]>
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> | |
78d904b4 | 7 | #include <linux/ftrace_irq.h> |
7a8e76a3 SR |
8 | #include <linux/spinlock.h> |
9 | #include <linux/debugfs.h> | |
10 | #include <linux/uaccess.h> | |
a81bd80a | 11 | #include <linux/hardirq.h> |
7a8e76a3 SR |
12 | #include <linux/module.h> |
13 | #include <linux/percpu.h> | |
14 | #include <linux/mutex.h> | |
15 | #include <linux/sched.h> /* used for sched_clock() (for now) */ | |
16 | #include <linux/init.h> | |
17 | #include <linux/hash.h> | |
18 | #include <linux/list.h> | |
19 | #include <linux/fs.h> | |
20 | ||
182e9f5f SR |
21 | #include "trace.h" |
22 | ||
033601a3 SR |
23 | /* |
24 | * A fast way to enable or disable all ring buffers is to | |
25 | * call tracing_on or tracing_off. Turning off the ring buffers | |
26 | * prevents all ring buffers from being recorded to. | |
27 | * Turning this switch on, makes it OK to write to the | |
28 | * ring buffer, if the ring buffer is enabled itself. | |
29 | * | |
30 | * There's three layers that must be on in order to write | |
31 | * to the ring buffer. | |
32 | * | |
33 | * 1) This global flag must be set. | |
34 | * 2) The ring buffer must be enabled for recording. | |
35 | * 3) The per cpu buffer must be enabled for recording. | |
36 | * | |
37 | * In case of an anomaly, this global flag has a bit set that | |
38 | * will permantly disable all ring buffers. | |
39 | */ | |
40 | ||
41 | /* | |
42 | * Global flag to disable all recording to ring buffers | |
43 | * This has two bits: ON, DISABLED | |
44 | * | |
45 | * ON DISABLED | |
46 | * ---- ---------- | |
47 | * 0 0 : ring buffers are off | |
48 | * 1 0 : ring buffers are on | |
49 | * X 1 : ring buffers are permanently disabled | |
50 | */ | |
51 | ||
52 | enum { | |
53 | RB_BUFFERS_ON_BIT = 0, | |
54 | RB_BUFFERS_DISABLED_BIT = 1, | |
55 | }; | |
56 | ||
57 | enum { | |
58 | RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT, | |
59 | RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT, | |
60 | }; | |
61 | ||
5e39841c | 62 | static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON; |
a3583244 SR |
63 | |
64 | /** | |
65 | * tracing_on - enable all tracing buffers | |
66 | * | |
67 | * This function enables all tracing buffers that may have been | |
68 | * disabled with tracing_off. | |
69 | */ | |
70 | void tracing_on(void) | |
71 | { | |
033601a3 | 72 | set_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags); |
a3583244 | 73 | } |
c4f50183 | 74 | EXPORT_SYMBOL_GPL(tracing_on); |
a3583244 SR |
75 | |
76 | /** | |
77 | * tracing_off - turn off all tracing buffers | |
78 | * | |
79 | * This function stops all tracing buffers from recording data. | |
80 | * It does not disable any overhead the tracers themselves may | |
81 | * be causing. This function simply causes all recording to | |
82 | * the ring buffers to fail. | |
83 | */ | |
84 | void tracing_off(void) | |
85 | { | |
033601a3 SR |
86 | clear_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags); |
87 | } | |
c4f50183 | 88 | EXPORT_SYMBOL_GPL(tracing_off); |
033601a3 SR |
89 | |
90 | /** | |
91 | * tracing_off_permanent - permanently disable ring buffers | |
92 | * | |
93 | * This function, once called, will disable all ring buffers | |
c3706f00 | 94 | * permanently. |
033601a3 SR |
95 | */ |
96 | void tracing_off_permanent(void) | |
97 | { | |
98 | set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags); | |
a3583244 SR |
99 | } |
100 | ||
988ae9d6 SR |
101 | /** |
102 | * tracing_is_on - show state of ring buffers enabled | |
103 | */ | |
104 | int tracing_is_on(void) | |
105 | { | |
106 | return ring_buffer_flags == RB_BUFFERS_ON; | |
107 | } | |
108 | EXPORT_SYMBOL_GPL(tracing_is_on); | |
109 | ||
d06bbd66 IM |
110 | #include "trace.h" |
111 | ||
7a8e76a3 SR |
112 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
113 | #define DEBUG_SHIFT 0 | |
114 | ||
115 | /* FIXME!!! */ | |
116 | u64 ring_buffer_time_stamp(int cpu) | |
117 | { | |
47e74f2b SR |
118 | u64 time; |
119 | ||
120 | preempt_disable_notrace(); | |
7a8e76a3 | 121 | /* shift to debug/test normalization and TIME_EXTENTS */ |
47e74f2b | 122 | time = sched_clock() << DEBUG_SHIFT; |
2c2d7329 | 123 | preempt_enable_no_resched_notrace(); |
47e74f2b SR |
124 | |
125 | return time; | |
7a8e76a3 | 126 | } |
c4f50183 | 127 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); |
7a8e76a3 SR |
128 | |
129 | void ring_buffer_normalize_time_stamp(int cpu, u64 *ts) | |
130 | { | |
131 | /* Just stupid testing the normalize function and deltas */ | |
132 | *ts >>= DEBUG_SHIFT; | |
133 | } | |
c4f50183 | 134 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); |
7a8e76a3 SR |
135 | |
136 | #define RB_EVNT_HDR_SIZE (sizeof(struct ring_buffer_event)) | |
67d34724 | 137 | #define RB_ALIGNMENT 4U |
7a8e76a3 SR |
138 | #define RB_MAX_SMALL_DATA 28 |
139 | ||
140 | enum { | |
141 | RB_LEN_TIME_EXTEND = 8, | |
142 | RB_LEN_TIME_STAMP = 16, | |
143 | }; | |
144 | ||
145 | /* inline for ring buffer fast paths */ | |
34a148bf | 146 | static unsigned |
7a8e76a3 SR |
147 | rb_event_length(struct ring_buffer_event *event) |
148 | { | |
149 | unsigned length; | |
150 | ||
151 | switch (event->type) { | |
152 | case RINGBUF_TYPE_PADDING: | |
153 | /* undefined */ | |
154 | return -1; | |
155 | ||
156 | case RINGBUF_TYPE_TIME_EXTEND: | |
157 | return RB_LEN_TIME_EXTEND; | |
158 | ||
159 | case RINGBUF_TYPE_TIME_STAMP: | |
160 | return RB_LEN_TIME_STAMP; | |
161 | ||
162 | case RINGBUF_TYPE_DATA: | |
163 | if (event->len) | |
67d34724 | 164 | length = event->len * RB_ALIGNMENT; |
7a8e76a3 SR |
165 | else |
166 | length = event->array[0]; | |
167 | return length + RB_EVNT_HDR_SIZE; | |
168 | default: | |
169 | BUG(); | |
170 | } | |
171 | /* not hit */ | |
172 | return 0; | |
173 | } | |
174 | ||
175 | /** | |
176 | * ring_buffer_event_length - return the length of the event | |
177 | * @event: the event to get the length of | |
178 | */ | |
179 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
180 | { | |
465634ad RR |
181 | unsigned length = rb_event_length(event); |
182 | if (event->type != RINGBUF_TYPE_DATA) | |
183 | return length; | |
184 | length -= RB_EVNT_HDR_SIZE; | |
185 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
186 | length -= sizeof(event->array[0]); | |
187 | return length; | |
7a8e76a3 | 188 | } |
c4f50183 | 189 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
190 | |
191 | /* inline for ring buffer fast paths */ | |
34a148bf | 192 | static void * |
7a8e76a3 SR |
193 | rb_event_data(struct ring_buffer_event *event) |
194 | { | |
195 | BUG_ON(event->type != RINGBUF_TYPE_DATA); | |
196 | /* If length is in len field, then array[0] has the data */ | |
197 | if (event->len) | |
198 | return (void *)&event->array[0]; | |
199 | /* Otherwise length is in array[0] and array[1] has the data */ | |
200 | return (void *)&event->array[1]; | |
201 | } | |
202 | ||
203 | /** | |
204 | * ring_buffer_event_data - return the data of the event | |
205 | * @event: the event to get the data from | |
206 | */ | |
207 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
208 | { | |
209 | return rb_event_data(event); | |
210 | } | |
c4f50183 | 211 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
212 | |
213 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 214 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
215 | |
216 | #define TS_SHIFT 27 | |
217 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
218 | #define TS_DELTA_TEST (~TS_MASK) | |
219 | ||
abc9b56d | 220 | struct buffer_data_page { |
e4c2ce82 | 221 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 222 | local_t commit; /* write committed index */ |
abc9b56d SR |
223 | unsigned char data[]; /* data of buffer page */ |
224 | }; | |
225 | ||
226 | struct buffer_page { | |
227 | local_t write; /* index for next write */ | |
6f807acd | 228 | unsigned read; /* index for next read */ |
e4c2ce82 | 229 | struct list_head list; /* list of free pages */ |
abc9b56d | 230 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
231 | }; |
232 | ||
044fa782 | 233 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 234 | { |
044fa782 | 235 | local_set(&bpage->commit, 0); |
abc9b56d SR |
236 | } |
237 | ||
ed56829c SR |
238 | /* |
239 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
240 | * this issue out. | |
241 | */ | |
34a148bf | 242 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 243 | { |
34a148bf | 244 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 245 | kfree(bpage); |
ed56829c SR |
246 | } |
247 | ||
7a8e76a3 SR |
248 | /* |
249 | * We need to fit the time_stamp delta into 27 bits. | |
250 | */ | |
251 | static inline int test_time_stamp(u64 delta) | |
252 | { | |
253 | if (delta & TS_DELTA_TEST) | |
254 | return 1; | |
255 | return 0; | |
256 | } | |
257 | ||
082605de | 258 | #define BUF_PAGE_SIZE (PAGE_SIZE - offsetof(struct buffer_data_page, data)) |
7a8e76a3 SR |
259 | |
260 | /* | |
261 | * head_page == tail_page && head == tail then buffer is empty. | |
262 | */ | |
263 | struct ring_buffer_per_cpu { | |
264 | int cpu; | |
265 | struct ring_buffer *buffer; | |
f83c9d0f | 266 | spinlock_t reader_lock; /* serialize readers */ |
3e03fb7f | 267 | raw_spinlock_t lock; |
7a8e76a3 SR |
268 | struct lock_class_key lock_key; |
269 | struct list_head pages; | |
6f807acd SR |
270 | struct buffer_page *head_page; /* read from head */ |
271 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 272 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 273 | struct buffer_page *reader_page; |
7a8e76a3 SR |
274 | unsigned long overrun; |
275 | unsigned long entries; | |
276 | u64 write_stamp; | |
277 | u64 read_stamp; | |
278 | atomic_t record_disabled; | |
279 | }; | |
280 | ||
281 | struct ring_buffer { | |
7a8e76a3 SR |
282 | unsigned pages; |
283 | unsigned flags; | |
284 | int cpus; | |
7a8e76a3 | 285 | atomic_t record_disabled; |
00f62f61 | 286 | cpumask_var_t cpumask; |
7a8e76a3 SR |
287 | |
288 | struct mutex mutex; | |
289 | ||
290 | struct ring_buffer_per_cpu **buffers; | |
291 | }; | |
292 | ||
293 | struct ring_buffer_iter { | |
294 | struct ring_buffer_per_cpu *cpu_buffer; | |
295 | unsigned long head; | |
296 | struct buffer_page *head_page; | |
297 | u64 read_stamp; | |
298 | }; | |
299 | ||
f536aafc | 300 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
bf41a158 | 301 | #define RB_WARN_ON(buffer, cond) \ |
3e89c7bb SR |
302 | ({ \ |
303 | int _____ret = unlikely(cond); \ | |
304 | if (_____ret) { \ | |
bf41a158 SR |
305 | atomic_inc(&buffer->record_disabled); \ |
306 | WARN_ON(1); \ | |
307 | } \ | |
3e89c7bb SR |
308 | _____ret; \ |
309 | }) | |
f536aafc | 310 | |
7a8e76a3 SR |
311 | /** |
312 | * check_pages - integrity check of buffer pages | |
313 | * @cpu_buffer: CPU buffer with pages to test | |
314 | * | |
c3706f00 | 315 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
316 | * been corrupted. |
317 | */ | |
318 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
319 | { | |
320 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 321 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 322 | |
3e89c7bb SR |
323 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
324 | return -1; | |
325 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
326 | return -1; | |
7a8e76a3 | 327 | |
044fa782 | 328 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 329 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 330 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
331 | return -1; |
332 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 333 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 334 | return -1; |
7a8e76a3 SR |
335 | } |
336 | ||
337 | return 0; | |
338 | } | |
339 | ||
7a8e76a3 SR |
340 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
341 | unsigned nr_pages) | |
342 | { | |
343 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 344 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
345 | unsigned long addr; |
346 | LIST_HEAD(pages); | |
347 | unsigned i; | |
348 | ||
349 | for (i = 0; i < nr_pages; i++) { | |
044fa782 | 350 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
aa1e0e3b | 351 | GFP_KERNEL, cpu_to_node(cpu_buffer->cpu)); |
044fa782 | 352 | if (!bpage) |
e4c2ce82 | 353 | goto free_pages; |
044fa782 | 354 | list_add(&bpage->list, &pages); |
e4c2ce82 | 355 | |
7a8e76a3 SR |
356 | addr = __get_free_page(GFP_KERNEL); |
357 | if (!addr) | |
358 | goto free_pages; | |
044fa782 SR |
359 | bpage->page = (void *)addr; |
360 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
361 | } |
362 | ||
363 | list_splice(&pages, head); | |
364 | ||
365 | rb_check_pages(cpu_buffer); | |
366 | ||
367 | return 0; | |
368 | ||
369 | free_pages: | |
044fa782 SR |
370 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
371 | list_del_init(&bpage->list); | |
372 | free_buffer_page(bpage); | |
7a8e76a3 SR |
373 | } |
374 | return -ENOMEM; | |
375 | } | |
376 | ||
377 | static struct ring_buffer_per_cpu * | |
378 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) | |
379 | { | |
380 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 381 | struct buffer_page *bpage; |
d769041f | 382 | unsigned long addr; |
7a8e76a3 SR |
383 | int ret; |
384 | ||
385 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
386 | GFP_KERNEL, cpu_to_node(cpu)); | |
387 | if (!cpu_buffer) | |
388 | return NULL; | |
389 | ||
390 | cpu_buffer->cpu = cpu; | |
391 | cpu_buffer->buffer = buffer; | |
f83c9d0f | 392 | spin_lock_init(&cpu_buffer->reader_lock); |
3e03fb7f | 393 | cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; |
7a8e76a3 SR |
394 | INIT_LIST_HEAD(&cpu_buffer->pages); |
395 | ||
044fa782 | 396 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 397 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 398 | if (!bpage) |
e4c2ce82 SR |
399 | goto fail_free_buffer; |
400 | ||
044fa782 | 401 | cpu_buffer->reader_page = bpage; |
d769041f SR |
402 | addr = __get_free_page(GFP_KERNEL); |
403 | if (!addr) | |
e4c2ce82 | 404 | goto fail_free_reader; |
044fa782 SR |
405 | bpage->page = (void *)addr; |
406 | rb_init_page(bpage->page); | |
e4c2ce82 | 407 | |
d769041f | 408 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 409 | |
7a8e76a3 SR |
410 | ret = rb_allocate_pages(cpu_buffer, buffer->pages); |
411 | if (ret < 0) | |
d769041f | 412 | goto fail_free_reader; |
7a8e76a3 SR |
413 | |
414 | cpu_buffer->head_page | |
415 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 416 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 SR |
417 | |
418 | return cpu_buffer; | |
419 | ||
d769041f SR |
420 | fail_free_reader: |
421 | free_buffer_page(cpu_buffer->reader_page); | |
422 | ||
7a8e76a3 SR |
423 | fail_free_buffer: |
424 | kfree(cpu_buffer); | |
425 | return NULL; | |
426 | } | |
427 | ||
428 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
429 | { | |
430 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 431 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 432 | |
d769041f SR |
433 | list_del_init(&cpu_buffer->reader_page->list); |
434 | free_buffer_page(cpu_buffer->reader_page); | |
435 | ||
044fa782 SR |
436 | list_for_each_entry_safe(bpage, tmp, head, list) { |
437 | list_del_init(&bpage->list); | |
438 | free_buffer_page(bpage); | |
7a8e76a3 SR |
439 | } |
440 | kfree(cpu_buffer); | |
441 | } | |
442 | ||
a7b13743 SR |
443 | /* |
444 | * Causes compile errors if the struct buffer_page gets bigger | |
445 | * than the struct page. | |
446 | */ | |
447 | extern int ring_buffer_page_too_big(void); | |
448 | ||
7a8e76a3 SR |
449 | /** |
450 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 451 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
452 | * @flags: attributes to set for the ring buffer. |
453 | * | |
454 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
455 | * flag. This flag means that the buffer will overwrite old data | |
456 | * when the buffer wraps. If this flag is not set, the buffer will | |
457 | * drop data when the tail hits the head. | |
458 | */ | |
459 | struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags) | |
460 | { | |
461 | struct ring_buffer *buffer; | |
462 | int bsize; | |
463 | int cpu; | |
464 | ||
a7b13743 SR |
465 | /* Paranoid! Optimizes out when all is well */ |
466 | if (sizeof(struct buffer_page) > sizeof(struct page)) | |
467 | ring_buffer_page_too_big(); | |
468 | ||
469 | ||
7a8e76a3 SR |
470 | /* keep it in its own cache line */ |
471 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
472 | GFP_KERNEL); | |
473 | if (!buffer) | |
474 | return NULL; | |
475 | ||
9e01c1b7 RR |
476 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
477 | goto fail_free_buffer; | |
478 | ||
7a8e76a3 SR |
479 | buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
480 | buffer->flags = flags; | |
481 | ||
482 | /* need at least two pages */ | |
483 | if (buffer->pages == 1) | |
484 | buffer->pages++; | |
485 | ||
9e01c1b7 | 486 | cpumask_copy(buffer->cpumask, cpu_possible_mask); |
7a8e76a3 SR |
487 | buffer->cpus = nr_cpu_ids; |
488 | ||
489 | bsize = sizeof(void *) * nr_cpu_ids; | |
490 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
491 | GFP_KERNEL); | |
492 | if (!buffer->buffers) | |
9e01c1b7 | 493 | goto fail_free_cpumask; |
7a8e76a3 SR |
494 | |
495 | for_each_buffer_cpu(buffer, cpu) { | |
496 | buffer->buffers[cpu] = | |
497 | rb_allocate_cpu_buffer(buffer, cpu); | |
498 | if (!buffer->buffers[cpu]) | |
499 | goto fail_free_buffers; | |
500 | } | |
501 | ||
502 | mutex_init(&buffer->mutex); | |
503 | ||
504 | return buffer; | |
505 | ||
506 | fail_free_buffers: | |
507 | for_each_buffer_cpu(buffer, cpu) { | |
508 | if (buffer->buffers[cpu]) | |
509 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
510 | } | |
511 | kfree(buffer->buffers); | |
512 | ||
9e01c1b7 RR |
513 | fail_free_cpumask: |
514 | free_cpumask_var(buffer->cpumask); | |
515 | ||
7a8e76a3 SR |
516 | fail_free_buffer: |
517 | kfree(buffer); | |
518 | return NULL; | |
519 | } | |
c4f50183 | 520 | EXPORT_SYMBOL_GPL(ring_buffer_alloc); |
7a8e76a3 SR |
521 | |
522 | /** | |
523 | * ring_buffer_free - free a ring buffer. | |
524 | * @buffer: the buffer to free. | |
525 | */ | |
526 | void | |
527 | ring_buffer_free(struct ring_buffer *buffer) | |
528 | { | |
529 | int cpu; | |
530 | ||
531 | for_each_buffer_cpu(buffer, cpu) | |
532 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
533 | ||
9e01c1b7 RR |
534 | free_cpumask_var(buffer->cpumask); |
535 | ||
7a8e76a3 SR |
536 | kfree(buffer); |
537 | } | |
c4f50183 | 538 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 SR |
539 | |
540 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); | |
541 | ||
542 | static void | |
543 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages) | |
544 | { | |
044fa782 | 545 | struct buffer_page *bpage; |
7a8e76a3 SR |
546 | struct list_head *p; |
547 | unsigned i; | |
548 | ||
549 | atomic_inc(&cpu_buffer->record_disabled); | |
550 | synchronize_sched(); | |
551 | ||
552 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
553 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
554 | return; | |
7a8e76a3 | 555 | p = cpu_buffer->pages.next; |
044fa782 SR |
556 | bpage = list_entry(p, struct buffer_page, list); |
557 | list_del_init(&bpage->list); | |
558 | free_buffer_page(bpage); | |
7a8e76a3 | 559 | } |
3e89c7bb SR |
560 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
561 | return; | |
7a8e76a3 SR |
562 | |
563 | rb_reset_cpu(cpu_buffer); | |
564 | ||
565 | rb_check_pages(cpu_buffer); | |
566 | ||
567 | atomic_dec(&cpu_buffer->record_disabled); | |
568 | ||
569 | } | |
570 | ||
571 | static void | |
572 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
573 | struct list_head *pages, unsigned nr_pages) | |
574 | { | |
044fa782 | 575 | struct buffer_page *bpage; |
7a8e76a3 SR |
576 | struct list_head *p; |
577 | unsigned i; | |
578 | ||
579 | atomic_inc(&cpu_buffer->record_disabled); | |
580 | synchronize_sched(); | |
581 | ||
582 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
583 | if (RB_WARN_ON(cpu_buffer, list_empty(pages))) |
584 | return; | |
7a8e76a3 | 585 | p = pages->next; |
044fa782 SR |
586 | bpage = list_entry(p, struct buffer_page, list); |
587 | list_del_init(&bpage->list); | |
588 | list_add_tail(&bpage->list, &cpu_buffer->pages); | |
7a8e76a3 SR |
589 | } |
590 | rb_reset_cpu(cpu_buffer); | |
591 | ||
592 | rb_check_pages(cpu_buffer); | |
593 | ||
594 | atomic_dec(&cpu_buffer->record_disabled); | |
595 | } | |
596 | ||
597 | /** | |
598 | * ring_buffer_resize - resize the ring buffer | |
599 | * @buffer: the buffer to resize. | |
600 | * @size: the new size. | |
601 | * | |
602 | * The tracer is responsible for making sure that the buffer is | |
603 | * not being used while changing the size. | |
604 | * Note: We may be able to change the above requirement by using | |
605 | * RCU synchronizations. | |
606 | * | |
607 | * Minimum size is 2 * BUF_PAGE_SIZE. | |
608 | * | |
609 | * Returns -1 on failure. | |
610 | */ | |
611 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) | |
612 | { | |
613 | struct ring_buffer_per_cpu *cpu_buffer; | |
614 | unsigned nr_pages, rm_pages, new_pages; | |
044fa782 | 615 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
616 | unsigned long buffer_size; |
617 | unsigned long addr; | |
618 | LIST_HEAD(pages); | |
619 | int i, cpu; | |
620 | ||
ee51a1de IM |
621 | /* |
622 | * Always succeed at resizing a non-existent buffer: | |
623 | */ | |
624 | if (!buffer) | |
625 | return size; | |
626 | ||
7a8e76a3 SR |
627 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
628 | size *= BUF_PAGE_SIZE; | |
629 | buffer_size = buffer->pages * BUF_PAGE_SIZE; | |
630 | ||
631 | /* we need a minimum of two pages */ | |
632 | if (size < BUF_PAGE_SIZE * 2) | |
633 | size = BUF_PAGE_SIZE * 2; | |
634 | ||
635 | if (size == buffer_size) | |
636 | return size; | |
637 | ||
638 | mutex_lock(&buffer->mutex); | |
639 | ||
640 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
641 | ||
642 | if (size < buffer_size) { | |
643 | ||
644 | /* easy case, just free pages */ | |
3e89c7bb SR |
645 | if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) { |
646 | mutex_unlock(&buffer->mutex); | |
647 | return -1; | |
648 | } | |
7a8e76a3 SR |
649 | |
650 | rm_pages = buffer->pages - nr_pages; | |
651 | ||
652 | for_each_buffer_cpu(buffer, cpu) { | |
653 | cpu_buffer = buffer->buffers[cpu]; | |
654 | rb_remove_pages(cpu_buffer, rm_pages); | |
655 | } | |
656 | goto out; | |
657 | } | |
658 | ||
659 | /* | |
660 | * This is a bit more difficult. We only want to add pages | |
661 | * when we can allocate enough for all CPUs. We do this | |
662 | * by allocating all the pages and storing them on a local | |
663 | * link list. If we succeed in our allocation, then we | |
664 | * add these pages to the cpu_buffers. Otherwise we just free | |
665 | * them all and return -ENOMEM; | |
666 | */ | |
3e89c7bb SR |
667 | if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) { |
668 | mutex_unlock(&buffer->mutex); | |
669 | return -1; | |
670 | } | |
f536aafc | 671 | |
7a8e76a3 SR |
672 | new_pages = nr_pages - buffer->pages; |
673 | ||
674 | for_each_buffer_cpu(buffer, cpu) { | |
675 | for (i = 0; i < new_pages; i++) { | |
044fa782 | 676 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), |
e4c2ce82 SR |
677 | cache_line_size()), |
678 | GFP_KERNEL, cpu_to_node(cpu)); | |
044fa782 | 679 | if (!bpage) |
e4c2ce82 | 680 | goto free_pages; |
044fa782 | 681 | list_add(&bpage->list, &pages); |
7a8e76a3 SR |
682 | addr = __get_free_page(GFP_KERNEL); |
683 | if (!addr) | |
684 | goto free_pages; | |
044fa782 SR |
685 | bpage->page = (void *)addr; |
686 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
687 | } |
688 | } | |
689 | ||
690 | for_each_buffer_cpu(buffer, cpu) { | |
691 | cpu_buffer = buffer->buffers[cpu]; | |
692 | rb_insert_pages(cpu_buffer, &pages, new_pages); | |
693 | } | |
694 | ||
3e89c7bb SR |
695 | if (RB_WARN_ON(buffer, !list_empty(&pages))) { |
696 | mutex_unlock(&buffer->mutex); | |
697 | return -1; | |
698 | } | |
7a8e76a3 SR |
699 | |
700 | out: | |
701 | buffer->pages = nr_pages; | |
702 | mutex_unlock(&buffer->mutex); | |
703 | ||
704 | return size; | |
705 | ||
706 | free_pages: | |
044fa782 SR |
707 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
708 | list_del_init(&bpage->list); | |
709 | free_buffer_page(bpage); | |
7a8e76a3 | 710 | } |
641d2f63 | 711 | mutex_unlock(&buffer->mutex); |
7a8e76a3 SR |
712 | return -ENOMEM; |
713 | } | |
c4f50183 | 714 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 715 | |
7a8e76a3 SR |
716 | static inline int rb_null_event(struct ring_buffer_event *event) |
717 | { | |
718 | return event->type == RINGBUF_TYPE_PADDING; | |
719 | } | |
720 | ||
8789a9e7 | 721 | static inline void * |
044fa782 | 722 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 723 | { |
044fa782 | 724 | return bpage->data + index; |
8789a9e7 SR |
725 | } |
726 | ||
044fa782 | 727 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 728 | { |
044fa782 | 729 | return bpage->page->data + index; |
7a8e76a3 SR |
730 | } |
731 | ||
732 | static inline struct ring_buffer_event * | |
d769041f | 733 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 734 | { |
6f807acd SR |
735 | return __rb_page_index(cpu_buffer->reader_page, |
736 | cpu_buffer->reader_page->read); | |
737 | } | |
738 | ||
739 | static inline struct ring_buffer_event * | |
740 | rb_head_event(struct ring_buffer_per_cpu *cpu_buffer) | |
741 | { | |
742 | return __rb_page_index(cpu_buffer->head_page, | |
743 | cpu_buffer->head_page->read); | |
7a8e76a3 SR |
744 | } |
745 | ||
746 | static inline struct ring_buffer_event * | |
747 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
748 | { | |
6f807acd | 749 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
750 | } |
751 | ||
bf41a158 SR |
752 | static inline unsigned rb_page_write(struct buffer_page *bpage) |
753 | { | |
754 | return local_read(&bpage->write); | |
755 | } | |
756 | ||
757 | static inline unsigned rb_page_commit(struct buffer_page *bpage) | |
758 | { | |
abc9b56d | 759 | return local_read(&bpage->page->commit); |
bf41a158 SR |
760 | } |
761 | ||
762 | /* Size is determined by what has been commited */ | |
763 | static inline unsigned rb_page_size(struct buffer_page *bpage) | |
764 | { | |
765 | return rb_page_commit(bpage); | |
766 | } | |
767 | ||
768 | static inline unsigned | |
769 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
770 | { | |
771 | return rb_page_commit(cpu_buffer->commit_page); | |
772 | } | |
773 | ||
774 | static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer) | |
775 | { | |
776 | return rb_page_commit(cpu_buffer->head_page); | |
777 | } | |
778 | ||
7a8e76a3 SR |
779 | /* |
780 | * When the tail hits the head and the buffer is in overwrite mode, | |
781 | * the head jumps to the next page and all content on the previous | |
782 | * page is discarded. But before doing so, we update the overrun | |
783 | * variable of the buffer. | |
784 | */ | |
785 | static void rb_update_overflow(struct ring_buffer_per_cpu *cpu_buffer) | |
786 | { | |
787 | struct ring_buffer_event *event; | |
788 | unsigned long head; | |
789 | ||
790 | for (head = 0; head < rb_head_size(cpu_buffer); | |
791 | head += rb_event_length(event)) { | |
792 | ||
6f807acd | 793 | event = __rb_page_index(cpu_buffer->head_page, head); |
3e89c7bb SR |
794 | if (RB_WARN_ON(cpu_buffer, rb_null_event(event))) |
795 | return; | |
7a8e76a3 SR |
796 | /* Only count data entries */ |
797 | if (event->type != RINGBUF_TYPE_DATA) | |
798 | continue; | |
799 | cpu_buffer->overrun++; | |
800 | cpu_buffer->entries--; | |
801 | } | |
802 | } | |
803 | ||
804 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
044fa782 | 805 | struct buffer_page **bpage) |
7a8e76a3 | 806 | { |
044fa782 | 807 | struct list_head *p = (*bpage)->list.next; |
7a8e76a3 SR |
808 | |
809 | if (p == &cpu_buffer->pages) | |
810 | p = p->next; | |
811 | ||
044fa782 | 812 | *bpage = list_entry(p, struct buffer_page, list); |
7a8e76a3 SR |
813 | } |
814 | ||
bf41a158 SR |
815 | static inline unsigned |
816 | rb_event_index(struct ring_buffer_event *event) | |
817 | { | |
818 | unsigned long addr = (unsigned long)event; | |
819 | ||
820 | return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE); | |
821 | } | |
822 | ||
34a148bf | 823 | static int |
bf41a158 SR |
824 | rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
825 | struct ring_buffer_event *event) | |
826 | { | |
827 | unsigned long addr = (unsigned long)event; | |
828 | unsigned long index; | |
829 | ||
830 | index = rb_event_index(event); | |
831 | addr &= PAGE_MASK; | |
832 | ||
833 | return cpu_buffer->commit_page->page == (void *)addr && | |
834 | rb_commit_index(cpu_buffer) == index; | |
835 | } | |
836 | ||
34a148bf | 837 | static void |
bf41a158 SR |
838 | rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer, |
839 | struct ring_buffer_event *event) | |
7a8e76a3 | 840 | { |
bf41a158 SR |
841 | unsigned long addr = (unsigned long)event; |
842 | unsigned long index; | |
843 | ||
844 | index = rb_event_index(event); | |
845 | addr &= PAGE_MASK; | |
846 | ||
847 | while (cpu_buffer->commit_page->page != (void *)addr) { | |
3e89c7bb SR |
848 | if (RB_WARN_ON(cpu_buffer, |
849 | cpu_buffer->commit_page == cpu_buffer->tail_page)) | |
850 | return; | |
abc9b56d | 851 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
852 | cpu_buffer->commit_page->write; |
853 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
854 | cpu_buffer->write_stamp = |
855 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
856 | } |
857 | ||
858 | /* Now set the commit to the event's index */ | |
abc9b56d | 859 | local_set(&cpu_buffer->commit_page->page->commit, index); |
7a8e76a3 SR |
860 | } |
861 | ||
34a148bf | 862 | static void |
bf41a158 | 863 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 864 | { |
bf41a158 SR |
865 | /* |
866 | * We only race with interrupts and NMIs on this CPU. | |
867 | * If we own the commit event, then we can commit | |
868 | * all others that interrupted us, since the interruptions | |
869 | * are in stack format (they finish before they come | |
870 | * back to us). This allows us to do a simple loop to | |
871 | * assign the commit to the tail. | |
872 | */ | |
a8ccf1d6 | 873 | again: |
bf41a158 | 874 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
abc9b56d | 875 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
876 | cpu_buffer->commit_page->write; |
877 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
878 | cpu_buffer->write_stamp = |
879 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
880 | /* add barrier to keep gcc from optimizing too much */ |
881 | barrier(); | |
882 | } | |
883 | while (rb_commit_index(cpu_buffer) != | |
884 | rb_page_write(cpu_buffer->commit_page)) { | |
abc9b56d | 885 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
886 | cpu_buffer->commit_page->write; |
887 | barrier(); | |
888 | } | |
a8ccf1d6 SR |
889 | |
890 | /* again, keep gcc from optimizing */ | |
891 | barrier(); | |
892 | ||
893 | /* | |
894 | * If an interrupt came in just after the first while loop | |
895 | * and pushed the tail page forward, we will be left with | |
896 | * a dangling commit that will never go forward. | |
897 | */ | |
898 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
899 | goto again; | |
7a8e76a3 SR |
900 | } |
901 | ||
d769041f | 902 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 903 | { |
abc9b56d | 904 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 905 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
906 | } |
907 | ||
34a148bf | 908 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
909 | { |
910 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
911 | ||
912 | /* | |
913 | * The iterator could be on the reader page (it starts there). | |
914 | * But the head could have moved, since the reader was | |
915 | * found. Check for this case and assign the iterator | |
916 | * to the head page instead of next. | |
917 | */ | |
918 | if (iter->head_page == cpu_buffer->reader_page) | |
919 | iter->head_page = cpu_buffer->head_page; | |
920 | else | |
921 | rb_inc_page(cpu_buffer, &iter->head_page); | |
922 | ||
abc9b56d | 923 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
924 | iter->head = 0; |
925 | } | |
926 | ||
927 | /** | |
928 | * ring_buffer_update_event - update event type and data | |
929 | * @event: the even to update | |
930 | * @type: the type of event | |
931 | * @length: the size of the event field in the ring buffer | |
932 | * | |
933 | * Update the type and data fields of the event. The length | |
934 | * is the actual size that is written to the ring buffer, | |
935 | * and with this, we can determine what to place into the | |
936 | * data field. | |
937 | */ | |
34a148bf | 938 | static void |
7a8e76a3 SR |
939 | rb_update_event(struct ring_buffer_event *event, |
940 | unsigned type, unsigned length) | |
941 | { | |
942 | event->type = type; | |
943 | ||
944 | switch (type) { | |
945 | ||
946 | case RINGBUF_TYPE_PADDING: | |
947 | break; | |
948 | ||
949 | case RINGBUF_TYPE_TIME_EXTEND: | |
67d34724 | 950 | event->len = DIV_ROUND_UP(RB_LEN_TIME_EXTEND, RB_ALIGNMENT); |
7a8e76a3 SR |
951 | break; |
952 | ||
953 | case RINGBUF_TYPE_TIME_STAMP: | |
67d34724 | 954 | event->len = DIV_ROUND_UP(RB_LEN_TIME_STAMP, RB_ALIGNMENT); |
7a8e76a3 SR |
955 | break; |
956 | ||
957 | case RINGBUF_TYPE_DATA: | |
958 | length -= RB_EVNT_HDR_SIZE; | |
959 | if (length > RB_MAX_SMALL_DATA) { | |
960 | event->len = 0; | |
961 | event->array[0] = length; | |
962 | } else | |
67d34724 | 963 | event->len = DIV_ROUND_UP(length, RB_ALIGNMENT); |
7a8e76a3 SR |
964 | break; |
965 | default: | |
966 | BUG(); | |
967 | } | |
968 | } | |
969 | ||
34a148bf | 970 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
971 | { |
972 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
973 | ||
974 | /* zero length can cause confusions */ | |
975 | if (!length) | |
976 | length = 1; | |
977 | ||
978 | if (length > RB_MAX_SMALL_DATA) | |
979 | length += sizeof(event.array[0]); | |
980 | ||
981 | length += RB_EVNT_HDR_SIZE; | |
982 | length = ALIGN(length, RB_ALIGNMENT); | |
983 | ||
984 | return length; | |
985 | } | |
986 | ||
987 | static struct ring_buffer_event * | |
988 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
989 | unsigned type, unsigned long length, u64 *ts) | |
990 | { | |
98db8df7 | 991 | struct buffer_page *tail_page, *head_page, *reader_page, *commit_page; |
bf41a158 | 992 | unsigned long tail, write; |
7a8e76a3 SR |
993 | struct ring_buffer *buffer = cpu_buffer->buffer; |
994 | struct ring_buffer_event *event; | |
bf41a158 | 995 | unsigned long flags; |
78d904b4 | 996 | bool lock_taken = false; |
7a8e76a3 | 997 | |
98db8df7 SR |
998 | commit_page = cpu_buffer->commit_page; |
999 | /* we just need to protect against interrupts */ | |
1000 | barrier(); | |
7a8e76a3 | 1001 | tail_page = cpu_buffer->tail_page; |
bf41a158 SR |
1002 | write = local_add_return(length, &tail_page->write); |
1003 | tail = write - length; | |
7a8e76a3 | 1004 | |
bf41a158 SR |
1005 | /* See if we shot pass the end of this buffer page */ |
1006 | if (write > BUF_PAGE_SIZE) { | |
7a8e76a3 SR |
1007 | struct buffer_page *next_page = tail_page; |
1008 | ||
3e03fb7f | 1009 | local_irq_save(flags); |
78d904b4 | 1010 | /* |
a81bd80a SR |
1011 | * Since the write to the buffer is still not |
1012 | * fully lockless, we must be careful with NMIs. | |
1013 | * The locks in the writers are taken when a write | |
1014 | * crosses to a new page. The locks protect against | |
1015 | * races with the readers (this will soon be fixed | |
1016 | * with a lockless solution). | |
1017 | * | |
1018 | * Because we can not protect against NMIs, and we | |
1019 | * want to keep traces reentrant, we need to manage | |
1020 | * what happens when we are in an NMI. | |
1021 | * | |
78d904b4 SR |
1022 | * NMIs can happen after we take the lock. |
1023 | * If we are in an NMI, only take the lock | |
1024 | * if it is not already taken. Otherwise | |
1025 | * simply fail. | |
1026 | */ | |
a81bd80a | 1027 | if (unlikely(in_nmi())) { |
78d904b4 | 1028 | if (!__raw_spin_trylock(&cpu_buffer->lock)) |
45141d46 | 1029 | goto out_reset; |
78d904b4 SR |
1030 | } else |
1031 | __raw_spin_lock(&cpu_buffer->lock); | |
1032 | ||
1033 | lock_taken = true; | |
bf41a158 | 1034 | |
7a8e76a3 SR |
1035 | rb_inc_page(cpu_buffer, &next_page); |
1036 | ||
d769041f SR |
1037 | head_page = cpu_buffer->head_page; |
1038 | reader_page = cpu_buffer->reader_page; | |
1039 | ||
1040 | /* we grabbed the lock before incrementing */ | |
3e89c7bb | 1041 | if (RB_WARN_ON(cpu_buffer, next_page == reader_page)) |
45141d46 | 1042 | goto out_reset; |
bf41a158 SR |
1043 | |
1044 | /* | |
1045 | * If for some reason, we had an interrupt storm that made | |
1046 | * it all the way around the buffer, bail, and warn | |
1047 | * about it. | |
1048 | */ | |
98db8df7 | 1049 | if (unlikely(next_page == commit_page)) { |
bf41a158 | 1050 | WARN_ON_ONCE(1); |
45141d46 | 1051 | goto out_reset; |
bf41a158 | 1052 | } |
d769041f | 1053 | |
7a8e76a3 | 1054 | if (next_page == head_page) { |
6f3b3440 | 1055 | if (!(buffer->flags & RB_FL_OVERWRITE)) |
45141d46 | 1056 | goto out_reset; |
7a8e76a3 | 1057 | |
bf41a158 SR |
1058 | /* tail_page has not moved yet? */ |
1059 | if (tail_page == cpu_buffer->tail_page) { | |
1060 | /* count overflows */ | |
1061 | rb_update_overflow(cpu_buffer); | |
1062 | ||
1063 | rb_inc_page(cpu_buffer, &head_page); | |
1064 | cpu_buffer->head_page = head_page; | |
1065 | cpu_buffer->head_page->read = 0; | |
1066 | } | |
1067 | } | |
7a8e76a3 | 1068 | |
bf41a158 SR |
1069 | /* |
1070 | * If the tail page is still the same as what we think | |
1071 | * it is, then it is up to us to update the tail | |
1072 | * pointer. | |
1073 | */ | |
1074 | if (tail_page == cpu_buffer->tail_page) { | |
1075 | local_set(&next_page->write, 0); | |
abc9b56d | 1076 | local_set(&next_page->page->commit, 0); |
bf41a158 SR |
1077 | cpu_buffer->tail_page = next_page; |
1078 | ||
1079 | /* reread the time stamp */ | |
1080 | *ts = ring_buffer_time_stamp(cpu_buffer->cpu); | |
abc9b56d | 1081 | cpu_buffer->tail_page->page->time_stamp = *ts; |
7a8e76a3 SR |
1082 | } |
1083 | ||
bf41a158 SR |
1084 | /* |
1085 | * The actual tail page has moved forward. | |
1086 | */ | |
1087 | if (tail < BUF_PAGE_SIZE) { | |
1088 | /* Mark the rest of the page with padding */ | |
6f807acd | 1089 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
1090 | event->type = RINGBUF_TYPE_PADDING; |
1091 | } | |
1092 | ||
bf41a158 SR |
1093 | if (tail <= BUF_PAGE_SIZE) |
1094 | /* Set the write back to the previous setting */ | |
1095 | local_set(&tail_page->write, tail); | |
1096 | ||
1097 | /* | |
1098 | * If this was a commit entry that failed, | |
1099 | * increment that too | |
1100 | */ | |
1101 | if (tail_page == cpu_buffer->commit_page && | |
1102 | tail == rb_commit_index(cpu_buffer)) { | |
1103 | rb_set_commit_to_write(cpu_buffer); | |
1104 | } | |
1105 | ||
3e03fb7f SR |
1106 | __raw_spin_unlock(&cpu_buffer->lock); |
1107 | local_irq_restore(flags); | |
bf41a158 SR |
1108 | |
1109 | /* fail and let the caller try again */ | |
1110 | return ERR_PTR(-EAGAIN); | |
7a8e76a3 SR |
1111 | } |
1112 | ||
bf41a158 SR |
1113 | /* We reserved something on the buffer */ |
1114 | ||
3e89c7bb SR |
1115 | if (RB_WARN_ON(cpu_buffer, write > BUF_PAGE_SIZE)) |
1116 | return NULL; | |
7a8e76a3 | 1117 | |
6f807acd | 1118 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
1119 | rb_update_event(event, type, length); |
1120 | ||
bf41a158 SR |
1121 | /* |
1122 | * If this is a commit and the tail is zero, then update | |
1123 | * this page's time stamp. | |
1124 | */ | |
1125 | if (!tail && rb_is_commit(cpu_buffer, event)) | |
abc9b56d | 1126 | cpu_buffer->commit_page->page->time_stamp = *ts; |
bf41a158 | 1127 | |
7a8e76a3 | 1128 | return event; |
bf41a158 | 1129 | |
45141d46 | 1130 | out_reset: |
6f3b3440 LJ |
1131 | /* reset write */ |
1132 | if (tail <= BUF_PAGE_SIZE) | |
1133 | local_set(&tail_page->write, tail); | |
1134 | ||
78d904b4 SR |
1135 | if (likely(lock_taken)) |
1136 | __raw_spin_unlock(&cpu_buffer->lock); | |
3e03fb7f | 1137 | local_irq_restore(flags); |
bf41a158 | 1138 | return NULL; |
7a8e76a3 SR |
1139 | } |
1140 | ||
1141 | static int | |
1142 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1143 | u64 *ts, u64 *delta) | |
1144 | { | |
1145 | struct ring_buffer_event *event; | |
1146 | static int once; | |
bf41a158 | 1147 | int ret; |
7a8e76a3 SR |
1148 | |
1149 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1150 | printk(KERN_WARNING "Delta way too big! %llu" | |
1151 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1152 | (unsigned long long)*delta, |
1153 | (unsigned long long)*ts, | |
1154 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1155 | WARN_ON(1); |
1156 | } | |
1157 | ||
1158 | /* | |
1159 | * The delta is too big, we to add a | |
1160 | * new timestamp. | |
1161 | */ | |
1162 | event = __rb_reserve_next(cpu_buffer, | |
1163 | RINGBUF_TYPE_TIME_EXTEND, | |
1164 | RB_LEN_TIME_EXTEND, | |
1165 | ts); | |
1166 | if (!event) | |
bf41a158 | 1167 | return -EBUSY; |
7a8e76a3 | 1168 | |
bf41a158 SR |
1169 | if (PTR_ERR(event) == -EAGAIN) |
1170 | return -EAGAIN; | |
1171 | ||
1172 | /* Only a commited time event can update the write stamp */ | |
1173 | if (rb_is_commit(cpu_buffer, event)) { | |
1174 | /* | |
1175 | * If this is the first on the page, then we need to | |
1176 | * update the page itself, and just put in a zero. | |
1177 | */ | |
1178 | if (rb_event_index(event)) { | |
1179 | event->time_delta = *delta & TS_MASK; | |
1180 | event->array[0] = *delta >> TS_SHIFT; | |
1181 | } else { | |
abc9b56d | 1182 | cpu_buffer->commit_page->page->time_stamp = *ts; |
bf41a158 SR |
1183 | event->time_delta = 0; |
1184 | event->array[0] = 0; | |
1185 | } | |
7a8e76a3 | 1186 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
1187 | /* let the caller know this was the commit */ |
1188 | ret = 1; | |
1189 | } else { | |
1190 | /* Darn, this is just wasted space */ | |
1191 | event->time_delta = 0; | |
1192 | event->array[0] = 0; | |
1193 | ret = 0; | |
7a8e76a3 SR |
1194 | } |
1195 | ||
bf41a158 SR |
1196 | *delta = 0; |
1197 | ||
1198 | return ret; | |
7a8e76a3 SR |
1199 | } |
1200 | ||
1201 | static struct ring_buffer_event * | |
1202 | rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, | |
1203 | unsigned type, unsigned long length) | |
1204 | { | |
1205 | struct ring_buffer_event *event; | |
1206 | u64 ts, delta; | |
bf41a158 | 1207 | int commit = 0; |
818e3dd3 | 1208 | int nr_loops = 0; |
7a8e76a3 | 1209 | |
bf41a158 | 1210 | again: |
818e3dd3 SR |
1211 | /* |
1212 | * We allow for interrupts to reenter here and do a trace. | |
1213 | * If one does, it will cause this original code to loop | |
1214 | * back here. Even with heavy interrupts happening, this | |
1215 | * should only happen a few times in a row. If this happens | |
1216 | * 1000 times in a row, there must be either an interrupt | |
1217 | * storm or we have something buggy. | |
1218 | * Bail! | |
1219 | */ | |
3e89c7bb | 1220 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
818e3dd3 | 1221 | return NULL; |
818e3dd3 | 1222 | |
7a8e76a3 SR |
1223 | ts = ring_buffer_time_stamp(cpu_buffer->cpu); |
1224 | ||
bf41a158 SR |
1225 | /* |
1226 | * Only the first commit can update the timestamp. | |
1227 | * Yes there is a race here. If an interrupt comes in | |
1228 | * just after the conditional and it traces too, then it | |
1229 | * will also check the deltas. More than one timestamp may | |
1230 | * also be made. But only the entry that did the actual | |
1231 | * commit will be something other than zero. | |
1232 | */ | |
1233 | if (cpu_buffer->tail_page == cpu_buffer->commit_page && | |
1234 | rb_page_write(cpu_buffer->tail_page) == | |
1235 | rb_commit_index(cpu_buffer)) { | |
1236 | ||
7a8e76a3 SR |
1237 | delta = ts - cpu_buffer->write_stamp; |
1238 | ||
bf41a158 SR |
1239 | /* make sure this delta is calculated here */ |
1240 | barrier(); | |
1241 | ||
1242 | /* Did the write stamp get updated already? */ | |
1243 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
4143c5cb | 1244 | delta = 0; |
bf41a158 | 1245 | |
7a8e76a3 | 1246 | if (test_time_stamp(delta)) { |
7a8e76a3 | 1247 | |
bf41a158 SR |
1248 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
1249 | ||
1250 | if (commit == -EBUSY) | |
7a8e76a3 | 1251 | return NULL; |
bf41a158 SR |
1252 | |
1253 | if (commit == -EAGAIN) | |
1254 | goto again; | |
1255 | ||
1256 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 1257 | } |
bf41a158 SR |
1258 | } else |
1259 | /* Non commits have zero deltas */ | |
7a8e76a3 | 1260 | delta = 0; |
7a8e76a3 SR |
1261 | |
1262 | event = __rb_reserve_next(cpu_buffer, type, length, &ts); | |
bf41a158 SR |
1263 | if (PTR_ERR(event) == -EAGAIN) |
1264 | goto again; | |
1265 | ||
1266 | if (!event) { | |
1267 | if (unlikely(commit)) | |
1268 | /* | |
1269 | * Ouch! We needed a timestamp and it was commited. But | |
1270 | * we didn't get our event reserved. | |
1271 | */ | |
1272 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 | 1273 | return NULL; |
bf41a158 | 1274 | } |
7a8e76a3 | 1275 | |
bf41a158 SR |
1276 | /* |
1277 | * If the timestamp was commited, make the commit our entry | |
1278 | * now so that we will update it when needed. | |
1279 | */ | |
1280 | if (commit) | |
1281 | rb_set_commit_event(cpu_buffer, event); | |
1282 | else if (!rb_is_commit(cpu_buffer, event)) | |
7a8e76a3 SR |
1283 | delta = 0; |
1284 | ||
1285 | event->time_delta = delta; | |
1286 | ||
1287 | return event; | |
1288 | } | |
1289 | ||
bf41a158 SR |
1290 | static DEFINE_PER_CPU(int, rb_need_resched); |
1291 | ||
7a8e76a3 SR |
1292 | /** |
1293 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
1294 | * @buffer: the ring buffer to reserve from | |
1295 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
1296 | * |
1297 | * Returns a reseverd event on the ring buffer to copy directly to. | |
1298 | * The user of this interface will need to get the body to write into | |
1299 | * and can use the ring_buffer_event_data() interface. | |
1300 | * | |
1301 | * The length is the length of the data needed, not the event length | |
1302 | * which also includes the event header. | |
1303 | * | |
1304 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
1305 | * If NULL is returned, then nothing has been allocated or locked. | |
1306 | */ | |
1307 | struct ring_buffer_event * | |
0a987751 | 1308 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
1309 | { |
1310 | struct ring_buffer_per_cpu *cpu_buffer; | |
1311 | struct ring_buffer_event *event; | |
bf41a158 | 1312 | int cpu, resched; |
7a8e76a3 | 1313 | |
033601a3 | 1314 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1315 | return NULL; |
1316 | ||
7a8e76a3 SR |
1317 | if (atomic_read(&buffer->record_disabled)) |
1318 | return NULL; | |
1319 | ||
bf41a158 | 1320 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 1321 | resched = ftrace_preempt_disable(); |
bf41a158 | 1322 | |
7a8e76a3 SR |
1323 | cpu = raw_smp_processor_id(); |
1324 | ||
9e01c1b7 | 1325 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1326 | goto out; |
7a8e76a3 SR |
1327 | |
1328 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1329 | |
1330 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 1331 | goto out; |
7a8e76a3 SR |
1332 | |
1333 | length = rb_calculate_event_length(length); | |
1334 | if (length > BUF_PAGE_SIZE) | |
bf41a158 | 1335 | goto out; |
7a8e76a3 SR |
1336 | |
1337 | event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length); | |
1338 | if (!event) | |
d769041f | 1339 | goto out; |
7a8e76a3 | 1340 | |
bf41a158 SR |
1341 | /* |
1342 | * Need to store resched state on this cpu. | |
1343 | * Only the first needs to. | |
1344 | */ | |
1345 | ||
1346 | if (preempt_count() == 1) | |
1347 | per_cpu(rb_need_resched, cpu) = resched; | |
1348 | ||
7a8e76a3 SR |
1349 | return event; |
1350 | ||
d769041f | 1351 | out: |
182e9f5f | 1352 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1353 | return NULL; |
1354 | } | |
c4f50183 | 1355 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 SR |
1356 | |
1357 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
1358 | struct ring_buffer_event *event) | |
1359 | { | |
7a8e76a3 | 1360 | cpu_buffer->entries++; |
bf41a158 SR |
1361 | |
1362 | /* Only process further if we own the commit */ | |
1363 | if (!rb_is_commit(cpu_buffer, event)) | |
1364 | return; | |
1365 | ||
1366 | cpu_buffer->write_stamp += event->time_delta; | |
1367 | ||
1368 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 SR |
1369 | } |
1370 | ||
1371 | /** | |
1372 | * ring_buffer_unlock_commit - commit a reserved | |
1373 | * @buffer: The buffer to commit to | |
1374 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
1375 | * |
1376 | * This commits the data to the ring buffer, and releases any locks held. | |
1377 | * | |
1378 | * Must be paired with ring_buffer_lock_reserve. | |
1379 | */ | |
1380 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 1381 | struct ring_buffer_event *event) |
7a8e76a3 SR |
1382 | { |
1383 | struct ring_buffer_per_cpu *cpu_buffer; | |
1384 | int cpu = raw_smp_processor_id(); | |
1385 | ||
1386 | cpu_buffer = buffer->buffers[cpu]; | |
1387 | ||
7a8e76a3 SR |
1388 | rb_commit(cpu_buffer, event); |
1389 | ||
bf41a158 SR |
1390 | /* |
1391 | * Only the last preempt count needs to restore preemption. | |
1392 | */ | |
182e9f5f SR |
1393 | if (preempt_count() == 1) |
1394 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1395 | else | |
bf41a158 | 1396 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
1397 | |
1398 | return 0; | |
1399 | } | |
c4f50183 | 1400 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 SR |
1401 | |
1402 | /** | |
1403 | * ring_buffer_write - write data to the buffer without reserving | |
1404 | * @buffer: The ring buffer to write to. | |
1405 | * @length: The length of the data being written (excluding the event header) | |
1406 | * @data: The data to write to the buffer. | |
1407 | * | |
1408 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
1409 | * one function. If you already have the data to write to the buffer, it | |
1410 | * may be easier to simply call this function. | |
1411 | * | |
1412 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
1413 | * and not the length of the event which would hold the header. | |
1414 | */ | |
1415 | int ring_buffer_write(struct ring_buffer *buffer, | |
1416 | unsigned long length, | |
1417 | void *data) | |
1418 | { | |
1419 | struct ring_buffer_per_cpu *cpu_buffer; | |
1420 | struct ring_buffer_event *event; | |
bf41a158 | 1421 | unsigned long event_length; |
7a8e76a3 SR |
1422 | void *body; |
1423 | int ret = -EBUSY; | |
bf41a158 | 1424 | int cpu, resched; |
7a8e76a3 | 1425 | |
033601a3 | 1426 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1427 | return -EBUSY; |
1428 | ||
7a8e76a3 SR |
1429 | if (atomic_read(&buffer->record_disabled)) |
1430 | return -EBUSY; | |
1431 | ||
182e9f5f | 1432 | resched = ftrace_preempt_disable(); |
bf41a158 | 1433 | |
7a8e76a3 SR |
1434 | cpu = raw_smp_processor_id(); |
1435 | ||
9e01c1b7 | 1436 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1437 | goto out; |
7a8e76a3 SR |
1438 | |
1439 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1440 | |
1441 | if (atomic_read(&cpu_buffer->record_disabled)) | |
1442 | goto out; | |
1443 | ||
1444 | event_length = rb_calculate_event_length(length); | |
1445 | event = rb_reserve_next_event(cpu_buffer, | |
1446 | RINGBUF_TYPE_DATA, event_length); | |
1447 | if (!event) | |
1448 | goto out; | |
1449 | ||
1450 | body = rb_event_data(event); | |
1451 | ||
1452 | memcpy(body, data, length); | |
1453 | ||
1454 | rb_commit(cpu_buffer, event); | |
1455 | ||
1456 | ret = 0; | |
1457 | out: | |
182e9f5f | 1458 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1459 | |
1460 | return ret; | |
1461 | } | |
c4f50183 | 1462 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 1463 | |
34a148bf | 1464 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
1465 | { |
1466 | struct buffer_page *reader = cpu_buffer->reader_page; | |
1467 | struct buffer_page *head = cpu_buffer->head_page; | |
1468 | struct buffer_page *commit = cpu_buffer->commit_page; | |
1469 | ||
1470 | return reader->read == rb_page_commit(reader) && | |
1471 | (commit == reader || | |
1472 | (commit == head && | |
1473 | head->read == rb_page_commit(commit))); | |
1474 | } | |
1475 | ||
7a8e76a3 SR |
1476 | /** |
1477 | * ring_buffer_record_disable - stop all writes into the buffer | |
1478 | * @buffer: The ring buffer to stop writes to. | |
1479 | * | |
1480 | * This prevents all writes to the buffer. Any attempt to write | |
1481 | * to the buffer after this will fail and return NULL. | |
1482 | * | |
1483 | * The caller should call synchronize_sched() after this. | |
1484 | */ | |
1485 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
1486 | { | |
1487 | atomic_inc(&buffer->record_disabled); | |
1488 | } | |
c4f50183 | 1489 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
1490 | |
1491 | /** | |
1492 | * ring_buffer_record_enable - enable writes to the buffer | |
1493 | * @buffer: The ring buffer to enable writes | |
1494 | * | |
1495 | * Note, multiple disables will need the same number of enables | |
1496 | * to truely enable the writing (much like preempt_disable). | |
1497 | */ | |
1498 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
1499 | { | |
1500 | atomic_dec(&buffer->record_disabled); | |
1501 | } | |
c4f50183 | 1502 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 SR |
1503 | |
1504 | /** | |
1505 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
1506 | * @buffer: The ring buffer to stop writes to. | |
1507 | * @cpu: The CPU buffer to stop | |
1508 | * | |
1509 | * This prevents all writes to the buffer. Any attempt to write | |
1510 | * to the buffer after this will fail and return NULL. | |
1511 | * | |
1512 | * The caller should call synchronize_sched() after this. | |
1513 | */ | |
1514 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
1515 | { | |
1516 | struct ring_buffer_per_cpu *cpu_buffer; | |
1517 | ||
9e01c1b7 | 1518 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
1519 | return; |
1520 | ||
1521 | cpu_buffer = buffer->buffers[cpu]; | |
1522 | atomic_inc(&cpu_buffer->record_disabled); | |
1523 | } | |
c4f50183 | 1524 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
1525 | |
1526 | /** | |
1527 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
1528 | * @buffer: The ring buffer to enable writes | |
1529 | * @cpu: The CPU to enable. | |
1530 | * | |
1531 | * Note, multiple disables will need the same number of enables | |
1532 | * to truely enable the writing (much like preempt_disable). | |
1533 | */ | |
1534 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
1535 | { | |
1536 | struct ring_buffer_per_cpu *cpu_buffer; | |
1537 | ||
9e01c1b7 | 1538 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
1539 | return; |
1540 | ||
1541 | cpu_buffer = buffer->buffers[cpu]; | |
1542 | atomic_dec(&cpu_buffer->record_disabled); | |
1543 | } | |
c4f50183 | 1544 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 SR |
1545 | |
1546 | /** | |
1547 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
1548 | * @buffer: The ring buffer | |
1549 | * @cpu: The per CPU buffer to get the entries from. | |
1550 | */ | |
1551 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
1552 | { | |
1553 | struct ring_buffer_per_cpu *cpu_buffer; | |
1554 | ||
9e01c1b7 | 1555 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
1556 | return 0; |
1557 | ||
1558 | cpu_buffer = buffer->buffers[cpu]; | |
1559 | return cpu_buffer->entries; | |
1560 | } | |
c4f50183 | 1561 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
1562 | |
1563 | /** | |
1564 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
1565 | * @buffer: The ring buffer | |
1566 | * @cpu: The per CPU buffer to get the number of overruns from | |
1567 | */ | |
1568 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
1569 | { | |
1570 | struct ring_buffer_per_cpu *cpu_buffer; | |
1571 | ||
9e01c1b7 | 1572 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
1573 | return 0; |
1574 | ||
1575 | cpu_buffer = buffer->buffers[cpu]; | |
1576 | return cpu_buffer->overrun; | |
1577 | } | |
c4f50183 | 1578 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 SR |
1579 | |
1580 | /** | |
1581 | * ring_buffer_entries - get the number of entries in a buffer | |
1582 | * @buffer: The ring buffer | |
1583 | * | |
1584 | * Returns the total number of entries in the ring buffer | |
1585 | * (all CPU entries) | |
1586 | */ | |
1587 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
1588 | { | |
1589 | struct ring_buffer_per_cpu *cpu_buffer; | |
1590 | unsigned long entries = 0; | |
1591 | int cpu; | |
1592 | ||
1593 | /* if you care about this being correct, lock the buffer */ | |
1594 | for_each_buffer_cpu(buffer, cpu) { | |
1595 | cpu_buffer = buffer->buffers[cpu]; | |
1596 | entries += cpu_buffer->entries; | |
1597 | } | |
1598 | ||
1599 | return entries; | |
1600 | } | |
c4f50183 | 1601 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
1602 | |
1603 | /** | |
1604 | * ring_buffer_overrun_cpu - get the number of overruns in buffer | |
1605 | * @buffer: The ring buffer | |
1606 | * | |
1607 | * Returns the total number of overruns in the ring buffer | |
1608 | * (all CPU entries) | |
1609 | */ | |
1610 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
1611 | { | |
1612 | struct ring_buffer_per_cpu *cpu_buffer; | |
1613 | unsigned long overruns = 0; | |
1614 | int cpu; | |
1615 | ||
1616 | /* if you care about this being correct, lock the buffer */ | |
1617 | for_each_buffer_cpu(buffer, cpu) { | |
1618 | cpu_buffer = buffer->buffers[cpu]; | |
1619 | overruns += cpu_buffer->overrun; | |
1620 | } | |
1621 | ||
1622 | return overruns; | |
1623 | } | |
c4f50183 | 1624 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 1625 | |
642edba5 | 1626 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
1627 | { |
1628 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1629 | ||
d769041f SR |
1630 | /* Iterator usage is expected to have record disabled */ |
1631 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
1632 | iter->head_page = cpu_buffer->head_page; | |
6f807acd | 1633 | iter->head = cpu_buffer->head_page->read; |
d769041f SR |
1634 | } else { |
1635 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 1636 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
1637 | } |
1638 | if (iter->head) | |
1639 | iter->read_stamp = cpu_buffer->read_stamp; | |
1640 | else | |
abc9b56d | 1641 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 1642 | } |
f83c9d0f | 1643 | |
642edba5 SR |
1644 | /** |
1645 | * ring_buffer_iter_reset - reset an iterator | |
1646 | * @iter: The iterator to reset | |
1647 | * | |
1648 | * Resets the iterator, so that it will start from the beginning | |
1649 | * again. | |
1650 | */ | |
1651 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
1652 | { | |
1653 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1654 | unsigned long flags; | |
1655 | ||
1656 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
1657 | rb_iter_reset(iter); | |
f83c9d0f | 1658 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 1659 | } |
c4f50183 | 1660 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
1661 | |
1662 | /** | |
1663 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
1664 | * @iter: The iterator to check | |
1665 | */ | |
1666 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
1667 | { | |
1668 | struct ring_buffer_per_cpu *cpu_buffer; | |
1669 | ||
1670 | cpu_buffer = iter->cpu_buffer; | |
1671 | ||
bf41a158 SR |
1672 | return iter->head_page == cpu_buffer->commit_page && |
1673 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 1674 | } |
c4f50183 | 1675 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
1676 | |
1677 | static void | |
1678 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1679 | struct ring_buffer_event *event) | |
1680 | { | |
1681 | u64 delta; | |
1682 | ||
1683 | switch (event->type) { | |
1684 | case RINGBUF_TYPE_PADDING: | |
1685 | return; | |
1686 | ||
1687 | case RINGBUF_TYPE_TIME_EXTEND: | |
1688 | delta = event->array[0]; | |
1689 | delta <<= TS_SHIFT; | |
1690 | delta += event->time_delta; | |
1691 | cpu_buffer->read_stamp += delta; | |
1692 | return; | |
1693 | ||
1694 | case RINGBUF_TYPE_TIME_STAMP: | |
1695 | /* FIXME: not implemented */ | |
1696 | return; | |
1697 | ||
1698 | case RINGBUF_TYPE_DATA: | |
1699 | cpu_buffer->read_stamp += event->time_delta; | |
1700 | return; | |
1701 | ||
1702 | default: | |
1703 | BUG(); | |
1704 | } | |
1705 | return; | |
1706 | } | |
1707 | ||
1708 | static void | |
1709 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
1710 | struct ring_buffer_event *event) | |
1711 | { | |
1712 | u64 delta; | |
1713 | ||
1714 | switch (event->type) { | |
1715 | case RINGBUF_TYPE_PADDING: | |
1716 | return; | |
1717 | ||
1718 | case RINGBUF_TYPE_TIME_EXTEND: | |
1719 | delta = event->array[0]; | |
1720 | delta <<= TS_SHIFT; | |
1721 | delta += event->time_delta; | |
1722 | iter->read_stamp += delta; | |
1723 | return; | |
1724 | ||
1725 | case RINGBUF_TYPE_TIME_STAMP: | |
1726 | /* FIXME: not implemented */ | |
1727 | return; | |
1728 | ||
1729 | case RINGBUF_TYPE_DATA: | |
1730 | iter->read_stamp += event->time_delta; | |
1731 | return; | |
1732 | ||
1733 | default: | |
1734 | BUG(); | |
1735 | } | |
1736 | return; | |
1737 | } | |
1738 | ||
d769041f SR |
1739 | static struct buffer_page * |
1740 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 1741 | { |
d769041f SR |
1742 | struct buffer_page *reader = NULL; |
1743 | unsigned long flags; | |
818e3dd3 | 1744 | int nr_loops = 0; |
d769041f | 1745 | |
3e03fb7f SR |
1746 | local_irq_save(flags); |
1747 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
1748 | |
1749 | again: | |
818e3dd3 SR |
1750 | /* |
1751 | * This should normally only loop twice. But because the | |
1752 | * start of the reader inserts an empty page, it causes | |
1753 | * a case where we will loop three times. There should be no | |
1754 | * reason to loop four times (that I know of). | |
1755 | */ | |
3e89c7bb | 1756 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
1757 | reader = NULL; |
1758 | goto out; | |
1759 | } | |
1760 | ||
d769041f SR |
1761 | reader = cpu_buffer->reader_page; |
1762 | ||
1763 | /* If there's more to read, return this page */ | |
bf41a158 | 1764 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
1765 | goto out; |
1766 | ||
1767 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
1768 | if (RB_WARN_ON(cpu_buffer, |
1769 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
1770 | goto out; | |
d769041f SR |
1771 | |
1772 | /* check if we caught up to the tail */ | |
1773 | reader = NULL; | |
bf41a158 | 1774 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 1775 | goto out; |
7a8e76a3 SR |
1776 | |
1777 | /* | |
d769041f SR |
1778 | * Splice the empty reader page into the list around the head. |
1779 | * Reset the reader page to size zero. | |
7a8e76a3 | 1780 | */ |
7a8e76a3 | 1781 | |
d769041f SR |
1782 | reader = cpu_buffer->head_page; |
1783 | cpu_buffer->reader_page->list.next = reader->list.next; | |
1784 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 SR |
1785 | |
1786 | local_set(&cpu_buffer->reader_page->write, 0); | |
abc9b56d | 1787 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
7a8e76a3 | 1788 | |
d769041f SR |
1789 | /* Make the reader page now replace the head */ |
1790 | reader->list.prev->next = &cpu_buffer->reader_page->list; | |
1791 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
7a8e76a3 SR |
1792 | |
1793 | /* | |
d769041f SR |
1794 | * If the tail is on the reader, then we must set the head |
1795 | * to the inserted page, otherwise we set it one before. | |
7a8e76a3 | 1796 | */ |
d769041f | 1797 | cpu_buffer->head_page = cpu_buffer->reader_page; |
7a8e76a3 | 1798 | |
bf41a158 | 1799 | if (cpu_buffer->commit_page != reader) |
d769041f SR |
1800 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
1801 | ||
1802 | /* Finally update the reader page to the new head */ | |
1803 | cpu_buffer->reader_page = reader; | |
1804 | rb_reset_reader_page(cpu_buffer); | |
1805 | ||
1806 | goto again; | |
1807 | ||
1808 | out: | |
3e03fb7f SR |
1809 | __raw_spin_unlock(&cpu_buffer->lock); |
1810 | local_irq_restore(flags); | |
d769041f SR |
1811 | |
1812 | return reader; | |
1813 | } | |
1814 | ||
1815 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
1816 | { | |
1817 | struct ring_buffer_event *event; | |
1818 | struct buffer_page *reader; | |
1819 | unsigned length; | |
1820 | ||
1821 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 1822 | |
d769041f | 1823 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
1824 | if (RB_WARN_ON(cpu_buffer, !reader)) |
1825 | return; | |
7a8e76a3 | 1826 | |
d769041f SR |
1827 | event = rb_reader_event(cpu_buffer); |
1828 | ||
1829 | if (event->type == RINGBUF_TYPE_DATA) | |
1830 | cpu_buffer->entries--; | |
1831 | ||
1832 | rb_update_read_stamp(cpu_buffer, event); | |
1833 | ||
1834 | length = rb_event_length(event); | |
6f807acd | 1835 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
1836 | } |
1837 | ||
1838 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
1839 | { | |
1840 | struct ring_buffer *buffer; | |
1841 | struct ring_buffer_per_cpu *cpu_buffer; | |
1842 | struct ring_buffer_event *event; | |
1843 | unsigned length; | |
1844 | ||
1845 | cpu_buffer = iter->cpu_buffer; | |
1846 | buffer = cpu_buffer->buffer; | |
1847 | ||
1848 | /* | |
1849 | * Check if we are at the end of the buffer. | |
1850 | */ | |
bf41a158 | 1851 | if (iter->head >= rb_page_size(iter->head_page)) { |
3e89c7bb SR |
1852 | if (RB_WARN_ON(buffer, |
1853 | iter->head_page == cpu_buffer->commit_page)) | |
1854 | return; | |
d769041f | 1855 | rb_inc_iter(iter); |
7a8e76a3 SR |
1856 | return; |
1857 | } | |
1858 | ||
1859 | event = rb_iter_head_event(iter); | |
1860 | ||
1861 | length = rb_event_length(event); | |
1862 | ||
1863 | /* | |
1864 | * This should not be called to advance the header if we are | |
1865 | * at the tail of the buffer. | |
1866 | */ | |
3e89c7bb | 1867 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 1868 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
1869 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
1870 | return; | |
7a8e76a3 SR |
1871 | |
1872 | rb_update_iter_read_stamp(iter, event); | |
1873 | ||
1874 | iter->head += length; | |
1875 | ||
1876 | /* check for end of page padding */ | |
bf41a158 SR |
1877 | if ((iter->head >= rb_page_size(iter->head_page)) && |
1878 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
1879 | rb_advance_iter(iter); |
1880 | } | |
1881 | ||
f83c9d0f SR |
1882 | static struct ring_buffer_event * |
1883 | rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
7a8e76a3 SR |
1884 | { |
1885 | struct ring_buffer_per_cpu *cpu_buffer; | |
1886 | struct ring_buffer_event *event; | |
d769041f | 1887 | struct buffer_page *reader; |
818e3dd3 | 1888 | int nr_loops = 0; |
7a8e76a3 | 1889 | |
9e01c1b7 | 1890 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
1891 | return NULL; |
1892 | ||
1893 | cpu_buffer = buffer->buffers[cpu]; | |
1894 | ||
1895 | again: | |
818e3dd3 SR |
1896 | /* |
1897 | * We repeat when a timestamp is encountered. It is possible | |
1898 | * to get multiple timestamps from an interrupt entering just | |
1899 | * as one timestamp is about to be written. The max times | |
1900 | * that this can happen is the number of nested interrupts we | |
1901 | * can have. Nesting 10 deep of interrupts is clearly | |
1902 | * an anomaly. | |
1903 | */ | |
3e89c7bb | 1904 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 1905 | return NULL; |
818e3dd3 | 1906 | |
d769041f SR |
1907 | reader = rb_get_reader_page(cpu_buffer); |
1908 | if (!reader) | |
7a8e76a3 SR |
1909 | return NULL; |
1910 | ||
d769041f | 1911 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 SR |
1912 | |
1913 | switch (event->type) { | |
1914 | case RINGBUF_TYPE_PADDING: | |
bf41a158 | 1915 | RB_WARN_ON(cpu_buffer, 1); |
d769041f SR |
1916 | rb_advance_reader(cpu_buffer); |
1917 | return NULL; | |
7a8e76a3 SR |
1918 | |
1919 | case RINGBUF_TYPE_TIME_EXTEND: | |
1920 | /* Internal data, OK to advance */ | |
d769041f | 1921 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
1922 | goto again; |
1923 | ||
1924 | case RINGBUF_TYPE_TIME_STAMP: | |
1925 | /* FIXME: not implemented */ | |
d769041f | 1926 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
1927 | goto again; |
1928 | ||
1929 | case RINGBUF_TYPE_DATA: | |
1930 | if (ts) { | |
1931 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
1932 | ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts); | |
1933 | } | |
1934 | return event; | |
1935 | ||
1936 | default: | |
1937 | BUG(); | |
1938 | } | |
1939 | ||
1940 | return NULL; | |
1941 | } | |
c4f50183 | 1942 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 1943 | |
f83c9d0f SR |
1944 | static struct ring_buffer_event * |
1945 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
1946 | { |
1947 | struct ring_buffer *buffer; | |
1948 | struct ring_buffer_per_cpu *cpu_buffer; | |
1949 | struct ring_buffer_event *event; | |
818e3dd3 | 1950 | int nr_loops = 0; |
7a8e76a3 SR |
1951 | |
1952 | if (ring_buffer_iter_empty(iter)) | |
1953 | return NULL; | |
1954 | ||
1955 | cpu_buffer = iter->cpu_buffer; | |
1956 | buffer = cpu_buffer->buffer; | |
1957 | ||
1958 | again: | |
818e3dd3 SR |
1959 | /* |
1960 | * We repeat when a timestamp is encountered. It is possible | |
1961 | * to get multiple timestamps from an interrupt entering just | |
1962 | * as one timestamp is about to be written. The max times | |
1963 | * that this can happen is the number of nested interrupts we | |
1964 | * can have. Nesting 10 deep of interrupts is clearly | |
1965 | * an anomaly. | |
1966 | */ | |
3e89c7bb | 1967 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 1968 | return NULL; |
818e3dd3 | 1969 | |
7a8e76a3 SR |
1970 | if (rb_per_cpu_empty(cpu_buffer)) |
1971 | return NULL; | |
1972 | ||
1973 | event = rb_iter_head_event(iter); | |
1974 | ||
1975 | switch (event->type) { | |
1976 | case RINGBUF_TYPE_PADDING: | |
d769041f | 1977 | rb_inc_iter(iter); |
7a8e76a3 SR |
1978 | goto again; |
1979 | ||
1980 | case RINGBUF_TYPE_TIME_EXTEND: | |
1981 | /* Internal data, OK to advance */ | |
1982 | rb_advance_iter(iter); | |
1983 | goto again; | |
1984 | ||
1985 | case RINGBUF_TYPE_TIME_STAMP: | |
1986 | /* FIXME: not implemented */ | |
1987 | rb_advance_iter(iter); | |
1988 | goto again; | |
1989 | ||
1990 | case RINGBUF_TYPE_DATA: | |
1991 | if (ts) { | |
1992 | *ts = iter->read_stamp + event->time_delta; | |
1993 | ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts); | |
1994 | } | |
1995 | return event; | |
1996 | ||
1997 | default: | |
1998 | BUG(); | |
1999 | } | |
2000 | ||
2001 | return NULL; | |
2002 | } | |
c4f50183 | 2003 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 2004 | |
f83c9d0f SR |
2005 | /** |
2006 | * ring_buffer_peek - peek at the next event to be read | |
2007 | * @buffer: The ring buffer to read | |
2008 | * @cpu: The cpu to peak at | |
2009 | * @ts: The timestamp counter of this event. | |
2010 | * | |
2011 | * This will return the event that will be read next, but does | |
2012 | * not consume the data. | |
2013 | */ | |
2014 | struct ring_buffer_event * | |
2015 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2016 | { | |
2017 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2018 | struct ring_buffer_event *event; | |
2019 | unsigned long flags; | |
2020 | ||
2021 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2022 | event = rb_buffer_peek(buffer, cpu, ts); | |
2023 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2024 | ||
2025 | return event; | |
2026 | } | |
2027 | ||
2028 | /** | |
2029 | * ring_buffer_iter_peek - peek at the next event to be read | |
2030 | * @iter: The ring buffer iterator | |
2031 | * @ts: The timestamp counter of this event. | |
2032 | * | |
2033 | * This will return the event that will be read next, but does | |
2034 | * not increment the iterator. | |
2035 | */ | |
2036 | struct ring_buffer_event * | |
2037 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
2038 | { | |
2039 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2040 | struct ring_buffer_event *event; | |
2041 | unsigned long flags; | |
2042 | ||
2043 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2044 | event = rb_iter_peek(iter, ts); | |
2045 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2046 | ||
2047 | return event; | |
2048 | } | |
2049 | ||
7a8e76a3 SR |
2050 | /** |
2051 | * ring_buffer_consume - return an event and consume it | |
2052 | * @buffer: The ring buffer to get the next event from | |
2053 | * | |
2054 | * Returns the next event in the ring buffer, and that event is consumed. | |
2055 | * Meaning, that sequential reads will keep returning a different event, | |
2056 | * and eventually empty the ring buffer if the producer is slower. | |
2057 | */ | |
2058 | struct ring_buffer_event * | |
2059 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2060 | { | |
f83c9d0f | 2061 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; |
7a8e76a3 | 2062 | struct ring_buffer_event *event; |
f83c9d0f | 2063 | unsigned long flags; |
7a8e76a3 | 2064 | |
9e01c1b7 | 2065 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
2066 | return NULL; |
2067 | ||
f83c9d0f SR |
2068 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2069 | ||
2070 | event = rb_buffer_peek(buffer, cpu, ts); | |
7a8e76a3 | 2071 | if (!event) |
f83c9d0f | 2072 | goto out; |
7a8e76a3 | 2073 | |
d769041f | 2074 | rb_advance_reader(cpu_buffer); |
7a8e76a3 | 2075 | |
f83c9d0f SR |
2076 | out: |
2077 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2078 | ||
7a8e76a3 SR |
2079 | return event; |
2080 | } | |
c4f50183 | 2081 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
2082 | |
2083 | /** | |
2084 | * ring_buffer_read_start - start a non consuming read of the buffer | |
2085 | * @buffer: The ring buffer to read from | |
2086 | * @cpu: The cpu buffer to iterate over | |
2087 | * | |
2088 | * This starts up an iteration through the buffer. It also disables | |
2089 | * the recording to the buffer until the reading is finished. | |
2090 | * This prevents the reading from being corrupted. This is not | |
2091 | * a consuming read, so a producer is not expected. | |
2092 | * | |
2093 | * Must be paired with ring_buffer_finish. | |
2094 | */ | |
2095 | struct ring_buffer_iter * | |
2096 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
2097 | { | |
2098 | struct ring_buffer_per_cpu *cpu_buffer; | |
2099 | struct ring_buffer_iter *iter; | |
d769041f | 2100 | unsigned long flags; |
7a8e76a3 | 2101 | |
9e01c1b7 | 2102 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
2103 | return NULL; |
2104 | ||
2105 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
2106 | if (!iter) | |
2107 | return NULL; | |
2108 | ||
2109 | cpu_buffer = buffer->buffers[cpu]; | |
2110 | ||
2111 | iter->cpu_buffer = cpu_buffer; | |
2112 | ||
2113 | atomic_inc(&cpu_buffer->record_disabled); | |
2114 | synchronize_sched(); | |
2115 | ||
f83c9d0f | 2116 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 2117 | __raw_spin_lock(&cpu_buffer->lock); |
642edba5 | 2118 | rb_iter_reset(iter); |
3e03fb7f | 2119 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 2120 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
2121 | |
2122 | return iter; | |
2123 | } | |
c4f50183 | 2124 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
2125 | |
2126 | /** | |
2127 | * ring_buffer_finish - finish reading the iterator of the buffer | |
2128 | * @iter: The iterator retrieved by ring_buffer_start | |
2129 | * | |
2130 | * This re-enables the recording to the buffer, and frees the | |
2131 | * iterator. | |
2132 | */ | |
2133 | void | |
2134 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
2135 | { | |
2136 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2137 | ||
2138 | atomic_dec(&cpu_buffer->record_disabled); | |
2139 | kfree(iter); | |
2140 | } | |
c4f50183 | 2141 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
2142 | |
2143 | /** | |
2144 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
2145 | * @iter: The ring buffer iterator | |
2146 | * @ts: The time stamp of the event read. | |
2147 | * | |
2148 | * This reads the next event in the ring buffer and increments the iterator. | |
2149 | */ | |
2150 | struct ring_buffer_event * | |
2151 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
2152 | { | |
2153 | struct ring_buffer_event *event; | |
f83c9d0f SR |
2154 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
2155 | unsigned long flags; | |
7a8e76a3 | 2156 | |
f83c9d0f SR |
2157 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2158 | event = rb_iter_peek(iter, ts); | |
7a8e76a3 | 2159 | if (!event) |
f83c9d0f | 2160 | goto out; |
7a8e76a3 SR |
2161 | |
2162 | rb_advance_iter(iter); | |
f83c9d0f SR |
2163 | out: |
2164 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 SR |
2165 | |
2166 | return event; | |
2167 | } | |
c4f50183 | 2168 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
2169 | |
2170 | /** | |
2171 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
2172 | * @buffer: The ring buffer. | |
2173 | */ | |
2174 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
2175 | { | |
2176 | return BUF_PAGE_SIZE * buffer->pages; | |
2177 | } | |
c4f50183 | 2178 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
2179 | |
2180 | static void | |
2181 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
2182 | { | |
2183 | cpu_buffer->head_page | |
2184 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 2185 | local_set(&cpu_buffer->head_page->write, 0); |
abc9b56d | 2186 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 2187 | |
6f807acd | 2188 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
2189 | |
2190 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
2191 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
2192 | ||
2193 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
2194 | local_set(&cpu_buffer->reader_page->write, 0); | |
abc9b56d | 2195 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 2196 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 2197 | |
7a8e76a3 SR |
2198 | cpu_buffer->overrun = 0; |
2199 | cpu_buffer->entries = 0; | |
69507c06 SR |
2200 | |
2201 | cpu_buffer->write_stamp = 0; | |
2202 | cpu_buffer->read_stamp = 0; | |
7a8e76a3 SR |
2203 | } |
2204 | ||
2205 | /** | |
2206 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
2207 | * @buffer: The ring buffer to reset a per cpu buffer of | |
2208 | * @cpu: The CPU buffer to be reset | |
2209 | */ | |
2210 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
2211 | { | |
2212 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2213 | unsigned long flags; | |
2214 | ||
9e01c1b7 | 2215 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
2216 | return; |
2217 | ||
f83c9d0f SR |
2218 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2219 | ||
3e03fb7f | 2220 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
2221 | |
2222 | rb_reset_cpu(cpu_buffer); | |
2223 | ||
3e03fb7f | 2224 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f SR |
2225 | |
2226 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 2227 | } |
c4f50183 | 2228 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
2229 | |
2230 | /** | |
2231 | * ring_buffer_reset - reset a ring buffer | |
2232 | * @buffer: The ring buffer to reset all cpu buffers | |
2233 | */ | |
2234 | void ring_buffer_reset(struct ring_buffer *buffer) | |
2235 | { | |
7a8e76a3 SR |
2236 | int cpu; |
2237 | ||
7a8e76a3 | 2238 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 2239 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 2240 | } |
c4f50183 | 2241 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
2242 | |
2243 | /** | |
2244 | * rind_buffer_empty - is the ring buffer empty? | |
2245 | * @buffer: The ring buffer to test | |
2246 | */ | |
2247 | int ring_buffer_empty(struct ring_buffer *buffer) | |
2248 | { | |
2249 | struct ring_buffer_per_cpu *cpu_buffer; | |
2250 | int cpu; | |
2251 | ||
2252 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
2253 | for_each_buffer_cpu(buffer, cpu) { | |
2254 | cpu_buffer = buffer->buffers[cpu]; | |
2255 | if (!rb_per_cpu_empty(cpu_buffer)) | |
2256 | return 0; | |
2257 | } | |
2258 | return 1; | |
2259 | } | |
c4f50183 | 2260 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
2261 | |
2262 | /** | |
2263 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
2264 | * @buffer: The ring buffer | |
2265 | * @cpu: The CPU buffer to test | |
2266 | */ | |
2267 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
2268 | { | |
2269 | struct ring_buffer_per_cpu *cpu_buffer; | |
2270 | ||
9e01c1b7 | 2271 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
2272 | return 1; |
2273 | ||
2274 | cpu_buffer = buffer->buffers[cpu]; | |
2275 | return rb_per_cpu_empty(cpu_buffer); | |
2276 | } | |
c4f50183 | 2277 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 SR |
2278 | |
2279 | /** | |
2280 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
2281 | * @buffer_a: One buffer to swap with | |
2282 | * @buffer_b: The other buffer to swap with | |
2283 | * | |
2284 | * This function is useful for tracers that want to take a "snapshot" | |
2285 | * of a CPU buffer and has another back up buffer lying around. | |
2286 | * it is expected that the tracer handles the cpu buffer not being | |
2287 | * used at the moment. | |
2288 | */ | |
2289 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
2290 | struct ring_buffer *buffer_b, int cpu) | |
2291 | { | |
2292 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
2293 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
2294 | ||
9e01c1b7 RR |
2295 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
2296 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
7a8e76a3 SR |
2297 | return -EINVAL; |
2298 | ||
2299 | /* At least make sure the two buffers are somewhat the same */ | |
6d102bc6 | 2300 | if (buffer_a->pages != buffer_b->pages) |
7a8e76a3 SR |
2301 | return -EINVAL; |
2302 | ||
97b17efe SR |
2303 | if (ring_buffer_flags != RB_BUFFERS_ON) |
2304 | return -EAGAIN; | |
2305 | ||
2306 | if (atomic_read(&buffer_a->record_disabled)) | |
2307 | return -EAGAIN; | |
2308 | ||
2309 | if (atomic_read(&buffer_b->record_disabled)) | |
2310 | return -EAGAIN; | |
2311 | ||
7a8e76a3 SR |
2312 | cpu_buffer_a = buffer_a->buffers[cpu]; |
2313 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
2314 | ||
97b17efe SR |
2315 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
2316 | return -EAGAIN; | |
2317 | ||
2318 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
2319 | return -EAGAIN; | |
2320 | ||
7a8e76a3 SR |
2321 | /* |
2322 | * We can't do a synchronize_sched here because this | |
2323 | * function can be called in atomic context. | |
2324 | * Normally this will be called from the same CPU as cpu. | |
2325 | * If not it's up to the caller to protect this. | |
2326 | */ | |
2327 | atomic_inc(&cpu_buffer_a->record_disabled); | |
2328 | atomic_inc(&cpu_buffer_b->record_disabled); | |
2329 | ||
2330 | buffer_a->buffers[cpu] = cpu_buffer_b; | |
2331 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
2332 | ||
2333 | cpu_buffer_b->buffer = buffer_a; | |
2334 | cpu_buffer_a->buffer = buffer_b; | |
2335 | ||
2336 | atomic_dec(&cpu_buffer_a->record_disabled); | |
2337 | atomic_dec(&cpu_buffer_b->record_disabled); | |
2338 | ||
2339 | return 0; | |
2340 | } | |
c4f50183 | 2341 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
7a8e76a3 | 2342 | |
8789a9e7 | 2343 | static void rb_remove_entries(struct ring_buffer_per_cpu *cpu_buffer, |
667d2412 LJ |
2344 | struct buffer_data_page *bpage, |
2345 | unsigned int offset) | |
8789a9e7 SR |
2346 | { |
2347 | struct ring_buffer_event *event; | |
2348 | unsigned long head; | |
2349 | ||
2350 | __raw_spin_lock(&cpu_buffer->lock); | |
667d2412 | 2351 | for (head = offset; head < local_read(&bpage->commit); |
8789a9e7 SR |
2352 | head += rb_event_length(event)) { |
2353 | ||
044fa782 | 2354 | event = __rb_data_page_index(bpage, head); |
8789a9e7 SR |
2355 | if (RB_WARN_ON(cpu_buffer, rb_null_event(event))) |
2356 | return; | |
2357 | /* Only count data entries */ | |
2358 | if (event->type != RINGBUF_TYPE_DATA) | |
2359 | continue; | |
2360 | cpu_buffer->entries--; | |
2361 | } | |
2362 | __raw_spin_unlock(&cpu_buffer->lock); | |
2363 | } | |
2364 | ||
2365 | /** | |
2366 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
2367 | * @buffer: the buffer to allocate for. | |
2368 | * | |
2369 | * This function is used in conjunction with ring_buffer_read_page. | |
2370 | * When reading a full page from the ring buffer, these functions | |
2371 | * can be used to speed up the process. The calling function should | |
2372 | * allocate a few pages first with this function. Then when it | |
2373 | * needs to get pages from the ring buffer, it passes the result | |
2374 | * of this function into ring_buffer_read_page, which will swap | |
2375 | * the page that was allocated, with the read page of the buffer. | |
2376 | * | |
2377 | * Returns: | |
2378 | * The page allocated, or NULL on error. | |
2379 | */ | |
2380 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer) | |
2381 | { | |
2382 | unsigned long addr; | |
044fa782 | 2383 | struct buffer_data_page *bpage; |
8789a9e7 SR |
2384 | |
2385 | addr = __get_free_page(GFP_KERNEL); | |
2386 | if (!addr) | |
2387 | return NULL; | |
2388 | ||
044fa782 | 2389 | bpage = (void *)addr; |
8789a9e7 | 2390 | |
044fa782 | 2391 | return bpage; |
8789a9e7 SR |
2392 | } |
2393 | ||
2394 | /** | |
2395 | * ring_buffer_free_read_page - free an allocated read page | |
2396 | * @buffer: the buffer the page was allocate for | |
2397 | * @data: the page to free | |
2398 | * | |
2399 | * Free a page allocated from ring_buffer_alloc_read_page. | |
2400 | */ | |
2401 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
2402 | { | |
2403 | free_page((unsigned long)data); | |
2404 | } | |
2405 | ||
2406 | /** | |
2407 | * ring_buffer_read_page - extract a page from the ring buffer | |
2408 | * @buffer: buffer to extract from | |
2409 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
2410 | * @cpu: the cpu of the buffer to extract | |
2411 | * @full: should the extraction only happen when the page is full. | |
2412 | * | |
2413 | * This function will pull out a page from the ring buffer and consume it. | |
2414 | * @data_page must be the address of the variable that was returned | |
2415 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
2416 | * to swap with a page in the ring buffer. | |
2417 | * | |
2418 | * for example: | |
b85fa01e | 2419 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
2420 | * if (!rpage) |
2421 | * return error; | |
2422 | * ret = ring_buffer_read_page(buffer, &rpage, cpu, 0); | |
667d2412 LJ |
2423 | * if (ret >= 0) |
2424 | * process_page(rpage, ret); | |
8789a9e7 SR |
2425 | * |
2426 | * When @full is set, the function will not return true unless | |
2427 | * the writer is off the reader page. | |
2428 | * | |
2429 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
2430 | * The ring buffer can be used anywhere in the kernel and can not | |
2431 | * blindly call wake_up. The layer that uses the ring buffer must be | |
2432 | * responsible for that. | |
2433 | * | |
2434 | * Returns: | |
667d2412 LJ |
2435 | * >=0 if data has been transferred, returns the offset of consumed data. |
2436 | * <0 if no data has been transferred. | |
8789a9e7 SR |
2437 | */ |
2438 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
2439 | void **data_page, int cpu, int full) | |
2440 | { | |
2441 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2442 | struct ring_buffer_event *event; | |
044fa782 | 2443 | struct buffer_data_page *bpage; |
8789a9e7 | 2444 | unsigned long flags; |
667d2412 LJ |
2445 | unsigned int read; |
2446 | int ret = -1; | |
8789a9e7 SR |
2447 | |
2448 | if (!data_page) | |
2449 | return 0; | |
2450 | ||
044fa782 SR |
2451 | bpage = *data_page; |
2452 | if (!bpage) | |
8789a9e7 SR |
2453 | return 0; |
2454 | ||
2455 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2456 | ||
2457 | /* | |
2458 | * rb_buffer_peek will get the next ring buffer if | |
2459 | * the current reader page is empty. | |
2460 | */ | |
2461 | event = rb_buffer_peek(buffer, cpu, NULL); | |
2462 | if (!event) | |
2463 | goto out; | |
2464 | ||
2465 | /* check for data */ | |
2466 | if (!local_read(&cpu_buffer->reader_page->page->commit)) | |
2467 | goto out; | |
667d2412 LJ |
2468 | |
2469 | read = cpu_buffer->reader_page->read; | |
8789a9e7 SR |
2470 | /* |
2471 | * If the writer is already off of the read page, then simply | |
2472 | * switch the read page with the given page. Otherwise | |
2473 | * we need to copy the data from the reader to the writer. | |
2474 | */ | |
2475 | if (cpu_buffer->reader_page == cpu_buffer->commit_page) { | |
b85fa01e | 2476 | unsigned int commit = rb_page_commit(cpu_buffer->reader_page); |
667d2412 | 2477 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
8789a9e7 SR |
2478 | |
2479 | if (full) | |
2480 | goto out; | |
2481 | /* The writer is still on the reader page, we must copy */ | |
667d2412 | 2482 | memcpy(bpage->data + read, rpage->data + read, commit - read); |
8789a9e7 SR |
2483 | |
2484 | /* consume what was read */ | |
b85fa01e | 2485 | cpu_buffer->reader_page->read = commit; |
667d2412 LJ |
2486 | |
2487 | /* update bpage */ | |
2488 | local_set(&bpage->commit, commit); | |
2489 | if (!read) | |
2490 | bpage->time_stamp = rpage->time_stamp; | |
8789a9e7 SR |
2491 | } else { |
2492 | /* swap the pages */ | |
044fa782 SR |
2493 | rb_init_page(bpage); |
2494 | bpage = cpu_buffer->reader_page->page; | |
8789a9e7 SR |
2495 | cpu_buffer->reader_page->page = *data_page; |
2496 | cpu_buffer->reader_page->read = 0; | |
044fa782 | 2497 | *data_page = bpage; |
8789a9e7 | 2498 | } |
667d2412 | 2499 | ret = read; |
8789a9e7 SR |
2500 | |
2501 | /* update the entry counter */ | |
667d2412 | 2502 | rb_remove_entries(cpu_buffer, bpage, read); |
8789a9e7 SR |
2503 | out: |
2504 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2505 | ||
2506 | return ret; | |
2507 | } | |
2508 | ||
a3583244 SR |
2509 | static ssize_t |
2510 | rb_simple_read(struct file *filp, char __user *ubuf, | |
2511 | size_t cnt, loff_t *ppos) | |
2512 | { | |
5e39841c | 2513 | unsigned long *p = filp->private_data; |
a3583244 SR |
2514 | char buf[64]; |
2515 | int r; | |
2516 | ||
033601a3 SR |
2517 | if (test_bit(RB_BUFFERS_DISABLED_BIT, p)) |
2518 | r = sprintf(buf, "permanently disabled\n"); | |
2519 | else | |
2520 | r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p)); | |
a3583244 SR |
2521 | |
2522 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
2523 | } | |
2524 | ||
2525 | static ssize_t | |
2526 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
2527 | size_t cnt, loff_t *ppos) | |
2528 | { | |
5e39841c | 2529 | unsigned long *p = filp->private_data; |
a3583244 | 2530 | char buf[64]; |
5e39841c | 2531 | unsigned long val; |
a3583244 SR |
2532 | int ret; |
2533 | ||
2534 | if (cnt >= sizeof(buf)) | |
2535 | return -EINVAL; | |
2536 | ||
2537 | if (copy_from_user(&buf, ubuf, cnt)) | |
2538 | return -EFAULT; | |
2539 | ||
2540 | buf[cnt] = 0; | |
2541 | ||
2542 | ret = strict_strtoul(buf, 10, &val); | |
2543 | if (ret < 0) | |
2544 | return ret; | |
2545 | ||
033601a3 SR |
2546 | if (val) |
2547 | set_bit(RB_BUFFERS_ON_BIT, p); | |
2548 | else | |
2549 | clear_bit(RB_BUFFERS_ON_BIT, p); | |
a3583244 SR |
2550 | |
2551 | (*ppos)++; | |
2552 | ||
2553 | return cnt; | |
2554 | } | |
2555 | ||
2556 | static struct file_operations rb_simple_fops = { | |
2557 | .open = tracing_open_generic, | |
2558 | .read = rb_simple_read, | |
2559 | .write = rb_simple_write, | |
2560 | }; | |
2561 | ||
2562 | ||
2563 | static __init int rb_init_debugfs(void) | |
2564 | { | |
2565 | struct dentry *d_tracer; | |
2566 | struct dentry *entry; | |
2567 | ||
2568 | d_tracer = tracing_init_dentry(); | |
2569 | ||
2570 | entry = debugfs_create_file("tracing_on", 0644, d_tracer, | |
033601a3 | 2571 | &ring_buffer_flags, &rb_simple_fops); |
a3583244 SR |
2572 | if (!entry) |
2573 | pr_warning("Could not create debugfs 'tracing_on' entry\n"); | |
2574 | ||
2575 | return 0; | |
2576 | } | |
2577 | ||
2578 | fs_initcall(rb_init_debugfs); |