]>
Commit | Line | Data |
---|---|---|
7a8e76a3 SR |
1 | /* |
2 | * Generic ring buffer | |
3 | * | |
4 | * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com> | |
5 | */ | |
6 | #include <linux/ring_buffer.h> | |
7 | #include <linux/spinlock.h> | |
8 | #include <linux/debugfs.h> | |
9 | #include <linux/uaccess.h> | |
10 | #include <linux/module.h> | |
11 | #include <linux/percpu.h> | |
12 | #include <linux/mutex.h> | |
13 | #include <linux/sched.h> /* used for sched_clock() (for now) */ | |
14 | #include <linux/init.h> | |
15 | #include <linux/hash.h> | |
16 | #include <linux/list.h> | |
17 | #include <linux/fs.h> | |
18 | ||
182e9f5f SR |
19 | #include "trace.h" |
20 | ||
7a8e76a3 SR |
21 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
22 | #define DEBUG_SHIFT 0 | |
23 | ||
24 | /* FIXME!!! */ | |
25 | u64 ring_buffer_time_stamp(int cpu) | |
26 | { | |
27 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
28 | return sched_clock() << DEBUG_SHIFT; | |
29 | } | |
30 | ||
31 | void ring_buffer_normalize_time_stamp(int cpu, u64 *ts) | |
32 | { | |
33 | /* Just stupid testing the normalize function and deltas */ | |
34 | *ts >>= DEBUG_SHIFT; | |
35 | } | |
36 | ||
37 | #define RB_EVNT_HDR_SIZE (sizeof(struct ring_buffer_event)) | |
38 | #define RB_ALIGNMENT_SHIFT 2 | |
39 | #define RB_ALIGNMENT (1 << RB_ALIGNMENT_SHIFT) | |
40 | #define RB_MAX_SMALL_DATA 28 | |
41 | ||
42 | enum { | |
43 | RB_LEN_TIME_EXTEND = 8, | |
44 | RB_LEN_TIME_STAMP = 16, | |
45 | }; | |
46 | ||
47 | /* inline for ring buffer fast paths */ | |
48 | static inline unsigned | |
49 | rb_event_length(struct ring_buffer_event *event) | |
50 | { | |
51 | unsigned length; | |
52 | ||
53 | switch (event->type) { | |
54 | case RINGBUF_TYPE_PADDING: | |
55 | /* undefined */ | |
56 | return -1; | |
57 | ||
58 | case RINGBUF_TYPE_TIME_EXTEND: | |
59 | return RB_LEN_TIME_EXTEND; | |
60 | ||
61 | case RINGBUF_TYPE_TIME_STAMP: | |
62 | return RB_LEN_TIME_STAMP; | |
63 | ||
64 | case RINGBUF_TYPE_DATA: | |
65 | if (event->len) | |
66 | length = event->len << RB_ALIGNMENT_SHIFT; | |
67 | else | |
68 | length = event->array[0]; | |
69 | return length + RB_EVNT_HDR_SIZE; | |
70 | default: | |
71 | BUG(); | |
72 | } | |
73 | /* not hit */ | |
74 | return 0; | |
75 | } | |
76 | ||
77 | /** | |
78 | * ring_buffer_event_length - return the length of the event | |
79 | * @event: the event to get the length of | |
80 | */ | |
81 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
82 | { | |
83 | return rb_event_length(event); | |
84 | } | |
85 | ||
86 | /* inline for ring buffer fast paths */ | |
87 | static inline void * | |
88 | rb_event_data(struct ring_buffer_event *event) | |
89 | { | |
90 | BUG_ON(event->type != RINGBUF_TYPE_DATA); | |
91 | /* If length is in len field, then array[0] has the data */ | |
92 | if (event->len) | |
93 | return (void *)&event->array[0]; | |
94 | /* Otherwise length is in array[0] and array[1] has the data */ | |
95 | return (void *)&event->array[1]; | |
96 | } | |
97 | ||
98 | /** | |
99 | * ring_buffer_event_data - return the data of the event | |
100 | * @event: the event to get the data from | |
101 | */ | |
102 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
103 | { | |
104 | return rb_event_data(event); | |
105 | } | |
106 | ||
107 | #define for_each_buffer_cpu(buffer, cpu) \ | |
108 | for_each_cpu_mask(cpu, buffer->cpumask) | |
109 | ||
110 | #define TS_SHIFT 27 | |
111 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
112 | #define TS_DELTA_TEST (~TS_MASK) | |
113 | ||
114 | /* | |
115 | * This hack stolen from mm/slob.c. | |
116 | * We can store per page timing information in the page frame of the page. | |
117 | * Thanks to Peter Zijlstra for suggesting this idea. | |
118 | */ | |
119 | struct buffer_page { | |
e4c2ce82 | 120 | u64 time_stamp; /* page time stamp */ |
bf41a158 SR |
121 | local_t write; /* index for next write */ |
122 | local_t commit; /* write commited index */ | |
6f807acd | 123 | unsigned read; /* index for next read */ |
e4c2ce82 SR |
124 | struct list_head list; /* list of free pages */ |
125 | void *page; /* Actual data page */ | |
7a8e76a3 SR |
126 | }; |
127 | ||
ed56829c SR |
128 | /* |
129 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
130 | * this issue out. | |
131 | */ | |
132 | static inline void free_buffer_page(struct buffer_page *bpage) | |
133 | { | |
e4c2ce82 | 134 | if (bpage->page) |
6ae2a076 | 135 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 136 | kfree(bpage); |
ed56829c SR |
137 | } |
138 | ||
7a8e76a3 SR |
139 | /* |
140 | * We need to fit the time_stamp delta into 27 bits. | |
141 | */ | |
142 | static inline int test_time_stamp(u64 delta) | |
143 | { | |
144 | if (delta & TS_DELTA_TEST) | |
145 | return 1; | |
146 | return 0; | |
147 | } | |
148 | ||
149 | #define BUF_PAGE_SIZE PAGE_SIZE | |
150 | ||
151 | /* | |
152 | * head_page == tail_page && head == tail then buffer is empty. | |
153 | */ | |
154 | struct ring_buffer_per_cpu { | |
155 | int cpu; | |
156 | struct ring_buffer *buffer; | |
f83c9d0f | 157 | spinlock_t reader_lock; /* serialize readers */ |
3e03fb7f | 158 | raw_spinlock_t lock; |
7a8e76a3 SR |
159 | struct lock_class_key lock_key; |
160 | struct list_head pages; | |
6f807acd SR |
161 | struct buffer_page *head_page; /* read from head */ |
162 | struct buffer_page *tail_page; /* write to tail */ | |
bf41a158 | 163 | struct buffer_page *commit_page; /* commited pages */ |
d769041f | 164 | struct buffer_page *reader_page; |
7a8e76a3 SR |
165 | unsigned long overrun; |
166 | unsigned long entries; | |
167 | u64 write_stamp; | |
168 | u64 read_stamp; | |
169 | atomic_t record_disabled; | |
170 | }; | |
171 | ||
172 | struct ring_buffer { | |
173 | unsigned long size; | |
174 | unsigned pages; | |
175 | unsigned flags; | |
176 | int cpus; | |
177 | cpumask_t cpumask; | |
178 | atomic_t record_disabled; | |
179 | ||
180 | struct mutex mutex; | |
181 | ||
182 | struct ring_buffer_per_cpu **buffers; | |
183 | }; | |
184 | ||
185 | struct ring_buffer_iter { | |
186 | struct ring_buffer_per_cpu *cpu_buffer; | |
187 | unsigned long head; | |
188 | struct buffer_page *head_page; | |
189 | u64 read_stamp; | |
190 | }; | |
191 | ||
f536aafc | 192 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
bf41a158 SR |
193 | #define RB_WARN_ON(buffer, cond) \ |
194 | do { \ | |
195 | if (unlikely(cond)) { \ | |
196 | atomic_inc(&buffer->record_disabled); \ | |
197 | WARN_ON(1); \ | |
198 | } \ | |
199 | } while (0) | |
200 | ||
201 | #define RB_WARN_ON_RET(buffer, cond) \ | |
f536aafc SR |
202 | do { \ |
203 | if (unlikely(cond)) { \ | |
204 | atomic_inc(&buffer->record_disabled); \ | |
205 | WARN_ON(1); \ | |
206 | return; \ | |
207 | } \ | |
208 | } while (0) | |
209 | ||
210 | #define RB_WARN_ON_RET_INT(buffer, cond) \ | |
bf41a158 SR |
211 | do { \ |
212 | if (unlikely(cond)) { \ | |
213 | atomic_inc(&buffer->record_disabled); \ | |
214 | WARN_ON(1); \ | |
215 | return -1; \ | |
216 | } \ | |
217 | } while (0) | |
218 | ||
f536aafc SR |
219 | #define RB_WARN_ON_RET_NULL(buffer, cond) \ |
220 | do { \ | |
221 | if (unlikely(cond)) { \ | |
222 | atomic_inc(&buffer->record_disabled); \ | |
223 | WARN_ON(1); \ | |
224 | return NULL; \ | |
225 | } \ | |
226 | } while (0) | |
227 | ||
bf41a158 SR |
228 | #define RB_WARN_ON_ONCE(buffer, cond) \ |
229 | do { \ | |
230 | static int once; \ | |
231 | if (unlikely(cond) && !once) { \ | |
232 | once++; \ | |
233 | atomic_inc(&buffer->record_disabled); \ | |
234 | WARN_ON(1); \ | |
235 | } \ | |
236 | } while (0) | |
7a8e76a3 | 237 | |
f536aafc SR |
238 | /* buffer must be ring_buffer not per_cpu */ |
239 | #define RB_WARN_ON_UNLOCK(buffer, cond) \ | |
240 | do { \ | |
241 | if (unlikely(cond)) { \ | |
242 | mutex_unlock(&buffer->mutex); \ | |
243 | atomic_inc(&buffer->record_disabled); \ | |
244 | WARN_ON(1); \ | |
245 | return -1; \ | |
246 | } \ | |
247 | } while (0) | |
248 | ||
7a8e76a3 SR |
249 | /** |
250 | * check_pages - integrity check of buffer pages | |
251 | * @cpu_buffer: CPU buffer with pages to test | |
252 | * | |
253 | * As a safty measure we check to make sure the data pages have not | |
254 | * been corrupted. | |
255 | */ | |
256 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
257 | { | |
258 | struct list_head *head = &cpu_buffer->pages; | |
259 | struct buffer_page *page, *tmp; | |
260 | ||
f536aafc SR |
261 | RB_WARN_ON_RET_INT(cpu_buffer, head->next->prev != head); |
262 | RB_WARN_ON_RET_INT(cpu_buffer, head->prev->next != head); | |
7a8e76a3 SR |
263 | |
264 | list_for_each_entry_safe(page, tmp, head, list) { | |
f536aafc | 265 | RB_WARN_ON_RET_INT(cpu_buffer, |
bf41a158 | 266 | page->list.next->prev != &page->list); |
f536aafc | 267 | RB_WARN_ON_RET_INT(cpu_buffer, |
bf41a158 | 268 | page->list.prev->next != &page->list); |
7a8e76a3 SR |
269 | } |
270 | ||
271 | return 0; | |
272 | } | |
273 | ||
7a8e76a3 SR |
274 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
275 | unsigned nr_pages) | |
276 | { | |
277 | struct list_head *head = &cpu_buffer->pages; | |
278 | struct buffer_page *page, *tmp; | |
279 | unsigned long addr; | |
280 | LIST_HEAD(pages); | |
281 | unsigned i; | |
282 | ||
283 | for (i = 0; i < nr_pages; i++) { | |
e4c2ce82 | 284 | page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()), |
aa1e0e3b | 285 | GFP_KERNEL, cpu_to_node(cpu_buffer->cpu)); |
e4c2ce82 SR |
286 | if (!page) |
287 | goto free_pages; | |
288 | list_add(&page->list, &pages); | |
289 | ||
7a8e76a3 SR |
290 | addr = __get_free_page(GFP_KERNEL); |
291 | if (!addr) | |
292 | goto free_pages; | |
e4c2ce82 | 293 | page->page = (void *)addr; |
7a8e76a3 SR |
294 | } |
295 | ||
296 | list_splice(&pages, head); | |
297 | ||
298 | rb_check_pages(cpu_buffer); | |
299 | ||
300 | return 0; | |
301 | ||
302 | free_pages: | |
303 | list_for_each_entry_safe(page, tmp, &pages, list) { | |
304 | list_del_init(&page->list); | |
ed56829c | 305 | free_buffer_page(page); |
7a8e76a3 SR |
306 | } |
307 | return -ENOMEM; | |
308 | } | |
309 | ||
310 | static struct ring_buffer_per_cpu * | |
311 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) | |
312 | { | |
313 | struct ring_buffer_per_cpu *cpu_buffer; | |
e4c2ce82 | 314 | struct buffer_page *page; |
d769041f | 315 | unsigned long addr; |
7a8e76a3 SR |
316 | int ret; |
317 | ||
318 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
319 | GFP_KERNEL, cpu_to_node(cpu)); | |
320 | if (!cpu_buffer) | |
321 | return NULL; | |
322 | ||
323 | cpu_buffer->cpu = cpu; | |
324 | cpu_buffer->buffer = buffer; | |
f83c9d0f | 325 | spin_lock_init(&cpu_buffer->reader_lock); |
3e03fb7f | 326 | cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; |
7a8e76a3 SR |
327 | INIT_LIST_HEAD(&cpu_buffer->pages); |
328 | ||
e4c2ce82 SR |
329 | page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()), |
330 | GFP_KERNEL, cpu_to_node(cpu)); | |
331 | if (!page) | |
332 | goto fail_free_buffer; | |
333 | ||
334 | cpu_buffer->reader_page = page; | |
d769041f SR |
335 | addr = __get_free_page(GFP_KERNEL); |
336 | if (!addr) | |
e4c2ce82 SR |
337 | goto fail_free_reader; |
338 | page->page = (void *)addr; | |
339 | ||
d769041f | 340 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 341 | |
7a8e76a3 SR |
342 | ret = rb_allocate_pages(cpu_buffer, buffer->pages); |
343 | if (ret < 0) | |
d769041f | 344 | goto fail_free_reader; |
7a8e76a3 SR |
345 | |
346 | cpu_buffer->head_page | |
347 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 348 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 SR |
349 | |
350 | return cpu_buffer; | |
351 | ||
d769041f SR |
352 | fail_free_reader: |
353 | free_buffer_page(cpu_buffer->reader_page); | |
354 | ||
7a8e76a3 SR |
355 | fail_free_buffer: |
356 | kfree(cpu_buffer); | |
357 | return NULL; | |
358 | } | |
359 | ||
360 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
361 | { | |
362 | struct list_head *head = &cpu_buffer->pages; | |
363 | struct buffer_page *page, *tmp; | |
364 | ||
d769041f SR |
365 | list_del_init(&cpu_buffer->reader_page->list); |
366 | free_buffer_page(cpu_buffer->reader_page); | |
367 | ||
7a8e76a3 SR |
368 | list_for_each_entry_safe(page, tmp, head, list) { |
369 | list_del_init(&page->list); | |
ed56829c | 370 | free_buffer_page(page); |
7a8e76a3 SR |
371 | } |
372 | kfree(cpu_buffer); | |
373 | } | |
374 | ||
a7b13743 SR |
375 | /* |
376 | * Causes compile errors if the struct buffer_page gets bigger | |
377 | * than the struct page. | |
378 | */ | |
379 | extern int ring_buffer_page_too_big(void); | |
380 | ||
7a8e76a3 SR |
381 | /** |
382 | * ring_buffer_alloc - allocate a new ring_buffer | |
383 | * @size: the size in bytes that is needed. | |
384 | * @flags: attributes to set for the ring buffer. | |
385 | * | |
386 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
387 | * flag. This flag means that the buffer will overwrite old data | |
388 | * when the buffer wraps. If this flag is not set, the buffer will | |
389 | * drop data when the tail hits the head. | |
390 | */ | |
391 | struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags) | |
392 | { | |
393 | struct ring_buffer *buffer; | |
394 | int bsize; | |
395 | int cpu; | |
396 | ||
a7b13743 SR |
397 | /* Paranoid! Optimizes out when all is well */ |
398 | if (sizeof(struct buffer_page) > sizeof(struct page)) | |
399 | ring_buffer_page_too_big(); | |
400 | ||
401 | ||
7a8e76a3 SR |
402 | /* keep it in its own cache line */ |
403 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
404 | GFP_KERNEL); | |
405 | if (!buffer) | |
406 | return NULL; | |
407 | ||
408 | buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
409 | buffer->flags = flags; | |
410 | ||
411 | /* need at least two pages */ | |
412 | if (buffer->pages == 1) | |
413 | buffer->pages++; | |
414 | ||
415 | buffer->cpumask = cpu_possible_map; | |
416 | buffer->cpus = nr_cpu_ids; | |
417 | ||
418 | bsize = sizeof(void *) * nr_cpu_ids; | |
419 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
420 | GFP_KERNEL); | |
421 | if (!buffer->buffers) | |
422 | goto fail_free_buffer; | |
423 | ||
424 | for_each_buffer_cpu(buffer, cpu) { | |
425 | buffer->buffers[cpu] = | |
426 | rb_allocate_cpu_buffer(buffer, cpu); | |
427 | if (!buffer->buffers[cpu]) | |
428 | goto fail_free_buffers; | |
429 | } | |
430 | ||
431 | mutex_init(&buffer->mutex); | |
432 | ||
433 | return buffer; | |
434 | ||
435 | fail_free_buffers: | |
436 | for_each_buffer_cpu(buffer, cpu) { | |
437 | if (buffer->buffers[cpu]) | |
438 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
439 | } | |
440 | kfree(buffer->buffers); | |
441 | ||
442 | fail_free_buffer: | |
443 | kfree(buffer); | |
444 | return NULL; | |
445 | } | |
446 | ||
447 | /** | |
448 | * ring_buffer_free - free a ring buffer. | |
449 | * @buffer: the buffer to free. | |
450 | */ | |
451 | void | |
452 | ring_buffer_free(struct ring_buffer *buffer) | |
453 | { | |
454 | int cpu; | |
455 | ||
456 | for_each_buffer_cpu(buffer, cpu) | |
457 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
458 | ||
459 | kfree(buffer); | |
460 | } | |
461 | ||
462 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); | |
463 | ||
464 | static void | |
465 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages) | |
466 | { | |
467 | struct buffer_page *page; | |
468 | struct list_head *p; | |
469 | unsigned i; | |
470 | ||
471 | atomic_inc(&cpu_buffer->record_disabled); | |
472 | synchronize_sched(); | |
473 | ||
474 | for (i = 0; i < nr_pages; i++) { | |
f536aafc | 475 | RB_WARN_ON_RET(cpu_buffer, list_empty(&cpu_buffer->pages)); |
7a8e76a3 SR |
476 | p = cpu_buffer->pages.next; |
477 | page = list_entry(p, struct buffer_page, list); | |
478 | list_del_init(&page->list); | |
ed56829c | 479 | free_buffer_page(page); |
7a8e76a3 | 480 | } |
f536aafc | 481 | RB_WARN_ON_RET(cpu_buffer, list_empty(&cpu_buffer->pages)); |
7a8e76a3 SR |
482 | |
483 | rb_reset_cpu(cpu_buffer); | |
484 | ||
485 | rb_check_pages(cpu_buffer); | |
486 | ||
487 | atomic_dec(&cpu_buffer->record_disabled); | |
488 | ||
489 | } | |
490 | ||
491 | static void | |
492 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
493 | struct list_head *pages, unsigned nr_pages) | |
494 | { | |
495 | struct buffer_page *page; | |
496 | struct list_head *p; | |
497 | unsigned i; | |
498 | ||
499 | atomic_inc(&cpu_buffer->record_disabled); | |
500 | synchronize_sched(); | |
501 | ||
502 | for (i = 0; i < nr_pages; i++) { | |
f536aafc | 503 | RB_WARN_ON_RET(cpu_buffer, list_empty(pages)); |
7a8e76a3 SR |
504 | p = pages->next; |
505 | page = list_entry(p, struct buffer_page, list); | |
506 | list_del_init(&page->list); | |
507 | list_add_tail(&page->list, &cpu_buffer->pages); | |
508 | } | |
509 | rb_reset_cpu(cpu_buffer); | |
510 | ||
511 | rb_check_pages(cpu_buffer); | |
512 | ||
513 | atomic_dec(&cpu_buffer->record_disabled); | |
514 | } | |
515 | ||
516 | /** | |
517 | * ring_buffer_resize - resize the ring buffer | |
518 | * @buffer: the buffer to resize. | |
519 | * @size: the new size. | |
520 | * | |
521 | * The tracer is responsible for making sure that the buffer is | |
522 | * not being used while changing the size. | |
523 | * Note: We may be able to change the above requirement by using | |
524 | * RCU synchronizations. | |
525 | * | |
526 | * Minimum size is 2 * BUF_PAGE_SIZE. | |
527 | * | |
528 | * Returns -1 on failure. | |
529 | */ | |
530 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) | |
531 | { | |
532 | struct ring_buffer_per_cpu *cpu_buffer; | |
533 | unsigned nr_pages, rm_pages, new_pages; | |
534 | struct buffer_page *page, *tmp; | |
535 | unsigned long buffer_size; | |
536 | unsigned long addr; | |
537 | LIST_HEAD(pages); | |
538 | int i, cpu; | |
539 | ||
540 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
541 | size *= BUF_PAGE_SIZE; | |
542 | buffer_size = buffer->pages * BUF_PAGE_SIZE; | |
543 | ||
544 | /* we need a minimum of two pages */ | |
545 | if (size < BUF_PAGE_SIZE * 2) | |
546 | size = BUF_PAGE_SIZE * 2; | |
547 | ||
548 | if (size == buffer_size) | |
549 | return size; | |
550 | ||
551 | mutex_lock(&buffer->mutex); | |
552 | ||
553 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
554 | ||
555 | if (size < buffer_size) { | |
556 | ||
557 | /* easy case, just free pages */ | |
f536aafc | 558 | RB_WARN_ON_UNLOCK(buffer, nr_pages >= buffer->pages); |
7a8e76a3 SR |
559 | |
560 | rm_pages = buffer->pages - nr_pages; | |
561 | ||
562 | for_each_buffer_cpu(buffer, cpu) { | |
563 | cpu_buffer = buffer->buffers[cpu]; | |
564 | rb_remove_pages(cpu_buffer, rm_pages); | |
565 | } | |
566 | goto out; | |
567 | } | |
568 | ||
569 | /* | |
570 | * This is a bit more difficult. We only want to add pages | |
571 | * when we can allocate enough for all CPUs. We do this | |
572 | * by allocating all the pages and storing them on a local | |
573 | * link list. If we succeed in our allocation, then we | |
574 | * add these pages to the cpu_buffers. Otherwise we just free | |
575 | * them all and return -ENOMEM; | |
576 | */ | |
f536aafc SR |
577 | RB_WARN_ON_UNLOCK(buffer, nr_pages <= buffer->pages); |
578 | ||
7a8e76a3 SR |
579 | new_pages = nr_pages - buffer->pages; |
580 | ||
581 | for_each_buffer_cpu(buffer, cpu) { | |
582 | for (i = 0; i < new_pages; i++) { | |
e4c2ce82 SR |
583 | page = kzalloc_node(ALIGN(sizeof(*page), |
584 | cache_line_size()), | |
585 | GFP_KERNEL, cpu_to_node(cpu)); | |
586 | if (!page) | |
587 | goto free_pages; | |
588 | list_add(&page->list, &pages); | |
7a8e76a3 SR |
589 | addr = __get_free_page(GFP_KERNEL); |
590 | if (!addr) | |
591 | goto free_pages; | |
e4c2ce82 | 592 | page->page = (void *)addr; |
7a8e76a3 SR |
593 | } |
594 | } | |
595 | ||
596 | for_each_buffer_cpu(buffer, cpu) { | |
597 | cpu_buffer = buffer->buffers[cpu]; | |
598 | rb_insert_pages(cpu_buffer, &pages, new_pages); | |
599 | } | |
600 | ||
f536aafc | 601 | RB_WARN_ON_UNLOCK(buffer, !list_empty(&pages)); |
7a8e76a3 SR |
602 | |
603 | out: | |
604 | buffer->pages = nr_pages; | |
605 | mutex_unlock(&buffer->mutex); | |
606 | ||
607 | return size; | |
608 | ||
609 | free_pages: | |
610 | list_for_each_entry_safe(page, tmp, &pages, list) { | |
611 | list_del_init(&page->list); | |
ed56829c | 612 | free_buffer_page(page); |
7a8e76a3 SR |
613 | } |
614 | return -ENOMEM; | |
615 | } | |
616 | ||
7a8e76a3 SR |
617 | static inline int rb_null_event(struct ring_buffer_event *event) |
618 | { | |
619 | return event->type == RINGBUF_TYPE_PADDING; | |
620 | } | |
621 | ||
6f807acd | 622 | static inline void *__rb_page_index(struct buffer_page *page, unsigned index) |
7a8e76a3 | 623 | { |
e4c2ce82 | 624 | return page->page + index; |
7a8e76a3 SR |
625 | } |
626 | ||
627 | static inline struct ring_buffer_event * | |
d769041f | 628 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 629 | { |
6f807acd SR |
630 | return __rb_page_index(cpu_buffer->reader_page, |
631 | cpu_buffer->reader_page->read); | |
632 | } | |
633 | ||
634 | static inline struct ring_buffer_event * | |
635 | rb_head_event(struct ring_buffer_per_cpu *cpu_buffer) | |
636 | { | |
637 | return __rb_page_index(cpu_buffer->head_page, | |
638 | cpu_buffer->head_page->read); | |
7a8e76a3 SR |
639 | } |
640 | ||
641 | static inline struct ring_buffer_event * | |
642 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
643 | { | |
6f807acd | 644 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
645 | } |
646 | ||
bf41a158 SR |
647 | static inline unsigned rb_page_write(struct buffer_page *bpage) |
648 | { | |
649 | return local_read(&bpage->write); | |
650 | } | |
651 | ||
652 | static inline unsigned rb_page_commit(struct buffer_page *bpage) | |
653 | { | |
654 | return local_read(&bpage->commit); | |
655 | } | |
656 | ||
657 | /* Size is determined by what has been commited */ | |
658 | static inline unsigned rb_page_size(struct buffer_page *bpage) | |
659 | { | |
660 | return rb_page_commit(bpage); | |
661 | } | |
662 | ||
663 | static inline unsigned | |
664 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
665 | { | |
666 | return rb_page_commit(cpu_buffer->commit_page); | |
667 | } | |
668 | ||
669 | static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer) | |
670 | { | |
671 | return rb_page_commit(cpu_buffer->head_page); | |
672 | } | |
673 | ||
7a8e76a3 SR |
674 | /* |
675 | * When the tail hits the head and the buffer is in overwrite mode, | |
676 | * the head jumps to the next page and all content on the previous | |
677 | * page is discarded. But before doing so, we update the overrun | |
678 | * variable of the buffer. | |
679 | */ | |
680 | static void rb_update_overflow(struct ring_buffer_per_cpu *cpu_buffer) | |
681 | { | |
682 | struct ring_buffer_event *event; | |
683 | unsigned long head; | |
684 | ||
685 | for (head = 0; head < rb_head_size(cpu_buffer); | |
686 | head += rb_event_length(event)) { | |
687 | ||
6f807acd | 688 | event = __rb_page_index(cpu_buffer->head_page, head); |
f536aafc | 689 | RB_WARN_ON_RET(cpu_buffer, rb_null_event(event)); |
7a8e76a3 SR |
690 | /* Only count data entries */ |
691 | if (event->type != RINGBUF_TYPE_DATA) | |
692 | continue; | |
693 | cpu_buffer->overrun++; | |
694 | cpu_buffer->entries--; | |
695 | } | |
696 | } | |
697 | ||
698 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
699 | struct buffer_page **page) | |
700 | { | |
701 | struct list_head *p = (*page)->list.next; | |
702 | ||
703 | if (p == &cpu_buffer->pages) | |
704 | p = p->next; | |
705 | ||
706 | *page = list_entry(p, struct buffer_page, list); | |
707 | } | |
708 | ||
bf41a158 SR |
709 | static inline unsigned |
710 | rb_event_index(struct ring_buffer_event *event) | |
711 | { | |
712 | unsigned long addr = (unsigned long)event; | |
713 | ||
714 | return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE); | |
715 | } | |
716 | ||
717 | static inline int | |
718 | rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
719 | struct ring_buffer_event *event) | |
720 | { | |
721 | unsigned long addr = (unsigned long)event; | |
722 | unsigned long index; | |
723 | ||
724 | index = rb_event_index(event); | |
725 | addr &= PAGE_MASK; | |
726 | ||
727 | return cpu_buffer->commit_page->page == (void *)addr && | |
728 | rb_commit_index(cpu_buffer) == index; | |
729 | } | |
730 | ||
7a8e76a3 | 731 | static inline void |
bf41a158 SR |
732 | rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer, |
733 | struct ring_buffer_event *event) | |
7a8e76a3 | 734 | { |
bf41a158 SR |
735 | unsigned long addr = (unsigned long)event; |
736 | unsigned long index; | |
737 | ||
738 | index = rb_event_index(event); | |
739 | addr &= PAGE_MASK; | |
740 | ||
741 | while (cpu_buffer->commit_page->page != (void *)addr) { | |
742 | RB_WARN_ON(cpu_buffer, | |
743 | cpu_buffer->commit_page == cpu_buffer->tail_page); | |
744 | cpu_buffer->commit_page->commit = | |
745 | cpu_buffer->commit_page->write; | |
746 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
747 | cpu_buffer->write_stamp = cpu_buffer->commit_page->time_stamp; | |
748 | } | |
749 | ||
750 | /* Now set the commit to the event's index */ | |
751 | local_set(&cpu_buffer->commit_page->commit, index); | |
7a8e76a3 SR |
752 | } |
753 | ||
bf41a158 SR |
754 | static inline void |
755 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 756 | { |
bf41a158 SR |
757 | /* |
758 | * We only race with interrupts and NMIs on this CPU. | |
759 | * If we own the commit event, then we can commit | |
760 | * all others that interrupted us, since the interruptions | |
761 | * are in stack format (they finish before they come | |
762 | * back to us). This allows us to do a simple loop to | |
763 | * assign the commit to the tail. | |
764 | */ | |
765 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { | |
766 | cpu_buffer->commit_page->commit = | |
767 | cpu_buffer->commit_page->write; | |
768 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
769 | cpu_buffer->write_stamp = cpu_buffer->commit_page->time_stamp; | |
770 | /* add barrier to keep gcc from optimizing too much */ | |
771 | barrier(); | |
772 | } | |
773 | while (rb_commit_index(cpu_buffer) != | |
774 | rb_page_write(cpu_buffer->commit_page)) { | |
775 | cpu_buffer->commit_page->commit = | |
776 | cpu_buffer->commit_page->write; | |
777 | barrier(); | |
778 | } | |
7a8e76a3 SR |
779 | } |
780 | ||
d769041f | 781 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 782 | { |
d769041f | 783 | cpu_buffer->read_stamp = cpu_buffer->reader_page->time_stamp; |
6f807acd | 784 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
785 | } |
786 | ||
787 | static inline void rb_inc_iter(struct ring_buffer_iter *iter) | |
788 | { | |
789 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
790 | ||
791 | /* | |
792 | * The iterator could be on the reader page (it starts there). | |
793 | * But the head could have moved, since the reader was | |
794 | * found. Check for this case and assign the iterator | |
795 | * to the head page instead of next. | |
796 | */ | |
797 | if (iter->head_page == cpu_buffer->reader_page) | |
798 | iter->head_page = cpu_buffer->head_page; | |
799 | else | |
800 | rb_inc_page(cpu_buffer, &iter->head_page); | |
801 | ||
7a8e76a3 SR |
802 | iter->read_stamp = iter->head_page->time_stamp; |
803 | iter->head = 0; | |
804 | } | |
805 | ||
806 | /** | |
807 | * ring_buffer_update_event - update event type and data | |
808 | * @event: the even to update | |
809 | * @type: the type of event | |
810 | * @length: the size of the event field in the ring buffer | |
811 | * | |
812 | * Update the type and data fields of the event. The length | |
813 | * is the actual size that is written to the ring buffer, | |
814 | * and with this, we can determine what to place into the | |
815 | * data field. | |
816 | */ | |
817 | static inline void | |
818 | rb_update_event(struct ring_buffer_event *event, | |
819 | unsigned type, unsigned length) | |
820 | { | |
821 | event->type = type; | |
822 | ||
823 | switch (type) { | |
824 | ||
825 | case RINGBUF_TYPE_PADDING: | |
826 | break; | |
827 | ||
828 | case RINGBUF_TYPE_TIME_EXTEND: | |
829 | event->len = | |
830 | (RB_LEN_TIME_EXTEND + (RB_ALIGNMENT-1)) | |
831 | >> RB_ALIGNMENT_SHIFT; | |
832 | break; | |
833 | ||
834 | case RINGBUF_TYPE_TIME_STAMP: | |
835 | event->len = | |
836 | (RB_LEN_TIME_STAMP + (RB_ALIGNMENT-1)) | |
837 | >> RB_ALIGNMENT_SHIFT; | |
838 | break; | |
839 | ||
840 | case RINGBUF_TYPE_DATA: | |
841 | length -= RB_EVNT_HDR_SIZE; | |
842 | if (length > RB_MAX_SMALL_DATA) { | |
843 | event->len = 0; | |
844 | event->array[0] = length; | |
845 | } else | |
846 | event->len = | |
847 | (length + (RB_ALIGNMENT-1)) | |
848 | >> RB_ALIGNMENT_SHIFT; | |
849 | break; | |
850 | default: | |
851 | BUG(); | |
852 | } | |
853 | } | |
854 | ||
855 | static inline unsigned rb_calculate_event_length(unsigned length) | |
856 | { | |
857 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
858 | ||
859 | /* zero length can cause confusions */ | |
860 | if (!length) | |
861 | length = 1; | |
862 | ||
863 | if (length > RB_MAX_SMALL_DATA) | |
864 | length += sizeof(event.array[0]); | |
865 | ||
866 | length += RB_EVNT_HDR_SIZE; | |
867 | length = ALIGN(length, RB_ALIGNMENT); | |
868 | ||
869 | return length; | |
870 | } | |
871 | ||
872 | static struct ring_buffer_event * | |
873 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
874 | unsigned type, unsigned long length, u64 *ts) | |
875 | { | |
d769041f | 876 | struct buffer_page *tail_page, *head_page, *reader_page; |
bf41a158 | 877 | unsigned long tail, write; |
7a8e76a3 SR |
878 | struct ring_buffer *buffer = cpu_buffer->buffer; |
879 | struct ring_buffer_event *event; | |
bf41a158 | 880 | unsigned long flags; |
7a8e76a3 SR |
881 | |
882 | tail_page = cpu_buffer->tail_page; | |
bf41a158 SR |
883 | write = local_add_return(length, &tail_page->write); |
884 | tail = write - length; | |
7a8e76a3 | 885 | |
bf41a158 SR |
886 | /* See if we shot pass the end of this buffer page */ |
887 | if (write > BUF_PAGE_SIZE) { | |
7a8e76a3 SR |
888 | struct buffer_page *next_page = tail_page; |
889 | ||
3e03fb7f SR |
890 | local_irq_save(flags); |
891 | __raw_spin_lock(&cpu_buffer->lock); | |
bf41a158 | 892 | |
7a8e76a3 SR |
893 | rb_inc_page(cpu_buffer, &next_page); |
894 | ||
d769041f SR |
895 | head_page = cpu_buffer->head_page; |
896 | reader_page = cpu_buffer->reader_page; | |
897 | ||
898 | /* we grabbed the lock before incrementing */ | |
bf41a158 SR |
899 | RB_WARN_ON(cpu_buffer, next_page == reader_page); |
900 | ||
901 | /* | |
902 | * If for some reason, we had an interrupt storm that made | |
903 | * it all the way around the buffer, bail, and warn | |
904 | * about it. | |
905 | */ | |
906 | if (unlikely(next_page == cpu_buffer->commit_page)) { | |
907 | WARN_ON_ONCE(1); | |
908 | goto out_unlock; | |
909 | } | |
d769041f | 910 | |
7a8e76a3 | 911 | if (next_page == head_page) { |
d769041f | 912 | if (!(buffer->flags & RB_FL_OVERWRITE)) { |
bf41a158 SR |
913 | /* reset write */ |
914 | if (tail <= BUF_PAGE_SIZE) | |
915 | local_set(&tail_page->write, tail); | |
916 | goto out_unlock; | |
d769041f | 917 | } |
7a8e76a3 | 918 | |
bf41a158 SR |
919 | /* tail_page has not moved yet? */ |
920 | if (tail_page == cpu_buffer->tail_page) { | |
921 | /* count overflows */ | |
922 | rb_update_overflow(cpu_buffer); | |
923 | ||
924 | rb_inc_page(cpu_buffer, &head_page); | |
925 | cpu_buffer->head_page = head_page; | |
926 | cpu_buffer->head_page->read = 0; | |
927 | } | |
928 | } | |
7a8e76a3 | 929 | |
bf41a158 SR |
930 | /* |
931 | * If the tail page is still the same as what we think | |
932 | * it is, then it is up to us to update the tail | |
933 | * pointer. | |
934 | */ | |
935 | if (tail_page == cpu_buffer->tail_page) { | |
936 | local_set(&next_page->write, 0); | |
937 | local_set(&next_page->commit, 0); | |
938 | cpu_buffer->tail_page = next_page; | |
939 | ||
940 | /* reread the time stamp */ | |
941 | *ts = ring_buffer_time_stamp(cpu_buffer->cpu); | |
942 | cpu_buffer->tail_page->time_stamp = *ts; | |
7a8e76a3 SR |
943 | } |
944 | ||
bf41a158 SR |
945 | /* |
946 | * The actual tail page has moved forward. | |
947 | */ | |
948 | if (tail < BUF_PAGE_SIZE) { | |
949 | /* Mark the rest of the page with padding */ | |
6f807acd | 950 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
951 | event->type = RINGBUF_TYPE_PADDING; |
952 | } | |
953 | ||
bf41a158 SR |
954 | if (tail <= BUF_PAGE_SIZE) |
955 | /* Set the write back to the previous setting */ | |
956 | local_set(&tail_page->write, tail); | |
957 | ||
958 | /* | |
959 | * If this was a commit entry that failed, | |
960 | * increment that too | |
961 | */ | |
962 | if (tail_page == cpu_buffer->commit_page && | |
963 | tail == rb_commit_index(cpu_buffer)) { | |
964 | rb_set_commit_to_write(cpu_buffer); | |
965 | } | |
966 | ||
3e03fb7f SR |
967 | __raw_spin_unlock(&cpu_buffer->lock); |
968 | local_irq_restore(flags); | |
bf41a158 SR |
969 | |
970 | /* fail and let the caller try again */ | |
971 | return ERR_PTR(-EAGAIN); | |
7a8e76a3 SR |
972 | } |
973 | ||
bf41a158 SR |
974 | /* We reserved something on the buffer */ |
975 | ||
f536aafc | 976 | RB_WARN_ON_RET_NULL(cpu_buffer, write > BUF_PAGE_SIZE); |
7a8e76a3 | 977 | |
6f807acd | 978 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
979 | rb_update_event(event, type, length); |
980 | ||
bf41a158 SR |
981 | /* |
982 | * If this is a commit and the tail is zero, then update | |
983 | * this page's time stamp. | |
984 | */ | |
985 | if (!tail && rb_is_commit(cpu_buffer, event)) | |
986 | cpu_buffer->commit_page->time_stamp = *ts; | |
987 | ||
7a8e76a3 | 988 | return event; |
bf41a158 SR |
989 | |
990 | out_unlock: | |
3e03fb7f SR |
991 | __raw_spin_unlock(&cpu_buffer->lock); |
992 | local_irq_restore(flags); | |
bf41a158 | 993 | return NULL; |
7a8e76a3 SR |
994 | } |
995 | ||
996 | static int | |
997 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
998 | u64 *ts, u64 *delta) | |
999 | { | |
1000 | struct ring_buffer_event *event; | |
1001 | static int once; | |
bf41a158 | 1002 | int ret; |
7a8e76a3 SR |
1003 | |
1004 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1005 | printk(KERN_WARNING "Delta way too big! %llu" | |
1006 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1007 | (unsigned long long)*delta, |
1008 | (unsigned long long)*ts, | |
1009 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1010 | WARN_ON(1); |
1011 | } | |
1012 | ||
1013 | /* | |
1014 | * The delta is too big, we to add a | |
1015 | * new timestamp. | |
1016 | */ | |
1017 | event = __rb_reserve_next(cpu_buffer, | |
1018 | RINGBUF_TYPE_TIME_EXTEND, | |
1019 | RB_LEN_TIME_EXTEND, | |
1020 | ts); | |
1021 | if (!event) | |
bf41a158 | 1022 | return -EBUSY; |
7a8e76a3 | 1023 | |
bf41a158 SR |
1024 | if (PTR_ERR(event) == -EAGAIN) |
1025 | return -EAGAIN; | |
1026 | ||
1027 | /* Only a commited time event can update the write stamp */ | |
1028 | if (rb_is_commit(cpu_buffer, event)) { | |
1029 | /* | |
1030 | * If this is the first on the page, then we need to | |
1031 | * update the page itself, and just put in a zero. | |
1032 | */ | |
1033 | if (rb_event_index(event)) { | |
1034 | event->time_delta = *delta & TS_MASK; | |
1035 | event->array[0] = *delta >> TS_SHIFT; | |
1036 | } else { | |
1037 | cpu_buffer->commit_page->time_stamp = *ts; | |
1038 | event->time_delta = 0; | |
1039 | event->array[0] = 0; | |
1040 | } | |
7a8e76a3 | 1041 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
1042 | /* let the caller know this was the commit */ |
1043 | ret = 1; | |
1044 | } else { | |
1045 | /* Darn, this is just wasted space */ | |
1046 | event->time_delta = 0; | |
1047 | event->array[0] = 0; | |
1048 | ret = 0; | |
7a8e76a3 SR |
1049 | } |
1050 | ||
bf41a158 SR |
1051 | *delta = 0; |
1052 | ||
1053 | return ret; | |
7a8e76a3 SR |
1054 | } |
1055 | ||
1056 | static struct ring_buffer_event * | |
1057 | rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, | |
1058 | unsigned type, unsigned long length) | |
1059 | { | |
1060 | struct ring_buffer_event *event; | |
1061 | u64 ts, delta; | |
bf41a158 | 1062 | int commit = 0; |
818e3dd3 | 1063 | int nr_loops = 0; |
7a8e76a3 | 1064 | |
bf41a158 | 1065 | again: |
818e3dd3 SR |
1066 | /* |
1067 | * We allow for interrupts to reenter here and do a trace. | |
1068 | * If one does, it will cause this original code to loop | |
1069 | * back here. Even with heavy interrupts happening, this | |
1070 | * should only happen a few times in a row. If this happens | |
1071 | * 1000 times in a row, there must be either an interrupt | |
1072 | * storm or we have something buggy. | |
1073 | * Bail! | |
1074 | */ | |
1075 | if (unlikely(++nr_loops > 1000)) { | |
1076 | RB_WARN_ON(cpu_buffer, 1); | |
1077 | return NULL; | |
1078 | } | |
1079 | ||
7a8e76a3 SR |
1080 | ts = ring_buffer_time_stamp(cpu_buffer->cpu); |
1081 | ||
bf41a158 SR |
1082 | /* |
1083 | * Only the first commit can update the timestamp. | |
1084 | * Yes there is a race here. If an interrupt comes in | |
1085 | * just after the conditional and it traces too, then it | |
1086 | * will also check the deltas. More than one timestamp may | |
1087 | * also be made. But only the entry that did the actual | |
1088 | * commit will be something other than zero. | |
1089 | */ | |
1090 | if (cpu_buffer->tail_page == cpu_buffer->commit_page && | |
1091 | rb_page_write(cpu_buffer->tail_page) == | |
1092 | rb_commit_index(cpu_buffer)) { | |
1093 | ||
7a8e76a3 SR |
1094 | delta = ts - cpu_buffer->write_stamp; |
1095 | ||
bf41a158 SR |
1096 | /* make sure this delta is calculated here */ |
1097 | barrier(); | |
1098 | ||
1099 | /* Did the write stamp get updated already? */ | |
1100 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
4143c5cb | 1101 | delta = 0; |
bf41a158 | 1102 | |
7a8e76a3 | 1103 | if (test_time_stamp(delta)) { |
7a8e76a3 | 1104 | |
bf41a158 SR |
1105 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
1106 | ||
1107 | if (commit == -EBUSY) | |
7a8e76a3 | 1108 | return NULL; |
bf41a158 SR |
1109 | |
1110 | if (commit == -EAGAIN) | |
1111 | goto again; | |
1112 | ||
1113 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 1114 | } |
bf41a158 SR |
1115 | } else |
1116 | /* Non commits have zero deltas */ | |
7a8e76a3 | 1117 | delta = 0; |
7a8e76a3 SR |
1118 | |
1119 | event = __rb_reserve_next(cpu_buffer, type, length, &ts); | |
bf41a158 SR |
1120 | if (PTR_ERR(event) == -EAGAIN) |
1121 | goto again; | |
1122 | ||
1123 | if (!event) { | |
1124 | if (unlikely(commit)) | |
1125 | /* | |
1126 | * Ouch! We needed a timestamp and it was commited. But | |
1127 | * we didn't get our event reserved. | |
1128 | */ | |
1129 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 | 1130 | return NULL; |
bf41a158 | 1131 | } |
7a8e76a3 | 1132 | |
bf41a158 SR |
1133 | /* |
1134 | * If the timestamp was commited, make the commit our entry | |
1135 | * now so that we will update it when needed. | |
1136 | */ | |
1137 | if (commit) | |
1138 | rb_set_commit_event(cpu_buffer, event); | |
1139 | else if (!rb_is_commit(cpu_buffer, event)) | |
7a8e76a3 SR |
1140 | delta = 0; |
1141 | ||
1142 | event->time_delta = delta; | |
1143 | ||
1144 | return event; | |
1145 | } | |
1146 | ||
bf41a158 SR |
1147 | static DEFINE_PER_CPU(int, rb_need_resched); |
1148 | ||
7a8e76a3 SR |
1149 | /** |
1150 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
1151 | * @buffer: the ring buffer to reserve from | |
1152 | * @length: the length of the data to reserve (excluding event header) | |
1153 | * @flags: a pointer to save the interrupt flags | |
1154 | * | |
1155 | * Returns a reseverd event on the ring buffer to copy directly to. | |
1156 | * The user of this interface will need to get the body to write into | |
1157 | * and can use the ring_buffer_event_data() interface. | |
1158 | * | |
1159 | * The length is the length of the data needed, not the event length | |
1160 | * which also includes the event header. | |
1161 | * | |
1162 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
1163 | * If NULL is returned, then nothing has been allocated or locked. | |
1164 | */ | |
1165 | struct ring_buffer_event * | |
1166 | ring_buffer_lock_reserve(struct ring_buffer *buffer, | |
1167 | unsigned long length, | |
1168 | unsigned long *flags) | |
1169 | { | |
1170 | struct ring_buffer_per_cpu *cpu_buffer; | |
1171 | struct ring_buffer_event *event; | |
bf41a158 | 1172 | int cpu, resched; |
7a8e76a3 SR |
1173 | |
1174 | if (atomic_read(&buffer->record_disabled)) | |
1175 | return NULL; | |
1176 | ||
bf41a158 | 1177 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 1178 | resched = ftrace_preempt_disable(); |
bf41a158 | 1179 | |
7a8e76a3 SR |
1180 | cpu = raw_smp_processor_id(); |
1181 | ||
1182 | if (!cpu_isset(cpu, buffer->cpumask)) | |
d769041f | 1183 | goto out; |
7a8e76a3 SR |
1184 | |
1185 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1186 | |
1187 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 1188 | goto out; |
7a8e76a3 SR |
1189 | |
1190 | length = rb_calculate_event_length(length); | |
1191 | if (length > BUF_PAGE_SIZE) | |
bf41a158 | 1192 | goto out; |
7a8e76a3 SR |
1193 | |
1194 | event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length); | |
1195 | if (!event) | |
d769041f | 1196 | goto out; |
7a8e76a3 | 1197 | |
bf41a158 SR |
1198 | /* |
1199 | * Need to store resched state on this cpu. | |
1200 | * Only the first needs to. | |
1201 | */ | |
1202 | ||
1203 | if (preempt_count() == 1) | |
1204 | per_cpu(rb_need_resched, cpu) = resched; | |
1205 | ||
7a8e76a3 SR |
1206 | return event; |
1207 | ||
d769041f | 1208 | out: |
182e9f5f | 1209 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1210 | return NULL; |
1211 | } | |
1212 | ||
1213 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
1214 | struct ring_buffer_event *event) | |
1215 | { | |
7a8e76a3 | 1216 | cpu_buffer->entries++; |
bf41a158 SR |
1217 | |
1218 | /* Only process further if we own the commit */ | |
1219 | if (!rb_is_commit(cpu_buffer, event)) | |
1220 | return; | |
1221 | ||
1222 | cpu_buffer->write_stamp += event->time_delta; | |
1223 | ||
1224 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 SR |
1225 | } |
1226 | ||
1227 | /** | |
1228 | * ring_buffer_unlock_commit - commit a reserved | |
1229 | * @buffer: The buffer to commit to | |
1230 | * @event: The event pointer to commit. | |
1231 | * @flags: the interrupt flags received from ring_buffer_lock_reserve. | |
1232 | * | |
1233 | * This commits the data to the ring buffer, and releases any locks held. | |
1234 | * | |
1235 | * Must be paired with ring_buffer_lock_reserve. | |
1236 | */ | |
1237 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
1238 | struct ring_buffer_event *event, | |
1239 | unsigned long flags) | |
1240 | { | |
1241 | struct ring_buffer_per_cpu *cpu_buffer; | |
1242 | int cpu = raw_smp_processor_id(); | |
1243 | ||
1244 | cpu_buffer = buffer->buffers[cpu]; | |
1245 | ||
7a8e76a3 SR |
1246 | rb_commit(cpu_buffer, event); |
1247 | ||
bf41a158 SR |
1248 | /* |
1249 | * Only the last preempt count needs to restore preemption. | |
1250 | */ | |
182e9f5f SR |
1251 | if (preempt_count() == 1) |
1252 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1253 | else | |
bf41a158 | 1254 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
1255 | |
1256 | return 0; | |
1257 | } | |
1258 | ||
1259 | /** | |
1260 | * ring_buffer_write - write data to the buffer without reserving | |
1261 | * @buffer: The ring buffer to write to. | |
1262 | * @length: The length of the data being written (excluding the event header) | |
1263 | * @data: The data to write to the buffer. | |
1264 | * | |
1265 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
1266 | * one function. If you already have the data to write to the buffer, it | |
1267 | * may be easier to simply call this function. | |
1268 | * | |
1269 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
1270 | * and not the length of the event which would hold the header. | |
1271 | */ | |
1272 | int ring_buffer_write(struct ring_buffer *buffer, | |
1273 | unsigned long length, | |
1274 | void *data) | |
1275 | { | |
1276 | struct ring_buffer_per_cpu *cpu_buffer; | |
1277 | struct ring_buffer_event *event; | |
bf41a158 | 1278 | unsigned long event_length; |
7a8e76a3 SR |
1279 | void *body; |
1280 | int ret = -EBUSY; | |
bf41a158 | 1281 | int cpu, resched; |
7a8e76a3 SR |
1282 | |
1283 | if (atomic_read(&buffer->record_disabled)) | |
1284 | return -EBUSY; | |
1285 | ||
182e9f5f | 1286 | resched = ftrace_preempt_disable(); |
bf41a158 | 1287 | |
7a8e76a3 SR |
1288 | cpu = raw_smp_processor_id(); |
1289 | ||
1290 | if (!cpu_isset(cpu, buffer->cpumask)) | |
d769041f | 1291 | goto out; |
7a8e76a3 SR |
1292 | |
1293 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1294 | |
1295 | if (atomic_read(&cpu_buffer->record_disabled)) | |
1296 | goto out; | |
1297 | ||
1298 | event_length = rb_calculate_event_length(length); | |
1299 | event = rb_reserve_next_event(cpu_buffer, | |
1300 | RINGBUF_TYPE_DATA, event_length); | |
1301 | if (!event) | |
1302 | goto out; | |
1303 | ||
1304 | body = rb_event_data(event); | |
1305 | ||
1306 | memcpy(body, data, length); | |
1307 | ||
1308 | rb_commit(cpu_buffer, event); | |
1309 | ||
1310 | ret = 0; | |
1311 | out: | |
182e9f5f | 1312 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1313 | |
1314 | return ret; | |
1315 | } | |
1316 | ||
bf41a158 SR |
1317 | static inline int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
1318 | { | |
1319 | struct buffer_page *reader = cpu_buffer->reader_page; | |
1320 | struct buffer_page *head = cpu_buffer->head_page; | |
1321 | struct buffer_page *commit = cpu_buffer->commit_page; | |
1322 | ||
1323 | return reader->read == rb_page_commit(reader) && | |
1324 | (commit == reader || | |
1325 | (commit == head && | |
1326 | head->read == rb_page_commit(commit))); | |
1327 | } | |
1328 | ||
7a8e76a3 SR |
1329 | /** |
1330 | * ring_buffer_record_disable - stop all writes into the buffer | |
1331 | * @buffer: The ring buffer to stop writes to. | |
1332 | * | |
1333 | * This prevents all writes to the buffer. Any attempt to write | |
1334 | * to the buffer after this will fail and return NULL. | |
1335 | * | |
1336 | * The caller should call synchronize_sched() after this. | |
1337 | */ | |
1338 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
1339 | { | |
1340 | atomic_inc(&buffer->record_disabled); | |
1341 | } | |
1342 | ||
1343 | /** | |
1344 | * ring_buffer_record_enable - enable writes to the buffer | |
1345 | * @buffer: The ring buffer to enable writes | |
1346 | * | |
1347 | * Note, multiple disables will need the same number of enables | |
1348 | * to truely enable the writing (much like preempt_disable). | |
1349 | */ | |
1350 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
1351 | { | |
1352 | atomic_dec(&buffer->record_disabled); | |
1353 | } | |
1354 | ||
1355 | /** | |
1356 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
1357 | * @buffer: The ring buffer to stop writes to. | |
1358 | * @cpu: The CPU buffer to stop | |
1359 | * | |
1360 | * This prevents all writes to the buffer. Any attempt to write | |
1361 | * to the buffer after this will fail and return NULL. | |
1362 | * | |
1363 | * The caller should call synchronize_sched() after this. | |
1364 | */ | |
1365 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
1366 | { | |
1367 | struct ring_buffer_per_cpu *cpu_buffer; | |
1368 | ||
1369 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1370 | return; | |
1371 | ||
1372 | cpu_buffer = buffer->buffers[cpu]; | |
1373 | atomic_inc(&cpu_buffer->record_disabled); | |
1374 | } | |
1375 | ||
1376 | /** | |
1377 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
1378 | * @buffer: The ring buffer to enable writes | |
1379 | * @cpu: The CPU to enable. | |
1380 | * | |
1381 | * Note, multiple disables will need the same number of enables | |
1382 | * to truely enable the writing (much like preempt_disable). | |
1383 | */ | |
1384 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
1385 | { | |
1386 | struct ring_buffer_per_cpu *cpu_buffer; | |
1387 | ||
1388 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1389 | return; | |
1390 | ||
1391 | cpu_buffer = buffer->buffers[cpu]; | |
1392 | atomic_dec(&cpu_buffer->record_disabled); | |
1393 | } | |
1394 | ||
1395 | /** | |
1396 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
1397 | * @buffer: The ring buffer | |
1398 | * @cpu: The per CPU buffer to get the entries from. | |
1399 | */ | |
1400 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
1401 | { | |
1402 | struct ring_buffer_per_cpu *cpu_buffer; | |
1403 | ||
1404 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1405 | return 0; | |
1406 | ||
1407 | cpu_buffer = buffer->buffers[cpu]; | |
1408 | return cpu_buffer->entries; | |
1409 | } | |
1410 | ||
1411 | /** | |
1412 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
1413 | * @buffer: The ring buffer | |
1414 | * @cpu: The per CPU buffer to get the number of overruns from | |
1415 | */ | |
1416 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
1417 | { | |
1418 | struct ring_buffer_per_cpu *cpu_buffer; | |
1419 | ||
1420 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1421 | return 0; | |
1422 | ||
1423 | cpu_buffer = buffer->buffers[cpu]; | |
1424 | return cpu_buffer->overrun; | |
1425 | } | |
1426 | ||
1427 | /** | |
1428 | * ring_buffer_entries - get the number of entries in a buffer | |
1429 | * @buffer: The ring buffer | |
1430 | * | |
1431 | * Returns the total number of entries in the ring buffer | |
1432 | * (all CPU entries) | |
1433 | */ | |
1434 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
1435 | { | |
1436 | struct ring_buffer_per_cpu *cpu_buffer; | |
1437 | unsigned long entries = 0; | |
1438 | int cpu; | |
1439 | ||
1440 | /* if you care about this being correct, lock the buffer */ | |
1441 | for_each_buffer_cpu(buffer, cpu) { | |
1442 | cpu_buffer = buffer->buffers[cpu]; | |
1443 | entries += cpu_buffer->entries; | |
1444 | } | |
1445 | ||
1446 | return entries; | |
1447 | } | |
1448 | ||
1449 | /** | |
1450 | * ring_buffer_overrun_cpu - get the number of overruns in buffer | |
1451 | * @buffer: The ring buffer | |
1452 | * | |
1453 | * Returns the total number of overruns in the ring buffer | |
1454 | * (all CPU entries) | |
1455 | */ | |
1456 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
1457 | { | |
1458 | struct ring_buffer_per_cpu *cpu_buffer; | |
1459 | unsigned long overruns = 0; | |
1460 | int cpu; | |
1461 | ||
1462 | /* if you care about this being correct, lock the buffer */ | |
1463 | for_each_buffer_cpu(buffer, cpu) { | |
1464 | cpu_buffer = buffer->buffers[cpu]; | |
1465 | overruns += cpu_buffer->overrun; | |
1466 | } | |
1467 | ||
1468 | return overruns; | |
1469 | } | |
1470 | ||
1471 | /** | |
1472 | * ring_buffer_iter_reset - reset an iterator | |
1473 | * @iter: The iterator to reset | |
1474 | * | |
1475 | * Resets the iterator, so that it will start from the beginning | |
1476 | * again. | |
1477 | */ | |
1478 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
1479 | { | |
1480 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
f83c9d0f SR |
1481 | unsigned long flags; |
1482 | ||
1483 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 1484 | |
d769041f SR |
1485 | /* Iterator usage is expected to have record disabled */ |
1486 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
1487 | iter->head_page = cpu_buffer->head_page; | |
6f807acd | 1488 | iter->head = cpu_buffer->head_page->read; |
d769041f SR |
1489 | } else { |
1490 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 1491 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
1492 | } |
1493 | if (iter->head) | |
1494 | iter->read_stamp = cpu_buffer->read_stamp; | |
1495 | else | |
1496 | iter->read_stamp = iter->head_page->time_stamp; | |
f83c9d0f SR |
1497 | |
1498 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 SR |
1499 | } |
1500 | ||
1501 | /** | |
1502 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
1503 | * @iter: The iterator to check | |
1504 | */ | |
1505 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
1506 | { | |
1507 | struct ring_buffer_per_cpu *cpu_buffer; | |
1508 | ||
1509 | cpu_buffer = iter->cpu_buffer; | |
1510 | ||
bf41a158 SR |
1511 | return iter->head_page == cpu_buffer->commit_page && |
1512 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 SR |
1513 | } |
1514 | ||
1515 | static void | |
1516 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1517 | struct ring_buffer_event *event) | |
1518 | { | |
1519 | u64 delta; | |
1520 | ||
1521 | switch (event->type) { | |
1522 | case RINGBUF_TYPE_PADDING: | |
1523 | return; | |
1524 | ||
1525 | case RINGBUF_TYPE_TIME_EXTEND: | |
1526 | delta = event->array[0]; | |
1527 | delta <<= TS_SHIFT; | |
1528 | delta += event->time_delta; | |
1529 | cpu_buffer->read_stamp += delta; | |
1530 | return; | |
1531 | ||
1532 | case RINGBUF_TYPE_TIME_STAMP: | |
1533 | /* FIXME: not implemented */ | |
1534 | return; | |
1535 | ||
1536 | case RINGBUF_TYPE_DATA: | |
1537 | cpu_buffer->read_stamp += event->time_delta; | |
1538 | return; | |
1539 | ||
1540 | default: | |
1541 | BUG(); | |
1542 | } | |
1543 | return; | |
1544 | } | |
1545 | ||
1546 | static void | |
1547 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
1548 | struct ring_buffer_event *event) | |
1549 | { | |
1550 | u64 delta; | |
1551 | ||
1552 | switch (event->type) { | |
1553 | case RINGBUF_TYPE_PADDING: | |
1554 | return; | |
1555 | ||
1556 | case RINGBUF_TYPE_TIME_EXTEND: | |
1557 | delta = event->array[0]; | |
1558 | delta <<= TS_SHIFT; | |
1559 | delta += event->time_delta; | |
1560 | iter->read_stamp += delta; | |
1561 | return; | |
1562 | ||
1563 | case RINGBUF_TYPE_TIME_STAMP: | |
1564 | /* FIXME: not implemented */ | |
1565 | return; | |
1566 | ||
1567 | case RINGBUF_TYPE_DATA: | |
1568 | iter->read_stamp += event->time_delta; | |
1569 | return; | |
1570 | ||
1571 | default: | |
1572 | BUG(); | |
1573 | } | |
1574 | return; | |
1575 | } | |
1576 | ||
d769041f SR |
1577 | static struct buffer_page * |
1578 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 1579 | { |
d769041f SR |
1580 | struct buffer_page *reader = NULL; |
1581 | unsigned long flags; | |
818e3dd3 | 1582 | int nr_loops = 0; |
d769041f | 1583 | |
3e03fb7f SR |
1584 | local_irq_save(flags); |
1585 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
1586 | |
1587 | again: | |
818e3dd3 SR |
1588 | /* |
1589 | * This should normally only loop twice. But because the | |
1590 | * start of the reader inserts an empty page, it causes | |
1591 | * a case where we will loop three times. There should be no | |
1592 | * reason to loop four times (that I know of). | |
1593 | */ | |
1594 | if (unlikely(++nr_loops > 3)) { | |
1595 | RB_WARN_ON(cpu_buffer, 1); | |
1596 | reader = NULL; | |
1597 | goto out; | |
1598 | } | |
1599 | ||
d769041f SR |
1600 | reader = cpu_buffer->reader_page; |
1601 | ||
1602 | /* If there's more to read, return this page */ | |
bf41a158 | 1603 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
1604 | goto out; |
1605 | ||
1606 | /* Never should we have an index greater than the size */ | |
bf41a158 SR |
1607 | RB_WARN_ON(cpu_buffer, |
1608 | cpu_buffer->reader_page->read > rb_page_size(reader)); | |
d769041f SR |
1609 | |
1610 | /* check if we caught up to the tail */ | |
1611 | reader = NULL; | |
bf41a158 | 1612 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 1613 | goto out; |
7a8e76a3 SR |
1614 | |
1615 | /* | |
d769041f SR |
1616 | * Splice the empty reader page into the list around the head. |
1617 | * Reset the reader page to size zero. | |
7a8e76a3 | 1618 | */ |
7a8e76a3 | 1619 | |
d769041f SR |
1620 | reader = cpu_buffer->head_page; |
1621 | cpu_buffer->reader_page->list.next = reader->list.next; | |
1622 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 SR |
1623 | |
1624 | local_set(&cpu_buffer->reader_page->write, 0); | |
1625 | local_set(&cpu_buffer->reader_page->commit, 0); | |
7a8e76a3 | 1626 | |
d769041f SR |
1627 | /* Make the reader page now replace the head */ |
1628 | reader->list.prev->next = &cpu_buffer->reader_page->list; | |
1629 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
7a8e76a3 SR |
1630 | |
1631 | /* | |
d769041f SR |
1632 | * If the tail is on the reader, then we must set the head |
1633 | * to the inserted page, otherwise we set it one before. | |
7a8e76a3 | 1634 | */ |
d769041f | 1635 | cpu_buffer->head_page = cpu_buffer->reader_page; |
7a8e76a3 | 1636 | |
bf41a158 | 1637 | if (cpu_buffer->commit_page != reader) |
d769041f SR |
1638 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
1639 | ||
1640 | /* Finally update the reader page to the new head */ | |
1641 | cpu_buffer->reader_page = reader; | |
1642 | rb_reset_reader_page(cpu_buffer); | |
1643 | ||
1644 | goto again; | |
1645 | ||
1646 | out: | |
3e03fb7f SR |
1647 | __raw_spin_unlock(&cpu_buffer->lock); |
1648 | local_irq_restore(flags); | |
d769041f SR |
1649 | |
1650 | return reader; | |
1651 | } | |
1652 | ||
1653 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
1654 | { | |
1655 | struct ring_buffer_event *event; | |
1656 | struct buffer_page *reader; | |
1657 | unsigned length; | |
1658 | ||
1659 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 1660 | |
d769041f | 1661 | /* This function should not be called when buffer is empty */ |
f536aafc | 1662 | RB_WARN_ON_RET(cpu_buffer, !reader); |
7a8e76a3 | 1663 | |
d769041f SR |
1664 | event = rb_reader_event(cpu_buffer); |
1665 | ||
1666 | if (event->type == RINGBUF_TYPE_DATA) | |
1667 | cpu_buffer->entries--; | |
1668 | ||
1669 | rb_update_read_stamp(cpu_buffer, event); | |
1670 | ||
1671 | length = rb_event_length(event); | |
6f807acd | 1672 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
1673 | } |
1674 | ||
1675 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
1676 | { | |
1677 | struct ring_buffer *buffer; | |
1678 | struct ring_buffer_per_cpu *cpu_buffer; | |
1679 | struct ring_buffer_event *event; | |
1680 | unsigned length; | |
1681 | ||
1682 | cpu_buffer = iter->cpu_buffer; | |
1683 | buffer = cpu_buffer->buffer; | |
1684 | ||
1685 | /* | |
1686 | * Check if we are at the end of the buffer. | |
1687 | */ | |
bf41a158 | 1688 | if (iter->head >= rb_page_size(iter->head_page)) { |
f536aafc SR |
1689 | RB_WARN_ON_RET(buffer, |
1690 | iter->head_page == cpu_buffer->commit_page); | |
d769041f | 1691 | rb_inc_iter(iter); |
7a8e76a3 SR |
1692 | return; |
1693 | } | |
1694 | ||
1695 | event = rb_iter_head_event(iter); | |
1696 | ||
1697 | length = rb_event_length(event); | |
1698 | ||
1699 | /* | |
1700 | * This should not be called to advance the header if we are | |
1701 | * at the tail of the buffer. | |
1702 | */ | |
f536aafc SR |
1703 | RB_WARN_ON_RET(cpu_buffer, |
1704 | (iter->head_page == cpu_buffer->commit_page) && | |
1705 | (iter->head + length > rb_commit_index(cpu_buffer))); | |
7a8e76a3 SR |
1706 | |
1707 | rb_update_iter_read_stamp(iter, event); | |
1708 | ||
1709 | iter->head += length; | |
1710 | ||
1711 | /* check for end of page padding */ | |
bf41a158 SR |
1712 | if ((iter->head >= rb_page_size(iter->head_page)) && |
1713 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
1714 | rb_advance_iter(iter); |
1715 | } | |
1716 | ||
f83c9d0f SR |
1717 | static struct ring_buffer_event * |
1718 | rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
7a8e76a3 SR |
1719 | { |
1720 | struct ring_buffer_per_cpu *cpu_buffer; | |
1721 | struct ring_buffer_event *event; | |
d769041f | 1722 | struct buffer_page *reader; |
818e3dd3 | 1723 | int nr_loops = 0; |
7a8e76a3 SR |
1724 | |
1725 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1726 | return NULL; | |
1727 | ||
1728 | cpu_buffer = buffer->buffers[cpu]; | |
1729 | ||
1730 | again: | |
818e3dd3 SR |
1731 | /* |
1732 | * We repeat when a timestamp is encountered. It is possible | |
1733 | * to get multiple timestamps from an interrupt entering just | |
1734 | * as one timestamp is about to be written. The max times | |
1735 | * that this can happen is the number of nested interrupts we | |
1736 | * can have. Nesting 10 deep of interrupts is clearly | |
1737 | * an anomaly. | |
1738 | */ | |
1739 | if (unlikely(++nr_loops > 10)) { | |
1740 | RB_WARN_ON(cpu_buffer, 1); | |
1741 | return NULL; | |
1742 | } | |
1743 | ||
d769041f SR |
1744 | reader = rb_get_reader_page(cpu_buffer); |
1745 | if (!reader) | |
7a8e76a3 SR |
1746 | return NULL; |
1747 | ||
d769041f | 1748 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 SR |
1749 | |
1750 | switch (event->type) { | |
1751 | case RINGBUF_TYPE_PADDING: | |
bf41a158 | 1752 | RB_WARN_ON(cpu_buffer, 1); |
d769041f SR |
1753 | rb_advance_reader(cpu_buffer); |
1754 | return NULL; | |
7a8e76a3 SR |
1755 | |
1756 | case RINGBUF_TYPE_TIME_EXTEND: | |
1757 | /* Internal data, OK to advance */ | |
d769041f | 1758 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
1759 | goto again; |
1760 | ||
1761 | case RINGBUF_TYPE_TIME_STAMP: | |
1762 | /* FIXME: not implemented */ | |
d769041f | 1763 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
1764 | goto again; |
1765 | ||
1766 | case RINGBUF_TYPE_DATA: | |
1767 | if (ts) { | |
1768 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
1769 | ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts); | |
1770 | } | |
1771 | return event; | |
1772 | ||
1773 | default: | |
1774 | BUG(); | |
1775 | } | |
1776 | ||
1777 | return NULL; | |
1778 | } | |
1779 | ||
f83c9d0f SR |
1780 | static struct ring_buffer_event * |
1781 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
1782 | { |
1783 | struct ring_buffer *buffer; | |
1784 | struct ring_buffer_per_cpu *cpu_buffer; | |
1785 | struct ring_buffer_event *event; | |
818e3dd3 | 1786 | int nr_loops = 0; |
7a8e76a3 SR |
1787 | |
1788 | if (ring_buffer_iter_empty(iter)) | |
1789 | return NULL; | |
1790 | ||
1791 | cpu_buffer = iter->cpu_buffer; | |
1792 | buffer = cpu_buffer->buffer; | |
1793 | ||
1794 | again: | |
818e3dd3 SR |
1795 | /* |
1796 | * We repeat when a timestamp is encountered. It is possible | |
1797 | * to get multiple timestamps from an interrupt entering just | |
1798 | * as one timestamp is about to be written. The max times | |
1799 | * that this can happen is the number of nested interrupts we | |
1800 | * can have. Nesting 10 deep of interrupts is clearly | |
1801 | * an anomaly. | |
1802 | */ | |
1803 | if (unlikely(++nr_loops > 10)) { | |
1804 | RB_WARN_ON(cpu_buffer, 1); | |
1805 | return NULL; | |
1806 | } | |
1807 | ||
7a8e76a3 SR |
1808 | if (rb_per_cpu_empty(cpu_buffer)) |
1809 | return NULL; | |
1810 | ||
1811 | event = rb_iter_head_event(iter); | |
1812 | ||
1813 | switch (event->type) { | |
1814 | case RINGBUF_TYPE_PADDING: | |
d769041f | 1815 | rb_inc_iter(iter); |
7a8e76a3 SR |
1816 | goto again; |
1817 | ||
1818 | case RINGBUF_TYPE_TIME_EXTEND: | |
1819 | /* Internal data, OK to advance */ | |
1820 | rb_advance_iter(iter); | |
1821 | goto again; | |
1822 | ||
1823 | case RINGBUF_TYPE_TIME_STAMP: | |
1824 | /* FIXME: not implemented */ | |
1825 | rb_advance_iter(iter); | |
1826 | goto again; | |
1827 | ||
1828 | case RINGBUF_TYPE_DATA: | |
1829 | if (ts) { | |
1830 | *ts = iter->read_stamp + event->time_delta; | |
1831 | ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts); | |
1832 | } | |
1833 | return event; | |
1834 | ||
1835 | default: | |
1836 | BUG(); | |
1837 | } | |
1838 | ||
1839 | return NULL; | |
1840 | } | |
1841 | ||
f83c9d0f SR |
1842 | /** |
1843 | * ring_buffer_peek - peek at the next event to be read | |
1844 | * @buffer: The ring buffer to read | |
1845 | * @cpu: The cpu to peak at | |
1846 | * @ts: The timestamp counter of this event. | |
1847 | * | |
1848 | * This will return the event that will be read next, but does | |
1849 | * not consume the data. | |
1850 | */ | |
1851 | struct ring_buffer_event * | |
1852 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
1853 | { | |
1854 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
1855 | struct ring_buffer_event *event; | |
1856 | unsigned long flags; | |
1857 | ||
1858 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
1859 | event = rb_buffer_peek(buffer, cpu, ts); | |
1860 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
1861 | ||
1862 | return event; | |
1863 | } | |
1864 | ||
1865 | /** | |
1866 | * ring_buffer_iter_peek - peek at the next event to be read | |
1867 | * @iter: The ring buffer iterator | |
1868 | * @ts: The timestamp counter of this event. | |
1869 | * | |
1870 | * This will return the event that will be read next, but does | |
1871 | * not increment the iterator. | |
1872 | */ | |
1873 | struct ring_buffer_event * | |
1874 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
1875 | { | |
1876 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1877 | struct ring_buffer_event *event; | |
1878 | unsigned long flags; | |
1879 | ||
1880 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
1881 | event = rb_iter_peek(iter, ts); | |
1882 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
1883 | ||
1884 | return event; | |
1885 | } | |
1886 | ||
7a8e76a3 SR |
1887 | /** |
1888 | * ring_buffer_consume - return an event and consume it | |
1889 | * @buffer: The ring buffer to get the next event from | |
1890 | * | |
1891 | * Returns the next event in the ring buffer, and that event is consumed. | |
1892 | * Meaning, that sequential reads will keep returning a different event, | |
1893 | * and eventually empty the ring buffer if the producer is slower. | |
1894 | */ | |
1895 | struct ring_buffer_event * | |
1896 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
1897 | { | |
f83c9d0f | 1898 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; |
7a8e76a3 | 1899 | struct ring_buffer_event *event; |
f83c9d0f | 1900 | unsigned long flags; |
7a8e76a3 SR |
1901 | |
1902 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1903 | return NULL; | |
1904 | ||
f83c9d0f SR |
1905 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
1906 | ||
1907 | event = rb_buffer_peek(buffer, cpu, ts); | |
7a8e76a3 | 1908 | if (!event) |
f83c9d0f | 1909 | goto out; |
7a8e76a3 | 1910 | |
d769041f | 1911 | rb_advance_reader(cpu_buffer); |
7a8e76a3 | 1912 | |
f83c9d0f SR |
1913 | out: |
1914 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
1915 | ||
7a8e76a3 SR |
1916 | return event; |
1917 | } | |
1918 | ||
1919 | /** | |
1920 | * ring_buffer_read_start - start a non consuming read of the buffer | |
1921 | * @buffer: The ring buffer to read from | |
1922 | * @cpu: The cpu buffer to iterate over | |
1923 | * | |
1924 | * This starts up an iteration through the buffer. It also disables | |
1925 | * the recording to the buffer until the reading is finished. | |
1926 | * This prevents the reading from being corrupted. This is not | |
1927 | * a consuming read, so a producer is not expected. | |
1928 | * | |
1929 | * Must be paired with ring_buffer_finish. | |
1930 | */ | |
1931 | struct ring_buffer_iter * | |
1932 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
1933 | { | |
1934 | struct ring_buffer_per_cpu *cpu_buffer; | |
1935 | struct ring_buffer_iter *iter; | |
d769041f | 1936 | unsigned long flags; |
7a8e76a3 SR |
1937 | |
1938 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1939 | return NULL; | |
1940 | ||
1941 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
1942 | if (!iter) | |
1943 | return NULL; | |
1944 | ||
1945 | cpu_buffer = buffer->buffers[cpu]; | |
1946 | ||
1947 | iter->cpu_buffer = cpu_buffer; | |
1948 | ||
1949 | atomic_inc(&cpu_buffer->record_disabled); | |
1950 | synchronize_sched(); | |
1951 | ||
f83c9d0f | 1952 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 1953 | __raw_spin_lock(&cpu_buffer->lock); |
d769041f | 1954 | ring_buffer_iter_reset(iter); |
3e03fb7f | 1955 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 1956 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
1957 | |
1958 | return iter; | |
1959 | } | |
1960 | ||
1961 | /** | |
1962 | * ring_buffer_finish - finish reading the iterator of the buffer | |
1963 | * @iter: The iterator retrieved by ring_buffer_start | |
1964 | * | |
1965 | * This re-enables the recording to the buffer, and frees the | |
1966 | * iterator. | |
1967 | */ | |
1968 | void | |
1969 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
1970 | { | |
1971 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1972 | ||
1973 | atomic_dec(&cpu_buffer->record_disabled); | |
1974 | kfree(iter); | |
1975 | } | |
1976 | ||
1977 | /** | |
1978 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
1979 | * @iter: The ring buffer iterator | |
1980 | * @ts: The time stamp of the event read. | |
1981 | * | |
1982 | * This reads the next event in the ring buffer and increments the iterator. | |
1983 | */ | |
1984 | struct ring_buffer_event * | |
1985 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
1986 | { | |
1987 | struct ring_buffer_event *event; | |
f83c9d0f SR |
1988 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
1989 | unsigned long flags; | |
7a8e76a3 | 1990 | |
f83c9d0f SR |
1991 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
1992 | event = rb_iter_peek(iter, ts); | |
7a8e76a3 | 1993 | if (!event) |
f83c9d0f | 1994 | goto out; |
7a8e76a3 SR |
1995 | |
1996 | rb_advance_iter(iter); | |
f83c9d0f SR |
1997 | out: |
1998 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 SR |
1999 | |
2000 | return event; | |
2001 | } | |
2002 | ||
2003 | /** | |
2004 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
2005 | * @buffer: The ring buffer. | |
2006 | */ | |
2007 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
2008 | { | |
2009 | return BUF_PAGE_SIZE * buffer->pages; | |
2010 | } | |
2011 | ||
2012 | static void | |
2013 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
2014 | { | |
2015 | cpu_buffer->head_page | |
2016 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 SR |
2017 | local_set(&cpu_buffer->head_page->write, 0); |
2018 | local_set(&cpu_buffer->head_page->commit, 0); | |
d769041f | 2019 | |
6f807acd | 2020 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
2021 | |
2022 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
2023 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
2024 | ||
2025 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
2026 | local_set(&cpu_buffer->reader_page->write, 0); | |
2027 | local_set(&cpu_buffer->reader_page->commit, 0); | |
6f807acd | 2028 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 2029 | |
7a8e76a3 SR |
2030 | cpu_buffer->overrun = 0; |
2031 | cpu_buffer->entries = 0; | |
2032 | } | |
2033 | ||
2034 | /** | |
2035 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
2036 | * @buffer: The ring buffer to reset a per cpu buffer of | |
2037 | * @cpu: The CPU buffer to be reset | |
2038 | */ | |
2039 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
2040 | { | |
2041 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2042 | unsigned long flags; | |
2043 | ||
2044 | if (!cpu_isset(cpu, buffer->cpumask)) | |
2045 | return; | |
2046 | ||
f83c9d0f SR |
2047 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2048 | ||
3e03fb7f | 2049 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
2050 | |
2051 | rb_reset_cpu(cpu_buffer); | |
2052 | ||
3e03fb7f | 2053 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f SR |
2054 | |
2055 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 SR |
2056 | } |
2057 | ||
2058 | /** | |
2059 | * ring_buffer_reset - reset a ring buffer | |
2060 | * @buffer: The ring buffer to reset all cpu buffers | |
2061 | */ | |
2062 | void ring_buffer_reset(struct ring_buffer *buffer) | |
2063 | { | |
7a8e76a3 SR |
2064 | int cpu; |
2065 | ||
7a8e76a3 | 2066 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 2067 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 SR |
2068 | } |
2069 | ||
2070 | /** | |
2071 | * rind_buffer_empty - is the ring buffer empty? | |
2072 | * @buffer: The ring buffer to test | |
2073 | */ | |
2074 | int ring_buffer_empty(struct ring_buffer *buffer) | |
2075 | { | |
2076 | struct ring_buffer_per_cpu *cpu_buffer; | |
2077 | int cpu; | |
2078 | ||
2079 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
2080 | for_each_buffer_cpu(buffer, cpu) { | |
2081 | cpu_buffer = buffer->buffers[cpu]; | |
2082 | if (!rb_per_cpu_empty(cpu_buffer)) | |
2083 | return 0; | |
2084 | } | |
2085 | return 1; | |
2086 | } | |
2087 | ||
2088 | /** | |
2089 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
2090 | * @buffer: The ring buffer | |
2091 | * @cpu: The CPU buffer to test | |
2092 | */ | |
2093 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
2094 | { | |
2095 | struct ring_buffer_per_cpu *cpu_buffer; | |
2096 | ||
2097 | if (!cpu_isset(cpu, buffer->cpumask)) | |
2098 | return 1; | |
2099 | ||
2100 | cpu_buffer = buffer->buffers[cpu]; | |
2101 | return rb_per_cpu_empty(cpu_buffer); | |
2102 | } | |
2103 | ||
2104 | /** | |
2105 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
2106 | * @buffer_a: One buffer to swap with | |
2107 | * @buffer_b: The other buffer to swap with | |
2108 | * | |
2109 | * This function is useful for tracers that want to take a "snapshot" | |
2110 | * of a CPU buffer and has another back up buffer lying around. | |
2111 | * it is expected that the tracer handles the cpu buffer not being | |
2112 | * used at the moment. | |
2113 | */ | |
2114 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
2115 | struct ring_buffer *buffer_b, int cpu) | |
2116 | { | |
2117 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
2118 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
2119 | ||
2120 | if (!cpu_isset(cpu, buffer_a->cpumask) || | |
2121 | !cpu_isset(cpu, buffer_b->cpumask)) | |
2122 | return -EINVAL; | |
2123 | ||
2124 | /* At least make sure the two buffers are somewhat the same */ | |
2125 | if (buffer_a->size != buffer_b->size || | |
2126 | buffer_a->pages != buffer_b->pages) | |
2127 | return -EINVAL; | |
2128 | ||
2129 | cpu_buffer_a = buffer_a->buffers[cpu]; | |
2130 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
2131 | ||
2132 | /* | |
2133 | * We can't do a synchronize_sched here because this | |
2134 | * function can be called in atomic context. | |
2135 | * Normally this will be called from the same CPU as cpu. | |
2136 | * If not it's up to the caller to protect this. | |
2137 | */ | |
2138 | atomic_inc(&cpu_buffer_a->record_disabled); | |
2139 | atomic_inc(&cpu_buffer_b->record_disabled); | |
2140 | ||
2141 | buffer_a->buffers[cpu] = cpu_buffer_b; | |
2142 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
2143 | ||
2144 | cpu_buffer_b->buffer = buffer_a; | |
2145 | cpu_buffer_a->buffer = buffer_b; | |
2146 | ||
2147 | atomic_dec(&cpu_buffer_a->record_disabled); | |
2148 | atomic_dec(&cpu_buffer_b->record_disabled); | |
2149 | ||
2150 | return 0; | |
2151 | } | |
2152 |