600 local_inc(l: &cpu_buffer->dropped_events);
3601 goto out_reset;
3602 }
3603
3604 ret = rb_handle_head_page(cpu_buffer,
3605 tail_page,
3606 next_page);
3607 if (ret < 0)
3608 goto out_reset;
3609 if (ret)
3610 goto out_again;
3611 } else {
3612 /*
3613 * We need to be careful here too. The
3614 * commit page could still be on the reader
3615 * page. We could have a small buffer, and
3616 * have filled up the buffer with events
3617 * from interrupts and such, and wrapped.
3618 *
3619 * Note, if the tail page is also on the
3620 * reader_page, we let it move out.
3621 */
3622 if (unlikely((cpu_buffer->commit_page !=
3623 cpu_buffer->tail_page) &&
3624 (cpu_buffer->commit_page ==
3625 cpu_buffer->reader_page))) {
3626 local_inc(l: &cpu_buffer->commit_overrun);
3627 goto out_reset;
3628 }
3629 }
3630 }
3631
3632 rb_tail_page_update(cpu_buffer, tail_page, next_page);
3633
3634 out_again:
3635
3636 rb_reset_tail(cpu_buffer, tail, info);
3637
3638 /* Commit what we have for now. */
3639 rb_end_commit(cpu_buffer);
3640 /* rb_end_commit() decs committing */
3641 local_inc(l: &cpu_buffer->committing);
3642
3643 /* fail and let the caller try again */
3644 return ERR_PTR(error: -EAGAIN);
3645
3646 out_reset:
3647 /* reset write */
3648 rb_reset_tail(cpu_buffer, tail, info);
3649
3650 return NULL;
3651}
3652
3653/* Slow path */
3654static struct ring_buffer_event *
3655rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer,
3656 struct ring_buffer_event *event, u64 delta, bool abs)
3657{
3658 if (abs)
3659 event->type_len = RINGBUF_TYPE_TIME_STAMP;
3660 else
3661 event->type_len = RINGBUF_TYPE_TIME_EXTEND;
3662
3663 /* Not the first event on the page, or not delta? */
3664 if (abs || rb_event_index(cpu_buffer, event)) {
3665 event->time_delta = delta & TS_MASK;
3666 event->array[0] = delta >> TS_SHIFT;
3667 } else {
3668 /* nope, just zero it */
3669 event->time_delta = 0;
3670 event->array[0] = 0;
3671 }
3672
3673 return skip_time_extend(event);
3674}
3675
3676#ifndef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
3677static inline bool sched_clock_stable(void)
3678{
3679 return true;
3680}
3681#endif
3682
3683static void
3684rb_check_timestamp(struct ring_buffer_per_cpu *cpu_buffer,
3685 struct rb_event_info *info)
3686{
3687 u64 write_stamp;
3688
3689 WARN_ONCE(1, "Delta way too big! %llu ts=%llu before=%llu after=%llu write stamp=%llu\n%s",
3690 (unsigned long long)info->delta,
3691 (unsigned long long)info->ts,
3692 (unsigned long long)info->before,
3693 (unsigned long long)info->after,
3694 (unsigned long long)({rb_time_read(&cpu_buffer->write_stamp, &write_stamp); write_stamp;}),
3695 sched_clock_stable() ? "" :
3696 "If you just came from a suspend/resume,\n"
3697 "please switch to the trace global clock:\n"
3698 " echo global > /sys/kernel/tracing/trace_clock\n"
3699 "or add trace_clock=global to the kernel command line\n");
3700}
3701
3702static void rb_add_timestamp(struct ring_buffer_per_cpu *cpu_buffer,
3703 struct ring_buffer_event **event,
3704 struct rb_event_info *info,
3705 u64 *delta,
3706 unsigned int *length)
3707{
3708 bool abs = info->add_timestamp &
3709 (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE);
3710
3711 if (unlikely(info->delta > (1ULL << 59))) {
3712 /*
3713 * Some timers can use more than 59 bits, and when a timestamp
3714 * is added to the buffer, it will lose those bits.
3715 */
3716 if (abs && (info->ts & TS_MSB)) {
3717 info->delta &= ABS_TS_MASK;
3718
3719 /* did the clock go backwards */
3720 } else if (info->before == info->after && info->before > info->ts) {
3721 /* not interrupted */
3722 static int once;
3723
3724 /*
3725 * This is possible with a recalibrating of the TSC.
3726 * Do not produce a call stack, but just report it.
3727 */
3728 if (!once) {
3729 once++;
3730 pr_warn("Ring buffer clock went backwards: %llu -> %llu\n",
3731 info->before, info->ts);
3732 }
3733 } else
3734 rb_check_timestamp(cpu_buffer, info);
3735 if (!abs)
3736 info->delta = 0;
3737 }
3738 *event = rb_add_time_stamp(cpu_buffer, event: *event, delta: info->delta, abs);
3739 *length -= RB_LEN_TIME_EXTEND;
3740 *delta = 0;
3741}
3742
3743/**
3744 * rb_update_event - update event type and data
3745 * @cpu_buffer: The per cpu buffer of the @event
3746 * @event: the event to update
3747 * @info: The info to update the @event with (contains length and delta)
3748 *
3749 * Update the type and data fields of the @event. The length
3750 * is the actual size that is written to the ring buffer,
3751 * and with this, we can determine what to place into the
3752 * data field.
3753 */
3754static void
3755rb_update_event(struct ring_buffer_per_cpu *cpu_buffer,
3756 struct ring_buffer_event *event,
3757 struct rb_event_info *info)
3758{
3759 unsigned length = info->length;
3760 u64 delta = info->delta;
3761 unsigned int nest = local_read(&cpu_buffer->committing) - 1;
3762
3763 if (!WARN_ON_ONCE(nest >= MAX_NEST))
3764 cpu_buffer->event_stamp[nest] = info->ts;
3765
3766 /*
3767 * If we need to add a timestamp, then we
3768 * add it to the start of the reserved space.
3769 */
3770 if (unlikely(info->add_timestamp))
3771 rb_add_timestamp(cpu_buffer, event: &event, info, delta: &delta, length: &length);
3772
3773 event->time_delta = delta;
3774 length -= RB_EVNT_HDR_SIZE;
3775 if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) {
3776 event->type_len = 0;
3777 event->array[0] = length;
3778 } else
3779 event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT);
3780}
3781
3782static unsigned rb_calculate_event_length(unsigned length)
3783{
3784 struct ring_buffer_event event; /* Used only for sizeof array */
3785
3786 /* zero length can cause confusions */
3787 if (!length)
3788 length++;
3789
3790 if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT)
3791 length += sizeof(event.array[0]);
3792
3793 length += RB_EVNT_HDR_SIZE;
3794 length = ALIGN(length, RB_ARCH_ALIGNMENT);
3795
3796 /*
3797 * In case the time delta is larger than the 27 bits for it
3798 * in the header, we need to add a timestamp. If another
3799 * event comes in when trying to discard this one to increase
3800 * the length, then the timestamp will be added in the allocated
3801 * space of this event. If length is bigger than the size needed
3802 * for the TIME_EXTEND, then padding has to be used. The events
3803 * length must be either RB_LEN_TIME_EXTEND, or greater than or equal
3804 * to RB_LEN_TIME_EXTEND + 8, as 8 is the minimum size for padding.
3805 * As length is a multiple of 4, we only need to worry if it
3806 * is 12 (RB_LEN_TIME_EXTEND + 4).
3807 */
3808 if (length == RB_LEN_TIME_EXTEND + RB_ALIGNMENT)
3809 length += RB_ALIGNMENT;
3810
3811 return length;
3812}
3813
3814static inline bool
3815rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer,
3816 struct ring_buffer_event *event)
3817{
3818 unsigned long new_index, old_index;
3819 struct buffer_page *bpage;
3820 unsigned long addr;
3821
3822 new_index = rb_event_index(cpu_buffer, event);
3823 old_index = new_index + rb_event_ts_length(event);
3824 addr = (unsigned long)event;
3825 addr &= ~((PAGE_SIZE << cpu_buffer->buffer->subbuf_order) - 1);
3826
3827 bpage = READ_ONCE(cpu_buffer->tail_page);
3828
3829 /*
3830 * Make sure the tail_page is still the same and
3831 * the next write location is the end of this event
3832 */
3833 if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) {
3834 unsigned long write_mask =
3835 local_read(&bpage->write) & ~RB_WRITE_MASK;
3836 unsigned long event_length = rb_event_length(event);
3837
3838 /*
3839 * For the before_stamp to be different than the write_stamp
3840 * to make sure that the next event adds an absolute
3841 * value and does not rely on the saved write stamp, which
3842 * is now going to be bogus.
3843 *
3844 * By setting the before_stamp to zero, the next event
3845 * is not going to use the write_stamp and will instead
3846 * create an absolute timestamp. This means there's no
3847 * reason to update the wirte_stamp!
3848 */
3849 rb_time_set(t: &cpu_buffer->before_stamp, val: 0);
3850
3851 /*
3852 * If an event were to come in now, it would see that the
3853 * write_stamp and the before_stamp are different, and assume
3854 * that this event just added itself before updating
3855 * the write stamp. The interrupting event will fix the
3856 * write stamp for us, and use an absolute timestamp.
3857 */
3858
3859 /*
3860 * This is on the tail page. It is possible that
3861 * a write could come in and move the tail page
3862 * and write to the next page. That is fine
3863 * because we just shorten what is on this page.
3864 */
3865 old_index += write_mask;
3866 new_index += write_mask;
3867
3868 /* caution: old_index gets updated on cmpxchg failure */
3869 if (local_try_cmpxchg(l: &bpage->write, old: &old_index, new: new_index)) {
3870 /* update counters */
3871 local_sub(i: event_length, l: &cpu_buffer->entries_bytes);
3872 return true;
3873 }
3874 }
3875
3876 /* could not discard */
3877 return false;
3878}
3879
3880static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer)
3881{
3882 local_inc(l: &cpu_buffer->committing);
3883 local_inc(l: &cpu_buffer->commits);
3884}
3885
3886static __always_inline void
3887rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
3888{
3889 unsigned long max_count;
3890
3891 /*
3892 * We only race with interrupts and NMIs on this CPU.
3893 * If we own the commit event, then we can commit
3894 * all others that interrupted us, since the interruptions
3895 * are in stack format (they finish before they come
3896 * back to us). This allows us to do a simple loop to
3897 * assign the commit to the tail.
3898 */
3899 again:
3900 max_count = cpu_buffer->nr_pages * 100;
3901
3902 while (cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)) {
3903 if (RB_WARN_ON(cpu_buffer, !(--max_count)))
3904 return;
3905 if (RB_WARN_ON(cpu_buffer,
3906 rb_is_reader_page(cpu_buffer->tail_page)))
3907 return;
3908 /*
3909 * No need for a memory barrier here, as the update
3910 * of the tail_page did it for this page.
3911 */
3912 local_set(&cpu_buffer->commit_page->page->commit,
3913 rb_page_write(cpu_buffer->commit_page));
3914 rb_inc_page(bpage: &cpu_buffer->commit_page);
3915 if (cpu_buffer->ring_meta) {
3916 struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
3917 meta->commit_buffer = (unsigned long)cpu_buffer->commit_page->page;
3918 }
3919 /* add barrier to keep gcc from optimizing too much */
3920 barrier();
3921 }
3922 while (rb_commit_index(cpu_buffer) !=
3923 rb_page_write(bpage: cpu_buffer->commit_page)) {
3924
3925 /* Make sure the readers see the content of what is committed. */
3926 smp_wmb();
3927 local_set(&cpu_buffer->commit_page->page->commit,
3928 rb_page_write(cpu_buffer->commit_page));
3929 RB_WARN_ON(cpu_buffer,
3930 local_read(&cpu_buffer->commit_page->page->commit) &
3931 ~RB_WRITE_MASK);
3932 barrier();
3933 }
3934
3935 /* again, keep gcc from optimizing */
3936 barrier();
3937
3938 /*
3939 * If an interrupt came in just after the first while loop
3940 * and pushed the tail page forward, we will be left with
3941 * a dangling commit that will never go forward.
3942 */
3943 if (unlikely(cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)))
3944 goto again;
3945}
3946
3947static __always_inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer)
3948{
3949 unsigned long commits;
3950
3951 if (RB_WARN_ON(cpu_buffer,
3952 !local_read(&cpu_buffer->committing)))
3953 return;
3954
3955 again:
3956 commits = local_read(&cpu_buffer->commits);
3957 /* synchronize with interrupts */
3958 barrier();
3959 if (local_read(&cpu_buffer->committing) == 1)
3960 rb_set_commit_to_write(cpu_buffer);
3961
3962 local_dec(l: &cpu_buffer->committing);
3963
3964 /* synchronize with interrupts */
3965 barrier();
3966
3967 /*
3968 * Need to account for interrupts coming in between the
3969 * updating of the commit page and the clearing of the
3970 * committing counter.
3971 */
3972 if (unlikely(local_read(&cpu_buffer->commits) != commits) &&
3973 !local_read(&cpu_buffer->committing)) {
3974 local_inc(l: &cpu_buffer->committing);
3975 goto again;
3976 }
3977}
3978
3979static inline void rb_event_discard(struct ring_buffer_event *event)
3980{
3981 if (extended_time(event))
3982 event = skip_time_extend(event);
3983
3984 /* array[0] holds the actual length for the discarded event */
3985 event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE;
3986 event->type_len = RINGBUF_TYPE_PADDING;
3987 /* time delta must be non zero */
3988 if (!event->time_delta)
3989 event->time_delta = 1;
3990}
3991
3992static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer)
3993{
3994 local_inc(l: &cpu_buffer->entries);
3995 rb_end_commit(cpu_buffer);
3996}
3997
3998static __always_inline void
3999rb_wakeups(struct trace_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer)
4000{
4001 if (buffer->irq_work.waiters_pending) {
4002 buffer->irq_work.waiters_pending = false;
4003 /* irq_work_queue() supplies it's own memory barriers */
4004 irq_work_queue(work: &buffer->irq_work.work);
4005 }
4006
4007 if (cpu_buffer->irq_work.waiters_pending) {
4008 cpu_buffer->irq_work.waiters_pending = false;
4009 /* irq_work_queue() supplies it's own memory barriers */
4010 irq_work_queue(work: &cpu_buffer->irq_work.work);
4011 }
4012
4013 if (cpu_buffer->last_pages_touch == local_read(&cpu_buffer->pages_touched))
4014 return;
4015
4016 if (cpu_buffer->reader_page == cpu_buffer->commit_page)
4017 return;
4018
4019 if (!cpu_buffer->irq_work.full_waiters_pending)
4020 return;
4021
4022 cpu_buffer->last_pages_touch = local_read(&cpu_buffer->pages_touched);
4023
4024 if (!full_hit(buffer, cpu: cpu_buffer->cpu, full: cpu_buffer->shortest_full))
4025 return;
4026
4027 cpu_buffer->irq_work.wakeup_full = true;
4028 cpu_buffer->irq_work.full_waiters_pending = false;
4029 /* irq_work_queue() supplies it's own memory barriers */
4030 irq_work_queue(work: &cpu_buffer->irq_work.work);
4031}
4032
4033#ifdef CONFIG_RING_BUFFER_RECORD_RECURSION
4034# define do_ring_buffer_record_recursion() \
4035 do_ftrace_record_recursion(_THIS_IP_, _RET_IP_)
4036#else
4037# define do_ring_buffer_record_recursion() do { } while (0)
4038#endif
4039
4040/*
4041 * The lock and unlock are done within a preempt disable section.
4042 * The current_context per_cpu variable can only be modified
4043 * by the current task between lock and unlock. But it can
4044 * be modified more than once via an interrupt. To pass this
4045 * information from the lock to the unlock without having to
4046 * access the 'in_interrupt()' functions again (which do show
4047 * a bit of overhead in something as critical as function tracing,
4048 * we use a bitmask trick.
4049 *
4050 * bit 1 = NMI context
4051 * bit 2 = IRQ context
4052 * bit 3 = SoftIRQ context
4053 * bit 4 = normal context.
4054 *
4055 * This works because this is the order of contexts that can
4056 * preempt other contexts. A SoftIRQ never preempts an IRQ
4057 * context.
4058 *
4059 * When the context is determined, the corresponding bit is
4060 * checked and set (if it was set, then a recursion of that context
4061 * happened).
4062 *
4063 * On unlock, we need to clear this bit. To do so, just subtract
4064 * 1 from the current_context and AND it to itself.
4065 *
4066 * (binary)
4067 * 101 - 1 = 100
4068 * 101 & 100 = 100 (clearing bit zero)
4069 *
4070 * 1010 - 1 = 1001
4071 * 1010 & 1001 = 1000 (clearing bit 1)
4072 *
4073 * The least significant bit can be cleared this way, and it
4074 * just so happens that it is the same bit corresponding to
4075 * the current context.
4076 *
4077 * Now the TRANSITION bit breaks the above slightly. The TRANSITION bit
4078 * is set when a recursion is detected at the current context, and if
4079 * the TRANSITION bit is already set, it will fail the recursion.
4080 * This is needed because there's a lag between the changing of
4081 * interrupt context and updating the preempt count. In this case,
4082 * a false positive will be found. To handle this, one extra recursion
4083 * is allowed, and this is done by the TRANSITION bit. If the TRANSITION
4084 * bit is already set, then it is considered a recursion and the function
4085 * ends. Otherwise, the TRANSITION bit is set, and that bit is returned.
4086 *
4087 * On the trace_recursive_unlock(), the TRANSITION bit will be the first
4088 * to be cleared. Even if it wasn't the context that set it. That is,
4089 * if an interrupt comes in while NORMAL bit is set and the ring buffer
4090 * is called before preempt_count() is updated, since the check will
4091 * be on the NORMAL bit, the TRANSITION bit will then be set. If an
4092 * NMI then comes in, it will set the NMI bit, but when the NMI code
4093 * does the trace_recursive_unlock() it will clear the TRANSITION bit
4094 * and leave the NMI bit set. But this is fine, because the interrupt
4095 * code that set the TRANSITION bit will then clear the NMI bit when it
4096 * calls trace_recursive_unlock(). If another NMI comes in, it will
4097 * set the TRANSITION bit and continue.
4098 *
4099 * Note: The TRANSITION bit only handles a single transition between context.
4100 */
4101
4102static __always_inline bool
4103trace_recursive_lock(struct ring_buffer_per_cpu *cpu_buffer)
4104{
4105 unsigned int val = cpu_buffer->current_context;
4106 int bit = interrupt_context_level();
4107
4108 bit = RB_CTX_NORMAL - bit;
4109
4110 if (unlikely(val & (1 << (bit + cpu_buffer->nest)))) {
4111 /*
4112 * It is possible that this was called by transitioning
4113 * between interrupt context, and preempt_count() has not
4114 * been updated yet. In this case, use the TRANSITION bit.
4115 */
4116 bit = RB_CTX_TRANSITION;
4117 if (val & (1 << (bit + cpu_buffer->nest))) {
4118 do_ring_buffer_record_recursion();
4119 return true;
4120 }
4121 }
4122
4123 val |= (1 << (bit + cpu_buffer->nest));
4124 cpu_buffer->current_context = val;
4125
4126 return false;
4127}
4128
4129static __always_inline void
4130trace_recursive_unlock(struct ring_buffer_per_cpu *cpu_buffer)
4131{
4132 cpu_buffer->current_context &=
4133 cpu_buffer->current_context - (1 << cpu_buffer->nest);
4134}
4135
4136/* The recursive locking above uses 5 bits */
4137#define NESTED_BITS 5
4138
4139/**
4140 * ring_buffer_nest_start - Allow to trace while nested
4141 * @buffer: The ring buffer to modify
4142 *
4143 * The ring buffer has a safety mechanism to prevent recursion.
4144 * But there may be a case where a trace needs to be done while
4145 * tracing something else. In this case, calling this function
4146 * will allow this function to nest within a currently active
4147 * ring_buffer_lock_reserve().
4148 *
4149 * Call this function before calling another ring_buffer_lock_reserve() and
4150 * call ring_buffer_nest_end() after the nested ring_buffer_unlock_commit().
4151 */
4152void ring_buffer_nest_start(struct trace_buffer *buffer)
4153{
4154 struct ring_buffer_per_cpu *cpu_buffer;
4155 int cpu;
4156
4157 /* Enabled by ring_buffer_nest_end() */
4158 preempt_disable_notrace();
4159 cpu = raw_smp_processor_id();
4160 cpu_buffer = buffer->buffers[cpu];
4161 /* This is the shift value for the above recursive locking */
4162 cpu_buffer->nest += NESTED_BITS;
4163}
4164
4165/**
4166 * ring_buffer_nest_end - Allow to trace while nested
4167 * @buffer: The ring buffer to modify
4168 *
4169 * Must be called after ring_buffer_nest_start() and after the
4170 * ring_buffer_unlock_commit().
4171 */
4172void ring_buffer_nest_end(struct trace_buffer *buffer)
4173{
4174 struct ring_buffer_per_cpu *cpu_buffer;
4175 int cpu;
4176
4177 /* disabled by ring_buffer_nest_start() */
4178 cpu = raw_smp_processor_id();
4179 cpu_buffer = buffer->buffers[cpu];
4180 /* This is the shift value for the above recursive locking */
4181 cpu_buffer->nest -= NESTED_BITS;
4182 preempt_enable_notrace();
4183}
4184
4185/**
4186 * ring_buffer_unlock_commit - commit a reserved
4187 * @buffer: The buffer to commit to
4188 *
4189 * This commits the data to the ring buffer, and releases any locks held.
4190 *
4191 * Must be paired with ring_buffer_lock_reserve.
4192 */
4193int ring_buffer_unlock_commit(struct trace_buffer *buffer)
4194{
4195 struct ring_buffer_per_cpu *cpu_buffer;
4196 int cpu = raw_smp_processor_id();
4197
4198 cpu_buffer = buffer->buffers[cpu];
4199
4200 rb_commit(cpu_buffer);
4201
4202 rb_wakeups(buffer, cpu_buffer);
4203
4204 trace_recursive_unlock(cpu_buffer);
4205
4206 preempt_enable_notrace();
4207
4208 return 0;
4209}
4210EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit);
4211
4212/* Special value to validate all deltas on a page. */
4213#define CHECK_FULL_PAGE 1L
4214
4215#ifdef CONFIG_RING_BUFFER_VALIDATE_TIME_DELTAS
4216
4217static const char *show_irq_str(int bits)
4218{
4219 static const char * type[] = {
4220 ".", // 0
4221 "s", // 1
4222 "h", // 2
4223 "Hs", // 3
4224 "n", // 4
4225 "Ns", // 5
4226 "Nh", // 6
4227 "NHs", // 7
4228 };
4229
4230 return type[bits];
4231}
4232
4233/* Assume this is a trace event */
4234static const char *show_flags(struct ring_buffer_event *event)
4235{
4236 struct trace_entry *entry;
4237 int bits = 0;
4238
4239 if (rb_event_data_length(event) - RB_EVNT_HDR_SIZE < sizeof(*entry))
4240 return "X";
4241
4242 entry = ring_buffer_event_data(event);
4243
4244 if (entry->flags & TRACE_FLAG_SOFTIRQ)
4245 bits |= 1;
4246
4247 if (entry->flags & TRACE_FLAG_HARDIRQ)
4248 bits |= 2;
4249
4250 if (entry->flags & TRACE_FLAG_NMI)
4251 bits |= 4;
4252
4253 return show_irq_str(bits);
4254}
4255
4256static const char *show_irq(struct ring_buffer_event *event)
4257{
4258 struct trace_entry *entry;
4259
4260 if (rb_event_data_length(event) - RB_EVNT_HDR_SIZE < sizeof(*entry))
4261 return "";
4262
4263 entry = ring_buffer_event_data(event);
4264 if (entry->flags & TRACE_FLAG_IRQS_OFF)
4265 return "d";
4266 return "";
4267}
4268
4269static const char *show_interrupt_level(void)
4270{
4271 unsigned long pc = preempt_count();
4272 unsigned char level = 0;
4273
4274 if (pc & SOFTIRQ_OFFSET)
4275 level |= 1;
4276
4277 if (pc & HARDIRQ_MASK)
4278 level |= 2;
4279
4280 if (pc & NMI_MASK)
4281 level |= 4;
4282
4283 return show_irq_str(level);
4284}
4285
4286static void dump_buffer_page(struct buffer_data_page *bpage,
4287 struct rb_event_info *info,
4288 unsigned long tail)
4289{
4290 struct ring_buffer_event *event;
4291 u64 ts, delta;
4292 int e;
4293
4294 ts = bpage->time_stamp;
4295 pr_warn(" [%lld] PAGE TIME STAMP\n", ts);
4296
4297 for (e = 0; e < tail; e += rb_event_length(event)) {
4298
4299 event = (struct ring_buffer_event *)(bpage->data + e);
4300
4301 switch (event->type_len) {
4302
4303 case RINGBUF_TYPE_TIME_EXTEND:
4304 delta = rb_event_time_stamp(event);
4305 ts += delta;
4306 pr_warn(" 0x%x: [%lld] delta:%lld TIME EXTEND\n",
4307 e, ts, delta);
4308 break;
4309
4310 case RINGBUF_TYPE_TIME_STAMP:
4311 delta = rb_event_time_stamp(event);
4312 ts = rb_fix_abs_ts(delta, ts);
4313 pr_warn(" 0x%x: [%lld] absolute:%lld TIME STAMP\n",
4314 e, ts, delta);
4315 break;
4316
4317 case RINGBUF_TYPE_PADDING:
4318 ts += event->time_delta;
4319 pr_warn(" 0x%x: [%lld] delta:%d PADDING\n",
4320 e, ts, event->time_delta);
4321 break;
4322
4323 case RINGBUF_TYPE_DATA:
4324 ts += event->time_delta;
4325 pr_warn(" 0x%x: [%lld] delta:%d %s%s\n",
4326 e, ts, event->time_delta,
4327 show_flags(event), show_irq(event));
4328 break;
4329
4330 default:
4331 break;
4332 }
4333 }
4334 pr_warn("expected end:0x%lx last event actually ended at:0x%x\n", tail, e);
4335}
4336
4337static DEFINE_PER_CPU(atomic_t, checking);
4338static atomic_t ts_dump;
4339
4340#define buffer_warn_return(fmt, ...) \
4341 do { \
4342 /* If another report is happening, ignore this one */ \
4343 if (atomic_inc_return(&ts_dump) != 1) { \
4344 atomic_dec(&ts_dump); \
4345 goto out; \
4346 } \
4347 atomic_inc(&cpu_buffer->record_disabled); \
4348 pr_warn(fmt, ##__VA_ARGS__); \
4349 dump_buffer_page(bpage, info, tail); \
4350 atomic_dec(&ts_dump); \
4351 /* There's some cases in boot up that this can happen */ \
4352 if (WARN_ON_ONCE(system_state != SYSTEM_BOOTING)) \
4353 /* Do not re-enable checking */ \
4354 return; \
4355 } while (0)
4356
4357/*
4358 * Check if the current event time stamp matches the deltas on
4359 * the buffer page.
4360 */
4361static void check_buffer(struct ring_buffer_per_cpu *cpu_buffer,
4362 struct rb_event_info *info,
4363 unsigned long tail)
4364{
4365 struct buffer_data_page *bpage;
4366 u64 ts, delta;
4367 bool full = false;
4368 int ret;
4369
4370 bpage = info->tail_page->page;
4371
4372 if (tail == CHECK_FULL_PAGE) {
4373 full = true;
4374 tail = local_read(&bpage->commit);
4375 } else if (info->add_timestamp &
4376 (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE)) {
4377 /* Ignore events with absolute time stamps */
4378 return;
4379 }
4380
4381 /*
4382 * Do not check the first event (skip possible extends too).
4383 * Also do not check if previous events have not been committed.
4384 */
4385 if (tail <= 8 || tail > local_read(&bpage->commit))
4386 return;
4387
4388 /*
4389 * If this interrupted another event,
4390 */
4391 if (atomic_inc_return(this_cpu_ptr(&checking)) != 1)
4392 goto out;
4393
4394 ret = rb_read_data_buffer(bpage, tail, cpu_buffer->cpu, &ts, &delta);
4395 if (ret < 0) {
4396 if (delta < ts) {
4397 buffer_warn_return("[CPU: %d]ABSOLUTE TIME WENT BACKWARDS: last ts: %lld absolute ts: %lld\n",
4398 cpu_buffer->cpu, ts, delta);
4399 goto out;
4400 }
4401 }
4402 if ((full && ts > info->ts) ||
4403 (!full && ts + info->delta != info->ts)) {
4404 buffer_warn_return("[CPU: %d]TIME DOES NOT MATCH expected:%lld actual:%lld delta:%lld before:%lld after:%lld%s context:%s\n",
4405 cpu_buffer->cpu,
4406 ts + info->delta, info->ts, info->delta,
4407 info->before, info->after,
4408 full ? " (full)" : "", show_interrupt_level());
4409 }
4410out:
4411 atomic_dec(this_cpu_ptr(&checking));
4412}
4413#else
4414static inline void check_buffer(struct ring_buffer_per_cpu *cpu_buffer,
4415 struct rb_event_info *info,
4416 unsigned long tail)
4417{
4418}
4419#endif /* CONFIG_RING_BUFFER_VALIDATE_TIME_DELTAS */
4420
4421static struct ring_buffer_event *
4422__rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
4423 struct rb_event_info *info)
4424{
4425 struct ring_buffer_event *event;
4426 struct buffer_page *tail_page;
4427 unsigned long tail, write, w;
4428
4429 /* Don't let the compiler play games with cpu_buffer->tail_page */
4430 tail_page = info->tail_page = READ_ONCE(cpu_buffer->tail_page);
4431
4432 /*A*/ w = local_read(&tail_page->write) & RB_WRITE_MASK;
4433 barrier();
4434 rb_time_read(t: &cpu_buffer->before_stamp, ret: &info->before);
4435 rb_time_read(t: &cpu_buffer->write_stamp, ret: &info->after);
4436 barrier();
4437 info->ts = rb_time_stamp(buffer: cpu_buffer->buffer);
4438
4439 if ((info->add_timestamp & RB_ADD_STAMP_ABSOLUTE)) {
4440 info->delta = info->ts;
4441 } else {
4442 /*
4443 * If interrupting an event time update, we may need an
4444 * absolute timestamp.
4445 * Don't bother if this is the start of a new page (w == 0).
4446 */
4447 if (!w) {
4448 /* Use the sub-buffer timestamp */
4449 info->delta = 0;
4450 } else if (unlikely(info->before != info->after)) {
4451 info->add_timestamp |= RB_ADD_STAMP_FORCE | RB_ADD_STAMP_EXTEND;
4452 info->length += RB_LEN_TIME_EXTEND;
4453 } else {
4454 info->delta = info->ts - info->after;
4455 if (unlikely(test_time_stamp(info->delta))) {
4456 info->add_timestamp |= RB_ADD_STAMP_EXTEND;
4457 info->length += RB_LEN_TIME_EXTEND;
4458 }
4459 }
4460 }
4461
4462 /*B*/ rb_time_set(t: &cpu_buffer->before_stamp, val: info->ts);
4463
4464 /*C*/ write = local_add_return(i: info->length, l: &tail_page->write);
4465
4466 /* set write to only the index of the write */
4467 write &= RB_WRITE_MASK;
4468
4469 tail = write - info->length;
4470
4471 /* See if we shot pass the end of this buffer page */
4472 if (unlikely(write > cpu_buffer->buffer->subbuf_size)) {
4473 check_buffer(cpu_buffer, info, CHECK_FULL_PAGE);
4474 return rb_move_tail(cpu_buffer, tail, info);
4475 }
4476
4477 if (likely(tail == w)) {
4478 /* Nothing interrupted us between A and C */
4479 /*D*/ rb_time_set(t: &cpu_buffer->write_stamp, val: info->ts);
4480 /*
4481 * If something came in between C and D, the write stamp
4482 * may now not be in sync. But that's fine as the before_stamp
4483 * will be different and then next event will just be forced
4484 * to use an absolute timestamp.
4485 */
4486 if (likely(!(info->add_timestamp &
4487 (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE))))
4488 /* This did not interrupt any time update */
4489 info->delta = info->ts - info->after;
4490 else
4491 /* Just use full timestamp for interrupting event */
4492 info->delta = info->ts;
4493 check_buffer(cpu_buffer, info, tail);
4494 } else {
4495 u64 ts;
4496 /* SLOW PATH - Interrupted between A and C */
4497
4498 /* Save the old before_stamp */
4499 rb_time_read(t: &cpu_buffer->before_stamp, ret: &info->before);
4500
4501 /*
4502 * Read a new timestamp and update the before_stamp to make
4503 * the next event after this one force using an absolute
4504 * timestamp. This is in case an interrupt were to come in
4505 * between E and F.
4506 */
4507 ts = rb_time_stamp(buffer: cpu_buffer->buffer);
4508 rb_time_set(t: &cpu_buffer->before_stamp, val: ts);
4509
4510 barrier();
4511 /*E*/ rb_time_read(t: &cpu_buffer->write_stamp, ret: &info->after);
4512 barrier();
4513 /*F*/ if (write == (local_read(&tail_page->write) & RB_WRITE_MASK) &&
4514 info->after == info->before && info->after < ts) {
4515 /*
4516 * Nothing came after this event between C and F, it is
4517 * safe to use info->after for the delta as it
4518 * matched info->before and is still valid.
4519 */
4520 info->delta = ts - info->after;
4521 } else {
4522 /*
4523 * Interrupted between C and F:
4524 * Lost the previous events time stamp. Just set the
4525 * delta to zero, and this will be the same time as
4526 * the event this event interrupted. And the events that
4527 * came after this will still be correct (as they would
4528 * have built their delta on the previous event.
4529 */
4530 info->delta = 0;
4531 }
4532 info->ts = ts;
4533 info->add_timestamp &= ~RB_ADD_STAMP_FORCE;
4534 }
4535
4536 /*
4537 * If this is the first commit on the page, then it has the same
4538 * timestamp as the page itself.
4539 */
4540 if (unlikely(!tail && !(info->add_timestamp &
4541 (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_ABSOLUTE))))
4542 info->delta = 0;
4543
4544 /* We reserved something on the buffer */
4545
4546 event = __rb_page_index(bpage: tail_page, index: tail);
4547 rb_update_event(cpu_buffer, event, info);
4548
4549 local_inc(l: &tail_page->entries);
4550
4551 /*
4552 * If this is the first commit on the page, then update
4553 * its timestamp.
4554 */
4555 if (unlikely(!tail))
4556 tail_page->page->time_stamp = info->ts;
4557
4558 /* account for these added bytes */
4559 local_add(i: info->length, l: &cpu_buffer->entries_bytes);
4560
4561 return event;
4562}
4563
4564static __always_inline struct ring_buffer_event *
4565rb_reserve_next_event(struct trace_buffer *buffer,
4566 struct ring_buffer_per_cpu *cpu_buffer,
4567 unsigned long length)
4568{
4569 struct ring_buffer_event *event;
4570 struct rb_event_info info;
4571 int nr_loops = 0;
4572 int add_ts_default;
4573
4574 /*
4575 * ring buffer does cmpxchg as well as atomic64 operations
4576 * (which some archs use locking for atomic64), make sure this
4577 * is safe in NMI context
4578 */
4579 if ((!IS_ENABLED(CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG) ||
4580 IS_ENABLED(CONFIG_GENERIC_ATOMIC64)) &&
4581 (unlikely(in_nmi()))) {
4582 return NULL;
4583 }
4584
4585 rb_start_commit(cpu_buffer);
4586 /* The commit page can not change after this */
4587
4588#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4589 /*
4590 * Due to the ability to swap a cpu buffer from a buffer
4591 * it is possible it was swapped before we committed.
4592 * (committing stops a swap). We check for it here and
4593 * if it happened, we have to fail the write.
4594 */
4595 barrier();
4596 if (unlikely(READ_ONCE(cpu_buffer->buffer) != buffer)) {
4597 local_dec(&cpu_buffer->committing);
4598 local_dec(&cpu_buffer->commits);
4599 return NULL;
4600 }
4601#endif
4602
4603 info.length = rb_calculate_event_length(length);
4604
4605 if (ring_buffer_time_stamp_abs(buffer: cpu_buffer->buffer)) {
4606 add_ts_default = RB_ADD_STAMP_ABSOLUTE;
4607 info.length += RB_LEN_TIME_EXTEND;
4608 if (info.length > cpu_buffer->buffer->max_data_size)
4609 goto out_fail;
4610 } else {
4611 add_ts_default = RB_ADD_STAMP_NONE;
4612 }
4613
4614 again:
4615 info.add_timestamp = add_ts_default;
4616 info.delta = 0;
4617
4618 /*
4619 * We allow for interrupts to reenter here and do a trace.
4620 * If one does, it will cause this original code to loop
4621 * back here. Even with heavy interrupts happening, this
4622 * should only happen a few times in a row. If this happens
4623 * 1000 times in a row, there must be either an interrupt
4624 * storm or we have something buggy.
4625 * Bail!
4626 */
4627 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
4628 goto out_fail;
4629
4630 event = __rb_reserve_next(cpu_buffer, info: &info);
4631
4632 if (unlikely(PTR_ERR(event) == -EAGAIN)) {
4633 if (info.add_timestamp & (RB_ADD_STAMP_FORCE | RB_ADD_STAMP_EXTEND))
4634 info.length -= RB_LEN_TIME_EXTEND;
4635 goto again;
4636 }
4637
4638 if (likely(event))
4639 return event;
4640 out_fail:
4641 rb_end_commit(cpu_buffer);
4642 return NULL;
4643}
4644
4645/**
4646 * ring_buffer_lock_reserve - reserve a part of the buffer
4647 * @buffer: the ring buffer to reserve from
4648 * @length: the length of the data to reserve (excluding event header)
4649 *
4650 * Returns a reserved event on the ring buffer to copy directly to.
4651 * The user of this interface will need to get the body to write into
4652 * and can use the ring_buffer_event_data() interface.
4653 *
4654 * The length is the length of the data needed, not the event length
4655 * which also includes the event header.
4656 *
4657 * Must be paired with ring_buffer_unlock_commit, unless NULL is returned.
4658 * If NULL is returned, then nothing has been allocated or locked.
4659 */
4660struct ring_buffer_event *
4661ring_buffer_lock_reserve(struct trace_buffer *buffer, unsigned long length)
4662{
4663 struct ring_buffer_per_cpu *cpu_buffer;
4664 struct ring_buffer_event *event;
4665 int cpu;
4666
4667 /* If we are tracing schedule, we don't want to recurse */
4668 preempt_disable_notrace();
4669
4670 if (unlikely(atomic_read(&buffer->record_disabled)))
4671 goto out;
4672
4673 cpu = raw_smp_processor_id();
4674
4675 if (unlikely(!cpumask_test_cpu(cpu, buffer->cpumask)))
4676 goto out;
4677
4678 cpu_buffer = buffer->buffers[cpu];
4679
4680 if (unlikely(atomic_read(&cpu_buffer->record_disabled)))
4681 goto out;
4682
4683 if (unlikely(length > buffer->max_data_size))
4684 goto out;
4685
4686 if (unlikely(trace_recursive_lock(cpu_buffer)))
4687 goto out;
4688
4689 event = rb_reserve_next_event(buffer, cpu_buffer, length);
4690 if (!event)
4691 goto out_unlock;
4692
4693 return event;
4694
4695 out_unlock:
4696 trace_recursive_unlock(cpu_buffer);
4697 out:
4698 preempt_enable_notrace();
4699 return NULL;
4700}
4701EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve);
4702
4703/*
4704 * Decrement the entries to the page that an event is on.
4705 * The event does not even need to exist, only the pointer
4706 * to the page it is on. This may only be called before the commit
4707 * takes place.
4708 */
4709static inline void
4710rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer,
4711 struct ring_buffer_event *event)
4712{
4713 unsigned long addr = (unsigned long)event;
4714 struct buffer_page *bpage = cpu_buffer->commit_page;
4715 struct buffer_page *start;
4716
4717 addr &= ~((PAGE_SIZE << cpu_buffer->buffer->subbuf_order) - 1);
4718
4719 /* Do the likely case first */
4720 if (likely(bpage->page == (void *)addr)) {
4721 local_dec(l: &bpage->entries);
4722 return;
4723 }
4724
4725 /*
4726 * Because the commit page may be on the reader page we
4727 * start with the next page and check the end loop there.
4728 */
4729 rb_inc_page(bpage: &bpage);
4730 start = bpage;
4731 do {
4732 if (bpage->page == (void *)addr) {
4733 local_dec(l: &bpage->entries);
4734 return;
4735 }
4736 rb_inc_page(bpage: &bpage);
4737 } while (bpage != start);
4738
4739 /* commit not part of this buffer?? */
4740 RB_WARN_ON(cpu_buffer, 1);
4741}
4742
4743/**
4744 * ring_buffer_discard_commit - discard an event that has not been committed
4745 * @buffer: the ring buffer
4746 * @event: non committed event to discard
4747 *
4748 * Sometimes an event that is in the ring buffer needs to be ignored.
4749 * This function lets the user discard an event in the ring buffer
4750 * and then that event will not be read later.
4751 *
4752 * This function only works if it is called before the item has been
4753 * committed. It will try to free the event from the ring buffer
4754 * if another event has not been added behind it.
4755 *
4756 * If another event has been added behind it, it will set the event
4757 * up as discarded, and perform the commit.
4758 *
4759 * If this function is called, do not call ring_buffer_unlock_commit on
4760 * the event.
4761 */
4762void ring_buffer_discard_commit(struct trace_buffer *buffer,
4763 struct ring_buffer_event *event)
4764{
4765 struct ring_buffer_per_cpu *cpu_buffer;
4766 int cpu;
4767
4768 /* The event is discarded regardless */
4769 rb_event_discard(event);
4770
4771 cpu = smp_processor_id();
4772 cpu_buffer = buffer->buffers[cpu];
4773
4774 /*
4775 * This must only be called if the event has not been
4776 * committed yet. Thus we can assume that preemption
4777 * is still disabled.
4778 */
4779 RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing));
4780
4781 rb_decrement_entry(cpu_buffer, event);
4782 rb_try_to_discard(cpu_buffer, event);
4783 rb_end_commit(cpu_buffer);
4784
4785 trace_recursive_unlock(cpu_buffer);
4786
4787 preempt_enable_notrace();
4788
4789}
4790EXPORT_SYMBOL_GPL(ring_buffer_discard_commit);
4791
4792/**
4793 * ring_buffer_write - write data to the buffer without reserving
4794 * @buffer: The ring buffer to write to.
4795 * @length: The length of the data being written (excluding the event header)
4796 * @data: The data to write to the buffer.
4797 *
4798 * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as
4799 * one function. If you already have the data to write to the buffer, it
4800 * may be easier to simply call this function.
4801 *
4802 * Note, like ring_buffer_lock_reserve, the length is the length of the data
4803 * and not the length of the event which would hold the header.
4804 */
4805int ring_buffer_write(struct trace_buffer *buffer,
4806 unsigned long length,
4807 void *data)
4808{
4809 struct ring_buffer_per_cpu *cpu_buffer;
4810 struct ring_buffer_event *event;
4811 void *body;
4812 int ret = -EBUSY;
4813 int cpu;
4814
4815 guard(preempt_notrace)();
4816
4817 if (atomic_read(v: &buffer->record_disabled))
4818 return -EBUSY;
4819
4820 cpu = raw_smp_processor_id();
4821
4822 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
4823 return -EBUSY;
4824
4825 cpu_buffer = buffer->buffers[cpu];
4826
4827 if (atomic_read(v: &cpu_buffer->record_disabled))
4828 return -EBUSY;
4829
4830 if (length > buffer->max_data_size)
4831 return -EBUSY;
4832
4833 if (unlikely(trace_recursive_lock(cpu_buffer)))
4834 return -EBUSY;
4835
4836 event = rb_reserve_next_event(buffer, cpu_buffer, length);
4837 if (!event)
4838 goto out_unlock;
4839
4840 body = rb_event_data(event);
4841
4842 memcpy(to: body, from: data, len: length);
4843
4844 rb_commit(cpu_buffer);
4845
4846 rb_wakeups(buffer, cpu_buffer);
4847
4848 ret = 0;
4849
4850 out_unlock:
4851 trace_recursive_unlock(cpu_buffer);
4852 return ret;
4853}
4854EXPORT_SYMBOL_GPL(ring_buffer_write);
4855
4856/*
4857 * The total entries in the ring buffer is the running counter
4858 * of entries entered into the ring buffer, minus the sum of
4859 * the entries read from the ring buffer and the number of
4860 * entries that were overwritten.
4861 */
4862static inline unsigned long
4863rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer)
4864{
4865 return local_read(&cpu_buffer->entries) -
4866 (local_read(&cpu_buffer->overrun) + cpu_buffer->read);
4867}
4868
4869static bool rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
4870{
4871 return !rb_num_of_entries(cpu_buffer);
4872}
4873
4874/**
4875 * ring_buffer_record_disable - stop all writes into the buffer
4876 * @buffer: The ring buffer to stop writes to.
4877 *
4878 * This prevents all writes to the buffer. Any attempt to write
4879 * to the buffer after this will fail and return NULL.
4880 *
4881 * The caller should call synchronize_rcu() after this.
4882 */
4883void ring_buffer_record_disable(struct trace_buffer *buffer)
4884{
4885 atomic_inc(v: &buffer->record_disabled);
4886}
4887EXPORT_SYMBOL_GPL(ring_buffer_record_disable);
4888
4889/**
4890 * ring_buffer_record_enable - enable writes to the buffer
4891 * @buffer: The ring buffer to enable writes
4892 *
4893 * Note, multiple disables will need the same number of enables
4894 * to truly enable the writing (much like preempt_disable).
4895 */
4896void ring_buffer_record_enable(struct trace_buffer *buffer)
4897{
4898 atomic_dec(v: &buffer->record_disabled);
4899}
4900EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
4901
4902/**
4903 * ring_buffer_record_off - stop all writes into the buffer
4904 * @buffer: The ring buffer to stop writes to.
4905 *
4906 * This prevents all writes to the buffer. Any attempt to write
4907 * to the buffer after this will fail and return NULL.
4908 *
4909 * This is different than ring_buffer_record_disable() as
4910 * it works like an on/off switch, where as the disable() version
4911 * must be paired with a enable().
4912 */
4913void ring_buffer_record_off(struct trace_buffer *buffer)
4914{
4915 unsigned int rd;
4916 unsigned int new_rd;
4917
4918 rd = atomic_read(v: &buffer->record_disabled);
4919 do {
4920 new_rd = rd | RB_BUFFER_OFF;
4921 } while (!atomic_try_cmpxchg(v: &buffer->record_disabled, old: &rd, new: new_rd));
4922}
4923EXPORT_SYMBOL_GPL(ring_buffer_record_off);
4924
4925/**
4926 * ring_buffer_record_on - restart writes into the buffer
4927 * @buffer: The ring buffer to start writes to.
4928 *
4929 * This enables all writes to the buffer that was disabled by
4930 * ring_buffer_record_off().
4931 *
4932 * This is different than ring_buffer_record_enable() as
4933 * it works like an on/off switch, where as the enable() version
4934 * must be paired with a disable().
4935 */
4936void ring_buffer_record_on(struct trace_buffer *buffer)
4937{
4938 unsigned int rd;
4939 unsigned int new_rd;
4940
4941 rd = atomic_read(v: &buffer->record_disabled);
4942 do {
4943 new_rd = rd & ~RB_BUFFER_OFF;
4944 } while (!atomic_try_cmpxchg(v: &buffer->record_disabled, old: &rd, new: new_rd));
4945}
4946EXPORT_SYMBOL_GPL(ring_buffer_record_on);
4947
4948/**
4949 * ring_buffer_record_is_on - return true if the ring buffer can write
4950 * @buffer: The ring buffer to see if write is enabled
4951 *
4952 * Returns true if the ring buffer is in a state that it accepts writes.
4953 */
4954bool ring_buffer_record_is_on(struct trace_buffer *buffer)
4955{
4956 return !atomic_read(v: &buffer->record_disabled);
4957}
4958
4959/**
4960 * ring_buffer_record_is_set_on - return true if the ring buffer is set writable
4961 * @buffer: The ring buffer to see if write is set enabled
4962 *
4963 * Returns true if the ring buffer is set writable by ring_buffer_record_on().
4964 * Note that this does NOT mean it is in a writable state.
4965 *
4966 * It may return true when the ring buffer has been disabled by
4967 * ring_buffer_record_disable(), as that is a temporary disabling of
4968 * the ring buffer.
4969 */
4970bool ring_buffer_record_is_set_on(struct trace_buffer *buffer)
4971{
4972 return !(atomic_read(v: &buffer->record_disabled) & RB_BUFFER_OFF);
4973}
4974
4975/**
4976 * ring_buffer_record_is_on_cpu - return true if the ring buffer can write
4977 * @buffer: The ring buffer to see if write is enabled
4978 * @cpu: The CPU to test if the ring buffer can write too
4979 *
4980 * Returns true if the ring buffer is in a state that it accepts writes
4981 * for a particular CPU.
4982 */
4983bool ring_buffer_record_is_on_cpu(struct trace_buffer *buffer, int cpu)
4984{
4985 struct ring_buffer_per_cpu *cpu_buffer;
4986
4987 cpu_buffer = buffer->buffers[cpu];
4988
4989 return ring_buffer_record_is_set_on(buffer) &&
4990 !atomic_read(v: &cpu_buffer->record_disabled);
4991}
4992
4993/**
4994 * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer
4995 * @buffer: The ring buffer to stop writes to.
4996 * @cpu: The CPU buffer to stop
4997 *
4998 * This prevents all writes to the buffer. Any attempt to write
4999 * to the buffer after this will fail and return NULL.
5000 *
5001 * The caller should call synchronize_rcu() after this.
5002 */
5003void ring_buffer_record_disable_cpu(struct trace_buffer *buffer, int cpu)
5004{
5005 struct ring_buffer_per_cpu *cpu_buffer;
5006
5007 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5008 return;
5009
5010 cpu_buffer = buffer->buffers[cpu];
5011 atomic_inc(v: &cpu_buffer->record_disabled);
5012}
5013EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu);
5014
5015/**
5016 * ring_buffer_record_enable_cpu - enable writes to the buffer
5017 * @buffer: The ring buffer to enable writes
5018 * @cpu: The CPU to enable.
5019 *
5020 * Note, multiple disables will need the same number of enables
5021 * to truly enable the writing (much like preempt_disable).
5022 */
5023void ring_buffer_record_enable_cpu(struct trace_buffer *buffer, int cpu)
5024{
5025 struct ring_buffer_per_cpu *cpu_buffer;
5026
5027 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5028 return;
5029
5030 cpu_buffer = buffer->buffers[cpu];
5031 atomic_dec(v: &cpu_buffer->record_disabled);
5032}
5033EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
5034
5035/**
5036 * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer
5037 * @buffer: The ring buffer
5038 * @cpu: The per CPU buffer to read from.
5039 */
5040u64 ring_buffer_oldest_event_ts(struct trace_buffer *buffer, int cpu)
5041{
5042 unsigned long flags;
5043 struct ring_buffer_per_cpu *cpu_buffer;
5044 struct buffer_page *bpage;
5045 u64 ret = 0;
5046
5047 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5048 return 0;
5049
5050 cpu_buffer = buffer->buffers[cpu];
5051 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5052 /*
5053 * if the tail is on reader_page, oldest time stamp is on the reader
5054 * page
5055 */
5056 if (cpu_buffer->tail_page == cpu_buffer->reader_page)
5057 bpage = cpu_buffer->reader_page;
5058 else
5059 bpage = rb_set_head_page(cpu_buffer);
5060 if (bpage)
5061 ret = bpage->page->time_stamp;
5062 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5063
5064 return ret;
5065}
5066EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts);
5067
5068/**
5069 * ring_buffer_bytes_cpu - get the number of bytes unconsumed in a cpu buffer
5070 * @buffer: The ring buffer
5071 * @cpu: The per CPU buffer to read from.
5072 */
5073unsigned long ring_buffer_bytes_cpu(struct trace_buffer *buffer, int cpu)
5074{
5075 struct ring_buffer_per_cpu *cpu_buffer;
5076 unsigned long ret;
5077
5078 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5079 return 0;
5080
5081 cpu_buffer = buffer->buffers[cpu];
5082 ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes;
5083
5084 return ret;
5085}
5086EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu);
5087
5088/**
5089 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
5090 * @buffer: The ring buffer
5091 * @cpu: The per CPU buffer to get the entries from.
5092 */
5093unsigned long ring_buffer_entries_cpu(struct trace_buffer *buffer, int cpu)
5094{
5095 struct ring_buffer_per_cpu *cpu_buffer;
5096
5097 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5098 return 0;
5099
5100 cpu_buffer = buffer->buffers[cpu];
5101
5102 return rb_num_of_entries(cpu_buffer);
5103}
5104EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
5105
5106/**
5107 * ring_buffer_overrun_cpu - get the number of overruns caused by the ring
5108 * buffer wrapping around (only if RB_FL_OVERWRITE is on).
5109 * @buffer: The ring buffer
5110 * @cpu: The per CPU buffer to get the number of overruns from
5111 */
5112unsigned long ring_buffer_overrun_cpu(struct trace_buffer *buffer, int cpu)
5113{
5114 struct ring_buffer_per_cpu *cpu_buffer;
5115 unsigned long ret;
5116
5117 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5118 return 0;
5119
5120 cpu_buffer = buffer->buffers[cpu];
5121 ret = local_read(&cpu_buffer->overrun);
5122
5123 return ret;
5124}
5125EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
5126
5127/**
5128 * ring_buffer_commit_overrun_cpu - get the number of overruns caused by
5129 * commits failing due to the buffer wrapping around while there are uncommitted
5130 * events, such as during an interrupt storm.
5131 * @buffer: The ring buffer
5132 * @cpu: The per CPU buffer to get the number of overruns from
5133 */
5134unsigned long
5135ring_buffer_commit_overrun_cpu(struct trace_buffer *buffer, int cpu)
5136{
5137 struct ring_buffer_per_cpu *cpu_buffer;
5138 unsigned long ret;
5139
5140 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5141 return 0;
5142
5143 cpu_buffer = buffer->buffers[cpu];
5144 ret = local_read(&cpu_buffer->commit_overrun);
5145
5146 return ret;
5147}
5148EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu);
5149
5150/**
5151 * ring_buffer_dropped_events_cpu - get the number of dropped events caused by
5152 * the ring buffer filling up (only if RB_FL_OVERWRITE is off).
5153 * @buffer: The ring buffer
5154 * @cpu: The per CPU buffer to get the number of overruns from
5155 */
5156unsigned long
5157ring_buffer_dropped_events_cpu(struct trace_buffer *buffer, int cpu)
5158{
5159 struct ring_buffer_per_cpu *cpu_buffer;
5160 unsigned long ret;
5161
5162 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5163 return 0;
5164
5165 cpu_buffer = buffer->buffers[cpu];
5166 ret = local_read(&cpu_buffer->dropped_events);
5167
5168 return ret;
5169}
5170EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu);
5171
5172/**
5173 * ring_buffer_read_events_cpu - get the number of events successfully read
5174 * @buffer: The ring buffer
5175 * @cpu: The per CPU buffer to get the number of events read
5176 */
5177unsigned long
5178ring_buffer_read_events_cpu(struct trace_buffer *buffer, int cpu)
5179{
5180 struct ring_buffer_per_cpu *cpu_buffer;
5181
5182 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5183 return 0;
5184
5185 cpu_buffer = buffer->buffers[cpu];
5186 return cpu_buffer->read;
5187}
5188EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu);
5189
5190/**
5191 * ring_buffer_entries - get the number of entries in a buffer
5192 * @buffer: The ring buffer
5193 *
5194 * Returns the total number of entries in the ring buffer
5195 * (all CPU entries)
5196 */
5197unsigned long ring_buffer_entries(struct trace_buffer *buffer)
5198{
5199 struct ring_buffer_per_cpu *cpu_buffer;
5200 unsigned long entries = 0;
5201 int cpu;
5202
5203 /* if you care about this being correct, lock the buffer */
5204 for_each_buffer_cpu(buffer, cpu) {
5205 cpu_buffer = buffer->buffers[cpu];
5206 entries += rb_num_of_entries(cpu_buffer);
5207 }
5208
5209 return entries;
5210}
5211EXPORT_SYMBOL_GPL(ring_buffer_entries);
5212
5213/**
5214 * ring_buffer_overruns - get the number of overruns in buffer
5215 * @buffer: The ring buffer
5216 *
5217 * Returns the total number of overruns in the ring buffer
5218 * (all CPU entries)
5219 */
5220unsigned long ring_buffer_overruns(struct trace_buffer *buffer)
5221{
5222 struct ring_buffer_per_cpu *cpu_buffer;
5223 unsigned long overruns = 0;
5224 int cpu;
5225
5226 /* if you care about this being correct, lock the buffer */
5227 for_each_buffer_cpu(buffer, cpu) {
5228 cpu_buffer = buffer->buffers[cpu];
5229 overruns += local_read(&cpu_buffer->overrun);
5230 }
5231
5232 return overruns;
5233}
5234EXPORT_SYMBOL_GPL(ring_buffer_overruns);
5235
5236static void rb_iter_reset(struct ring_buffer_iter *iter)
5237{
5238 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
5239
5240 /* Iterator usage is expected to have record disabled */
5241 iter->head_page = cpu_buffer->reader_page;
5242 iter->head = cpu_buffer->reader_page->read;
5243 iter->next_event = iter->head;
5244
5245 iter->cache_reader_page = iter->head_page;
5246 iter->cache_read = cpu_buffer->read;
5247 iter->cache_pages_removed = cpu_buffer->pages_removed;
5248
5249 if (iter->head) {
5250 iter->read_stamp = cpu_buffer->read_stamp;
5251 iter->page_stamp = cpu_buffer->reader_page->page->time_stamp;
5252 } else {
5253 iter->read_stamp = iter->head_page->page->time_stamp;
5254 iter->page_stamp = iter->read_stamp;
5255 }
5256}
5257
5258/**
5259 * ring_buffer_iter_reset - reset an iterator
5260 * @iter: The iterator to reset
5261 *
5262 * Resets the iterator, so that it will start from the beginning
5263 * again.
5264 */
5265void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
5266{
5267 struct ring_buffer_per_cpu *cpu_buffer;
5268 unsigned long flags;
5269
5270 if (!iter)
5271 return;
5272
5273 cpu_buffer = iter->cpu_buffer;
5274
5275 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5276 rb_iter_reset(iter);
5277 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5278}
5279EXPORT_SYMBOL_GPL(ring_buffer_iter_reset);
5280
5281/**
5282 * ring_buffer_iter_empty - check if an iterator has no more to read
5283 * @iter: The iterator to check
5284 */
5285int ring_buffer_iter_empty(struct ring_buffer_iter *iter)
5286{
5287 struct ring_buffer_per_cpu *cpu_buffer;
5288 struct buffer_page *reader;
5289 struct buffer_page *head_page;
5290 struct buffer_page *commit_page;
5291 struct buffer_page *curr_commit_page;
5292 unsigned commit;
5293 u64 curr_commit_ts;
5294 u64 commit_ts;
5295
5296 cpu_buffer = iter->cpu_buffer;
5297 reader = cpu_buffer->reader_page;
5298 head_page = cpu_buffer->head_page;
5299 commit_page = READ_ONCE(cpu_buffer->commit_page);
5300 commit_ts = commit_page->page->time_stamp;
5301
5302 /*
5303 * When the writer goes across pages, it issues a cmpxchg which
5304 * is a mb(), which will synchronize with the rmb here.
5305 * (see rb_tail_page_update())
5306 */
5307 smp_rmb();
5308 commit = rb_page_commit(bpage: commit_page);
5309 /* We want to make sure that the commit page doesn't change */
5310 smp_rmb();
5311
5312 /* Make sure commit page didn't change */
5313 curr_commit_page = READ_ONCE(cpu_buffer->commit_page);
5314 curr_commit_ts = READ_ONCE(curr_commit_page->page->time_stamp);
5315
5316 /* If the commit page changed, then there's more data */
5317 if (curr_commit_page != commit_page ||
5318 curr_commit_ts != commit_ts)
5319 return 0;
5320
5321 /* Still racy, as it may return a false positive, but that's OK */
5322 return ((iter->head_page == commit_page && iter->head >= commit) ||
5323 (iter->head_page == reader && commit_page == head_page &&
5324 head_page->read == commit &&
5325 iter->head == rb_page_size(bpage: cpu_buffer->reader_page)));
5326}
5327EXPORT_SYMBOL_GPL(ring_buffer_iter_empty);
5328
5329static void
5330rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer,
5331 struct ring_buffer_event *event)
5332{
5333 u64 delta;
5334
5335 switch (event->type_len) {
5336 case RINGBUF_TYPE_PADDING:
5337 return;
5338
5339 case RINGBUF_TYPE_TIME_EXTEND:
5340 delta = rb_event_time_stamp(event);
5341 cpu_buffer->read_stamp += delta;
5342 return;
5343
5344 case RINGBUF_TYPE_TIME_STAMP:
5345 delta = rb_event_time_stamp(event);
5346 delta = rb_fix_abs_ts(abs: delta, save_ts: cpu_buffer->read_stamp);
5347 cpu_buffer->read_stamp = delta;
5348 return;
5349
5350 case RINGBUF_TYPE_DATA:
5351 cpu_buffer->read_stamp += event->time_delta;
5352 return;
5353
5354 default:
5355 RB_WARN_ON(cpu_buffer, 1);
5356 }
5357}
5358
5359static void
5360rb_update_iter_read_stamp(struct ring_buffer_iter *iter,
5361 struct ring_buffer_event *event)
5362{
5363 u64 delta;
5364
5365 switch (event->type_len) {
5366 case RINGBUF_TYPE_PADDING:
5367 return;
5368
5369 case RINGBUF_TYPE_TIME_EXTEND:
5370 delta = rb_event_time_stamp(event);
5371 iter->read_stamp += delta;
5372 return;
5373
5374 case RINGBUF_TYPE_TIME_STAMP:
5375 delta = rb_event_time_stamp(event);
5376 delta = rb_fix_abs_ts(abs: delta, save_ts: iter->read_stamp);
5377 iter->read_stamp = delta;
5378 return;
5379
5380 case RINGBUF_TYPE_DATA:
5381 iter->read_stamp += event->time_delta;
5382 return;
5383
5384 default:
5385 RB_WARN_ON(iter->cpu_buffer, 1);
5386 }
5387}
5388
5389static struct buffer_page *
5390rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
5391{
5392 struct buffer_page *reader = NULL;
5393 unsigned long bsize = READ_ONCE(cpu_buffer->buffer->subbuf_size);
5394 unsigned long overwrite;
5395 unsigned long flags;
5396 int nr_loops = 0;
5397 bool ret;
5398
5399 local_irq_save(flags);
5400 arch_spin_lock(&cpu_buffer->lock);
5401
5402 again:
5403 /*
5404 * This should normally only loop twice. But because the
5405 * start of the reader inserts an empty page, it causes
5406 * a case where we will loop three times. There should be no
5407 * reason to loop four times (that I know of).
5408 */
5409 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) {
5410 reader = NULL;
5411 goto out;
5412 }
5413
5414 reader = cpu_buffer->reader_page;
5415
5416 /* If there's more to read, return this page */
5417 if (cpu_buffer->reader_page->read < rb_page_size(bpage: reader))
5418 goto out;
5419
5420 /* Never should we have an index greater than the size */
5421 if (RB_WARN_ON(cpu_buffer,
5422 cpu_buffer->reader_page->read > rb_page_size(reader)))
5423 goto out;
5424
5425 /* check if we caught up to the tail */
5426 reader = NULL;
5427 if (cpu_buffer->commit_page == cpu_buffer->reader_page)
5428 goto out;
5429
5430 /* Don't bother swapping if the ring buffer is empty */
5431 if (rb_num_of_entries(cpu_buffer) == 0)
5432 goto out;
5433
5434 /*
5435 * Reset the reader page to size zero.
5436 */
5437 local_set(&cpu_buffer->reader_page->write, 0);
5438 local_set(&cpu_buffer->reader_page->entries, 0);
5439 cpu_buffer->reader_page->real_end = 0;
5440
5441 spin:
5442 /*
5443 * Splice the empty reader page into the list around the head.
5444 */
5445 reader = rb_set_head_page(cpu_buffer);
5446 if (!reader)
5447 goto out;
5448 cpu_buffer->reader_page->list.next = rb_list_head(list: reader->list.next);
5449 cpu_buffer->reader_page->list.prev = reader->list.prev;
5450
5451 /*
5452 * cpu_buffer->pages just needs to point to the buffer, it
5453 * has no specific buffer page to point to. Lets move it out
5454 * of our way so we don't accidentally swap it.
5455 */
5456 cpu_buffer->pages = reader->list.prev;
5457
5458 /* The reader page will be pointing to the new head */
5459 rb_set_list_to_head(list: &cpu_buffer->reader_page->list);
5460
5461 /*
5462 * We want to make sure we read the overruns after we set up our
5463 * pointers to the next object. The writer side does a
5464 * cmpxchg to cross pages which acts as the mb on the writer
5465 * side. Note, the reader will constantly fail the swap
5466 * while the writer is updating the pointers, so this
5467 * guarantees that the overwrite recorded here is the one we
5468 * want to compare with the last_overrun.
5469 */
5470 smp_mb();
5471 overwrite = local_read(&(cpu_buffer->overrun));
5472
5473 /*
5474 * Here's the tricky part.
5475 *
5476 * We need to move the pointer past the header page.
5477 * But we can only do that if a writer is not currently
5478 * moving it. The page before the header page has the
5479 * flag bit '1' set if it is pointing to the page we want.
5480 * but if the writer is in the process of moving it
5481 * then it will be '2' or already moved '0'.
5482 */
5483
5484 ret = rb_head_page_replace(old: reader, new: cpu_buffer->reader_page);
5485
5486 /*
5487 * If we did not convert it, then we must try again.
5488 */
5489 if (!ret)
5490 goto spin;
5491
5492 if (cpu_buffer->ring_meta)
5493 rb_update_meta_reader(cpu_buffer, reader);
5494
5495 /*
5496 * Yay! We succeeded in replacing the page.
5497 *
5498 * Now make the new head point back to the reader page.
5499 */
5500 rb_list_head(list: reader->list.next)->prev = &cpu_buffer->reader_page->list;
5501 rb_inc_page(bpage: &cpu_buffer->head_page);
5502
5503 cpu_buffer->cnt++;
5504 local_inc(l: &cpu_buffer->pages_read);
5505
5506 /* Finally update the reader page to the new head */
5507 cpu_buffer->reader_page = reader;
5508 cpu_buffer->reader_page->read = 0;
5509
5510 if (overwrite != cpu_buffer->last_overrun) {
5511 cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun;
5512 cpu_buffer->last_overrun = overwrite;
5513 }
5514
5515 goto again;
5516
5517 out:
5518 /* Update the read_stamp on the first event */
5519 if (reader && reader->read == 0)
5520 cpu_buffer->read_stamp = reader->page->time_stamp;
5521
5522 arch_spin_unlock(&cpu_buffer->lock);
5523 local_irq_restore(flags);
5524
5525 /*
5526 * The writer has preempt disable, wait for it. But not forever
5527 * Although, 1 second is pretty much "forever"
5528 */
5529#define USECS_WAIT 1000000
5530 for (nr_loops = 0; nr_loops < USECS_WAIT; nr_loops++) {
5531 /* If the write is past the end of page, a writer is still updating it */
5532 if (likely(!reader || rb_page_write(reader) <= bsize))
5533 break;
5534
5535 udelay(usec: 1);
5536
5537 /* Get the latest version of the reader write value */
5538 smp_rmb();
5539 }
5540
5541 /* The writer is not moving forward? Something is wrong */
5542 if (RB_WARN_ON(cpu_buffer, nr_loops == USECS_WAIT))
5543 reader = NULL;
5544
5545 /*
5546 * Make sure we see any padding after the write update
5547 * (see rb_reset_tail()).
5548 *
5549 * In addition, a writer may be writing on the reader page
5550 * if the page has not been fully filled, so the read barrier
5551 * is also needed to make sure we see the content of what is
5552 * committed by the writer (see rb_set_commit_to_write()).
5553 */
5554 smp_rmb();
5555
5556
5557 return reader;
5558}
5559
5560static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
5561{
5562 struct ring_buffer_event *event;
5563 struct buffer_page *reader;
5564 unsigned length;
5565
5566 reader = rb_get_reader_page(cpu_buffer);
5567
5568 /* This function should not be called when buffer is empty */
5569 if (RB_WARN_ON(cpu_buffer, !reader))
5570 return;
5571
5572 event = rb_reader_event(cpu_buffer);
5573
5574 if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
5575 cpu_buffer->read++;
5576
5577 rb_update_read_stamp(cpu_buffer, event);
5578
5579 length = rb_event_length(event);
5580 cpu_buffer->reader_page->read += length;
5581 cpu_buffer->read_bytes += length;
5582}
5583
5584static void rb_advance_iter(struct ring_buffer_iter *iter)
5585{
5586 struct ring_buffer_per_cpu *cpu_buffer;
5587
5588 cpu_buffer = iter->cpu_buffer;
5589
5590 /* If head == next_event then we need to jump to the next event */
5591 if (iter->head == iter->next_event) {
5592 /* If the event gets overwritten again, there's nothing to do */
5593 if (rb_iter_head_event(iter) == NULL)
5594 return;
5595 }
5596
5597 iter->head = iter->next_event;
5598
5599 /*
5600 * Check if we are at the end of the buffer.
5601 */
5602 if (iter->next_event >= rb_page_size(bpage: iter->head_page)) {
5603 /* discarded commits can make the page empty */
5604 if (iter->head_page == cpu_buffer->commit_page)
5605 return;
5606 rb_inc_iter(iter);
5607 return;
5608 }
5609
5610 rb_update_iter_read_stamp(iter, event: iter->event);
5611}
5612
5613static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer)
5614{
5615 return cpu_buffer->lost_events;
5616}
5617
5618static struct ring_buffer_event *
5619rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts,
5620 unsigned long *lost_events)
5621{
5622 struct ring_buffer_event *event;
5623 struct buffer_page *reader;
5624 int nr_loops = 0;
5625
5626 if (ts)
5627 *ts = 0;
5628 again:
5629 /*
5630 * We repeat when a time extend is encountered.
5631 * Since the time extend is always attached to a data event,
5632 * we should never loop more than once.
5633 * (We never hit the following condition more than twice).
5634 */
5635 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2))
5636 return NULL;
5637
5638 reader = rb_get_reader_page(cpu_buffer);
5639 if (!reader)
5640 return NULL;
5641
5642 event = rb_reader_event(cpu_buffer);
5643
5644 switch (event->type_len) {
5645 case RINGBUF_TYPE_PADDING:
5646 if (rb_null_event(event))
5647 RB_WARN_ON(cpu_buffer, 1);
5648 /*
5649 * Because the writer could be discarding every
5650 * event it creates (which would probably be bad)
5651 * if we were to go back to "again" then we may never
5652 * catch up, and will trigger the warn on, or lock
5653 * the box. Return the padding, and we will release
5654 * the current locks, and try again.
5655 */
5656 return event;
5657
5658 case RINGBUF_TYPE_TIME_EXTEND:
5659 /* Internal data, OK to advance */
5660 rb_advance_reader(cpu_buffer);
5661 goto again;
5662
5663 case RINGBUF_TYPE_TIME_STAMP:
5664 if (ts) {
5665 *ts = rb_event_time_stamp(event);
5666 *ts = rb_fix_abs_ts(abs: *ts, save_ts: reader->page->time_stamp);
5667 ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
5668 cpu_buffer->cpu, ts);
5669 }
5670 /* Internal data, OK to advance */
5671 rb_advance_reader(cpu_buffer);
5672 goto again;
5673
5674 case RINGBUF_TYPE_DATA:
5675 if (ts && !(*ts)) {
5676 *ts = cpu_buffer->read_stamp + event->time_delta;
5677 ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
5678 cpu_buffer->cpu, ts);
5679 }
5680 if (lost_events)
5681 *lost_events = rb_lost_events(cpu_buffer);
5682 return event;
5683
5684 default:
5685 RB_WARN_ON(cpu_buffer, 1);
5686 }
5687
5688 return NULL;
5689}
5690EXPORT_SYMBOL_GPL(ring_buffer_peek);
5691
5692static struct ring_buffer_event *
5693rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
5694{
5695 struct trace_buffer *buffer;
5696 struct ring_buffer_per_cpu *cpu_buffer;
5697 struct ring_buffer_event *event;
5698 int nr_loops = 0;
5699
5700 if (ts)
5701 *ts = 0;
5702
5703 cpu_buffer = iter->cpu_buffer;
5704 buffer = cpu_buffer->buffer;
5705
5706 /*
5707 * Check if someone performed a consuming read to the buffer
5708 * or removed some pages from the buffer. In these cases,
5709 * iterator was invalidated and we need to reset it.
5710 */
5711 if (unlikely(iter->cache_read != cpu_buffer->read ||
5712 iter->cache_reader_page != cpu_buffer->reader_page ||
5713 iter->cache_pages_removed != cpu_buffer->pages_removed))
5714 rb_iter_reset(iter);
5715
5716 again:
5717 if (ring_buffer_iter_empty(iter))
5718 return NULL;
5719
5720 /*
5721 * As the writer can mess with what the iterator is trying
5722 * to read, just give up if we fail to get an event after
5723 * three tries. The iterator is not as reliable when reading
5724 * the ring buffer with an active write as the consumer is.
5725 * Do not warn if the three failures is reached.
5726 */
5727 if (++nr_loops > 3)
5728 return NULL;
5729
5730 if (rb_per_cpu_empty(cpu_buffer))
5731 return NULL;
5732
5733 if (iter->head >= rb_page_size(bpage: iter->head_page)) {
5734 rb_inc_iter(iter);
5735 goto again;
5736 }
5737
5738 event = rb_iter_head_event(iter);
5739 if (!event)
5740 goto again;
5741
5742 switch (event->type_len) {
5743 case RINGBUF_TYPE_PADDING:
5744 if (rb_null_event(event)) {
5745 rb_inc_iter(iter);
5746 goto again;
5747 }
5748 rb_advance_iter(iter);
5749 return event;
5750
5751 case RINGBUF_TYPE_TIME_EXTEND:
5752 /* Internal data, OK to advance */
5753 rb_advance_iter(iter);
5754 goto again;
5755
5756 case RINGBUF_TYPE_TIME_STAMP:
5757 if (ts) {
5758 *ts = rb_event_time_stamp(event);
5759 *ts = rb_fix_abs_ts(abs: *ts, save_ts: iter->head_page->page->time_stamp);
5760 ring_buffer_normalize_time_stamp(cpu_buffer->buffer,
5761 cpu_buffer->cpu, ts);
5762 }
5763 /* Internal data, OK to advance */
5764 rb_advance_iter(iter);
5765 goto again;
5766
5767 case RINGBUF_TYPE_DATA:
5768 if (ts && !(*ts)) {
5769 *ts = iter->read_stamp + event->time_delta;
5770 ring_buffer_normalize_time_stamp(buffer,
5771 cpu_buffer->cpu, ts);
5772 }
5773 return event;
5774
5775 default:
5776 RB_WARN_ON(cpu_buffer, 1);
5777 }
5778
5779 return NULL;
5780}
5781EXPORT_SYMBOL_GPL(ring_buffer_iter_peek);
5782
5783static inline bool rb_reader_lock(struct ring_buffer_per_cpu *cpu_buffer)
5784{
5785 if (likely(!in_nmi())) {
5786 raw_spin_lock(&cpu_buffer->reader_lock);
5787 return true;
5788 }
5789
5790 /*
5791 * If an NMI die dumps out the content of the ring buffer
5792 * trylock must be used to prevent a deadlock if the NMI
5793 * preempted a task that holds the ring buffer locks. If
5794 * we get the lock then all is fine, if not, then continue
5795 * to do the read, but this can corrupt the ring buffer,
5796 * so it must be permanently disabled from future writes.
5797 * Reading from NMI is a oneshot deal.
5798 */
5799 if (raw_spin_trylock(&cpu_buffer->reader_lock))
5800 return true;
5801
5802 /* Continue without locking, but disable the ring buffer */
5803 atomic_inc(v: &cpu_buffer->record_disabled);
5804 return false;
5805}
5806
5807static inline void
5808rb_reader_unlock(struct ring_buffer_per_cpu *cpu_buffer, bool locked)
5809{
5810 if (likely(locked))
5811 raw_spin_unlock(&cpu_buffer->reader_lock);
5812}
5813
5814/**
5815 * ring_buffer_peek - peek at the next event to be read
5816 * @buffer: The ring buffer to read
5817 * @cpu: The cpu to peak at
5818 * @ts: The timestamp counter of this event.
5819 * @lost_events: a variable to store if events were lost (may be NULL)
5820 *
5821 * This will return the event that will be read next, but does
5822 * not consume the data.
5823 */
5824struct ring_buffer_event *
5825ring_buffer_peek(struct trace_buffer *buffer, int cpu, u64 *ts,
5826 unsigned long *lost_events)
5827{
5828 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
5829 struct ring_buffer_event *event;
5830 unsigned long flags;
5831 bool dolock;
5832
5833 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5834 return NULL;
5835
5836 again:
5837 local_irq_save(flags);
5838 dolock = rb_reader_lock(cpu_buffer);
5839 event = rb_buffer_peek(cpu_buffer, ts, lost_events);
5840 if (event && event->type_len == RINGBUF_TYPE_PADDING)
5841 rb_advance_reader(cpu_buffer);
5842 rb_reader_unlock(cpu_buffer, locked: dolock);
5843 local_irq_restore(flags);
5844
5845 if (event && event->type_len == RINGBUF_TYPE_PADDING)
5846 goto again;
5847
5848 return event;
5849}
5850
5851/** ring_buffer_iter_dropped - report if there are dropped events
5852 * @iter: The ring buffer iterator
5853 *
5854 * Returns true if there was dropped events since the last peek.
5855 */
5856bool ring_buffer_iter_dropped(struct ring_buffer_iter *iter)
5857{
5858 bool ret = iter->missed_events != 0;
5859
5860 iter->missed_events = 0;
5861 return ret;
5862}
5863EXPORT_SYMBOL_GPL(ring_buffer_iter_dropped);
5864
5865/**
5866 * ring_buffer_iter_peek - peek at the next event to be read
5867 * @iter: The ring buffer iterator
5868 * @ts: The timestamp counter of this event.
5869 *
5870 * This will return the event that will be read next, but does
5871 * not increment the iterator.
5872 */
5873struct ring_buffer_event *
5874ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
5875{
5876 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
5877 struct ring_buffer_event *event;
5878 unsigned long flags;
5879
5880 again:
5881 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
5882 event = rb_iter_peek(iter, ts);
5883 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
5884
5885 if (event && event->type_len == RINGBUF_TYPE_PADDING)
5886 goto again;
5887
5888 return event;
5889}
5890
5891/**
5892 * ring_buffer_consume - return an event and consume it
5893 * @buffer: The ring buffer to get the next event from
5894 * @cpu: the cpu to read the buffer from
5895 * @ts: a variable to store the timestamp (may be NULL)
5896 * @lost_events: a variable to store if events were lost (may be NULL)
5897 *
5898 * Returns the next event in the ring buffer, and that event is consumed.
5899 * Meaning, that sequential reads will keep returning a different event,
5900 * and eventually empty the ring buffer if the producer is slower.
5901 */
5902struct ring_buffer_event *
5903ring_buffer_consume(struct trace_buffer *buffer, int cpu, u64 *ts,
5904 unsigned long *lost_events)
5905{
5906 struct ring_buffer_per_cpu *cpu_buffer;
5907 struct ring_buffer_event *event = NULL;
5908 unsigned long flags;
5909 bool dolock;
5910
5911 again:
5912 /* might be called in atomic */
5913 preempt_disable();
5914
5915 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5916 goto out;
5917
5918 cpu_buffer = buffer->buffers[cpu];
5919 local_irq_save(flags);
5920 dolock = rb_reader_lock(cpu_buffer);
5921
5922 event = rb_buffer_peek(cpu_buffer, ts, lost_events);
5923 if (event) {
5924 cpu_buffer->lost_events = 0;
5925 rb_advance_reader(cpu_buffer);
5926 }
5927
5928 rb_reader_unlock(cpu_buffer, locked: dolock);
5929 local_irq_restore(flags);
5930
5931 out:
5932 preempt_enable();
5933
5934 if (event && event->type_len == RINGBUF_TYPE_PADDING)
5935 goto again;
5936
5937 return event;
5938}
5939EXPORT_SYMBOL_GPL(ring_buffer_consume);
5940
5941/**
5942 * ring_buffer_read_start - start a non consuming read of the buffer
5943 * @buffer: The ring buffer to read from
5944 * @cpu: The cpu buffer to iterate over
5945 * @flags: gfp flags to use for memory allocation
5946 *
5947 * This creates an iterator to allow non-consuming iteration through
5948 * the buffer. If the buffer is disabled for writing, it will produce
5949 * the same information each time, but if the buffer is still writing
5950 * then the first hit of a write will cause the iteration to stop.
5951 *
5952 * Must be paired with ring_buffer_read_finish.
5953 */
5954struct ring_buffer_iter *
5955ring_buffer_read_start(struct trace_buffer *buffer, int cpu, gfp_t flags)
5956{
5957 struct ring_buffer_per_cpu *cpu_buffer;
5958 struct ring_buffer_iter *iter;
5959
5960 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
5961 return NULL;
5962
5963 iter = kzalloc(sizeof(*iter), flags);
5964 if (!iter)
5965 return NULL;
5966
5967 /* Holds the entire event: data and meta data */
5968 iter->event_size = buffer->subbuf_size;
5969 iter->event = kmalloc(iter->event_size, flags);
5970 if (!iter->event) {
5971 kfree(objp: iter);
5972 return NULL;
5973 }
5974
5975 cpu_buffer = buffer->buffers[cpu];
5976
5977 iter->cpu_buffer = cpu_buffer;
5978
5979 atomic_inc(v: &cpu_buffer->resize_disabled);
5980
5981 guard(raw_spinlock_irqsave)(l: &cpu_buffer->reader_lock);
5982 arch_spin_lock(&cpu_buffer->lock);
5983 rb_iter_reset(iter);
5984 arch_spin_unlock(&cpu_buffer->lock);
5985
5986 return iter;
5987}
5988EXPORT_SYMBOL_GPL(ring_buffer_read_start);
5989
5990/**
5991 * ring_buffer_read_finish - finish reading the iterator of the buffer
5992 * @iter: The iterator retrieved by ring_buffer_start
5993 *
5994 * This re-enables resizing of the buffer, and frees the iterator.
5995 */
5996void
5997ring_buffer_read_finish(struct ring_buffer_iter *iter)
5998{
5999 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
6000
6001 /* Use this opportunity to check the integrity of the ring buffer. */
6002 rb_check_pages(cpu_buffer);
6003
6004 atomic_dec(v: &cpu_buffer->resize_disabled);
6005 kfree(objp: iter->event);
6006 kfree(objp: iter);
6007}
6008EXPORT_SYMBOL_GPL(ring_buffer_read_finish);
6009
6010/**
6011 * ring_buffer_iter_advance - advance the iterator to the next location
6012 * @iter: The ring buffer iterator
6013 *
6014 * Move the location of the iterator such that the next read will
6015 * be the next location of the iterator.
6016 */
6017void ring_buffer_iter_advance(struct ring_buffer_iter *iter)
6018{
6019 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
6020 unsigned long flags;
6021
6022 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
6023
6024 rb_advance_iter(iter);
6025
6026 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
6027}
6028EXPORT_SYMBOL_GPL(ring_buffer_iter_advance);
6029
6030/**
6031 * ring_buffer_size - return the size of the ring buffer (in bytes)
6032 * @buffer: The ring buffer.
6033 * @cpu: The CPU to get ring buffer size from.
6034 */
6035unsigned long ring_buffer_size(struct trace_buffer *buffer, int cpu)
6036{
6037 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
6038 return 0;
6039
6040 return buffer->subbuf_size * buffer->buffers[cpu]->nr_pages;
6041}
6042EXPORT_SYMBOL_GPL(ring_buffer_size);
6043
6044/**
6045 * ring_buffer_max_event_size - return the max data size of an event
6046 * @buffer: The ring buffer.
6047 *
6048 * Returns the maximum size an event can be.
6049 */
6050unsigned long ring_buffer_max_event_size(struct trace_buffer *buffer)
6051{
6052 /* If abs timestamp is requested, events have a timestamp too */
6053 if (ring_buffer_time_stamp_abs(buffer))
6054 return buffer->max_data_size - RB_LEN_TIME_EXTEND;
6055 return buffer->max_data_size;
6056}
6057EXPORT_SYMBOL_GPL(ring_buffer_max_event_size);
6058
6059static void rb_clear_buffer_page(struct buffer_page *page)
6060{
6061 local_set(&page->write, 0);
6062 local_set(&page->entries, 0);
6063 rb_init_page(bpage: page->page);
6064 page->read = 0;
6065}
6066
6067/*
6068 * When the buffer is memory mapped to user space, each sub buffer
6069 * has a unique id that is used by the meta data to tell the user
6070 * where the current reader page is.
6071 *
6072 * For a normal allocated ring buffer, the id is saved in the buffer page
6073 * id field, and updated via this function.
6074 *
6075 * But for a fixed memory mapped buffer, the id is already assigned for
6076 * fixed memory ording in the memory layout and can not be used. Instead
6077 * the index of where the page lies in the memory layout is used.
6078 *
6079 * For the normal pages, set the buffer page id with the passed in @id
6080 * value and return that.
6081 *
6082 * For fixed memory mapped pages, get the page index in the memory layout
6083 * and return that as the id.
6084 */
6085static int rb_page_id(struct ring_buffer_per_cpu *cpu_buffer,
6086 struct buffer_page *bpage, int id)
6087{
6088 /*
6089 * For boot buffers, the id is the index,
6090 * otherwise, set the buffer page with this id
6091 */
6092 if (cpu_buffer->ring_meta)
6093 id = rb_meta_subbuf_idx(meta: cpu_buffer->ring_meta, subbuf: bpage->page);
6094 else
6095 bpage->id = id;
6096
6097 return id;
6098}
6099
6100static void rb_update_meta_page(struct ring_buffer_per_cpu *cpu_buffer)
6101{
6102 struct trace_buffer_meta *meta = cpu_buffer->meta_page;
6103
6104 if (!meta)
6105 return;
6106
6107 meta->reader.read = cpu_buffer->reader_page->read;
6108 meta->reader.id = rb_page_id(cpu_buffer, bpage: cpu_buffer->reader_page,
6109 id: cpu_buffer->reader_page->id);
6110
6111 meta->reader.lost_events = cpu_buffer->lost_events;
6112
6113 meta->entries = local_read(&cpu_buffer->entries);
6114 meta->overrun = local_read(&cpu_buffer->overrun);
6115 meta->read = cpu_buffer->read;
6116
6117 /* Some archs do not have data cache coherency between kernel and user-space */
6118 flush_kernel_vmap_range(vaddr: cpu_buffer->meta_page, PAGE_SIZE);
6119}
6120
6121static void
6122rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
6123{
6124 struct buffer_page *page;
6125
6126 rb_head_page_deactivate(cpu_buffer);
6127
6128 cpu_buffer->head_page
6129 = list_entry(cpu_buffer->pages, struct buffer_page, list);
6130 rb_clear_buffer_page(page: cpu_buffer->head_page);
6131 list_for_each_entry(page, cpu_buffer->pages, list) {
6132 rb_clear_buffer_page(page);
6133 }
6134
6135 cpu_buffer->tail_page = cpu_buffer->head_page;
6136 cpu_buffer->commit_page = cpu_buffer->head_page;
6137
6138 INIT_LIST_HEAD(list: &cpu_buffer->reader_page->list);
6139 INIT_LIST_HEAD(list: &cpu_buffer->new_pages);
6140 rb_clear_buffer_page(page: cpu_buffer->reader_page);
6141
6142 local_set(&cpu_buffer->entries_bytes, 0);
6143 local_set(&cpu_buffer->overrun, 0);
6144 local_set(&cpu_buffer->commit_overrun, 0);
6145 local_set(&cpu_buffer->dropped_events, 0);
6146 local_set(&cpu_buffer->entries, 0);
6147 local_set(&cpu_buffer->committing, 0);
6148 local_set(&cpu_buffer->commits, 0);
6149 local_set(&cpu_buffer->pages_touched, 0);
6150 local_set(&cpu_buffer->pages_lost, 0);
6151 local_set(&cpu_buffer->pages_read, 0);
6152 cpu_buffer->last_pages_touch = 0;
6153 cpu_buffer->shortest_full = 0;
6154 cpu_buffer->read = 0;
6155 cpu_buffer->read_bytes = 0;
6156
6157 rb_time_set(t: &cpu_buffer->write_stamp, val: 0);
6158 rb_time_set(t: &cpu_buffer->before_stamp, val: 0);
6159
6160 memset(s: cpu_buffer->event_stamp, c: 0, n: sizeof(cpu_buffer->event_stamp));
6161
6162 cpu_buffer->lost_events = 0;
6163 cpu_buffer->last_overrun = 0;
6164
6165 rb_head_page_activate(cpu_buffer);
6166 cpu_buffer->pages_removed = 0;
6167
6168 if (cpu_buffer->mapped) {
6169 rb_update_meta_page(cpu_buffer);
6170 if (cpu_buffer->ring_meta) {
6171 struct ring_buffer_cpu_meta *meta = cpu_buffer->ring_meta;
6172 meta->commit_buffer = meta->head_buffer;
6173 }
6174 }
6175}
6176
6177/* Must have disabled the cpu buffer then done a synchronize_rcu */
6178static void reset_disabled_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
6179{
6180 guard(raw_spinlock_irqsave)(l: &cpu_buffer->reader_lock);
6181
6182 if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing)))
6183 return;
6184
6185 arch_spin_lock(&cpu_buffer->lock);
6186
6187 rb_reset_cpu(cpu_buffer);
6188
6189 arch_spin_unlock(&cpu_buffer->lock);
6190}
6191
6192/**
6193 * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer
6194 * @buffer: The ring buffer to reset a per cpu buffer of
6195 * @cpu: The CPU buffer to be reset
6196 */
6197void ring_buffer_reset_cpu(struct trace_buffer *buffer, int cpu)
6198{
6199 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
6200
6201 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
6202 return;
6203
6204 /* prevent another thread from changing buffer sizes */
6205 mutex_lock(lock: &buffer->mutex);
6206
6207 atomic_inc(v: &cpu_buffer->resize_disabled);
6208 atomic_inc(v: &cpu_buffer->record_disabled);
6209
6210 /* Make sure all commits have finished */
6211 synchronize_rcu();
6212
6213 reset_disabled_cpu_buffer(cpu_buffer);
6214
6215 atomic_dec(v: &cpu_buffer->record_disabled);
6216 atomic_dec(v: &cpu_buffer->resize_disabled);
6217
6218 mutex_unlock(lock: &buffer->mutex);
6219}
6220EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
6221
6222/* Flag to ensure proper resetting of atomic variables */
6223#define RESET_BIT (1 << 30)
6224
6225/**
6226 * ring_buffer_reset_online_cpus - reset a ring buffer per CPU buffer
6227 * @buffer: The ring buffer to reset a per cpu buffer of
6228 */
6229void ring_buffer_reset_online_cpus(struct trace_buffer *buffer)
6230{
6231 struct ring_buffer_per_cpu *cpu_buffer;
6232 int cpu;
6233
6234 /* prevent another thread from changing buffer sizes */
6235 mutex_lock(lock: &buffer->mutex);
6236
6237 for_each_online_buffer_cpu(buffer, cpu) {
6238 cpu_buffer = buffer->buffers[cpu];
6239
6240 atomic_add(RESET_BIT, v: &cpu_buffer->resize_disabled);
6241 atomic_inc(v: &cpu_buffer->record_disabled);
6242 }
6243
6244 /* Make sure all commits have finished */
6245 synchronize_rcu();
6246
6247 for_each_buffer_cpu(buffer, cpu) {
6248 cpu_buffer = buffer->buffers[cpu];
6249
6250 /*
6251 * If a CPU came online during the synchronize_rcu(), then
6252 * ignore it.
6253 */
6254 if (!(atomic_read(v: &cpu_buffer->resize_disabled) & RESET_BIT))
6255 continue;
6256
6257 reset_disabled_cpu_buffer(cpu_buffer);
6258
6259 atomic_dec(v: &cpu_buffer->record_disabled);
6260 atomic_sub(RESET_BIT, v: &cpu_buffer->resize_disabled);
6261 }
6262
6263 mutex_unlock(lock: &buffer->mutex);
6264}
6265
6266/**
6267 * ring_buffer_reset - reset a ring buffer
6268 * @buffer: The ring buffer to reset all cpu buffers
6269 */
6270void ring_buffer_reset(struct trace_buffer *buffer)
6271{
6272 struct ring_buffer_per_cpu *cpu_buffer;
6273 int cpu;
6274
6275 /* prevent another thread from changing buffer sizes */
6276 mutex_lock(lock: &buffer->mutex);
6277
6278 for_each_buffer_cpu(buffer, cpu) {
6279 cpu_buffer = buffer->buffers[cpu];
6280
6281 atomic_inc(v: &cpu_buffer->resize_disabled);
6282 atomic_inc(v: &cpu_buffer->record_disabled);
6283 }
6284
6285 /* Make sure all commits have finished */
6286 synchronize_rcu();
6287
6288 for_each_buffer_cpu(buffer, cpu) {
6289 cpu_buffer = buffer->buffers[cpu];
6290
6291 reset_disabled_cpu_buffer(cpu_buffer);
6292
6293 atomic_dec(v: &cpu_buffer->record_disabled);
6294 atomic_dec(v: &cpu_buffer->resize_disabled);
6295 }
6296
6297 mutex_unlock(lock: &buffer->mutex);
6298}
6299EXPORT_SYMBOL_GPL(ring_buffer_reset);
6300
6301/**
6302 * ring_buffer_empty - is the ring buffer empty?
6303 * @buffer: The ring buffer to test
6304 */
6305bool ring_buffer_empty(struct trace_buffer *buffer)
6306{
6307 struct ring_buffer_per_cpu *cpu_buffer;
6308 unsigned long flags;
6309 bool dolock;
6310 bool ret;
6311 int cpu;
6312
6313 /* yes this is racy, but if you don't like the race, lock the buffer */
6314 for_each_buffer_cpu(buffer, cpu) {
6315 cpu_buffer = buffer->buffers[cpu];
6316 local_irq_save(flags);
6317 dolock = rb_reader_lock(cpu_buffer);
6318 ret = rb_per_cpu_empty(cpu_buffer);
6319 rb_reader_unlock(cpu_buffer, locked: dolock);
6320 local_irq_restore(flags);
6321
6322 if (!ret)
6323 return false;
6324 }
6325
6326 return true;
6327}
6328EXPORT_SYMBOL_GPL(ring_buffer_empty);
6329
6330/**
6331 * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty?
6332 * @buffer: The ring buffer
6333 * @cpu: The CPU buffer to test
6334 */
6335bool ring_buffer_empty_cpu(struct trace_buffer *buffer, int cpu)
6336{
6337 struct ring_buffer_per_cpu *cpu_buffer;
6338 unsigned long flags;
6339 bool dolock;
6340 bool ret;
6341
6342 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
6343 return true;
6344
6345 cpu_buffer = buffer->buffers[cpu];
6346 local_irq_save(flags);
6347 dolock = rb_reader_lock(cpu_buffer);
6348 ret = rb_per_cpu_empty(cpu_buffer);
6349 rb_reader_unlock(cpu_buffer, locked: dolock);
6350 local_irq_restore(flags);
6351
6352 return ret;
6353}
6354EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
6355
6356#ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
6357/**
6358 * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers
6359 * @buffer_a: One buffer to swap with
6360 * @buffer_b: The other buffer to swap with
6361 * @cpu: the CPU of the buffers to swap
6362 *
6363 * This function is useful for tracers that want to take a "snapshot"
6364 * of a CPU buffer and has another back up buffer lying around.
6365 * it is expected that the tracer handles the cpu buffer not being
6366 * used at the moment.
6367 */
6368int ring_buffer_swap_cpu(struct trace_buffer *buffer_a,
6369 struct trace_buffer *buffer_b, int cpu)
6370{
6371 struct ring_buffer_per_cpu *cpu_buffer_a;
6372 struct ring_buffer_per_cpu *cpu_buffer_b;
6373 int ret = -EINVAL;
6374
6375 if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
6376 !cpumask_test_cpu(cpu, buffer_b->cpumask))
6377 return -EINVAL;
6378
6379 cpu_buffer_a = buffer_a->buffers[cpu];
6380 cpu_buffer_b = buffer_b->buffers[cpu];
6381
6382 /* It's up to the callers to not try to swap mapped buffers */
6383 if (WARN_ON_ONCE(cpu_buffer_a->mapped || cpu_buffer_b->mapped))
6384 return -EBUSY;
6385
6386 /* At least make sure the two buffers are somewhat the same */
6387 if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages)
6388 return -EINVAL;
6389
6390 if (buffer_a->subbuf_order != buffer_b->subbuf_order)
6391 return -EINVAL;
6392
6393 if (atomic_read(&buffer_a->record_disabled))
6394 return -EAGAIN;
6395
6396 if (atomic_read(&buffer_b->record_disabled))
6397 return -EAGAIN;
6398
6399 if (atomic_read(&cpu_buffer_a->record_disabled))
6400 return -EAGAIN;
6401
6402 if (atomic_read(&cpu_buffer_b->record_disabled))
6403 return -EAGAIN;
6404
6405 /*
6406 * We can't do a synchronize_rcu here because this
6407 * function can be called in atomic context.
6408 * Normally this will be called from the same CPU as cpu.
6409 * If not it's up to the caller to protect this.
6410 */
6411 atomic_inc(&cpu_buffer_a->record_disabled);
6412 atomic_inc(&cpu_buffer_b->record_disabled);
6413
6414 ret = -EBUSY;
6415 if (local_read(&cpu_buffer_a->committing))
6416 goto out_dec;
6417 if (local_read(&cpu_buffer_b->committing))
6418 goto out_dec;
6419
6420 /*
6421 * When resize is in progress, we cannot swap it because
6422 * it will mess the state of the cpu buffer.
6423 */
6424 if (atomic_read(&buffer_a->resizing))
6425 goto out_dec;
6426 if (atomic_read(&buffer_b->resizing))
6427 goto out_dec;
6428
6429 buffer_a->buffers[cpu] = cpu_buffer_b;
6430 buffer_b->buffers[cpu] = cpu_buffer_a;
6431
6432 cpu_buffer_b->buffer = buffer_a;
6433 cpu_buffer_a->buffer = buffer_b;
6434
6435 ret = 0;
6436
6437out_dec:
6438 atomic_dec(&cpu_buffer_a->record_disabled);
6439 atomic_dec(&cpu_buffer_b->record_disabled);
6440 return ret;
6441}
6442EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
6443#endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */
6444
6445/**
6446 * ring_buffer_alloc_read_page - allocate a page to read from buffer
6447 * @buffer: the buffer to allocate for.
6448 * @cpu: the cpu buffer to allocate.
6449 *
6450 * This function is used in conjunction with ring_buffer_read_page.
6451 * When reading a full page from the ring buffer, these functions
6452 * can be used to speed up the process. The calling function should
6453 * allocate a few pages first with this function. Then when it
6454 * needs to get pages from the ring buffer, it passes the result
6455 * of this function into ring_buffer_read_page, which will swap
6456 * the page that was allocated, with the read page of the buffer.
6457 *
6458 * Returns:
6459 * The page allocated, or ERR_PTR
6460 */
6461struct buffer_data_read_page *
6462ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
6463{
6464 struct ring_buffer_per_cpu *cpu_buffer;
6465 struct buffer_data_read_page *bpage = NULL;
6466 unsigned long flags;
6467 struct page *page;
6468
6469 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
6470 return ERR_PTR(error: -ENODEV);
6471
6472 bpage = kzalloc(sizeof(*bpage), GFP_KERNEL);
6473 if (!bpage)
6474 return ERR_PTR(error: -ENOMEM);
6475
6476 bpage->order = buffer->subbuf_order;
6477 cpu_buffer = buffer->buffers[cpu];
6478 local_irq_save(flags);
6479 arch_spin_lock(&cpu_buffer->lock);
6480
6481 if (cpu_buffer->free_page) {
6482 bpage->data = cpu_buffer->free_page;
6483 cpu_buffer->free_page = NULL;
6484 }
6485
6486 arch_spin_unlock(&cpu_buffer->lock);
6487 local_irq_restore(flags);
6488
6489 if (bpage->data)
6490 goto out;
6491
6492 page = alloc_pages_node(cpu_to_node(cpu),
6493 GFP_KERNEL | __GFP_NORETRY | __GFP_COMP | __GFP_ZERO,
6494 cpu_buffer->buffer->subbuf_order);
6495 if (!page) {
6496 kfree(objp: bpage);
6497 return ERR_PTR(error: -ENOMEM);
6498 }
6499
6500 bpage->data = page_address(page);
6501
6502 out:
6503 rb_init_page(bpage: bpage->data);
6504
6505 return bpage;
6506}
6507EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page);
6508
6509/**
6510 * ring_buffer_free_read_page - free an allocated read page
6511 * @buffer: the buffer the page was allocate for
6512 * @cpu: the cpu buffer the page came from
6513 * @data_page: the page to free
6514 *
6515 * Free a page allocated from ring_buffer_alloc_read_page.
6516 */
6517void ring_buffer_free_read_page(struct trace_buffer *buffer, int cpu,
6518 struct buffer_data_read_page *data_page)
6519{
6520 struct ring_buffer_per_cpu *cpu_buffer;
6521 struct buffer_data_page *bpage = data_page->data;
6522 struct page *page = virt_to_page(bpage);
6523 unsigned long flags;
6524
6525 if (!buffer || !buffer->buffers || !buffer->buffers[cpu])
6526 return;
6527
6528 cpu_buffer = buffer->buffers[cpu];
6529
6530 /*
6531 * If the page is still in use someplace else, or order of the page
6532 * is different from the subbuffer order of the buffer -
6533 * we can't reuse it
6534 */
6535 if (page_ref_count(page) > 1 || data_page->order != buffer->subbuf_order)
6536 goto out;
6537
6538 local_irq_save(flags);
6539 arch_spin_lock(&cpu_buffer->lock);
6540
6541 if (!cpu_buffer->free_page) {
6542 cpu_buffer->free_page = bpage;
6543 bpage = NULL;
6544 }
6545
6546 arch_spin_unlock(&cpu_buffer->lock);
6547 local_irq_restore(flags);
6548
6549 out:
6550 free_pages(addr: (unsigned long)bpage, order: data_page->order);
6551 kfree(objp: data_page);
6552}
6553EXPORT_SYMBOL_GPL(ring_buffer_free_read_page);
6554
6555/**
6556 * ring_buffer_read_page - extract a page from the ring buffer
6557 * @buffer: buffer to extract from
6558 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
6559 * @len: amount to extract
6560 * @cpu: the cpu of the buffer to extract
6561 * @full: should the extraction only happen when the page is full.
6562 *
6563 * This function will pull out a page from the ring buffer and consume it.
6564 * @data_page must be the address of the variable that was returned
6565 * from ring_buffer_alloc_read_page. This is because the page might be used
6566 * to swap with a page in the ring buffer.
6567 *
6568 * for example:
6569 * rpage = ring_buffer_alloc_read_page(buffer, cpu);
6570 * if (IS_ERR(rpage))
6571 * return PTR_ERR(rpage);
6572 * ret = ring_buffer_read_page(buffer, rpage, len, cpu, 0);
6573 * if (ret >= 0)
6574 * process_page(ring_buffer_read_page_data(rpage), ret);
6575 * ring_buffer_free_read_page(buffer, cpu, rpage);
6576 *
6577 * When @full is set, the function will not return true unless
6578 * the writer is off the reader page.
6579 *
6580 * Note: it is up to the calling functions to handle sleeps and wakeups.
6581 * The ring buffer can be used anywhere in the kernel and can not
6582 * blindly call wake_up. The layer that uses the ring buffer must be
6583 * responsible for that.
6584 *
6585 * Returns:
6586 * >=0 if data has been transferred, returns the offset of consumed data.
6587 * <0 if no data has been transferred.
6588 */
6589int ring_buffer_read_page(struct trace_buffer *buffer,
6590 struct buffer_data_read_page *data_page,
6591 size_t len, int cpu, int full)
6592{
6593 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
6594 struct ring_buffer_event *event;
6595 struct buffer_data_page *bpage;
6596 struct buffer_page *reader;
6597 unsigned long missed_events;
6598 unsigned int commit;
6599 unsigned int read;
6600 u64 save_timestamp;
6601
6602 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
6603 return -1;
6604
6605 /*
6606 * If len is not big enough to hold the page header, then
6607 * we can not copy anything.
6608 */
6609 if (len <= BUF_PAGE_HDR_SIZE)
6610 return -1;
6611
6612 len -= BUF_PAGE_HDR_SIZE;
6613
6614 if (!data_page || !data_page->data)
6615 return -1;
6616
6617 if (data_page->order != buffer->subbuf_order)
6618 return -1;
6619
6620 bpage = data_page->data;
6621 if (!bpage)
6622 return -1;
6623
6624 guard(raw_spinlock_irqsave)(l: &cpu_buffer->reader_lock);
6625
6626 reader = rb_get_reader_page(cpu_buffer);
6627 if (!reader)
6628 return -1;
6629
6630 event = rb_reader_event(cpu_buffer);
6631
6632 read = reader->read;
6633 commit = rb_page_size(bpage: reader);
6634
6635 /* Check if any events were dropped */
6636 missed_events = cpu_buffer->lost_events;
6637
6638 /*
6639 * If this page has been partially read or
6640 * if len is not big enough to read the rest of the page or
6641 * a writer is still on the page, then
6642 * we must copy the data from the page to the buffer.
6643 * Otherwise, we can simply swap the page with the one passed in.
6644 */
6645 if (read || (len < (commit - read)) ||
6646 cpu_buffer->reader_page == cpu_buffer->commit_page ||
6647 cpu_buffer->mapped) {
6648 struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
6649 unsigned int rpos = read;
6650 unsigned int pos = 0;
6651 unsigned int size;
6652
6653 /*
6654 * If a full page is expected, this can still be returned
6655 * if there's been a previous partial read and the
6656 * rest of the page can be read and the commit page is off
6657 * the reader page.
6658 */
6659 if (full &&
6660 (!read || (len < (commit - read)) ||
6661 cpu_buffer->reader_page == cpu_buffer->commit_page))
6662 return -1;
6663
6664 if (len > (commit - read))
6665 len = (commit - read);
6666
6667 /* Always keep the time extend and data together */
6668 size = rb_event_ts_length(event);
6669
6670 if (len < size)
6671 return -1;
6672
6673 /* save the current timestamp, since the user will need it */
6674 save_timestamp = cpu_buffer->read_stamp;
6675
6676 /* Need to copy one event at a time */
6677 do {
6678 /* We need the size of one event, because
6679 * rb_advance_reader only advances by one event,
6680 * whereas rb_event_ts_length may include the size of
6681 * one or two events.
6682 * We have already ensured there's enough space if this
6683 * is a time extend. */
6684 size = rb_event_length(event);
6685 memcpy(to: bpage->data + pos, from: rpage->data + rpos, len: size);
6686
6687 len -= size;
6688
6689 rb_advance_reader(cpu_buffer);
6690 rpos = reader->read;
6691 pos += size;
6692
6693 if (rpos >= commit)
6694 break;
6695
6696 event = rb_reader_event(cpu_buffer);
6697 /* Always keep the time extend and data together */
6698 size = rb_event_ts_length(event);
6699 } while (len >= size);
6700
6701 /* update bpage */
6702 local_set(&bpage->commit, pos);
6703 bpage->time_stamp = save_timestamp;
6704
6705 /* we copied everything to the beginning */
6706 read = 0;
6707 } else {
6708 /* update the entry counter */
6709 cpu_buffer->read += rb_page_entries(bpage: reader);
6710 cpu_buffer->read_bytes += rb_page_size(bpage: reader);
6711
6712 /* swap the pages */
6713 rb_init_page(bpage);
6714 bpage = reader->page;
6715 reader->page = data_page->data;
6716 local_set(&reader->write, 0);
6717 local_set(&reader->entries, 0);
6718 reader->read = 0;
6719 data_page->data = bpage;
6720
6721 /*
6722 * Use the real_end for the data size,
6723 * This gives us a chance to store the lost events
6724 * on the page.
6725 */
6726 if (reader->real_end)
6727 local_set(&bpage->commit, reader->real_end);
6728 }
6729
6730 cpu_buffer->lost_events = 0;
6731
6732 commit = local_read(&bpage->commit);
6733 /*
6734 * Set a flag in the commit field if we lost events
6735 */
6736 if (missed_events) {
6737 /* If there is room at the end of the page to save the
6738 * missed events, then record it there.
6739 */
6740 if (buffer->subbuf_size - commit >= sizeof(missed_events)) {
6741 memcpy(to: &bpage->data[commit], from: &missed_events,
6742 len: sizeof(missed_events));
6743 local_add(RB_MISSED_STORED, l: &bpage->commit);
6744 commit += sizeof(missed_events);
6745 }
6746 local_add(RB_MISSED_EVENTS, l: &bpage->commit);
6747 }
6748
6749 /*
6750 * This page may be off to user land. Zero it out here.
6751 */
6752 if (commit < buffer->subbuf_size)
6753 memset(s: &bpage->data[commit], c: 0, n: buffer->subbuf_size - commit);
6754
6755 return read;
6756}
6757EXPORT_SYMBOL_GPL(ring_buffer_read_page);
6758
6759/**
6760 * ring_buffer_read_page_data - get pointer to the data in the page.
6761 * @page: the page to get the data from
6762 *
6763 * Returns pointer to the actual data in this page.
6764 */
6765void *ring_buffer_read_page_data(struct buffer_data_read_page *page)
6766{
6767 return page->data;
6768}
6769EXPORT_SYMBOL_GPL(ring_buffer_read_page_data);
6770
6771/**
6772 * ring_buffer_subbuf_size_get - get size of the sub buffer.
6773 * @buffer: the buffer to get the sub buffer size from
6774 *
6775 * Returns size of the sub buffer, in bytes.
6776 */
6777int ring_buffer_subbuf_size_get(struct trace_buffer *buffer)
6778{
6779 return buffer->subbuf_size + BUF_PAGE_HDR_SIZE;
6780}
6781EXPORT_SYMBOL_GPL(ring_buffer_subbuf_size_get);
6782
6783/**
6784 * ring_buffer_subbuf_order_get - get order of system sub pages in one buffer page.
6785 * @buffer: The ring_buffer to get the system sub page order from
6786 *
6787 * By default, one ring buffer sub page equals to one system page. This parameter
6788 * is configurable, per ring buffer. The size of the ring buffer sub page can be
6789 * extended, but must be an order of system page size.
6790 *
6791 * Returns the order of buffer sub page size, in system pages:
6792 * 0 means the sub buffer size is 1 system page and so forth.
6793 * In case of an error < 0 is returned.
6794 */
6795int ring_buffer_subbuf_order_get(struct trace_buffer *buffer)
6796{
6797 if (!buffer)
6798 return -EINVAL;
6799
6800 return buffer->subbuf_order;
6801}
6802EXPORT_SYMBOL_GPL(ring_buffer_subbuf_order_get);
6803
6804/**
6805 * ring_buffer_subbuf_order_set - set the size of ring buffer sub page.
6806 * @buffer: The ring_buffer to set the new page size.
6807 * @order: Order of the system pages in one sub buffer page
6808 *
6809 * By default, one ring buffer pages equals to one system page. This API can be
6810 * used to set new size of the ring buffer page. The size must be order of
6811 * system page size, that's why the input parameter @order is the order of
6812 * system pages that are allocated for one ring buffer page:
6813 * 0 - 1 system page
6814 * 1 - 2 system pages
6815 * 3 - 4 system pages
6816 * ...
6817 *
6818 * Returns 0 on success or < 0 in case of an error.
6819 */
6820int ring_buffer_subbuf_order_set(struct trace_buffer *buffer, int order)
6821{
6822 struct ring_buffer_per_cpu *cpu_buffer;
6823 struct buffer_page *bpage, *tmp;
6824 int old_order, old_size;
6825 int nr_pages;
6826 int psize;
6827 int err;
6828 int cpu;
6829
6830 if (!buffer || order < 0)
6831 return -EINVAL;
6832
6833 if (buffer->subbuf_order == order)
6834 return 0;
6835
6836 psize = (1 << order) * PAGE_SIZE;
6837 if (psize <= BUF_PAGE_HDR_SIZE)
6838 return -EINVAL;
6839
6840 /* Size of a subbuf cannot be greater than the write counter */
6841 if (psize > RB_WRITE_MASK + 1)
6842 return -EINVAL;
6843
6844 old_order = buffer->subbuf_order;
6845 old_size = buffer->subbuf_size;
6846
6847 /* prevent another thread from changing buffer sizes */
6848 guard(mutex)(T: &buffer->mutex);
6849 atomic_inc(v: &buffer->record_disabled);
6850
6851 /* Make sure all commits have finished */
6852 synchronize_rcu();
6853
6854 buffer->subbuf_order = order;
6855 buffer->subbuf_size = psize - BUF_PAGE_HDR_SIZE;
6856
6857 /* Make sure all new buffers are allocated, before deleting the old ones */
6858 for_each_buffer_cpu(buffer, cpu) {
6859
6860 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
6861 continue;
6862
6863 cpu_buffer = buffer->buffers[cpu];
6864
6865 if (cpu_buffer->mapped) {
6866 err = -EBUSY;
6867 goto error;
6868 }
6869
6870 /* Update the number of pages to match the new size */
6871 nr_pages = old_size * buffer->buffers[cpu]->nr_pages;
6872 nr_pages = DIV_ROUND_UP(nr_pages, buffer->subbuf_size);
6873
6874 /* we need a minimum of two pages */
6875 if (nr_pages < 2)
6876 nr_pages = 2;
6877
6878 cpu_buffer->nr_pages_to_update = nr_pages;
6879
6880 /* Include the reader page */
6881 nr_pages++;
6882
6883 /* Allocate the new size buffer */
6884 INIT_LIST_HEAD(list: &cpu_buffer->new_pages);
6885 if (__rb_allocate_pages(cpu_buffer, nr_pages,
6886 pages: &cpu_buffer->new_pages)) {
6887 /* not enough memory for new pages */
6888 err = -ENOMEM;
6889 goto error;
6890 }
6891 }
6892
6893 for_each_buffer_cpu(buffer, cpu) {
6894 struct buffer_data_page *old_free_data_page;
6895 struct list_head old_pages;
6896 unsigned long flags;
6897
6898 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
6899 continue;
6900
6901 cpu_buffer = buffer->buffers[cpu];
6902
6903 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
6904
6905 /* Clear the head bit to make the link list normal to read */
6906 rb_head_page_deactivate(cpu_buffer);
6907
6908 /*
6909 * Collect buffers from the cpu_buffer pages list and the
6910 * reader_page on old_pages, so they can be freed later when not
6911 * under a spinlock. The pages list is a linked list with no
6912 * head, adding old_pages turns it into a regular list with
6913 * old_pages being the head.
6914 */
6915 list_add(new: &old_pages, head: cpu_buffer->pages);
6916 list_add(new: &cpu_buffer->reader_page->list, head: &old_pages);
6917
6918 /* One page was allocated for the reader page */
6919 cpu_buffer->reader_page = list_entry(cpu_buffer->new_pages.next,
6920 struct buffer_page, list);
6921 list_del_init(entry: &cpu_buffer->reader_page->list);
6922
6923 /* Install the new pages, remove the head from the list */
6924 cpu_buffer->pages = cpu_buffer->new_pages.next;
6925 list_del_init(entry: &cpu_buffer->new_pages);
6926 cpu_buffer->cnt++;
6927
6928 cpu_buffer->head_page
6929 = list_entry(cpu_buffer->pages, struct buffer_page, list);
6930 cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
6931
6932 cpu_buffer->nr_pages = cpu_buffer->nr_pages_to_update;
6933 cpu_buffer->nr_pages_to_update = 0;
6934
6935 old_free_data_page = cpu_buffer->free_page;
6936 cpu_buffer->free_page = NULL;
6937
6938 rb_head_page_activate(cpu_buffer);
6939
6940 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
6941
6942 /* Free old sub buffers */
6943 list_for_each_entry_safe(bpage, tmp, &old_pages, list) {
6944 list_del_init(entry: &bpage->list);
6945 free_buffer_page(bpage);
6946 }
6947 free_pages(addr: (unsigned long)old_free_data_page, order: old_order);
6948
6949 rb_check_pages(cpu_buffer);
6950 }
6951
6952 atomic_dec(v: &buffer->record_disabled);
6953
6954 return 0;
6955
6956error:
6957 buffer->subbuf_order = old_order;
6958 buffer->subbuf_size = old_size;
6959
6960 atomic_dec(v: &buffer->record_disabled);
6961
6962 for_each_buffer_cpu(buffer, cpu) {
6963 cpu_buffer = buffer->buffers[cpu];
6964
6965 if (!cpu_buffer->nr_pages_to_update)
6966 continue;
6967
6968 list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, list) {
6969 list_del_init(entry: &bpage->list);
6970 free_buffer_page(bpage);
6971 }
6972 }
6973
6974 return err;
6975}
6976EXPORT_SYMBOL_GPL(ring_buffer_subbuf_order_set);
6977
6978static int rb_alloc_meta_page(struct ring_buffer_per_cpu *cpu_buffer)
6979{
6980 struct page *page;
6981
6982 if (cpu_buffer->meta_page)
6983 return 0;
6984
6985 page = alloc_page(GFP_USER | __GFP_ZERO);
6986 if (!page)
6987 return -ENOMEM;
6988
6989 cpu_buffer->meta_page = page_to_virt(page);
6990
6991 return 0;
6992}
6993
6994static void rb_free_meta_page(struct ring_buffer_per_cpu *cpu_buffer)
6995{
6996 unsigned long addr = (unsigned long)cpu_buffer->meta_page;
6997
6998 free_page(addr);
6999 cpu_buffer->meta_page = NULL;
7000}
7001
7002static void rb_setup_ids_meta_page(struct ring_buffer_per_cpu *cpu_buffer,
7003 unsigned long *subbuf_ids)
7004{
7005 struct trace_buffer_meta *meta = cpu_buffer->meta_page;
7006 unsigned int nr_subbufs = cpu_buffer->nr_pages + 1;
7007 struct buffer_page *first_subbuf, *subbuf;
7008 int cnt = 0;
7009 int id = 0;
7010
7011 id = rb_page_id(cpu_buffer, bpage: cpu_buffer->reader_page, id);
7012 subbuf_ids[id++] = (unsigned long)cpu_buffer->reader_page->page;
7013 cnt++;
7014
7015 first_subbuf = subbuf = rb_set_head_page(cpu_buffer);
7016 do {
7017 id = rb_page_id(cpu_buffer, bpage: subbuf, id);
7018
7019 if (WARN_ON(id >= nr_subbufs))
7020 break;
7021
7022 subbuf_ids[id] = (unsigned long)subbuf->page;
7023
7024 rb_inc_page(bpage: &subbuf);
7025 id++;
7026 cnt++;
7027 } while (subbuf != first_subbuf);
7028
7029 WARN_ON(cnt != nr_subbufs);
7030
7031 /* install subbuf ID to kern VA translation */
7032 cpu_buffer->subbuf_ids = subbuf_ids;
7033
7034 meta->meta_struct_len = sizeof(*meta);
7035 meta->nr_subbufs = nr_subbufs;
7036 meta->subbuf_size = cpu_buffer->buffer->subbuf_size + BUF_PAGE_HDR_SIZE;
7037 meta->meta_page_size = meta->subbuf_size;
7038
7039 rb_update_meta_page(cpu_buffer);
7040}
7041
7042static struct ring_buffer_per_cpu *
7043rb_get_mapped_buffer(struct trace_buffer *buffer, int cpu)
7044{
7045 struct ring_buffer_per_cpu *cpu_buffer;
7046
7047 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
7048 return ERR_PTR(error: -EINVAL);
7049
7050 cpu_buffer = buffer->buffers[cpu];
7051
7052 mutex_lock(lock: &cpu_buffer->mapping_lock);
7053
7054 if (!cpu_buffer->user_mapped) {
7055 mutex_unlock(lock: &cpu_buffer->mapping_lock);
7056 return ERR_PTR(error: -ENODEV);
7057 }
7058
7059 return cpu_buffer;
7060}
7061
7062static void rb_put_mapped_buffer(struct ring_buffer_per_cpu *cpu_buffer)
7063{
7064 mutex_unlock(lock: &cpu_buffer->mapping_lock);
7065}
7066
7067/*
7068 * Fast-path for rb_buffer_(un)map(). Called whenever the meta-page doesn't need
7069 * to be set-up or torn-down.
7070 */
7071static int __rb_inc_dec_mapped(struct ring_buffer_per_cpu *cpu_buffer,
7072 bool inc)
7073{
7074 unsigned long flags;
7075
7076 lockdep_assert_held(&cpu_buffer->mapping_lock);
7077
7078 /* mapped is always greater or equal to user_mapped */
7079 if (WARN_ON(cpu_buffer->mapped < cpu_buffer->user_mapped))
7080 return -EINVAL;
7081
7082 if (inc && cpu_buffer->mapped == UINT_MAX)
7083 return -EBUSY;
7084
7085 if (WARN_ON(!inc && cpu_buffer->user_mapped == 0))
7086 return -EINVAL;
7087
7088 mutex_lock(lock: &cpu_buffer->buffer->mutex);
7089 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
7090
7091 if (inc) {
7092 cpu_buffer->user_mapped++;
7093 cpu_buffer->mapped++;
7094 } else {
7095 cpu_buffer->user_mapped--;
7096 cpu_buffer->mapped--;
7097 }
7098
7099 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
7100 mutex_unlock(lock: &cpu_buffer->buffer->mutex);
7101
7102 return 0;
7103}
7104
7105/*
7106 * +--------------+ pgoff == 0
7107 * | meta page |
7108 * +--------------+ pgoff == 1
7109 * | subbuffer 0 |
7110 * | |
7111 * +--------------+ pgoff == (1 + (1 << subbuf_order))
7112 * | subbuffer 1 |
7113 * | |
7114 * ...
7115 */
7116#ifdef CONFIG_MMU
7117static int __rb_map_vma(struct ring_buffer_per_cpu *cpu_buffer,
7118 struct vm_area_struct *vma)
7119{
7120 unsigned long nr_subbufs, nr_pages, nr_vma_pages, pgoff = vma->vm_pgoff;
7121 unsigned int subbuf_pages, subbuf_order;
7122 struct page **pages __free(kfree) = NULL;
7123 int p = 0, s = 0;
7124 int err;
7125
7126 /* Refuse MP_PRIVATE or writable mappings */
7127 if (vma->vm_flags & VM_WRITE || vma->vm_flags & VM_EXEC ||
7128 !(vma->vm_flags & VM_MAYSHARE))
7129 return -EPERM;
7130
7131 subbuf_order = cpu_buffer->buffer->subbuf_order;
7132 subbuf_pages = 1 << subbuf_order;
7133
7134 if (subbuf_order && pgoff % subbuf_pages)
7135 return -EINVAL;
7136
7137 /*
7138 * Make sure the mapping cannot become writable later. Also tell the VM
7139 * to not touch these pages (VM_DONTCOPY | VM_DONTEXPAND).
7140 */
7141 vm_flags_mod(vma, VM_DONTCOPY | VM_DONTEXPAND | VM_DONTDUMP,
7142 VM_MAYWRITE);
7143
7144 lockdep_assert_held(&cpu_buffer->mapping_lock);
7145
7146 nr_subbufs = cpu_buffer->nr_pages + 1; /* + reader-subbuf */
7147 nr_pages = ((nr_subbufs + 1) << subbuf_order); /* + meta-page */
7148 if (nr_pages <= pgoff)
7149 return -EINVAL;
7150
7151 nr_pages -= pgoff;
7152
7153 nr_vma_pages = vma_pages(vma);
7154 if (!nr_vma_pages || nr_vma_pages > nr_pages)
7155 return -EINVAL;
7156
7157 nr_pages = nr_vma_pages;
7158
7159 pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
7160 if (!pages)
7161 return -ENOMEM;
7162
7163 if (!pgoff) {
7164 unsigned long meta_page_padding;
7165
7166 pages[p++] = virt_to_page(cpu_buffer->meta_page);
7167
7168 /*
7169 * Pad with the zero-page to align the meta-page with the
7170 * sub-buffers.
7171 */
7172 meta_page_padding = subbuf_pages - 1;
7173 while (meta_page_padding-- && p < nr_pages) {
7174 unsigned long __maybe_unused zero_addr =
7175 vma->vm_start + (PAGE_SIZE * p);
7176
7177 pages[p++] = ZERO_PAGE(zero_addr);
7178 }
7179 } else {
7180 /* Skip the meta-page */
7181 pgoff -= subbuf_pages;
7182
7183 s += pgoff / subbuf_pages;
7184 }
7185
7186 while (p < nr_pages) {
7187 struct page *page;
7188 int off = 0;
7189
7190 if (WARN_ON_ONCE(s >= nr_subbufs))
7191 return -EINVAL;
7192
7193 page = virt_to_page((void *)cpu_buffer->subbuf_ids[s]);
7194
7195 for (; off < (1 << (subbuf_order)); off++, page++) {
7196 if (p >= nr_pages)
7197 break;
7198
7199 pages[p++] = page;
7200 }
7201 s++;
7202 }
7203
7204 err = vm_insert_pages(vma, addr: vma->vm_start, pages, num: &nr_pages);
7205
7206 return err;
7207}
7208#else
7209static int __rb_map_vma(struct ring_buffer_per_cpu *cpu_buffer,
7210 struct vm_area_struct *vma)
7211{
7212 return -EOPNOTSUPP;
7213}
7214#endif
7215
7216int ring_buffer_map(struct trace_buffer *buffer, int cpu,
7217 struct vm_area_struct *vma)
7218{
7219 struct ring_buffer_per_cpu *cpu_buffer;
7220 unsigned long flags, *subbuf_ids;
7221 int err;
7222
7223 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
7224 return -EINVAL;
7225
7226 cpu_buffer = buffer->buffers[cpu];
7227
7228 guard(mutex)(T: &cpu_buffer->mapping_lock);
7229
7230 if (cpu_buffer->user_mapped) {
7231 err = __rb_map_vma(cpu_buffer, vma);
7232 if (!err)
7233 err = __rb_inc_dec_mapped(cpu_buffer, inc: true);
7234 return err;
7235 }
7236
7237 /* prevent another thread from changing buffer/sub-buffer sizes */
7238 guard(mutex)(T: &buffer->mutex);
7239
7240 err = rb_alloc_meta_page(cpu_buffer);
7241 if (err)
7242 return err;
7243
7244 /* subbuf_ids include the reader while nr_pages does not */
7245 subbuf_ids = kcalloc(cpu_buffer->nr_pages + 1, sizeof(*subbuf_ids), GFP_KERNEL);
7246 if (!subbuf_ids) {
7247 rb_free_meta_page(cpu_buffer);
7248 return -ENOMEM;
7249 }
7250
7251 atomic_inc(v: &cpu_buffer->resize_disabled);
7252
7253 /*
7254 * Lock all readers to block any subbuf swap until the subbuf IDs are
7255 * assigned.
7256 */
7257 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
7258 rb_setup_ids_meta_page(cpu_buffer, subbuf_ids);
7259
7260 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
7261
7262 err = __rb_map_vma(cpu_buffer, vma);
7263 if (!err) {
7264 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
7265 /* This is the first time it is mapped by user */
7266 cpu_buffer->mapped++;
7267 cpu_buffer->user_mapped = 1;
7268 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
7269 } else {
7270 kfree(objp: cpu_buffer->subbuf_ids);
7271 cpu_buffer->subbuf_ids = NULL;
7272 rb_free_meta_page(cpu_buffer);
7273 atomic_dec(v: &cpu_buffer->resize_disabled);
7274 }
7275
7276 return err;
7277}
7278
7279int ring_buffer_unmap(struct trace_buffer *buffer, int cpu)
7280{
7281 struct ring_buffer_per_cpu *cpu_buffer;
7282 unsigned long flags;
7283
7284 if (!cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
7285 return -EINVAL;
7286
7287 cpu_buffer = buffer->buffers[cpu];
7288
7289 guard(mutex)(T: &cpu_buffer->mapping_lock);
7290
7291 if (!cpu_buffer->user_mapped) {
7292 return -ENODEV;
7293 } else if (cpu_buffer->user_mapped > 1) {
7294 __rb_inc_dec_mapped(cpu_buffer, inc: false);
7295 return 0;
7296 }
7297
7298 guard(mutex)(T: &buffer->mutex);
7299 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
7300
7301 /* This is the last user space mapping */
7302 if (!WARN_ON_ONCE(cpu_buffer->mapped < cpu_buffer->user_mapped))
7303 cpu_buffer->mapped--;
7304 cpu_buffer->user_mapped = 0;
7305
7306 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
7307
7308 kfree(objp: cpu_buffer->subbuf_ids);
7309 cpu_buffer->subbuf_ids = NULL;
7310 rb_free_meta_page(cpu_buffer);
7311 atomic_dec(v: &cpu_buffer->resize_disabled);
7312
7313 return 0;
7314}
7315
7316int ring_buffer_map_get_reader(struct trace_buffer *buffer, int cpu)
7317{
7318 struct ring_buffer_per_cpu *cpu_buffer;
7319 struct buffer_page *reader;
7320 unsigned long missed_events;
7321 unsigned long reader_size;
7322 unsigned long flags;
7323
7324 cpu_buffer = rb_get_mapped_buffer(buffer, cpu);
7325 if (IS_ERR(ptr: cpu_buffer))
7326 return (int)PTR_ERR(ptr: cpu_buffer);
7327
7328 raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
7329
7330consume:
7331 if (rb_per_cpu_empty(cpu_buffer))
7332 goto out;
7333
7334 reader_size = rb_page_size(bpage: cpu_buffer->reader_page);
7335
7336 /*
7337 * There are data to be read on the current reader page, we can
7338 * return to the caller. But before that, we assume the latter will read
7339 * everything. Let's update the kernel reader accordingly.
7340 */
7341 if (cpu_buffer->reader_page->read < reader_size) {
7342 while (cpu_buffer->reader_page->read < reader_size)
7343 rb_advance_reader(cpu_buffer);
7344 goto out;
7345 }
7346
7347 reader = rb_get_reader_page(cpu_buffer);
7348 if (WARN_ON(!reader))
7349 goto out;
7350
7351 /* Check if any events were dropped */
7352 missed_events = cpu_buffer->lost_events;
7353
7354 if (missed_events) {
7355 if (cpu_buffer->reader_page != cpu_buffer->commit_page) {
7356 struct buffer_data_page *bpage = reader->page;
7357 unsigned int commit;
7358 /*
7359 * Use the real_end for the data size,
7360 * This gives us a chance to store the lost events
7361 * on the page.
7362 */
7363 if (reader->real_end)
7364 local_set(&bpage->commit, reader->real_end);
7365 /*
7366 * If there is room at the end of the page to save the
7367 * missed events, then record it there.
7368 */
7369 commit = rb_page_size(bpage: reader);
7370 if (buffer->subbuf_size - commit >= sizeof(missed_events)) {
7371 memcpy(to: &bpage->data[commit], from: &missed_events,
7372 len: sizeof(missed_events));
7373 local_add(RB_MISSED_STORED, l: &bpage->commit);
7374 }
7375 local_add(RB_MISSED_EVENTS, l: &bpage->commit);
7376 } else if (!WARN_ONCE(cpu_buffer->reader_page == cpu_buffer->tail_page,
7377 "Reader on commit with %ld missed events",
7378 missed_events)) {
7379 /*
7380 * There shouldn't be any missed events if the tail_page
7381 * is on the reader page. But if the tail page is not on the
7382 * reader page and the commit_page is, that would mean that
7383 * there's a commit_overrun (an interrupt preempted an
7384 * addition of an event and then filled the buffer
7385 * with new events). In this case it's not an
7386 * error, but it should still be reported.
7387 *
7388 * TODO: Add missed events to the page for user space to know.
7389 */
7390 pr_info("Ring buffer [%d] commit overrun lost %ld events at timestamp:%lld\n",
7391 cpu, missed_events, cpu_buffer->reader_page->page->time_stamp);
7392 }
7393 }
7394
7395 cpu_buffer->lost_events = 0;
7396
7397 goto consume;
7398
7399out:
7400 /* Some archs do not have data cache coherency between kernel and user-space */
7401 flush_kernel_vmap_range(vaddr: cpu_buffer->reader_page->page,
7402 size: buffer->subbuf_size + BUF_PAGE_HDR_SIZE);
7403
7404 rb_update_meta_page(cpu_buffer);
7405
7406 raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
7407 rb_put_mapped_buffer(cpu_buffer);
7408
7409 return 0;
7410}
7411
7412/*
7413 * We only allocate new buffers, never free them if the CPU goes down.
7414 * If we were to free the buffer, then the user would lose any trace that was in
7415 * the buffer.
7416 */
7417int trace_rb_cpu_prepare(unsigned int cpu, struct hlist_node *node)
7418{
7419 struct trace_buffer *buffer;
7420 long nr_pages_same;
7421 int cpu_i;
7422 unsigned long nr_pages;
7423
7424 buffer = container_of(node, struct trace_buffer, node);
7425 if (cpumask_test_cpu(cpu, cpumask: buffer->cpumask))
7426 return 0;
7427
7428 nr_pages = 0;
7429 nr_pages_same = 1;
7430 /* check if all cpu sizes are same */
7431 for_each_buffer_cpu(buffer, cpu_i) {
7432 /* fill in the size from first enabled cpu */
7433 if (nr_pages == 0)
7434 nr_pages = buffer->buffers[cpu_i]->nr_pages;
7435 if (nr_pages != buffer->buffers[cpu_i]->nr_pages) {
7436 nr_pages_same = 0;
7437 break;
7438 }
7439 }
7440 /* allocate minimum pages, user can later expand it */
7441 if (!nr_pages_same)
7442 nr_pages = 2;
7443 buffer->buffers[cpu] =
7444 rb_allocate_cpu_buffer(buffer, nr_pages, cpu);
7445 if (!buffer->buffers[cpu]) {
7446 WARN(1, "failed to allocate ring buffer on CPU %u\n",
7447 cpu);
7448 return -ENOMEM;
7449 }
7450 smp_wmb();
7451 cpumask_set_cpu(cpu, dstp: buffer->cpumask);
7452 return 0;
7453}
7454
7455#ifdef CONFIG_RING_BUFFER_STARTUP_TEST
7456/*
7457 * This is a basic integrity check of the ring buffer.
7458 * Late in the boot cycle this test will run when configured in.
7459 * It will kick off a thread per CPU that will go into a loop
7460 * writing to the per cpu ring buffer various sizes of data.
7461 * Some of the data will be large items, some small.
7462 *
7463 * Another thread is created that goes into a spin, sending out
7464 * IPIs to the other CPUs to also write into the ring buffer.
7465 * this is to test the nesting ability of the buffer.
7466 *
7467 * Basic stats are recorded and reported. If something in the
7468 * ring buffer should happen that's not expected, a big warning
7469 * is displayed and all ring buffers are disabled.
7470 */
7471static struct task_struct *rb_threads[NR_CPUS] __initdata;
7472
7473struct rb_test_data {
7474 struct trace_buffer *buffer;
7475 unsigned long events;
7476 unsigned long bytes_written;
7477 unsigned long bytes_alloc;
7478 unsigned long bytes_dropped;
7479 unsigned long events_nested;
7480 unsigned long bytes_written_nested;
7481 unsigned long bytes_alloc_nested;
7482 unsigned long bytes_dropped_nested;
7483 int min_size_nested;
7484 int max_size_nested;
7485 int max_size;
7486 int min_size;
7487 int cpu;
7488 int cnt;
7489};
7490
7491static struct rb_test_data rb_data[NR_CPUS] __initdata;
7492
7493/* 1 meg per cpu */
7494#define RB_TEST_BUFFER_SIZE 1048576
7495
7496static char rb_string[] __initdata =
7497 "abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\"
7498 "?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890"
7499 "!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv";
7500
7501static bool rb_test_started __initdata;
7502
7503struct rb_item {
7504 int size;
7505 char str[];
7506};
7507
7508static __init int rb_write_something(struct rb_test_data *data, bool nested)
7509{
7510 struct ring_buffer_event *event;
7511 struct rb_item *item;
7512 bool started;
7513 int event_len;
7514 int size;
7515 int len;
7516 int cnt;
7517
7518 /* Have nested writes different that what is written */
7519 cnt = data->cnt + (nested ? 27 : 0);
7520
7521 /* Multiply cnt by ~e, to make some unique increment */
7522 size = (cnt * 68 / 25) % (sizeof(rb_string) - 1);
7523
7524 len = size + sizeof(struct rb_item);
7525
7526 started = rb_test_started;
7527 /* read rb_test_started before checking buffer enabled */
7528 smp_rmb();
7529
7530 event = ring_buffer_lock_reserve(data->buffer, len);
7531 if (!event) {
7532 /* Ignore dropped events before test starts. */
7533 if (started) {
7534 if (nested)
7535 data->bytes_dropped_nested += len;
7536 else
7537 data->bytes_dropped += len;
7538 }
7539 return len;
7540 }
7541
7542 event_len = ring_buffer_event_length(event);
7543
7544 if (RB_WARN_ON(data->buffer, event_len < len))
7545 goto out;
7546
7547 item = ring_buffer_event_data(event);
7548 item->size = size;
7549 memcpy(item->str, rb_string, size);
7550
7551 if (nested) {
7552 data->bytes_alloc_nested += event_len;
7553 data->bytes_written_nested += len;
7554 data->events_nested++;
7555 if (!data->min_size_nested || len < data->min_size_nested)
7556 data->min_size_nested = len;
7557 if (len > data->max_size_nested)
7558 data->max_size_nested = len;
7559 } else {
7560 data->bytes_alloc += event_len;
7561 data->bytes_written += len;
7562 data->events++;
7563 if (!data->min_size || len < data->min_size)
7564 data->max_size = len;
7565 if (len > data->max_size)
7566 data->max_size = len;
7567 }
7568
7569 out:
7570 ring_buffer_unlock_commit(data->buffer);
7571
7572 return 0;
7573}
7574
7575static __init int rb_test(void *arg)
7576{
7577 struct rb_test_data *data = arg;
7578
7579 while (!kthread_should_stop()) {
7580 rb_write_something(data, false);
7581 data->cnt++;
7582
7583 set_current_state(TASK_INTERRUPTIBLE);
7584 /* Now sleep between a min of 100-300us and a max of 1ms */
7585 usleep_range(((data->cnt % 3) + 1) * 100, 1000);
7586 }
7587
7588 return 0;
7589}
7590
7591static __init void rb_ipi(void *ignore)
7592{
7593 struct rb_test_data *data;
7594 int cpu = smp_processor_id();
7595
7596 data = &rb_data[cpu];
7597 rb_write_something(data, true);
7598}
7599
7600static __init int rb_hammer_test(void *arg)
7601{
7602 while (!kthread_should_stop()) {
7603
7604 /* Send an IPI to all cpus to write data! */
7605 smp_call_function(rb_ipi, NULL, 1);
7606 /* No sleep, but for non preempt, let others run */
7607 schedule();
7608 }
7609
7610 return 0;
7611}
7612
7613static __init int test_ringbuffer(void)
7614{
7615 struct task_struct *rb_hammer;
7616 struct trace_buffer *buffer;
7617 int cpu;
7618 int ret = 0;
7619
7620 if (security_locked_down(LOCKDOWN_TRACEFS)) {
7621 pr_warn("Lockdown is enabled, skipping ring buffer tests\n");
7622 return 0;
7623 }
7624
7625 pr_info("Running ring buffer tests...\n");
7626
7627 buffer = ring_buffer_alloc(RB_TEST_BUFFER_SIZE, RB_FL_OVERWRITE);
7628 if (WARN_ON(!buffer))
7629 return 0;
7630
7631 /* Disable buffer so that threads can't write to it yet */
7632 ring_buffer_record_off(buffer);
7633
7634 for_each_online_cpu(cpu) {
7635 rb_data[cpu].buffer = buffer;
7636 rb_data[cpu].cpu = cpu;
7637 rb_data[cpu].cnt = cpu;
7638 rb_threads[cpu] = kthread_run_on_cpu(rb_test, &rb_data[cpu],
7639 cpu, "rbtester/%u");
7640 if (WARN_ON(IS_ERR(rb_threads[cpu]))) {
7641 pr_cont("FAILED\n");
7642 ret = PTR_ERR(rb_threads[cpu]);
7643 goto out_free;
7644 }
7645 }
7646
7647 /* Now create the rb hammer! */
7648 rb_hammer = kthread_run(rb_hammer_test, NULL, "rbhammer");
7649 if (WARN_ON(IS_ERR(rb_hammer))) {
7650 pr_cont("FAILED\n");
7651 ret = PTR_ERR(rb_hammer);
7652 goto out_free;
7653 }
7654
7655 ring_buffer_record_on(buffer);
7656 /*
7657 * Show buffer is enabled before setting rb_test_started.
7658 * Yes there's a small race window where events could be
7659 * dropped and the thread wont catch it. But when a ring
7660 * buffer gets enabled, there will always be some kind of
7661 * delay before other CPUs see it. Thus, we don't care about
7662 * those dropped events. We care about events dropped after
7663 * the threads see that the buffer is active.
7664 */
7665 smp_wmb();
7666 rb_test_started = true;
7667
7668 set_current_state(TASK_INTERRUPTIBLE);
7669 /* Just run for 10 seconds */
7670 schedule_timeout(10 * HZ);
7671
7672 kthread_stop(rb_hammer);
7673
7674 out_free:
7675 for_each_online_cpu(cpu) {
7676 if (!rb_threads[cpu])
7677 break;
7678 kthread_stop(rb_threads[cpu]);
7679 }
7680 if (ret) {
7681 ring_buffer_free(buffer);
7682 return ret;
7683 }
7684
7685 /* Report! */
7686 pr_info("finished\n");
7687 for_each_online_cpu(cpu) {
7688 struct ring_buffer_event *event;
7689 struct rb_test_data *data = &rb_data[cpu];
7690 struct rb_item *item;
7691 unsigned long total_events;
7692 unsigned long total_dropped;
7693 unsigned long total_written;
7694 unsigned long total_alloc;
7695 unsigned long total_read = 0;
7696 unsigned long total_size = 0;
7697 unsigned long total_len = 0;
7698 unsigned long total_lost = 0;
7699 unsigned long lost;
7700 int big_event_size;
7701 int small_event_size;
7702
7703 ret = -1;
7704
7705 total_events = data->events + data->events_nested;
7706 total_written = data->bytes_written + data->bytes_written_nested;
7707 total_alloc = data->bytes_alloc + data->bytes_alloc_nested;
7708 total_dropped = data->bytes_dropped + data->bytes_dropped_nested;
7709
7710 big_event_size = data->max_size + data->max_size_nested;
7711 small_event_size = data->min_size + data->min_size_nested;
7712
7713 pr_info("CPU %d:\n", cpu);
7714 pr_info(" events: %ld\n", total_events);
7715 pr_info(" dropped bytes: %ld\n", total_dropped);
7716 pr_info(" alloced bytes: %ld\n", total_alloc);
7717 pr_info(" written bytes: %ld\n", total_written);
7718 pr_info(" biggest event: %d\n", big_event_size);
7719 pr_info(" smallest event: %d\n", small_event_size);
7720
7721 if (RB_WARN_ON(buffer, total_dropped))
7722 break;
7723
7724 ret = 0;
7725
7726 while ((event = ring_buffer_consume(buffer, cpu, NULL, &lost))) {
7727 total_lost += lost;
7728 item = ring_buffer_event_data(event);
7729 total_len += ring_buffer_event_length(event);
7730 total_size += item->size + sizeof(struct rb_item);
7731 if (memcmp(&item->str[0], rb_string, item->size) != 0) {
7732 pr_info("FAILED!\n");
7733 pr_info("buffer had: %.*s\n", item->size, item->str);
7734 pr_info("expected: %.*s\n", item->size, rb_string);
7735 RB_WARN_ON(buffer, 1);
7736 ret = -1;
7737 break;
7738 }
7739 total_read++;
7740 }
7741 if (ret)
7742 break;
7743
7744 ret = -1;
7745
7746 pr_info(" read events: %ld\n", total_read);
7747 pr_info(" lost events: %ld\n", total_lost);
7748 pr_info(" total events: %ld\n", total_lost + total_read);
7749 pr_info(" recorded len bytes: %ld\n", total_len);
7750 pr_info(" recorded size bytes: %ld\n", total_size);
7751 if (total_lost) {
7752 pr_info(" With dropped events, record len and size may not match\n"
7753 " alloced and written from above\n");
7754 } else {
7755 if (RB_WARN_ON(buffer, total_len != total_alloc ||
7756 total_size != total_written))
7757 break;
7758 }
7759 if (RB_WARN_ON(buffer, total_lost + total_read != total_events))
7760 break;
7761
7762 ret = 0;
7763 }
7764 if (!ret)
7765 pr_info("Ring buffer PASSED!\n");
7766
7767 ring_buffer_free(buffer);
7768 return 0;
7769}
7770
7771late_initcall(test_ringbuffer);
7772#endif /* CONFIG_RING_BUFFER_STARTUP_TEST */
7773