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