]>
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 | ||
a3583244 SR |
21 | /* Global flag to disable all recording to ring buffers */ |
22 | static int ring_buffers_off __read_mostly; | |
23 | ||
24 | /** | |
25 | * tracing_on - enable all tracing buffers | |
26 | * | |
27 | * This function enables all tracing buffers that may have been | |
28 | * disabled with tracing_off. | |
29 | */ | |
30 | void tracing_on(void) | |
31 | { | |
32 | ring_buffers_off = 0; | |
33 | } | |
34 | ||
35 | /** | |
36 | * tracing_off - turn off all tracing buffers | |
37 | * | |
38 | * This function stops all tracing buffers from recording data. | |
39 | * It does not disable any overhead the tracers themselves may | |
40 | * be causing. This function simply causes all recording to | |
41 | * the ring buffers to fail. | |
42 | */ | |
43 | void tracing_off(void) | |
44 | { | |
45 | ring_buffers_off = 1; | |
46 | } | |
47 | ||
d06bbd66 IM |
48 | #include "trace.h" |
49 | ||
7a8e76a3 SR |
50 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
51 | #define DEBUG_SHIFT 0 | |
52 | ||
53 | /* FIXME!!! */ | |
54 | u64 ring_buffer_time_stamp(int cpu) | |
55 | { | |
47e74f2b SR |
56 | u64 time; |
57 | ||
58 | preempt_disable_notrace(); | |
7a8e76a3 | 59 | /* shift to debug/test normalization and TIME_EXTENTS */ |
47e74f2b SR |
60 | time = sched_clock() << DEBUG_SHIFT; |
61 | preempt_enable_notrace(); | |
62 | ||
63 | return time; | |
7a8e76a3 SR |
64 | } |
65 | ||
66 | void ring_buffer_normalize_time_stamp(int cpu, u64 *ts) | |
67 | { | |
68 | /* Just stupid testing the normalize function and deltas */ | |
69 | *ts >>= DEBUG_SHIFT; | |
70 | } | |
71 | ||
72 | #define RB_EVNT_HDR_SIZE (sizeof(struct ring_buffer_event)) | |
73 | #define RB_ALIGNMENT_SHIFT 2 | |
74 | #define RB_ALIGNMENT (1 << RB_ALIGNMENT_SHIFT) | |
75 | #define RB_MAX_SMALL_DATA 28 | |
76 | ||
77 | enum { | |
78 | RB_LEN_TIME_EXTEND = 8, | |
79 | RB_LEN_TIME_STAMP = 16, | |
80 | }; | |
81 | ||
82 | /* inline for ring buffer fast paths */ | |
83 | static inline unsigned | |
84 | rb_event_length(struct ring_buffer_event *event) | |
85 | { | |
86 | unsigned length; | |
87 | ||
88 | switch (event->type) { | |
89 | case RINGBUF_TYPE_PADDING: | |
90 | /* undefined */ | |
91 | return -1; | |
92 | ||
93 | case RINGBUF_TYPE_TIME_EXTEND: | |
94 | return RB_LEN_TIME_EXTEND; | |
95 | ||
96 | case RINGBUF_TYPE_TIME_STAMP: | |
97 | return RB_LEN_TIME_STAMP; | |
98 | ||
99 | case RINGBUF_TYPE_DATA: | |
100 | if (event->len) | |
101 | length = event->len << RB_ALIGNMENT_SHIFT; | |
102 | else | |
103 | length = event->array[0]; | |
104 | return length + RB_EVNT_HDR_SIZE; | |
105 | default: | |
106 | BUG(); | |
107 | } | |
108 | /* not hit */ | |
109 | return 0; | |
110 | } | |
111 | ||
112 | /** | |
113 | * ring_buffer_event_length - return the length of the event | |
114 | * @event: the event to get the length of | |
115 | */ | |
116 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
117 | { | |
118 | return rb_event_length(event); | |
119 | } | |
120 | ||
121 | /* inline for ring buffer fast paths */ | |
122 | static inline void * | |
123 | rb_event_data(struct ring_buffer_event *event) | |
124 | { | |
125 | BUG_ON(event->type != RINGBUF_TYPE_DATA); | |
126 | /* If length is in len field, then array[0] has the data */ | |
127 | if (event->len) | |
128 | return (void *)&event->array[0]; | |
129 | /* Otherwise length is in array[0] and array[1] has the data */ | |
130 | return (void *)&event->array[1]; | |
131 | } | |
132 | ||
133 | /** | |
134 | * ring_buffer_event_data - return the data of the event | |
135 | * @event: the event to get the data from | |
136 | */ | |
137 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
138 | { | |
139 | return rb_event_data(event); | |
140 | } | |
141 | ||
142 | #define for_each_buffer_cpu(buffer, cpu) \ | |
143 | for_each_cpu_mask(cpu, buffer->cpumask) | |
144 | ||
145 | #define TS_SHIFT 27 | |
146 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
147 | #define TS_DELTA_TEST (~TS_MASK) | |
148 | ||
149 | /* | |
150 | * This hack stolen from mm/slob.c. | |
151 | * We can store per page timing information in the page frame of the page. | |
152 | * Thanks to Peter Zijlstra for suggesting this idea. | |
153 | */ | |
154 | struct buffer_page { | |
e4c2ce82 | 155 | u64 time_stamp; /* page time stamp */ |
bf41a158 SR |
156 | local_t write; /* index for next write */ |
157 | local_t commit; /* write commited index */ | |
6f807acd | 158 | unsigned read; /* index for next read */ |
e4c2ce82 SR |
159 | struct list_head list; /* list of free pages */ |
160 | void *page; /* Actual data page */ | |
7a8e76a3 SR |
161 | }; |
162 | ||
ed56829c SR |
163 | /* |
164 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
165 | * this issue out. | |
166 | */ | |
167 | static inline void free_buffer_page(struct buffer_page *bpage) | |
168 | { | |
e4c2ce82 | 169 | if (bpage->page) |
6ae2a076 | 170 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 171 | kfree(bpage); |
ed56829c SR |
172 | } |
173 | ||
7a8e76a3 SR |
174 | /* |
175 | * We need to fit the time_stamp delta into 27 bits. | |
176 | */ | |
177 | static inline int test_time_stamp(u64 delta) | |
178 | { | |
179 | if (delta & TS_DELTA_TEST) | |
180 | return 1; | |
181 | return 0; | |
182 | } | |
183 | ||
184 | #define BUF_PAGE_SIZE PAGE_SIZE | |
185 | ||
186 | /* | |
187 | * head_page == tail_page && head == tail then buffer is empty. | |
188 | */ | |
189 | struct ring_buffer_per_cpu { | |
190 | int cpu; | |
191 | struct ring_buffer *buffer; | |
f83c9d0f | 192 | spinlock_t reader_lock; /* serialize readers */ |
3e03fb7f | 193 | raw_spinlock_t lock; |
7a8e76a3 SR |
194 | struct lock_class_key lock_key; |
195 | struct list_head pages; | |
6f807acd SR |
196 | struct buffer_page *head_page; /* read from head */ |
197 | struct buffer_page *tail_page; /* write to tail */ | |
bf41a158 | 198 | struct buffer_page *commit_page; /* commited pages */ |
d769041f | 199 | struct buffer_page *reader_page; |
7a8e76a3 SR |
200 | unsigned long overrun; |
201 | unsigned long entries; | |
202 | u64 write_stamp; | |
203 | u64 read_stamp; | |
204 | atomic_t record_disabled; | |
205 | }; | |
206 | ||
207 | struct ring_buffer { | |
208 | unsigned long size; | |
209 | unsigned pages; | |
210 | unsigned flags; | |
211 | int cpus; | |
212 | cpumask_t cpumask; | |
213 | atomic_t record_disabled; | |
214 | ||
215 | struct mutex mutex; | |
216 | ||
217 | struct ring_buffer_per_cpu **buffers; | |
218 | }; | |
219 | ||
220 | struct ring_buffer_iter { | |
221 | struct ring_buffer_per_cpu *cpu_buffer; | |
222 | unsigned long head; | |
223 | struct buffer_page *head_page; | |
224 | u64 read_stamp; | |
225 | }; | |
226 | ||
f536aafc | 227 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
bf41a158 | 228 | #define RB_WARN_ON(buffer, cond) \ |
3e89c7bb SR |
229 | ({ \ |
230 | int _____ret = unlikely(cond); \ | |
231 | if (_____ret) { \ | |
bf41a158 SR |
232 | atomic_inc(&buffer->record_disabled); \ |
233 | WARN_ON(1); \ | |
234 | } \ | |
3e89c7bb SR |
235 | _____ret; \ |
236 | }) | |
f536aafc | 237 | |
7a8e76a3 SR |
238 | /** |
239 | * check_pages - integrity check of buffer pages | |
240 | * @cpu_buffer: CPU buffer with pages to test | |
241 | * | |
242 | * As a safty measure we check to make sure the data pages have not | |
243 | * been corrupted. | |
244 | */ | |
245 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
246 | { | |
247 | struct list_head *head = &cpu_buffer->pages; | |
248 | struct buffer_page *page, *tmp; | |
249 | ||
3e89c7bb SR |
250 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
251 | return -1; | |
252 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
253 | return -1; | |
7a8e76a3 SR |
254 | |
255 | list_for_each_entry_safe(page, tmp, head, list) { | |
3e89c7bb SR |
256 | if (RB_WARN_ON(cpu_buffer, |
257 | page->list.next->prev != &page->list)) | |
258 | return -1; | |
259 | if (RB_WARN_ON(cpu_buffer, | |
260 | page->list.prev->next != &page->list)) | |
261 | return -1; | |
7a8e76a3 SR |
262 | } |
263 | ||
264 | return 0; | |
265 | } | |
266 | ||
7a8e76a3 SR |
267 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
268 | unsigned nr_pages) | |
269 | { | |
270 | struct list_head *head = &cpu_buffer->pages; | |
271 | struct buffer_page *page, *tmp; | |
272 | unsigned long addr; | |
273 | LIST_HEAD(pages); | |
274 | unsigned i; | |
275 | ||
276 | for (i = 0; i < nr_pages; i++) { | |
e4c2ce82 | 277 | page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()), |
aa1e0e3b | 278 | GFP_KERNEL, cpu_to_node(cpu_buffer->cpu)); |
e4c2ce82 SR |
279 | if (!page) |
280 | goto free_pages; | |
281 | list_add(&page->list, &pages); | |
282 | ||
7a8e76a3 SR |
283 | addr = __get_free_page(GFP_KERNEL); |
284 | if (!addr) | |
285 | goto free_pages; | |
e4c2ce82 | 286 | page->page = (void *)addr; |
7a8e76a3 SR |
287 | } |
288 | ||
289 | list_splice(&pages, head); | |
290 | ||
291 | rb_check_pages(cpu_buffer); | |
292 | ||
293 | return 0; | |
294 | ||
295 | free_pages: | |
296 | list_for_each_entry_safe(page, tmp, &pages, list) { | |
297 | list_del_init(&page->list); | |
ed56829c | 298 | free_buffer_page(page); |
7a8e76a3 SR |
299 | } |
300 | return -ENOMEM; | |
301 | } | |
302 | ||
303 | static struct ring_buffer_per_cpu * | |
304 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) | |
305 | { | |
306 | struct ring_buffer_per_cpu *cpu_buffer; | |
e4c2ce82 | 307 | struct buffer_page *page; |
d769041f | 308 | unsigned long addr; |
7a8e76a3 SR |
309 | int ret; |
310 | ||
311 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
312 | GFP_KERNEL, cpu_to_node(cpu)); | |
313 | if (!cpu_buffer) | |
314 | return NULL; | |
315 | ||
316 | cpu_buffer->cpu = cpu; | |
317 | cpu_buffer->buffer = buffer; | |
f83c9d0f | 318 | spin_lock_init(&cpu_buffer->reader_lock); |
3e03fb7f | 319 | cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; |
7a8e76a3 SR |
320 | INIT_LIST_HEAD(&cpu_buffer->pages); |
321 | ||
e4c2ce82 SR |
322 | page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()), |
323 | GFP_KERNEL, cpu_to_node(cpu)); | |
324 | if (!page) | |
325 | goto fail_free_buffer; | |
326 | ||
327 | cpu_buffer->reader_page = page; | |
d769041f SR |
328 | addr = __get_free_page(GFP_KERNEL); |
329 | if (!addr) | |
e4c2ce82 SR |
330 | goto fail_free_reader; |
331 | page->page = (void *)addr; | |
332 | ||
d769041f | 333 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 334 | |
7a8e76a3 SR |
335 | ret = rb_allocate_pages(cpu_buffer, buffer->pages); |
336 | if (ret < 0) | |
d769041f | 337 | goto fail_free_reader; |
7a8e76a3 SR |
338 | |
339 | cpu_buffer->head_page | |
340 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 341 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 SR |
342 | |
343 | return cpu_buffer; | |
344 | ||
d769041f SR |
345 | fail_free_reader: |
346 | free_buffer_page(cpu_buffer->reader_page); | |
347 | ||
7a8e76a3 SR |
348 | fail_free_buffer: |
349 | kfree(cpu_buffer); | |
350 | return NULL; | |
351 | } | |
352 | ||
353 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
354 | { | |
355 | struct list_head *head = &cpu_buffer->pages; | |
356 | struct buffer_page *page, *tmp; | |
357 | ||
d769041f SR |
358 | list_del_init(&cpu_buffer->reader_page->list); |
359 | free_buffer_page(cpu_buffer->reader_page); | |
360 | ||
7a8e76a3 SR |
361 | list_for_each_entry_safe(page, tmp, head, list) { |
362 | list_del_init(&page->list); | |
ed56829c | 363 | free_buffer_page(page); |
7a8e76a3 SR |
364 | } |
365 | kfree(cpu_buffer); | |
366 | } | |
367 | ||
a7b13743 SR |
368 | /* |
369 | * Causes compile errors if the struct buffer_page gets bigger | |
370 | * than the struct page. | |
371 | */ | |
372 | extern int ring_buffer_page_too_big(void); | |
373 | ||
7a8e76a3 SR |
374 | /** |
375 | * ring_buffer_alloc - allocate a new ring_buffer | |
376 | * @size: the size in bytes that is needed. | |
377 | * @flags: attributes to set for the ring buffer. | |
378 | * | |
379 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
380 | * flag. This flag means that the buffer will overwrite old data | |
381 | * when the buffer wraps. If this flag is not set, the buffer will | |
382 | * drop data when the tail hits the head. | |
383 | */ | |
384 | struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags) | |
385 | { | |
386 | struct ring_buffer *buffer; | |
387 | int bsize; | |
388 | int cpu; | |
389 | ||
a7b13743 SR |
390 | /* Paranoid! Optimizes out when all is well */ |
391 | if (sizeof(struct buffer_page) > sizeof(struct page)) | |
392 | ring_buffer_page_too_big(); | |
393 | ||
394 | ||
7a8e76a3 SR |
395 | /* keep it in its own cache line */ |
396 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
397 | GFP_KERNEL); | |
398 | if (!buffer) | |
399 | return NULL; | |
400 | ||
401 | buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
402 | buffer->flags = flags; | |
403 | ||
404 | /* need at least two pages */ | |
405 | if (buffer->pages == 1) | |
406 | buffer->pages++; | |
407 | ||
408 | buffer->cpumask = cpu_possible_map; | |
409 | buffer->cpus = nr_cpu_ids; | |
410 | ||
411 | bsize = sizeof(void *) * nr_cpu_ids; | |
412 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
413 | GFP_KERNEL); | |
414 | if (!buffer->buffers) | |
415 | goto fail_free_buffer; | |
416 | ||
417 | for_each_buffer_cpu(buffer, cpu) { | |
418 | buffer->buffers[cpu] = | |
419 | rb_allocate_cpu_buffer(buffer, cpu); | |
420 | if (!buffer->buffers[cpu]) | |
421 | goto fail_free_buffers; | |
422 | } | |
423 | ||
424 | mutex_init(&buffer->mutex); | |
425 | ||
426 | return buffer; | |
427 | ||
428 | fail_free_buffers: | |
429 | for_each_buffer_cpu(buffer, cpu) { | |
430 | if (buffer->buffers[cpu]) | |
431 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
432 | } | |
433 | kfree(buffer->buffers); | |
434 | ||
435 | fail_free_buffer: | |
436 | kfree(buffer); | |
437 | return NULL; | |
438 | } | |
439 | ||
440 | /** | |
441 | * ring_buffer_free - free a ring buffer. | |
442 | * @buffer: the buffer to free. | |
443 | */ | |
444 | void | |
445 | ring_buffer_free(struct ring_buffer *buffer) | |
446 | { | |
447 | int cpu; | |
448 | ||
449 | for_each_buffer_cpu(buffer, cpu) | |
450 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
451 | ||
452 | kfree(buffer); | |
453 | } | |
454 | ||
455 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); | |
456 | ||
457 | static void | |
458 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages) | |
459 | { | |
460 | struct buffer_page *page; | |
461 | struct list_head *p; | |
462 | unsigned i; | |
463 | ||
464 | atomic_inc(&cpu_buffer->record_disabled); | |
465 | synchronize_sched(); | |
466 | ||
467 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
468 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
469 | return; | |
7a8e76a3 SR |
470 | p = cpu_buffer->pages.next; |
471 | page = list_entry(p, struct buffer_page, list); | |
472 | list_del_init(&page->list); | |
ed56829c | 473 | free_buffer_page(page); |
7a8e76a3 | 474 | } |
3e89c7bb SR |
475 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
476 | return; | |
7a8e76a3 SR |
477 | |
478 | rb_reset_cpu(cpu_buffer); | |
479 | ||
480 | rb_check_pages(cpu_buffer); | |
481 | ||
482 | atomic_dec(&cpu_buffer->record_disabled); | |
483 | ||
484 | } | |
485 | ||
486 | static void | |
487 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
488 | struct list_head *pages, unsigned nr_pages) | |
489 | { | |
490 | struct buffer_page *page; | |
491 | struct list_head *p; | |
492 | unsigned i; | |
493 | ||
494 | atomic_inc(&cpu_buffer->record_disabled); | |
495 | synchronize_sched(); | |
496 | ||
497 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
498 | if (RB_WARN_ON(cpu_buffer, list_empty(pages))) |
499 | return; | |
7a8e76a3 SR |
500 | p = pages->next; |
501 | page = list_entry(p, struct buffer_page, list); | |
502 | list_del_init(&page->list); | |
503 | list_add_tail(&page->list, &cpu_buffer->pages); | |
504 | } | |
505 | rb_reset_cpu(cpu_buffer); | |
506 | ||
507 | rb_check_pages(cpu_buffer); | |
508 | ||
509 | atomic_dec(&cpu_buffer->record_disabled); | |
510 | } | |
511 | ||
512 | /** | |
513 | * ring_buffer_resize - resize the ring buffer | |
514 | * @buffer: the buffer to resize. | |
515 | * @size: the new size. | |
516 | * | |
517 | * The tracer is responsible for making sure that the buffer is | |
518 | * not being used while changing the size. | |
519 | * Note: We may be able to change the above requirement by using | |
520 | * RCU synchronizations. | |
521 | * | |
522 | * Minimum size is 2 * BUF_PAGE_SIZE. | |
523 | * | |
524 | * Returns -1 on failure. | |
525 | */ | |
526 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) | |
527 | { | |
528 | struct ring_buffer_per_cpu *cpu_buffer; | |
529 | unsigned nr_pages, rm_pages, new_pages; | |
530 | struct buffer_page *page, *tmp; | |
531 | unsigned long buffer_size; | |
532 | unsigned long addr; | |
533 | LIST_HEAD(pages); | |
534 | int i, cpu; | |
535 | ||
ee51a1de IM |
536 | /* |
537 | * Always succeed at resizing a non-existent buffer: | |
538 | */ | |
539 | if (!buffer) | |
540 | return size; | |
541 | ||
7a8e76a3 SR |
542 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
543 | size *= BUF_PAGE_SIZE; | |
544 | buffer_size = buffer->pages * BUF_PAGE_SIZE; | |
545 | ||
546 | /* we need a minimum of two pages */ | |
547 | if (size < BUF_PAGE_SIZE * 2) | |
548 | size = BUF_PAGE_SIZE * 2; | |
549 | ||
550 | if (size == buffer_size) | |
551 | return size; | |
552 | ||
553 | mutex_lock(&buffer->mutex); | |
554 | ||
555 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
556 | ||
557 | if (size < buffer_size) { | |
558 | ||
559 | /* easy case, just free pages */ | |
3e89c7bb SR |
560 | if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) { |
561 | mutex_unlock(&buffer->mutex); | |
562 | return -1; | |
563 | } | |
7a8e76a3 SR |
564 | |
565 | rm_pages = buffer->pages - nr_pages; | |
566 | ||
567 | for_each_buffer_cpu(buffer, cpu) { | |
568 | cpu_buffer = buffer->buffers[cpu]; | |
569 | rb_remove_pages(cpu_buffer, rm_pages); | |
570 | } | |
571 | goto out; | |
572 | } | |
573 | ||
574 | /* | |
575 | * This is a bit more difficult. We only want to add pages | |
576 | * when we can allocate enough for all CPUs. We do this | |
577 | * by allocating all the pages and storing them on a local | |
578 | * link list. If we succeed in our allocation, then we | |
579 | * add these pages to the cpu_buffers. Otherwise we just free | |
580 | * them all and return -ENOMEM; | |
581 | */ | |
3e89c7bb SR |
582 | if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) { |
583 | mutex_unlock(&buffer->mutex); | |
584 | return -1; | |
585 | } | |
f536aafc | 586 | |
7a8e76a3 SR |
587 | new_pages = nr_pages - buffer->pages; |
588 | ||
589 | for_each_buffer_cpu(buffer, cpu) { | |
590 | for (i = 0; i < new_pages; i++) { | |
e4c2ce82 SR |
591 | page = kzalloc_node(ALIGN(sizeof(*page), |
592 | cache_line_size()), | |
593 | GFP_KERNEL, cpu_to_node(cpu)); | |
594 | if (!page) | |
595 | goto free_pages; | |
596 | list_add(&page->list, &pages); | |
7a8e76a3 SR |
597 | addr = __get_free_page(GFP_KERNEL); |
598 | if (!addr) | |
599 | goto free_pages; | |
e4c2ce82 | 600 | page->page = (void *)addr; |
7a8e76a3 SR |
601 | } |
602 | } | |
603 | ||
604 | for_each_buffer_cpu(buffer, cpu) { | |
605 | cpu_buffer = buffer->buffers[cpu]; | |
606 | rb_insert_pages(cpu_buffer, &pages, new_pages); | |
607 | } | |
608 | ||
3e89c7bb SR |
609 | if (RB_WARN_ON(buffer, !list_empty(&pages))) { |
610 | mutex_unlock(&buffer->mutex); | |
611 | return -1; | |
612 | } | |
7a8e76a3 SR |
613 | |
614 | out: | |
615 | buffer->pages = nr_pages; | |
616 | mutex_unlock(&buffer->mutex); | |
617 | ||
618 | return size; | |
619 | ||
620 | free_pages: | |
621 | list_for_each_entry_safe(page, tmp, &pages, list) { | |
622 | list_del_init(&page->list); | |
ed56829c | 623 | free_buffer_page(page); |
7a8e76a3 | 624 | } |
641d2f63 | 625 | mutex_unlock(&buffer->mutex); |
7a8e76a3 SR |
626 | return -ENOMEM; |
627 | } | |
628 | ||
7a8e76a3 SR |
629 | static inline int rb_null_event(struct ring_buffer_event *event) |
630 | { | |
631 | return event->type == RINGBUF_TYPE_PADDING; | |
632 | } | |
633 | ||
6f807acd | 634 | static inline void *__rb_page_index(struct buffer_page *page, unsigned index) |
7a8e76a3 | 635 | { |
e4c2ce82 | 636 | return page->page + index; |
7a8e76a3 SR |
637 | } |
638 | ||
639 | static inline struct ring_buffer_event * | |
d769041f | 640 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 641 | { |
6f807acd SR |
642 | return __rb_page_index(cpu_buffer->reader_page, |
643 | cpu_buffer->reader_page->read); | |
644 | } | |
645 | ||
646 | static inline struct ring_buffer_event * | |
647 | rb_head_event(struct ring_buffer_per_cpu *cpu_buffer) | |
648 | { | |
649 | return __rb_page_index(cpu_buffer->head_page, | |
650 | cpu_buffer->head_page->read); | |
7a8e76a3 SR |
651 | } |
652 | ||
653 | static inline struct ring_buffer_event * | |
654 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
655 | { | |
6f807acd | 656 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
657 | } |
658 | ||
bf41a158 SR |
659 | static inline unsigned rb_page_write(struct buffer_page *bpage) |
660 | { | |
661 | return local_read(&bpage->write); | |
662 | } | |
663 | ||
664 | static inline unsigned rb_page_commit(struct buffer_page *bpage) | |
665 | { | |
666 | return local_read(&bpage->commit); | |
667 | } | |
668 | ||
669 | /* Size is determined by what has been commited */ | |
670 | static inline unsigned rb_page_size(struct buffer_page *bpage) | |
671 | { | |
672 | return rb_page_commit(bpage); | |
673 | } | |
674 | ||
675 | static inline unsigned | |
676 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
677 | { | |
678 | return rb_page_commit(cpu_buffer->commit_page); | |
679 | } | |
680 | ||
681 | static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer) | |
682 | { | |
683 | return rb_page_commit(cpu_buffer->head_page); | |
684 | } | |
685 | ||
7a8e76a3 SR |
686 | /* |
687 | * When the tail hits the head and the buffer is in overwrite mode, | |
688 | * the head jumps to the next page and all content on the previous | |
689 | * page is discarded. But before doing so, we update the overrun | |
690 | * variable of the buffer. | |
691 | */ | |
692 | static void rb_update_overflow(struct ring_buffer_per_cpu *cpu_buffer) | |
693 | { | |
694 | struct ring_buffer_event *event; | |
695 | unsigned long head; | |
696 | ||
697 | for (head = 0; head < rb_head_size(cpu_buffer); | |
698 | head += rb_event_length(event)) { | |
699 | ||
6f807acd | 700 | event = __rb_page_index(cpu_buffer->head_page, head); |
3e89c7bb SR |
701 | if (RB_WARN_ON(cpu_buffer, rb_null_event(event))) |
702 | return; | |
7a8e76a3 SR |
703 | /* Only count data entries */ |
704 | if (event->type != RINGBUF_TYPE_DATA) | |
705 | continue; | |
706 | cpu_buffer->overrun++; | |
707 | cpu_buffer->entries--; | |
708 | } | |
709 | } | |
710 | ||
711 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
712 | struct buffer_page **page) | |
713 | { | |
714 | struct list_head *p = (*page)->list.next; | |
715 | ||
716 | if (p == &cpu_buffer->pages) | |
717 | p = p->next; | |
718 | ||
719 | *page = list_entry(p, struct buffer_page, list); | |
720 | } | |
721 | ||
bf41a158 SR |
722 | static inline unsigned |
723 | rb_event_index(struct ring_buffer_event *event) | |
724 | { | |
725 | unsigned long addr = (unsigned long)event; | |
726 | ||
727 | return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE); | |
728 | } | |
729 | ||
730 | static inline int | |
731 | rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
732 | struct ring_buffer_event *event) | |
733 | { | |
734 | unsigned long addr = (unsigned long)event; | |
735 | unsigned long index; | |
736 | ||
737 | index = rb_event_index(event); | |
738 | addr &= PAGE_MASK; | |
739 | ||
740 | return cpu_buffer->commit_page->page == (void *)addr && | |
741 | rb_commit_index(cpu_buffer) == index; | |
742 | } | |
743 | ||
7a8e76a3 | 744 | static inline void |
bf41a158 SR |
745 | rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer, |
746 | struct ring_buffer_event *event) | |
7a8e76a3 | 747 | { |
bf41a158 SR |
748 | unsigned long addr = (unsigned long)event; |
749 | unsigned long index; | |
750 | ||
751 | index = rb_event_index(event); | |
752 | addr &= PAGE_MASK; | |
753 | ||
754 | while (cpu_buffer->commit_page->page != (void *)addr) { | |
3e89c7bb SR |
755 | if (RB_WARN_ON(cpu_buffer, |
756 | cpu_buffer->commit_page == cpu_buffer->tail_page)) | |
757 | return; | |
bf41a158 SR |
758 | cpu_buffer->commit_page->commit = |
759 | cpu_buffer->commit_page->write; | |
760 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
761 | cpu_buffer->write_stamp = cpu_buffer->commit_page->time_stamp; | |
762 | } | |
763 | ||
764 | /* Now set the commit to the event's index */ | |
765 | local_set(&cpu_buffer->commit_page->commit, index); | |
7a8e76a3 SR |
766 | } |
767 | ||
bf41a158 SR |
768 | static inline void |
769 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 770 | { |
bf41a158 SR |
771 | /* |
772 | * We only race with interrupts and NMIs on this CPU. | |
773 | * If we own the commit event, then we can commit | |
774 | * all others that interrupted us, since the interruptions | |
775 | * are in stack format (they finish before they come | |
776 | * back to us). This allows us to do a simple loop to | |
777 | * assign the commit to the tail. | |
778 | */ | |
779 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { | |
780 | cpu_buffer->commit_page->commit = | |
781 | cpu_buffer->commit_page->write; | |
782 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
783 | cpu_buffer->write_stamp = cpu_buffer->commit_page->time_stamp; | |
784 | /* add barrier to keep gcc from optimizing too much */ | |
785 | barrier(); | |
786 | } | |
787 | while (rb_commit_index(cpu_buffer) != | |
788 | rb_page_write(cpu_buffer->commit_page)) { | |
789 | cpu_buffer->commit_page->commit = | |
790 | cpu_buffer->commit_page->write; | |
791 | barrier(); | |
792 | } | |
7a8e76a3 SR |
793 | } |
794 | ||
d769041f | 795 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 796 | { |
d769041f | 797 | cpu_buffer->read_stamp = cpu_buffer->reader_page->time_stamp; |
6f807acd | 798 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
799 | } |
800 | ||
801 | static inline void rb_inc_iter(struct ring_buffer_iter *iter) | |
802 | { | |
803 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
804 | ||
805 | /* | |
806 | * The iterator could be on the reader page (it starts there). | |
807 | * But the head could have moved, since the reader was | |
808 | * found. Check for this case and assign the iterator | |
809 | * to the head page instead of next. | |
810 | */ | |
811 | if (iter->head_page == cpu_buffer->reader_page) | |
812 | iter->head_page = cpu_buffer->head_page; | |
813 | else | |
814 | rb_inc_page(cpu_buffer, &iter->head_page); | |
815 | ||
7a8e76a3 SR |
816 | iter->read_stamp = iter->head_page->time_stamp; |
817 | iter->head = 0; | |
818 | } | |
819 | ||
820 | /** | |
821 | * ring_buffer_update_event - update event type and data | |
822 | * @event: the even to update | |
823 | * @type: the type of event | |
824 | * @length: the size of the event field in the ring buffer | |
825 | * | |
826 | * Update the type and data fields of the event. The length | |
827 | * is the actual size that is written to the ring buffer, | |
828 | * and with this, we can determine what to place into the | |
829 | * data field. | |
830 | */ | |
831 | static inline void | |
832 | rb_update_event(struct ring_buffer_event *event, | |
833 | unsigned type, unsigned length) | |
834 | { | |
835 | event->type = type; | |
836 | ||
837 | switch (type) { | |
838 | ||
839 | case RINGBUF_TYPE_PADDING: | |
840 | break; | |
841 | ||
842 | case RINGBUF_TYPE_TIME_EXTEND: | |
843 | event->len = | |
844 | (RB_LEN_TIME_EXTEND + (RB_ALIGNMENT-1)) | |
845 | >> RB_ALIGNMENT_SHIFT; | |
846 | break; | |
847 | ||
848 | case RINGBUF_TYPE_TIME_STAMP: | |
849 | event->len = | |
850 | (RB_LEN_TIME_STAMP + (RB_ALIGNMENT-1)) | |
851 | >> RB_ALIGNMENT_SHIFT; | |
852 | break; | |
853 | ||
854 | case RINGBUF_TYPE_DATA: | |
855 | length -= RB_EVNT_HDR_SIZE; | |
856 | if (length > RB_MAX_SMALL_DATA) { | |
857 | event->len = 0; | |
858 | event->array[0] = length; | |
859 | } else | |
860 | event->len = | |
861 | (length + (RB_ALIGNMENT-1)) | |
862 | >> RB_ALIGNMENT_SHIFT; | |
863 | break; | |
864 | default: | |
865 | BUG(); | |
866 | } | |
867 | } | |
868 | ||
869 | static inline unsigned rb_calculate_event_length(unsigned length) | |
870 | { | |
871 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
872 | ||
873 | /* zero length can cause confusions */ | |
874 | if (!length) | |
875 | length = 1; | |
876 | ||
877 | if (length > RB_MAX_SMALL_DATA) | |
878 | length += sizeof(event.array[0]); | |
879 | ||
880 | length += RB_EVNT_HDR_SIZE; | |
881 | length = ALIGN(length, RB_ALIGNMENT); | |
882 | ||
883 | return length; | |
884 | } | |
885 | ||
886 | static struct ring_buffer_event * | |
887 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
888 | unsigned type, unsigned long length, u64 *ts) | |
889 | { | |
d769041f | 890 | struct buffer_page *tail_page, *head_page, *reader_page; |
bf41a158 | 891 | unsigned long tail, write; |
7a8e76a3 SR |
892 | struct ring_buffer *buffer = cpu_buffer->buffer; |
893 | struct ring_buffer_event *event; | |
bf41a158 | 894 | unsigned long flags; |
7a8e76a3 SR |
895 | |
896 | tail_page = cpu_buffer->tail_page; | |
bf41a158 SR |
897 | write = local_add_return(length, &tail_page->write); |
898 | tail = write - length; | |
7a8e76a3 | 899 | |
bf41a158 SR |
900 | /* See if we shot pass the end of this buffer page */ |
901 | if (write > BUF_PAGE_SIZE) { | |
7a8e76a3 SR |
902 | struct buffer_page *next_page = tail_page; |
903 | ||
3e03fb7f SR |
904 | local_irq_save(flags); |
905 | __raw_spin_lock(&cpu_buffer->lock); | |
bf41a158 | 906 | |
7a8e76a3 SR |
907 | rb_inc_page(cpu_buffer, &next_page); |
908 | ||
d769041f SR |
909 | head_page = cpu_buffer->head_page; |
910 | reader_page = cpu_buffer->reader_page; | |
911 | ||
912 | /* we grabbed the lock before incrementing */ | |
3e89c7bb SR |
913 | if (RB_WARN_ON(cpu_buffer, next_page == reader_page)) |
914 | goto out_unlock; | |
bf41a158 SR |
915 | |
916 | /* | |
917 | * If for some reason, we had an interrupt storm that made | |
918 | * it all the way around the buffer, bail, and warn | |
919 | * about it. | |
920 | */ | |
921 | if (unlikely(next_page == cpu_buffer->commit_page)) { | |
922 | WARN_ON_ONCE(1); | |
923 | goto out_unlock; | |
924 | } | |
d769041f | 925 | |
7a8e76a3 | 926 | if (next_page == head_page) { |
d769041f | 927 | if (!(buffer->flags & RB_FL_OVERWRITE)) { |
bf41a158 SR |
928 | /* reset write */ |
929 | if (tail <= BUF_PAGE_SIZE) | |
930 | local_set(&tail_page->write, tail); | |
931 | goto out_unlock; | |
d769041f | 932 | } |
7a8e76a3 | 933 | |
bf41a158 SR |
934 | /* tail_page has not moved yet? */ |
935 | if (tail_page == cpu_buffer->tail_page) { | |
936 | /* count overflows */ | |
937 | rb_update_overflow(cpu_buffer); | |
938 | ||
939 | rb_inc_page(cpu_buffer, &head_page); | |
940 | cpu_buffer->head_page = head_page; | |
941 | cpu_buffer->head_page->read = 0; | |
942 | } | |
943 | } | |
7a8e76a3 | 944 | |
bf41a158 SR |
945 | /* |
946 | * If the tail page is still the same as what we think | |
947 | * it is, then it is up to us to update the tail | |
948 | * pointer. | |
949 | */ | |
950 | if (tail_page == cpu_buffer->tail_page) { | |
951 | local_set(&next_page->write, 0); | |
952 | local_set(&next_page->commit, 0); | |
953 | cpu_buffer->tail_page = next_page; | |
954 | ||
955 | /* reread the time stamp */ | |
956 | *ts = ring_buffer_time_stamp(cpu_buffer->cpu); | |
957 | cpu_buffer->tail_page->time_stamp = *ts; | |
7a8e76a3 SR |
958 | } |
959 | ||
bf41a158 SR |
960 | /* |
961 | * The actual tail page has moved forward. | |
962 | */ | |
963 | if (tail < BUF_PAGE_SIZE) { | |
964 | /* Mark the rest of the page with padding */ | |
6f807acd | 965 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
966 | event->type = RINGBUF_TYPE_PADDING; |
967 | } | |
968 | ||
bf41a158 SR |
969 | if (tail <= BUF_PAGE_SIZE) |
970 | /* Set the write back to the previous setting */ | |
971 | local_set(&tail_page->write, tail); | |
972 | ||
973 | /* | |
974 | * If this was a commit entry that failed, | |
975 | * increment that too | |
976 | */ | |
977 | if (tail_page == cpu_buffer->commit_page && | |
978 | tail == rb_commit_index(cpu_buffer)) { | |
979 | rb_set_commit_to_write(cpu_buffer); | |
980 | } | |
981 | ||
3e03fb7f SR |
982 | __raw_spin_unlock(&cpu_buffer->lock); |
983 | local_irq_restore(flags); | |
bf41a158 SR |
984 | |
985 | /* fail and let the caller try again */ | |
986 | return ERR_PTR(-EAGAIN); | |
7a8e76a3 SR |
987 | } |
988 | ||
bf41a158 SR |
989 | /* We reserved something on the buffer */ |
990 | ||
3e89c7bb SR |
991 | if (RB_WARN_ON(cpu_buffer, write > BUF_PAGE_SIZE)) |
992 | return NULL; | |
7a8e76a3 | 993 | |
6f807acd | 994 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
995 | rb_update_event(event, type, length); |
996 | ||
bf41a158 SR |
997 | /* |
998 | * If this is a commit and the tail is zero, then update | |
999 | * this page's time stamp. | |
1000 | */ | |
1001 | if (!tail && rb_is_commit(cpu_buffer, event)) | |
1002 | cpu_buffer->commit_page->time_stamp = *ts; | |
1003 | ||
7a8e76a3 | 1004 | return event; |
bf41a158 SR |
1005 | |
1006 | out_unlock: | |
3e03fb7f SR |
1007 | __raw_spin_unlock(&cpu_buffer->lock); |
1008 | local_irq_restore(flags); | |
bf41a158 | 1009 | return NULL; |
7a8e76a3 SR |
1010 | } |
1011 | ||
1012 | static int | |
1013 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1014 | u64 *ts, u64 *delta) | |
1015 | { | |
1016 | struct ring_buffer_event *event; | |
1017 | static int once; | |
bf41a158 | 1018 | int ret; |
7a8e76a3 SR |
1019 | |
1020 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1021 | printk(KERN_WARNING "Delta way too big! %llu" | |
1022 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1023 | (unsigned long long)*delta, |
1024 | (unsigned long long)*ts, | |
1025 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1026 | WARN_ON(1); |
1027 | } | |
1028 | ||
1029 | /* | |
1030 | * The delta is too big, we to add a | |
1031 | * new timestamp. | |
1032 | */ | |
1033 | event = __rb_reserve_next(cpu_buffer, | |
1034 | RINGBUF_TYPE_TIME_EXTEND, | |
1035 | RB_LEN_TIME_EXTEND, | |
1036 | ts); | |
1037 | if (!event) | |
bf41a158 | 1038 | return -EBUSY; |
7a8e76a3 | 1039 | |
bf41a158 SR |
1040 | if (PTR_ERR(event) == -EAGAIN) |
1041 | return -EAGAIN; | |
1042 | ||
1043 | /* Only a commited time event can update the write stamp */ | |
1044 | if (rb_is_commit(cpu_buffer, event)) { | |
1045 | /* | |
1046 | * If this is the first on the page, then we need to | |
1047 | * update the page itself, and just put in a zero. | |
1048 | */ | |
1049 | if (rb_event_index(event)) { | |
1050 | event->time_delta = *delta & TS_MASK; | |
1051 | event->array[0] = *delta >> TS_SHIFT; | |
1052 | } else { | |
1053 | cpu_buffer->commit_page->time_stamp = *ts; | |
1054 | event->time_delta = 0; | |
1055 | event->array[0] = 0; | |
1056 | } | |
7a8e76a3 | 1057 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
1058 | /* let the caller know this was the commit */ |
1059 | ret = 1; | |
1060 | } else { | |
1061 | /* Darn, this is just wasted space */ | |
1062 | event->time_delta = 0; | |
1063 | event->array[0] = 0; | |
1064 | ret = 0; | |
7a8e76a3 SR |
1065 | } |
1066 | ||
bf41a158 SR |
1067 | *delta = 0; |
1068 | ||
1069 | return ret; | |
7a8e76a3 SR |
1070 | } |
1071 | ||
1072 | static struct ring_buffer_event * | |
1073 | rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, | |
1074 | unsigned type, unsigned long length) | |
1075 | { | |
1076 | struct ring_buffer_event *event; | |
1077 | u64 ts, delta; | |
bf41a158 | 1078 | int commit = 0; |
818e3dd3 | 1079 | int nr_loops = 0; |
7a8e76a3 | 1080 | |
bf41a158 | 1081 | again: |
818e3dd3 SR |
1082 | /* |
1083 | * We allow for interrupts to reenter here and do a trace. | |
1084 | * If one does, it will cause this original code to loop | |
1085 | * back here. Even with heavy interrupts happening, this | |
1086 | * should only happen a few times in a row. If this happens | |
1087 | * 1000 times in a row, there must be either an interrupt | |
1088 | * storm or we have something buggy. | |
1089 | * Bail! | |
1090 | */ | |
3e89c7bb | 1091 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
818e3dd3 | 1092 | return NULL; |
818e3dd3 | 1093 | |
7a8e76a3 SR |
1094 | ts = ring_buffer_time_stamp(cpu_buffer->cpu); |
1095 | ||
bf41a158 SR |
1096 | /* |
1097 | * Only the first commit can update the timestamp. | |
1098 | * Yes there is a race here. If an interrupt comes in | |
1099 | * just after the conditional and it traces too, then it | |
1100 | * will also check the deltas. More than one timestamp may | |
1101 | * also be made. But only the entry that did the actual | |
1102 | * commit will be something other than zero. | |
1103 | */ | |
1104 | if (cpu_buffer->tail_page == cpu_buffer->commit_page && | |
1105 | rb_page_write(cpu_buffer->tail_page) == | |
1106 | rb_commit_index(cpu_buffer)) { | |
1107 | ||
7a8e76a3 SR |
1108 | delta = ts - cpu_buffer->write_stamp; |
1109 | ||
bf41a158 SR |
1110 | /* make sure this delta is calculated here */ |
1111 | barrier(); | |
1112 | ||
1113 | /* Did the write stamp get updated already? */ | |
1114 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
4143c5cb | 1115 | delta = 0; |
bf41a158 | 1116 | |
7a8e76a3 | 1117 | if (test_time_stamp(delta)) { |
7a8e76a3 | 1118 | |
bf41a158 SR |
1119 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
1120 | ||
1121 | if (commit == -EBUSY) | |
7a8e76a3 | 1122 | return NULL; |
bf41a158 SR |
1123 | |
1124 | if (commit == -EAGAIN) | |
1125 | goto again; | |
1126 | ||
1127 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 1128 | } |
bf41a158 SR |
1129 | } else |
1130 | /* Non commits have zero deltas */ | |
7a8e76a3 | 1131 | delta = 0; |
7a8e76a3 SR |
1132 | |
1133 | event = __rb_reserve_next(cpu_buffer, type, length, &ts); | |
bf41a158 SR |
1134 | if (PTR_ERR(event) == -EAGAIN) |
1135 | goto again; | |
1136 | ||
1137 | if (!event) { | |
1138 | if (unlikely(commit)) | |
1139 | /* | |
1140 | * Ouch! We needed a timestamp and it was commited. But | |
1141 | * we didn't get our event reserved. | |
1142 | */ | |
1143 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 | 1144 | return NULL; |
bf41a158 | 1145 | } |
7a8e76a3 | 1146 | |
bf41a158 SR |
1147 | /* |
1148 | * If the timestamp was commited, make the commit our entry | |
1149 | * now so that we will update it when needed. | |
1150 | */ | |
1151 | if (commit) | |
1152 | rb_set_commit_event(cpu_buffer, event); | |
1153 | else if (!rb_is_commit(cpu_buffer, event)) | |
7a8e76a3 SR |
1154 | delta = 0; |
1155 | ||
1156 | event->time_delta = delta; | |
1157 | ||
1158 | return event; | |
1159 | } | |
1160 | ||
bf41a158 SR |
1161 | static DEFINE_PER_CPU(int, rb_need_resched); |
1162 | ||
7a8e76a3 SR |
1163 | /** |
1164 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
1165 | * @buffer: the ring buffer to reserve from | |
1166 | * @length: the length of the data to reserve (excluding event header) | |
1167 | * @flags: a pointer to save the interrupt flags | |
1168 | * | |
1169 | * Returns a reseverd event on the ring buffer to copy directly to. | |
1170 | * The user of this interface will need to get the body to write into | |
1171 | * and can use the ring_buffer_event_data() interface. | |
1172 | * | |
1173 | * The length is the length of the data needed, not the event length | |
1174 | * which also includes the event header. | |
1175 | * | |
1176 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
1177 | * If NULL is returned, then nothing has been allocated or locked. | |
1178 | */ | |
1179 | struct ring_buffer_event * | |
1180 | ring_buffer_lock_reserve(struct ring_buffer *buffer, | |
1181 | unsigned long length, | |
1182 | unsigned long *flags) | |
1183 | { | |
1184 | struct ring_buffer_per_cpu *cpu_buffer; | |
1185 | struct ring_buffer_event *event; | |
bf41a158 | 1186 | int cpu, resched; |
7a8e76a3 | 1187 | |
a3583244 SR |
1188 | if (ring_buffers_off) |
1189 | return NULL; | |
1190 | ||
7a8e76a3 SR |
1191 | if (atomic_read(&buffer->record_disabled)) |
1192 | return NULL; | |
1193 | ||
bf41a158 | 1194 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 1195 | resched = ftrace_preempt_disable(); |
bf41a158 | 1196 | |
7a8e76a3 SR |
1197 | cpu = raw_smp_processor_id(); |
1198 | ||
1199 | if (!cpu_isset(cpu, buffer->cpumask)) | |
d769041f | 1200 | goto out; |
7a8e76a3 SR |
1201 | |
1202 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1203 | |
1204 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 1205 | goto out; |
7a8e76a3 SR |
1206 | |
1207 | length = rb_calculate_event_length(length); | |
1208 | if (length > BUF_PAGE_SIZE) | |
bf41a158 | 1209 | goto out; |
7a8e76a3 SR |
1210 | |
1211 | event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length); | |
1212 | if (!event) | |
d769041f | 1213 | goto out; |
7a8e76a3 | 1214 | |
bf41a158 SR |
1215 | /* |
1216 | * Need to store resched state on this cpu. | |
1217 | * Only the first needs to. | |
1218 | */ | |
1219 | ||
1220 | if (preempt_count() == 1) | |
1221 | per_cpu(rb_need_resched, cpu) = resched; | |
1222 | ||
7a8e76a3 SR |
1223 | return event; |
1224 | ||
d769041f | 1225 | out: |
182e9f5f | 1226 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1227 | return NULL; |
1228 | } | |
1229 | ||
1230 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
1231 | struct ring_buffer_event *event) | |
1232 | { | |
7a8e76a3 | 1233 | cpu_buffer->entries++; |
bf41a158 SR |
1234 | |
1235 | /* Only process further if we own the commit */ | |
1236 | if (!rb_is_commit(cpu_buffer, event)) | |
1237 | return; | |
1238 | ||
1239 | cpu_buffer->write_stamp += event->time_delta; | |
1240 | ||
1241 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 SR |
1242 | } |
1243 | ||
1244 | /** | |
1245 | * ring_buffer_unlock_commit - commit a reserved | |
1246 | * @buffer: The buffer to commit to | |
1247 | * @event: The event pointer to commit. | |
1248 | * @flags: the interrupt flags received from ring_buffer_lock_reserve. | |
1249 | * | |
1250 | * This commits the data to the ring buffer, and releases any locks held. | |
1251 | * | |
1252 | * Must be paired with ring_buffer_lock_reserve. | |
1253 | */ | |
1254 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
1255 | struct ring_buffer_event *event, | |
1256 | unsigned long flags) | |
1257 | { | |
1258 | struct ring_buffer_per_cpu *cpu_buffer; | |
1259 | int cpu = raw_smp_processor_id(); | |
1260 | ||
1261 | cpu_buffer = buffer->buffers[cpu]; | |
1262 | ||
7a8e76a3 SR |
1263 | rb_commit(cpu_buffer, event); |
1264 | ||
bf41a158 SR |
1265 | /* |
1266 | * Only the last preempt count needs to restore preemption. | |
1267 | */ | |
182e9f5f SR |
1268 | if (preempt_count() == 1) |
1269 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1270 | else | |
bf41a158 | 1271 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
1272 | |
1273 | return 0; | |
1274 | } | |
1275 | ||
1276 | /** | |
1277 | * ring_buffer_write - write data to the buffer without reserving | |
1278 | * @buffer: The ring buffer to write to. | |
1279 | * @length: The length of the data being written (excluding the event header) | |
1280 | * @data: The data to write to the buffer. | |
1281 | * | |
1282 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
1283 | * one function. If you already have the data to write to the buffer, it | |
1284 | * may be easier to simply call this function. | |
1285 | * | |
1286 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
1287 | * and not the length of the event which would hold the header. | |
1288 | */ | |
1289 | int ring_buffer_write(struct ring_buffer *buffer, | |
1290 | unsigned long length, | |
1291 | void *data) | |
1292 | { | |
1293 | struct ring_buffer_per_cpu *cpu_buffer; | |
1294 | struct ring_buffer_event *event; | |
bf41a158 | 1295 | unsigned long event_length; |
7a8e76a3 SR |
1296 | void *body; |
1297 | int ret = -EBUSY; | |
bf41a158 | 1298 | int cpu, resched; |
7a8e76a3 | 1299 | |
a3583244 SR |
1300 | if (ring_buffers_off) |
1301 | return -EBUSY; | |
1302 | ||
7a8e76a3 SR |
1303 | if (atomic_read(&buffer->record_disabled)) |
1304 | return -EBUSY; | |
1305 | ||
182e9f5f | 1306 | resched = ftrace_preempt_disable(); |
bf41a158 | 1307 | |
7a8e76a3 SR |
1308 | cpu = raw_smp_processor_id(); |
1309 | ||
1310 | if (!cpu_isset(cpu, buffer->cpumask)) | |
d769041f | 1311 | goto out; |
7a8e76a3 SR |
1312 | |
1313 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1314 | |
1315 | if (atomic_read(&cpu_buffer->record_disabled)) | |
1316 | goto out; | |
1317 | ||
1318 | event_length = rb_calculate_event_length(length); | |
1319 | event = rb_reserve_next_event(cpu_buffer, | |
1320 | RINGBUF_TYPE_DATA, event_length); | |
1321 | if (!event) | |
1322 | goto out; | |
1323 | ||
1324 | body = rb_event_data(event); | |
1325 | ||
1326 | memcpy(body, data, length); | |
1327 | ||
1328 | rb_commit(cpu_buffer, event); | |
1329 | ||
1330 | ret = 0; | |
1331 | out: | |
182e9f5f | 1332 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1333 | |
1334 | return ret; | |
1335 | } | |
1336 | ||
bf41a158 SR |
1337 | static inline int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
1338 | { | |
1339 | struct buffer_page *reader = cpu_buffer->reader_page; | |
1340 | struct buffer_page *head = cpu_buffer->head_page; | |
1341 | struct buffer_page *commit = cpu_buffer->commit_page; | |
1342 | ||
1343 | return reader->read == rb_page_commit(reader) && | |
1344 | (commit == reader || | |
1345 | (commit == head && | |
1346 | head->read == rb_page_commit(commit))); | |
1347 | } | |
1348 | ||
7a8e76a3 SR |
1349 | /** |
1350 | * ring_buffer_record_disable - stop all writes into the buffer | |
1351 | * @buffer: The ring buffer to stop writes to. | |
1352 | * | |
1353 | * This prevents all writes to the buffer. Any attempt to write | |
1354 | * to the buffer after this will fail and return NULL. | |
1355 | * | |
1356 | * The caller should call synchronize_sched() after this. | |
1357 | */ | |
1358 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
1359 | { | |
1360 | atomic_inc(&buffer->record_disabled); | |
1361 | } | |
1362 | ||
1363 | /** | |
1364 | * ring_buffer_record_enable - enable writes to the buffer | |
1365 | * @buffer: The ring buffer to enable writes | |
1366 | * | |
1367 | * Note, multiple disables will need the same number of enables | |
1368 | * to truely enable the writing (much like preempt_disable). | |
1369 | */ | |
1370 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
1371 | { | |
1372 | atomic_dec(&buffer->record_disabled); | |
1373 | } | |
1374 | ||
1375 | /** | |
1376 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
1377 | * @buffer: The ring buffer to stop writes to. | |
1378 | * @cpu: The CPU buffer to stop | |
1379 | * | |
1380 | * This prevents all writes to the buffer. Any attempt to write | |
1381 | * to the buffer after this will fail and return NULL. | |
1382 | * | |
1383 | * The caller should call synchronize_sched() after this. | |
1384 | */ | |
1385 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
1386 | { | |
1387 | struct ring_buffer_per_cpu *cpu_buffer; | |
1388 | ||
1389 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1390 | return; | |
1391 | ||
1392 | cpu_buffer = buffer->buffers[cpu]; | |
1393 | atomic_inc(&cpu_buffer->record_disabled); | |
1394 | } | |
1395 | ||
1396 | /** | |
1397 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
1398 | * @buffer: The ring buffer to enable writes | |
1399 | * @cpu: The CPU to enable. | |
1400 | * | |
1401 | * Note, multiple disables will need the same number of enables | |
1402 | * to truely enable the writing (much like preempt_disable). | |
1403 | */ | |
1404 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
1405 | { | |
1406 | struct ring_buffer_per_cpu *cpu_buffer; | |
1407 | ||
1408 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1409 | return; | |
1410 | ||
1411 | cpu_buffer = buffer->buffers[cpu]; | |
1412 | atomic_dec(&cpu_buffer->record_disabled); | |
1413 | } | |
1414 | ||
1415 | /** | |
1416 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
1417 | * @buffer: The ring buffer | |
1418 | * @cpu: The per CPU buffer to get the entries from. | |
1419 | */ | |
1420 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
1421 | { | |
1422 | struct ring_buffer_per_cpu *cpu_buffer; | |
1423 | ||
1424 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1425 | return 0; | |
1426 | ||
1427 | cpu_buffer = buffer->buffers[cpu]; | |
1428 | return cpu_buffer->entries; | |
1429 | } | |
1430 | ||
1431 | /** | |
1432 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
1433 | * @buffer: The ring buffer | |
1434 | * @cpu: The per CPU buffer to get the number of overruns from | |
1435 | */ | |
1436 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
1437 | { | |
1438 | struct ring_buffer_per_cpu *cpu_buffer; | |
1439 | ||
1440 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1441 | return 0; | |
1442 | ||
1443 | cpu_buffer = buffer->buffers[cpu]; | |
1444 | return cpu_buffer->overrun; | |
1445 | } | |
1446 | ||
1447 | /** | |
1448 | * ring_buffer_entries - get the number of entries in a buffer | |
1449 | * @buffer: The ring buffer | |
1450 | * | |
1451 | * Returns the total number of entries in the ring buffer | |
1452 | * (all CPU entries) | |
1453 | */ | |
1454 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
1455 | { | |
1456 | struct ring_buffer_per_cpu *cpu_buffer; | |
1457 | unsigned long entries = 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 | entries += cpu_buffer->entries; | |
1464 | } | |
1465 | ||
1466 | return entries; | |
1467 | } | |
1468 | ||
1469 | /** | |
1470 | * ring_buffer_overrun_cpu - get the number of overruns in buffer | |
1471 | * @buffer: The ring buffer | |
1472 | * | |
1473 | * Returns the total number of overruns in the ring buffer | |
1474 | * (all CPU entries) | |
1475 | */ | |
1476 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
1477 | { | |
1478 | struct ring_buffer_per_cpu *cpu_buffer; | |
1479 | unsigned long overruns = 0; | |
1480 | int cpu; | |
1481 | ||
1482 | /* if you care about this being correct, lock the buffer */ | |
1483 | for_each_buffer_cpu(buffer, cpu) { | |
1484 | cpu_buffer = buffer->buffers[cpu]; | |
1485 | overruns += cpu_buffer->overrun; | |
1486 | } | |
1487 | ||
1488 | return overruns; | |
1489 | } | |
1490 | ||
642edba5 | 1491 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
1492 | { |
1493 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1494 | ||
d769041f SR |
1495 | /* Iterator usage is expected to have record disabled */ |
1496 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
1497 | iter->head_page = cpu_buffer->head_page; | |
6f807acd | 1498 | iter->head = cpu_buffer->head_page->read; |
d769041f SR |
1499 | } else { |
1500 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 1501 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
1502 | } |
1503 | if (iter->head) | |
1504 | iter->read_stamp = cpu_buffer->read_stamp; | |
1505 | else | |
1506 | iter->read_stamp = iter->head_page->time_stamp; | |
642edba5 | 1507 | } |
f83c9d0f | 1508 | |
642edba5 SR |
1509 | /** |
1510 | * ring_buffer_iter_reset - reset an iterator | |
1511 | * @iter: The iterator to reset | |
1512 | * | |
1513 | * Resets the iterator, so that it will start from the beginning | |
1514 | * again. | |
1515 | */ | |
1516 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
1517 | { | |
1518 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1519 | unsigned long flags; | |
1520 | ||
1521 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
1522 | rb_iter_reset(iter); | |
f83c9d0f | 1523 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
1524 | } |
1525 | ||
1526 | /** | |
1527 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
1528 | * @iter: The iterator to check | |
1529 | */ | |
1530 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
1531 | { | |
1532 | struct ring_buffer_per_cpu *cpu_buffer; | |
1533 | ||
1534 | cpu_buffer = iter->cpu_buffer; | |
1535 | ||
bf41a158 SR |
1536 | return iter->head_page == cpu_buffer->commit_page && |
1537 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 SR |
1538 | } |
1539 | ||
1540 | static void | |
1541 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1542 | struct ring_buffer_event *event) | |
1543 | { | |
1544 | u64 delta; | |
1545 | ||
1546 | switch (event->type) { | |
1547 | case RINGBUF_TYPE_PADDING: | |
1548 | return; | |
1549 | ||
1550 | case RINGBUF_TYPE_TIME_EXTEND: | |
1551 | delta = event->array[0]; | |
1552 | delta <<= TS_SHIFT; | |
1553 | delta += event->time_delta; | |
1554 | cpu_buffer->read_stamp += delta; | |
1555 | return; | |
1556 | ||
1557 | case RINGBUF_TYPE_TIME_STAMP: | |
1558 | /* FIXME: not implemented */ | |
1559 | return; | |
1560 | ||
1561 | case RINGBUF_TYPE_DATA: | |
1562 | cpu_buffer->read_stamp += event->time_delta; | |
1563 | return; | |
1564 | ||
1565 | default: | |
1566 | BUG(); | |
1567 | } | |
1568 | return; | |
1569 | } | |
1570 | ||
1571 | static void | |
1572 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
1573 | struct ring_buffer_event *event) | |
1574 | { | |
1575 | u64 delta; | |
1576 | ||
1577 | switch (event->type) { | |
1578 | case RINGBUF_TYPE_PADDING: | |
1579 | return; | |
1580 | ||
1581 | case RINGBUF_TYPE_TIME_EXTEND: | |
1582 | delta = event->array[0]; | |
1583 | delta <<= TS_SHIFT; | |
1584 | delta += event->time_delta; | |
1585 | iter->read_stamp += delta; | |
1586 | return; | |
1587 | ||
1588 | case RINGBUF_TYPE_TIME_STAMP: | |
1589 | /* FIXME: not implemented */ | |
1590 | return; | |
1591 | ||
1592 | case RINGBUF_TYPE_DATA: | |
1593 | iter->read_stamp += event->time_delta; | |
1594 | return; | |
1595 | ||
1596 | default: | |
1597 | BUG(); | |
1598 | } | |
1599 | return; | |
1600 | } | |
1601 | ||
d769041f SR |
1602 | static struct buffer_page * |
1603 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 1604 | { |
d769041f SR |
1605 | struct buffer_page *reader = NULL; |
1606 | unsigned long flags; | |
818e3dd3 | 1607 | int nr_loops = 0; |
d769041f | 1608 | |
3e03fb7f SR |
1609 | local_irq_save(flags); |
1610 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
1611 | |
1612 | again: | |
818e3dd3 SR |
1613 | /* |
1614 | * This should normally only loop twice. But because the | |
1615 | * start of the reader inserts an empty page, it causes | |
1616 | * a case where we will loop three times. There should be no | |
1617 | * reason to loop four times (that I know of). | |
1618 | */ | |
3e89c7bb | 1619 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
1620 | reader = NULL; |
1621 | goto out; | |
1622 | } | |
1623 | ||
d769041f SR |
1624 | reader = cpu_buffer->reader_page; |
1625 | ||
1626 | /* If there's more to read, return this page */ | |
bf41a158 | 1627 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
1628 | goto out; |
1629 | ||
1630 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
1631 | if (RB_WARN_ON(cpu_buffer, |
1632 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
1633 | goto out; | |
d769041f SR |
1634 | |
1635 | /* check if we caught up to the tail */ | |
1636 | reader = NULL; | |
bf41a158 | 1637 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 1638 | goto out; |
7a8e76a3 SR |
1639 | |
1640 | /* | |
d769041f SR |
1641 | * Splice the empty reader page into the list around the head. |
1642 | * Reset the reader page to size zero. | |
7a8e76a3 | 1643 | */ |
7a8e76a3 | 1644 | |
d769041f SR |
1645 | reader = cpu_buffer->head_page; |
1646 | cpu_buffer->reader_page->list.next = reader->list.next; | |
1647 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 SR |
1648 | |
1649 | local_set(&cpu_buffer->reader_page->write, 0); | |
1650 | local_set(&cpu_buffer->reader_page->commit, 0); | |
7a8e76a3 | 1651 | |
d769041f SR |
1652 | /* Make the reader page now replace the head */ |
1653 | reader->list.prev->next = &cpu_buffer->reader_page->list; | |
1654 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
7a8e76a3 SR |
1655 | |
1656 | /* | |
d769041f SR |
1657 | * If the tail is on the reader, then we must set the head |
1658 | * to the inserted page, otherwise we set it one before. | |
7a8e76a3 | 1659 | */ |
d769041f | 1660 | cpu_buffer->head_page = cpu_buffer->reader_page; |
7a8e76a3 | 1661 | |
bf41a158 | 1662 | if (cpu_buffer->commit_page != reader) |
d769041f SR |
1663 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
1664 | ||
1665 | /* Finally update the reader page to the new head */ | |
1666 | cpu_buffer->reader_page = reader; | |
1667 | rb_reset_reader_page(cpu_buffer); | |
1668 | ||
1669 | goto again; | |
1670 | ||
1671 | out: | |
3e03fb7f SR |
1672 | __raw_spin_unlock(&cpu_buffer->lock); |
1673 | local_irq_restore(flags); | |
d769041f SR |
1674 | |
1675 | return reader; | |
1676 | } | |
1677 | ||
1678 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
1679 | { | |
1680 | struct ring_buffer_event *event; | |
1681 | struct buffer_page *reader; | |
1682 | unsigned length; | |
1683 | ||
1684 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 1685 | |
d769041f | 1686 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
1687 | if (RB_WARN_ON(cpu_buffer, !reader)) |
1688 | return; | |
7a8e76a3 | 1689 | |
d769041f SR |
1690 | event = rb_reader_event(cpu_buffer); |
1691 | ||
1692 | if (event->type == RINGBUF_TYPE_DATA) | |
1693 | cpu_buffer->entries--; | |
1694 | ||
1695 | rb_update_read_stamp(cpu_buffer, event); | |
1696 | ||
1697 | length = rb_event_length(event); | |
6f807acd | 1698 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
1699 | } |
1700 | ||
1701 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
1702 | { | |
1703 | struct ring_buffer *buffer; | |
1704 | struct ring_buffer_per_cpu *cpu_buffer; | |
1705 | struct ring_buffer_event *event; | |
1706 | unsigned length; | |
1707 | ||
1708 | cpu_buffer = iter->cpu_buffer; | |
1709 | buffer = cpu_buffer->buffer; | |
1710 | ||
1711 | /* | |
1712 | * Check if we are at the end of the buffer. | |
1713 | */ | |
bf41a158 | 1714 | if (iter->head >= rb_page_size(iter->head_page)) { |
3e89c7bb SR |
1715 | if (RB_WARN_ON(buffer, |
1716 | iter->head_page == cpu_buffer->commit_page)) | |
1717 | return; | |
d769041f | 1718 | rb_inc_iter(iter); |
7a8e76a3 SR |
1719 | return; |
1720 | } | |
1721 | ||
1722 | event = rb_iter_head_event(iter); | |
1723 | ||
1724 | length = rb_event_length(event); | |
1725 | ||
1726 | /* | |
1727 | * This should not be called to advance the header if we are | |
1728 | * at the tail of the buffer. | |
1729 | */ | |
3e89c7bb | 1730 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 1731 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
1732 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
1733 | return; | |
7a8e76a3 SR |
1734 | |
1735 | rb_update_iter_read_stamp(iter, event); | |
1736 | ||
1737 | iter->head += length; | |
1738 | ||
1739 | /* check for end of page padding */ | |
bf41a158 SR |
1740 | if ((iter->head >= rb_page_size(iter->head_page)) && |
1741 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
1742 | rb_advance_iter(iter); |
1743 | } | |
1744 | ||
f83c9d0f SR |
1745 | static struct ring_buffer_event * |
1746 | rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
7a8e76a3 SR |
1747 | { |
1748 | struct ring_buffer_per_cpu *cpu_buffer; | |
1749 | struct ring_buffer_event *event; | |
d769041f | 1750 | struct buffer_page *reader; |
818e3dd3 | 1751 | int nr_loops = 0; |
7a8e76a3 SR |
1752 | |
1753 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1754 | return NULL; | |
1755 | ||
1756 | cpu_buffer = buffer->buffers[cpu]; | |
1757 | ||
1758 | again: | |
818e3dd3 SR |
1759 | /* |
1760 | * We repeat when a timestamp is encountered. It is possible | |
1761 | * to get multiple timestamps from an interrupt entering just | |
1762 | * as one timestamp is about to be written. The max times | |
1763 | * that this can happen is the number of nested interrupts we | |
1764 | * can have. Nesting 10 deep of interrupts is clearly | |
1765 | * an anomaly. | |
1766 | */ | |
3e89c7bb | 1767 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 1768 | return NULL; |
818e3dd3 | 1769 | |
d769041f SR |
1770 | reader = rb_get_reader_page(cpu_buffer); |
1771 | if (!reader) | |
7a8e76a3 SR |
1772 | return NULL; |
1773 | ||
d769041f | 1774 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 SR |
1775 | |
1776 | switch (event->type) { | |
1777 | case RINGBUF_TYPE_PADDING: | |
bf41a158 | 1778 | RB_WARN_ON(cpu_buffer, 1); |
d769041f SR |
1779 | rb_advance_reader(cpu_buffer); |
1780 | return NULL; | |
7a8e76a3 SR |
1781 | |
1782 | case RINGBUF_TYPE_TIME_EXTEND: | |
1783 | /* Internal data, OK to advance */ | |
d769041f | 1784 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
1785 | goto again; |
1786 | ||
1787 | case RINGBUF_TYPE_TIME_STAMP: | |
1788 | /* FIXME: not implemented */ | |
d769041f | 1789 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
1790 | goto again; |
1791 | ||
1792 | case RINGBUF_TYPE_DATA: | |
1793 | if (ts) { | |
1794 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
1795 | ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts); | |
1796 | } | |
1797 | return event; | |
1798 | ||
1799 | default: | |
1800 | BUG(); | |
1801 | } | |
1802 | ||
1803 | return NULL; | |
1804 | } | |
1805 | ||
f83c9d0f SR |
1806 | static struct ring_buffer_event * |
1807 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
1808 | { |
1809 | struct ring_buffer *buffer; | |
1810 | struct ring_buffer_per_cpu *cpu_buffer; | |
1811 | struct ring_buffer_event *event; | |
818e3dd3 | 1812 | int nr_loops = 0; |
7a8e76a3 SR |
1813 | |
1814 | if (ring_buffer_iter_empty(iter)) | |
1815 | return NULL; | |
1816 | ||
1817 | cpu_buffer = iter->cpu_buffer; | |
1818 | buffer = cpu_buffer->buffer; | |
1819 | ||
1820 | again: | |
818e3dd3 SR |
1821 | /* |
1822 | * We repeat when a timestamp is encountered. It is possible | |
1823 | * to get multiple timestamps from an interrupt entering just | |
1824 | * as one timestamp is about to be written. The max times | |
1825 | * that this can happen is the number of nested interrupts we | |
1826 | * can have. Nesting 10 deep of interrupts is clearly | |
1827 | * an anomaly. | |
1828 | */ | |
3e89c7bb | 1829 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 1830 | return NULL; |
818e3dd3 | 1831 | |
7a8e76a3 SR |
1832 | if (rb_per_cpu_empty(cpu_buffer)) |
1833 | return NULL; | |
1834 | ||
1835 | event = rb_iter_head_event(iter); | |
1836 | ||
1837 | switch (event->type) { | |
1838 | case RINGBUF_TYPE_PADDING: | |
d769041f | 1839 | rb_inc_iter(iter); |
7a8e76a3 SR |
1840 | goto again; |
1841 | ||
1842 | case RINGBUF_TYPE_TIME_EXTEND: | |
1843 | /* Internal data, OK to advance */ | |
1844 | rb_advance_iter(iter); | |
1845 | goto again; | |
1846 | ||
1847 | case RINGBUF_TYPE_TIME_STAMP: | |
1848 | /* FIXME: not implemented */ | |
1849 | rb_advance_iter(iter); | |
1850 | goto again; | |
1851 | ||
1852 | case RINGBUF_TYPE_DATA: | |
1853 | if (ts) { | |
1854 | *ts = iter->read_stamp + event->time_delta; | |
1855 | ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts); | |
1856 | } | |
1857 | return event; | |
1858 | ||
1859 | default: | |
1860 | BUG(); | |
1861 | } | |
1862 | ||
1863 | return NULL; | |
1864 | } | |
1865 | ||
f83c9d0f SR |
1866 | /** |
1867 | * ring_buffer_peek - peek at the next event to be read | |
1868 | * @buffer: The ring buffer to read | |
1869 | * @cpu: The cpu to peak at | |
1870 | * @ts: The timestamp counter of this event. | |
1871 | * | |
1872 | * This will return the event that will be read next, but does | |
1873 | * not consume the data. | |
1874 | */ | |
1875 | struct ring_buffer_event * | |
1876 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
1877 | { | |
1878 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
1879 | struct ring_buffer_event *event; | |
1880 | unsigned long flags; | |
1881 | ||
1882 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
1883 | event = rb_buffer_peek(buffer, cpu, ts); | |
1884 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
1885 | ||
1886 | return event; | |
1887 | } | |
1888 | ||
1889 | /** | |
1890 | * ring_buffer_iter_peek - peek at the next event to be read | |
1891 | * @iter: The ring buffer iterator | |
1892 | * @ts: The timestamp counter of this event. | |
1893 | * | |
1894 | * This will return the event that will be read next, but does | |
1895 | * not increment the iterator. | |
1896 | */ | |
1897 | struct ring_buffer_event * | |
1898 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
1899 | { | |
1900 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1901 | struct ring_buffer_event *event; | |
1902 | unsigned long flags; | |
1903 | ||
1904 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
1905 | event = rb_iter_peek(iter, ts); | |
1906 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
1907 | ||
1908 | return event; | |
1909 | } | |
1910 | ||
7a8e76a3 SR |
1911 | /** |
1912 | * ring_buffer_consume - return an event and consume it | |
1913 | * @buffer: The ring buffer to get the next event from | |
1914 | * | |
1915 | * Returns the next event in the ring buffer, and that event is consumed. | |
1916 | * Meaning, that sequential reads will keep returning a different event, | |
1917 | * and eventually empty the ring buffer if the producer is slower. | |
1918 | */ | |
1919 | struct ring_buffer_event * | |
1920 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
1921 | { | |
f83c9d0f | 1922 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; |
7a8e76a3 | 1923 | struct ring_buffer_event *event; |
f83c9d0f | 1924 | unsigned long flags; |
7a8e76a3 SR |
1925 | |
1926 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1927 | return NULL; | |
1928 | ||
f83c9d0f SR |
1929 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
1930 | ||
1931 | event = rb_buffer_peek(buffer, cpu, ts); | |
7a8e76a3 | 1932 | if (!event) |
f83c9d0f | 1933 | goto out; |
7a8e76a3 | 1934 | |
d769041f | 1935 | rb_advance_reader(cpu_buffer); |
7a8e76a3 | 1936 | |
f83c9d0f SR |
1937 | out: |
1938 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
1939 | ||
7a8e76a3 SR |
1940 | return event; |
1941 | } | |
1942 | ||
1943 | /** | |
1944 | * ring_buffer_read_start - start a non consuming read of the buffer | |
1945 | * @buffer: The ring buffer to read from | |
1946 | * @cpu: The cpu buffer to iterate over | |
1947 | * | |
1948 | * This starts up an iteration through the buffer. It also disables | |
1949 | * the recording to the buffer until the reading is finished. | |
1950 | * This prevents the reading from being corrupted. This is not | |
1951 | * a consuming read, so a producer is not expected. | |
1952 | * | |
1953 | * Must be paired with ring_buffer_finish. | |
1954 | */ | |
1955 | struct ring_buffer_iter * | |
1956 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
1957 | { | |
1958 | struct ring_buffer_per_cpu *cpu_buffer; | |
1959 | struct ring_buffer_iter *iter; | |
d769041f | 1960 | unsigned long flags; |
7a8e76a3 SR |
1961 | |
1962 | if (!cpu_isset(cpu, buffer->cpumask)) | |
1963 | return NULL; | |
1964 | ||
1965 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
1966 | if (!iter) | |
1967 | return NULL; | |
1968 | ||
1969 | cpu_buffer = buffer->buffers[cpu]; | |
1970 | ||
1971 | iter->cpu_buffer = cpu_buffer; | |
1972 | ||
1973 | atomic_inc(&cpu_buffer->record_disabled); | |
1974 | synchronize_sched(); | |
1975 | ||
f83c9d0f | 1976 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 1977 | __raw_spin_lock(&cpu_buffer->lock); |
642edba5 | 1978 | rb_iter_reset(iter); |
3e03fb7f | 1979 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 1980 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
1981 | |
1982 | return iter; | |
1983 | } | |
1984 | ||
1985 | /** | |
1986 | * ring_buffer_finish - finish reading the iterator of the buffer | |
1987 | * @iter: The iterator retrieved by ring_buffer_start | |
1988 | * | |
1989 | * This re-enables the recording to the buffer, and frees the | |
1990 | * iterator. | |
1991 | */ | |
1992 | void | |
1993 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
1994 | { | |
1995 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1996 | ||
1997 | atomic_dec(&cpu_buffer->record_disabled); | |
1998 | kfree(iter); | |
1999 | } | |
2000 | ||
2001 | /** | |
2002 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
2003 | * @iter: The ring buffer iterator | |
2004 | * @ts: The time stamp of the event read. | |
2005 | * | |
2006 | * This reads the next event in the ring buffer and increments the iterator. | |
2007 | */ | |
2008 | struct ring_buffer_event * | |
2009 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
2010 | { | |
2011 | struct ring_buffer_event *event; | |
f83c9d0f SR |
2012 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
2013 | unsigned long flags; | |
7a8e76a3 | 2014 | |
f83c9d0f SR |
2015 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2016 | event = rb_iter_peek(iter, ts); | |
7a8e76a3 | 2017 | if (!event) |
f83c9d0f | 2018 | goto out; |
7a8e76a3 SR |
2019 | |
2020 | rb_advance_iter(iter); | |
f83c9d0f SR |
2021 | out: |
2022 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 SR |
2023 | |
2024 | return event; | |
2025 | } | |
2026 | ||
2027 | /** | |
2028 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
2029 | * @buffer: The ring buffer. | |
2030 | */ | |
2031 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
2032 | { | |
2033 | return BUF_PAGE_SIZE * buffer->pages; | |
2034 | } | |
2035 | ||
2036 | static void | |
2037 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
2038 | { | |
2039 | cpu_buffer->head_page | |
2040 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 SR |
2041 | local_set(&cpu_buffer->head_page->write, 0); |
2042 | local_set(&cpu_buffer->head_page->commit, 0); | |
d769041f | 2043 | |
6f807acd | 2044 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
2045 | |
2046 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
2047 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
2048 | ||
2049 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
2050 | local_set(&cpu_buffer->reader_page->write, 0); | |
2051 | local_set(&cpu_buffer->reader_page->commit, 0); | |
6f807acd | 2052 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 2053 | |
7a8e76a3 SR |
2054 | cpu_buffer->overrun = 0; |
2055 | cpu_buffer->entries = 0; | |
2056 | } | |
2057 | ||
2058 | /** | |
2059 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
2060 | * @buffer: The ring buffer to reset a per cpu buffer of | |
2061 | * @cpu: The CPU buffer to be reset | |
2062 | */ | |
2063 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
2064 | { | |
2065 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2066 | unsigned long flags; | |
2067 | ||
2068 | if (!cpu_isset(cpu, buffer->cpumask)) | |
2069 | return; | |
2070 | ||
f83c9d0f SR |
2071 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2072 | ||
3e03fb7f | 2073 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
2074 | |
2075 | rb_reset_cpu(cpu_buffer); | |
2076 | ||
3e03fb7f | 2077 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f SR |
2078 | |
2079 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 SR |
2080 | } |
2081 | ||
2082 | /** | |
2083 | * ring_buffer_reset - reset a ring buffer | |
2084 | * @buffer: The ring buffer to reset all cpu buffers | |
2085 | */ | |
2086 | void ring_buffer_reset(struct ring_buffer *buffer) | |
2087 | { | |
7a8e76a3 SR |
2088 | int cpu; |
2089 | ||
7a8e76a3 | 2090 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 2091 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 SR |
2092 | } |
2093 | ||
2094 | /** | |
2095 | * rind_buffer_empty - is the ring buffer empty? | |
2096 | * @buffer: The ring buffer to test | |
2097 | */ | |
2098 | int ring_buffer_empty(struct ring_buffer *buffer) | |
2099 | { | |
2100 | struct ring_buffer_per_cpu *cpu_buffer; | |
2101 | int cpu; | |
2102 | ||
2103 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
2104 | for_each_buffer_cpu(buffer, cpu) { | |
2105 | cpu_buffer = buffer->buffers[cpu]; | |
2106 | if (!rb_per_cpu_empty(cpu_buffer)) | |
2107 | return 0; | |
2108 | } | |
2109 | return 1; | |
2110 | } | |
2111 | ||
2112 | /** | |
2113 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
2114 | * @buffer: The ring buffer | |
2115 | * @cpu: The CPU buffer to test | |
2116 | */ | |
2117 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
2118 | { | |
2119 | struct ring_buffer_per_cpu *cpu_buffer; | |
2120 | ||
2121 | if (!cpu_isset(cpu, buffer->cpumask)) | |
2122 | return 1; | |
2123 | ||
2124 | cpu_buffer = buffer->buffers[cpu]; | |
2125 | return rb_per_cpu_empty(cpu_buffer); | |
2126 | } | |
2127 | ||
2128 | /** | |
2129 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
2130 | * @buffer_a: One buffer to swap with | |
2131 | * @buffer_b: The other buffer to swap with | |
2132 | * | |
2133 | * This function is useful for tracers that want to take a "snapshot" | |
2134 | * of a CPU buffer and has another back up buffer lying around. | |
2135 | * it is expected that the tracer handles the cpu buffer not being | |
2136 | * used at the moment. | |
2137 | */ | |
2138 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
2139 | struct ring_buffer *buffer_b, int cpu) | |
2140 | { | |
2141 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
2142 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
2143 | ||
2144 | if (!cpu_isset(cpu, buffer_a->cpumask) || | |
2145 | !cpu_isset(cpu, buffer_b->cpumask)) | |
2146 | return -EINVAL; | |
2147 | ||
2148 | /* At least make sure the two buffers are somewhat the same */ | |
2149 | if (buffer_a->size != buffer_b->size || | |
2150 | buffer_a->pages != buffer_b->pages) | |
2151 | return -EINVAL; | |
2152 | ||
2153 | cpu_buffer_a = buffer_a->buffers[cpu]; | |
2154 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
2155 | ||
2156 | /* | |
2157 | * We can't do a synchronize_sched here because this | |
2158 | * function can be called in atomic context. | |
2159 | * Normally this will be called from the same CPU as cpu. | |
2160 | * If not it's up to the caller to protect this. | |
2161 | */ | |
2162 | atomic_inc(&cpu_buffer_a->record_disabled); | |
2163 | atomic_inc(&cpu_buffer_b->record_disabled); | |
2164 | ||
2165 | buffer_a->buffers[cpu] = cpu_buffer_b; | |
2166 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
2167 | ||
2168 | cpu_buffer_b->buffer = buffer_a; | |
2169 | cpu_buffer_a->buffer = buffer_b; | |
2170 | ||
2171 | atomic_dec(&cpu_buffer_a->record_disabled); | |
2172 | atomic_dec(&cpu_buffer_b->record_disabled); | |
2173 | ||
2174 | return 0; | |
2175 | } | |
2176 | ||
a3583244 SR |
2177 | static ssize_t |
2178 | rb_simple_read(struct file *filp, char __user *ubuf, | |
2179 | size_t cnt, loff_t *ppos) | |
2180 | { | |
2181 | int *p = filp->private_data; | |
2182 | char buf[64]; | |
2183 | int r; | |
2184 | ||
2185 | /* !ring_buffers_off == tracing_on */ | |
2186 | r = sprintf(buf, "%d\n", !*p); | |
2187 | ||
2188 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
2189 | } | |
2190 | ||
2191 | static ssize_t | |
2192 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
2193 | size_t cnt, loff_t *ppos) | |
2194 | { | |
2195 | int *p = filp->private_data; | |
2196 | char buf[64]; | |
2197 | long val; | |
2198 | int ret; | |
2199 | ||
2200 | if (cnt >= sizeof(buf)) | |
2201 | return -EINVAL; | |
2202 | ||
2203 | if (copy_from_user(&buf, ubuf, cnt)) | |
2204 | return -EFAULT; | |
2205 | ||
2206 | buf[cnt] = 0; | |
2207 | ||
2208 | ret = strict_strtoul(buf, 10, &val); | |
2209 | if (ret < 0) | |
2210 | return ret; | |
2211 | ||
2212 | /* !ring_buffers_off == tracing_on */ | |
2213 | *p = !val; | |
2214 | ||
2215 | (*ppos)++; | |
2216 | ||
2217 | return cnt; | |
2218 | } | |
2219 | ||
2220 | static struct file_operations rb_simple_fops = { | |
2221 | .open = tracing_open_generic, | |
2222 | .read = rb_simple_read, | |
2223 | .write = rb_simple_write, | |
2224 | }; | |
2225 | ||
2226 | ||
2227 | static __init int rb_init_debugfs(void) | |
2228 | { | |
2229 | struct dentry *d_tracer; | |
2230 | struct dentry *entry; | |
2231 | ||
2232 | d_tracer = tracing_init_dentry(); | |
2233 | ||
2234 | entry = debugfs_create_file("tracing_on", 0644, d_tracer, | |
2235 | &ring_buffers_off, &rb_simple_fops); | |
2236 | if (!entry) | |
2237 | pr_warning("Could not create debugfs 'tracing_on' entry\n"); | |
2238 | ||
2239 | return 0; | |
2240 | } | |
2241 | ||
2242 | fs_initcall(rb_init_debugfs); |