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