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> | |
14131f2f | 7 | #include <linux/trace_clock.h> |
78d904b4 | 8 | #include <linux/ftrace_irq.h> |
7a8e76a3 SR |
9 | #include <linux/spinlock.h> |
10 | #include <linux/debugfs.h> | |
11 | #include <linux/uaccess.h> | |
a81bd80a | 12 | #include <linux/hardirq.h> |
7a8e76a3 SR |
13 | #include <linux/module.h> |
14 | #include <linux/percpu.h> | |
15 | #include <linux/mutex.h> | |
7a8e76a3 SR |
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 | ||
7a8e76a3 SR |
115 | u64 ring_buffer_time_stamp(int cpu) |
116 | { | |
47e74f2b SR |
117 | u64 time; |
118 | ||
119 | preempt_disable_notrace(); | |
7a8e76a3 | 120 | /* shift to debug/test normalization and TIME_EXTENTS */ |
14131f2f | 121 | time = trace_clock_local() << DEBUG_SHIFT; |
2c2d7329 | 122 | preempt_enable_no_resched_notrace(); |
47e74f2b SR |
123 | |
124 | return time; | |
7a8e76a3 | 125 | } |
c4f50183 | 126 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); |
7a8e76a3 SR |
127 | |
128 | void ring_buffer_normalize_time_stamp(int cpu, u64 *ts) | |
129 | { | |
130 | /* Just stupid testing the normalize function and deltas */ | |
131 | *ts >>= DEBUG_SHIFT; | |
132 | } | |
c4f50183 | 133 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); |
7a8e76a3 SR |
134 | |
135 | #define RB_EVNT_HDR_SIZE (sizeof(struct ring_buffer_event)) | |
67d34724 | 136 | #define RB_ALIGNMENT 4U |
7a8e76a3 SR |
137 | #define RB_MAX_SMALL_DATA 28 |
138 | ||
139 | enum { | |
140 | RB_LEN_TIME_EXTEND = 8, | |
141 | RB_LEN_TIME_STAMP = 16, | |
142 | }; | |
143 | ||
144 | /* inline for ring buffer fast paths */ | |
34a148bf | 145 | static unsigned |
7a8e76a3 SR |
146 | rb_event_length(struct ring_buffer_event *event) |
147 | { | |
148 | unsigned length; | |
149 | ||
150 | switch (event->type) { | |
151 | case RINGBUF_TYPE_PADDING: | |
152 | /* undefined */ | |
153 | return -1; | |
154 | ||
155 | case RINGBUF_TYPE_TIME_EXTEND: | |
156 | return RB_LEN_TIME_EXTEND; | |
157 | ||
158 | case RINGBUF_TYPE_TIME_STAMP: | |
159 | return RB_LEN_TIME_STAMP; | |
160 | ||
161 | case RINGBUF_TYPE_DATA: | |
162 | if (event->len) | |
67d34724 | 163 | length = event->len * RB_ALIGNMENT; |
7a8e76a3 SR |
164 | else |
165 | length = event->array[0]; | |
166 | return length + RB_EVNT_HDR_SIZE; | |
167 | default: | |
168 | BUG(); | |
169 | } | |
170 | /* not hit */ | |
171 | return 0; | |
172 | } | |
173 | ||
174 | /** | |
175 | * ring_buffer_event_length - return the length of the event | |
176 | * @event: the event to get the length of | |
177 | */ | |
178 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
179 | { | |
465634ad RR |
180 | unsigned length = rb_event_length(event); |
181 | if (event->type != RINGBUF_TYPE_DATA) | |
182 | return length; | |
183 | length -= RB_EVNT_HDR_SIZE; | |
184 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
185 | length -= sizeof(event->array[0]); | |
186 | return length; | |
7a8e76a3 | 187 | } |
c4f50183 | 188 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
189 | |
190 | /* inline for ring buffer fast paths */ | |
34a148bf | 191 | static void * |
7a8e76a3 SR |
192 | rb_event_data(struct ring_buffer_event *event) |
193 | { | |
194 | BUG_ON(event->type != RINGBUF_TYPE_DATA); | |
195 | /* If length is in len field, then array[0] has the data */ | |
196 | if (event->len) | |
197 | return (void *)&event->array[0]; | |
198 | /* Otherwise length is in array[0] and array[1] has the data */ | |
199 | return (void *)&event->array[1]; | |
200 | } | |
201 | ||
202 | /** | |
203 | * ring_buffer_event_data - return the data of the event | |
204 | * @event: the event to get the data from | |
205 | */ | |
206 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
207 | { | |
208 | return rb_event_data(event); | |
209 | } | |
c4f50183 | 210 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
211 | |
212 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 213 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
214 | |
215 | #define TS_SHIFT 27 | |
216 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
217 | #define TS_DELTA_TEST (~TS_MASK) | |
218 | ||
abc9b56d | 219 | struct buffer_data_page { |
e4c2ce82 | 220 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 221 | local_t commit; /* write committed index */ |
abc9b56d SR |
222 | unsigned char data[]; /* data of buffer page */ |
223 | }; | |
224 | ||
225 | struct buffer_page { | |
226 | local_t write; /* index for next write */ | |
6f807acd | 227 | unsigned read; /* index for next read */ |
e4c2ce82 | 228 | struct list_head list; /* list of free pages */ |
abc9b56d | 229 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
230 | }; |
231 | ||
044fa782 | 232 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 233 | { |
044fa782 | 234 | local_set(&bpage->commit, 0); |
abc9b56d SR |
235 | } |
236 | ||
ed56829c SR |
237 | /* |
238 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
239 | * this issue out. | |
240 | */ | |
34a148bf | 241 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 242 | { |
34a148bf | 243 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 244 | kfree(bpage); |
ed56829c SR |
245 | } |
246 | ||
7a8e76a3 SR |
247 | /* |
248 | * We need to fit the time_stamp delta into 27 bits. | |
249 | */ | |
250 | static inline int test_time_stamp(u64 delta) | |
251 | { | |
252 | if (delta & TS_DELTA_TEST) | |
253 | return 1; | |
254 | return 0; | |
255 | } | |
256 | ||
082605de | 257 | #define BUF_PAGE_SIZE (PAGE_SIZE - offsetof(struct buffer_data_page, data)) |
7a8e76a3 SR |
258 | |
259 | /* | |
260 | * head_page == tail_page && head == tail then buffer is empty. | |
261 | */ | |
262 | struct ring_buffer_per_cpu { | |
263 | int cpu; | |
264 | struct ring_buffer *buffer; | |
f83c9d0f | 265 | spinlock_t reader_lock; /* serialize readers */ |
3e03fb7f | 266 | raw_spinlock_t lock; |
7a8e76a3 SR |
267 | struct lock_class_key lock_key; |
268 | struct list_head pages; | |
6f807acd SR |
269 | struct buffer_page *head_page; /* read from head */ |
270 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 271 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 272 | struct buffer_page *reader_page; |
7a8e76a3 SR |
273 | unsigned long overrun; |
274 | unsigned long entries; | |
275 | u64 write_stamp; | |
276 | u64 read_stamp; | |
277 | atomic_t record_disabled; | |
278 | }; | |
279 | ||
280 | struct ring_buffer { | |
7a8e76a3 SR |
281 | unsigned pages; |
282 | unsigned flags; | |
283 | int cpus; | |
7a8e76a3 | 284 | atomic_t record_disabled; |
00f62f61 | 285 | cpumask_var_t cpumask; |
7a8e76a3 SR |
286 | |
287 | struct mutex mutex; | |
288 | ||
289 | struct ring_buffer_per_cpu **buffers; | |
290 | }; | |
291 | ||
292 | struct ring_buffer_iter { | |
293 | struct ring_buffer_per_cpu *cpu_buffer; | |
294 | unsigned long head; | |
295 | struct buffer_page *head_page; | |
296 | u64 read_stamp; | |
297 | }; | |
298 | ||
f536aafc | 299 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
bf41a158 | 300 | #define RB_WARN_ON(buffer, cond) \ |
3e89c7bb SR |
301 | ({ \ |
302 | int _____ret = unlikely(cond); \ | |
303 | if (_____ret) { \ | |
bf41a158 SR |
304 | atomic_inc(&buffer->record_disabled); \ |
305 | WARN_ON(1); \ | |
306 | } \ | |
3e89c7bb SR |
307 | _____ret; \ |
308 | }) | |
f536aafc | 309 | |
7a8e76a3 SR |
310 | /** |
311 | * check_pages - integrity check of buffer pages | |
312 | * @cpu_buffer: CPU buffer with pages to test | |
313 | * | |
c3706f00 | 314 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
315 | * been corrupted. |
316 | */ | |
317 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
318 | { | |
319 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 320 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 321 | |
3e89c7bb SR |
322 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
323 | return -1; | |
324 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
325 | return -1; | |
7a8e76a3 | 326 | |
044fa782 | 327 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 328 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 329 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
330 | return -1; |
331 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 332 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 333 | return -1; |
7a8e76a3 SR |
334 | } |
335 | ||
336 | return 0; | |
337 | } | |
338 | ||
7a8e76a3 SR |
339 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
340 | unsigned nr_pages) | |
341 | { | |
342 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 343 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
344 | unsigned long addr; |
345 | LIST_HEAD(pages); | |
346 | unsigned i; | |
347 | ||
348 | for (i = 0; i < nr_pages; i++) { | |
044fa782 | 349 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
aa1e0e3b | 350 | GFP_KERNEL, cpu_to_node(cpu_buffer->cpu)); |
044fa782 | 351 | if (!bpage) |
e4c2ce82 | 352 | goto free_pages; |
044fa782 | 353 | list_add(&bpage->list, &pages); |
e4c2ce82 | 354 | |
7a8e76a3 SR |
355 | addr = __get_free_page(GFP_KERNEL); |
356 | if (!addr) | |
357 | goto free_pages; | |
044fa782 SR |
358 | bpage->page = (void *)addr; |
359 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
360 | } |
361 | ||
362 | list_splice(&pages, head); | |
363 | ||
364 | rb_check_pages(cpu_buffer); | |
365 | ||
366 | return 0; | |
367 | ||
368 | free_pages: | |
044fa782 SR |
369 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
370 | list_del_init(&bpage->list); | |
371 | free_buffer_page(bpage); | |
7a8e76a3 SR |
372 | } |
373 | return -ENOMEM; | |
374 | } | |
375 | ||
376 | static struct ring_buffer_per_cpu * | |
377 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) | |
378 | { | |
379 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 380 | struct buffer_page *bpage; |
d769041f | 381 | unsigned long addr; |
7a8e76a3 SR |
382 | int ret; |
383 | ||
384 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
385 | GFP_KERNEL, cpu_to_node(cpu)); | |
386 | if (!cpu_buffer) | |
387 | return NULL; | |
388 | ||
389 | cpu_buffer->cpu = cpu; | |
390 | cpu_buffer->buffer = buffer; | |
f83c9d0f | 391 | spin_lock_init(&cpu_buffer->reader_lock); |
3e03fb7f | 392 | cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; |
7a8e76a3 SR |
393 | INIT_LIST_HEAD(&cpu_buffer->pages); |
394 | ||
044fa782 | 395 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 396 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 397 | if (!bpage) |
e4c2ce82 SR |
398 | goto fail_free_buffer; |
399 | ||
044fa782 | 400 | cpu_buffer->reader_page = bpage; |
d769041f SR |
401 | addr = __get_free_page(GFP_KERNEL); |
402 | if (!addr) | |
e4c2ce82 | 403 | goto fail_free_reader; |
044fa782 SR |
404 | bpage->page = (void *)addr; |
405 | rb_init_page(bpage->page); | |
e4c2ce82 | 406 | |
d769041f | 407 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 408 | |
7a8e76a3 SR |
409 | ret = rb_allocate_pages(cpu_buffer, buffer->pages); |
410 | if (ret < 0) | |
d769041f | 411 | goto fail_free_reader; |
7a8e76a3 SR |
412 | |
413 | cpu_buffer->head_page | |
414 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 415 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 SR |
416 | |
417 | return cpu_buffer; | |
418 | ||
d769041f SR |
419 | fail_free_reader: |
420 | free_buffer_page(cpu_buffer->reader_page); | |
421 | ||
7a8e76a3 SR |
422 | fail_free_buffer: |
423 | kfree(cpu_buffer); | |
424 | return NULL; | |
425 | } | |
426 | ||
427 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
428 | { | |
429 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 430 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 431 | |
d769041f SR |
432 | list_del_init(&cpu_buffer->reader_page->list); |
433 | free_buffer_page(cpu_buffer->reader_page); | |
434 | ||
044fa782 SR |
435 | list_for_each_entry_safe(bpage, tmp, head, list) { |
436 | list_del_init(&bpage->list); | |
437 | free_buffer_page(bpage); | |
7a8e76a3 SR |
438 | } |
439 | kfree(cpu_buffer); | |
440 | } | |
441 | ||
a7b13743 SR |
442 | /* |
443 | * Causes compile errors if the struct buffer_page gets bigger | |
444 | * than the struct page. | |
445 | */ | |
446 | extern int ring_buffer_page_too_big(void); | |
447 | ||
7a8e76a3 SR |
448 | /** |
449 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 450 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
451 | * @flags: attributes to set for the ring buffer. |
452 | * | |
453 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
454 | * flag. This flag means that the buffer will overwrite old data | |
455 | * when the buffer wraps. If this flag is not set, the buffer will | |
456 | * drop data when the tail hits the head. | |
457 | */ | |
458 | struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags) | |
459 | { | |
460 | struct ring_buffer *buffer; | |
461 | int bsize; | |
462 | int cpu; | |
463 | ||
a7b13743 SR |
464 | /* Paranoid! Optimizes out when all is well */ |
465 | if (sizeof(struct buffer_page) > sizeof(struct page)) | |
466 | ring_buffer_page_too_big(); | |
467 | ||
468 | ||
7a8e76a3 SR |
469 | /* keep it in its own cache line */ |
470 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
471 | GFP_KERNEL); | |
472 | if (!buffer) | |
473 | return NULL; | |
474 | ||
9e01c1b7 RR |
475 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
476 | goto fail_free_buffer; | |
477 | ||
7a8e76a3 SR |
478 | buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
479 | buffer->flags = flags; | |
480 | ||
481 | /* need at least two pages */ | |
482 | if (buffer->pages == 1) | |
483 | buffer->pages++; | |
484 | ||
9e01c1b7 | 485 | cpumask_copy(buffer->cpumask, cpu_possible_mask); |
7a8e76a3 SR |
486 | buffer->cpus = nr_cpu_ids; |
487 | ||
488 | bsize = sizeof(void *) * nr_cpu_ids; | |
489 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
490 | GFP_KERNEL); | |
491 | if (!buffer->buffers) | |
9e01c1b7 | 492 | goto fail_free_cpumask; |
7a8e76a3 SR |
493 | |
494 | for_each_buffer_cpu(buffer, cpu) { | |
495 | buffer->buffers[cpu] = | |
496 | rb_allocate_cpu_buffer(buffer, cpu); | |
497 | if (!buffer->buffers[cpu]) | |
498 | goto fail_free_buffers; | |
499 | } | |
500 | ||
501 | mutex_init(&buffer->mutex); | |
502 | ||
503 | return buffer; | |
504 | ||
505 | fail_free_buffers: | |
506 | for_each_buffer_cpu(buffer, cpu) { | |
507 | if (buffer->buffers[cpu]) | |
508 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
509 | } | |
510 | kfree(buffer->buffers); | |
511 | ||
9e01c1b7 RR |
512 | fail_free_cpumask: |
513 | free_cpumask_var(buffer->cpumask); | |
514 | ||
7a8e76a3 SR |
515 | fail_free_buffer: |
516 | kfree(buffer); | |
517 | return NULL; | |
518 | } | |
c4f50183 | 519 | EXPORT_SYMBOL_GPL(ring_buffer_alloc); |
7a8e76a3 SR |
520 | |
521 | /** | |
522 | * ring_buffer_free - free a ring buffer. | |
523 | * @buffer: the buffer to free. | |
524 | */ | |
525 | void | |
526 | ring_buffer_free(struct ring_buffer *buffer) | |
527 | { | |
528 | int cpu; | |
529 | ||
530 | for_each_buffer_cpu(buffer, cpu) | |
531 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
532 | ||
9e01c1b7 RR |
533 | free_cpumask_var(buffer->cpumask); |
534 | ||
7a8e76a3 SR |
535 | kfree(buffer); |
536 | } | |
c4f50183 | 537 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 SR |
538 | |
539 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); | |
540 | ||
541 | static void | |
542 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages) | |
543 | { | |
044fa782 | 544 | struct buffer_page *bpage; |
7a8e76a3 SR |
545 | struct list_head *p; |
546 | unsigned i; | |
547 | ||
548 | atomic_inc(&cpu_buffer->record_disabled); | |
549 | synchronize_sched(); | |
550 | ||
551 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
552 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
553 | return; | |
7a8e76a3 | 554 | p = cpu_buffer->pages.next; |
044fa782 SR |
555 | bpage = list_entry(p, struct buffer_page, list); |
556 | list_del_init(&bpage->list); | |
557 | free_buffer_page(bpage); | |
7a8e76a3 | 558 | } |
3e89c7bb SR |
559 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
560 | return; | |
7a8e76a3 SR |
561 | |
562 | rb_reset_cpu(cpu_buffer); | |
563 | ||
564 | rb_check_pages(cpu_buffer); | |
565 | ||
566 | atomic_dec(&cpu_buffer->record_disabled); | |
567 | ||
568 | } | |
569 | ||
570 | static void | |
571 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
572 | struct list_head *pages, unsigned nr_pages) | |
573 | { | |
044fa782 | 574 | struct buffer_page *bpage; |
7a8e76a3 SR |
575 | struct list_head *p; |
576 | unsigned i; | |
577 | ||
578 | atomic_inc(&cpu_buffer->record_disabled); | |
579 | synchronize_sched(); | |
580 | ||
581 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
582 | if (RB_WARN_ON(cpu_buffer, list_empty(pages))) |
583 | return; | |
7a8e76a3 | 584 | p = pages->next; |
044fa782 SR |
585 | bpage = list_entry(p, struct buffer_page, list); |
586 | list_del_init(&bpage->list); | |
587 | list_add_tail(&bpage->list, &cpu_buffer->pages); | |
7a8e76a3 SR |
588 | } |
589 | rb_reset_cpu(cpu_buffer); | |
590 | ||
591 | rb_check_pages(cpu_buffer); | |
592 | ||
593 | atomic_dec(&cpu_buffer->record_disabled); | |
594 | } | |
595 | ||
596 | /** | |
597 | * ring_buffer_resize - resize the ring buffer | |
598 | * @buffer: the buffer to resize. | |
599 | * @size: the new size. | |
600 | * | |
601 | * The tracer is responsible for making sure that the buffer is | |
602 | * not being used while changing the size. | |
603 | * Note: We may be able to change the above requirement by using | |
604 | * RCU synchronizations. | |
605 | * | |
606 | * Minimum size is 2 * BUF_PAGE_SIZE. | |
607 | * | |
608 | * Returns -1 on failure. | |
609 | */ | |
610 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) | |
611 | { | |
612 | struct ring_buffer_per_cpu *cpu_buffer; | |
613 | unsigned nr_pages, rm_pages, new_pages; | |
044fa782 | 614 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
615 | unsigned long buffer_size; |
616 | unsigned long addr; | |
617 | LIST_HEAD(pages); | |
618 | int i, cpu; | |
619 | ||
ee51a1de IM |
620 | /* |
621 | * Always succeed at resizing a non-existent buffer: | |
622 | */ | |
623 | if (!buffer) | |
624 | return size; | |
625 | ||
7a8e76a3 SR |
626 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
627 | size *= BUF_PAGE_SIZE; | |
628 | buffer_size = buffer->pages * BUF_PAGE_SIZE; | |
629 | ||
630 | /* we need a minimum of two pages */ | |
631 | if (size < BUF_PAGE_SIZE * 2) | |
632 | size = BUF_PAGE_SIZE * 2; | |
633 | ||
634 | if (size == buffer_size) | |
635 | return size; | |
636 | ||
637 | mutex_lock(&buffer->mutex); | |
638 | ||
639 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
640 | ||
641 | if (size < buffer_size) { | |
642 | ||
643 | /* easy case, just free pages */ | |
3e89c7bb SR |
644 | if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) { |
645 | mutex_unlock(&buffer->mutex); | |
646 | return -1; | |
647 | } | |
7a8e76a3 SR |
648 | |
649 | rm_pages = buffer->pages - nr_pages; | |
650 | ||
651 | for_each_buffer_cpu(buffer, cpu) { | |
652 | cpu_buffer = buffer->buffers[cpu]; | |
653 | rb_remove_pages(cpu_buffer, rm_pages); | |
654 | } | |
655 | goto out; | |
656 | } | |
657 | ||
658 | /* | |
659 | * This is a bit more difficult. We only want to add pages | |
660 | * when we can allocate enough for all CPUs. We do this | |
661 | * by allocating all the pages and storing them on a local | |
662 | * link list. If we succeed in our allocation, then we | |
663 | * add these pages to the cpu_buffers. Otherwise we just free | |
664 | * them all and return -ENOMEM; | |
665 | */ | |
3e89c7bb SR |
666 | if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) { |
667 | mutex_unlock(&buffer->mutex); | |
668 | return -1; | |
669 | } | |
f536aafc | 670 | |
7a8e76a3 SR |
671 | new_pages = nr_pages - buffer->pages; |
672 | ||
673 | for_each_buffer_cpu(buffer, cpu) { | |
674 | for (i = 0; i < new_pages; i++) { | |
044fa782 | 675 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), |
e4c2ce82 SR |
676 | cache_line_size()), |
677 | GFP_KERNEL, cpu_to_node(cpu)); | |
044fa782 | 678 | if (!bpage) |
e4c2ce82 | 679 | goto free_pages; |
044fa782 | 680 | list_add(&bpage->list, &pages); |
7a8e76a3 SR |
681 | addr = __get_free_page(GFP_KERNEL); |
682 | if (!addr) | |
683 | goto free_pages; | |
044fa782 SR |
684 | bpage->page = (void *)addr; |
685 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
686 | } |
687 | } | |
688 | ||
689 | for_each_buffer_cpu(buffer, cpu) { | |
690 | cpu_buffer = buffer->buffers[cpu]; | |
691 | rb_insert_pages(cpu_buffer, &pages, new_pages); | |
692 | } | |
693 | ||
3e89c7bb SR |
694 | if (RB_WARN_ON(buffer, !list_empty(&pages))) { |
695 | mutex_unlock(&buffer->mutex); | |
696 | return -1; | |
697 | } | |
7a8e76a3 SR |
698 | |
699 | out: | |
700 | buffer->pages = nr_pages; | |
701 | mutex_unlock(&buffer->mutex); | |
702 | ||
703 | return size; | |
704 | ||
705 | free_pages: | |
044fa782 SR |
706 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
707 | list_del_init(&bpage->list); | |
708 | free_buffer_page(bpage); | |
7a8e76a3 | 709 | } |
641d2f63 | 710 | mutex_unlock(&buffer->mutex); |
7a8e76a3 SR |
711 | return -ENOMEM; |
712 | } | |
c4f50183 | 713 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 714 | |
7a8e76a3 SR |
715 | static inline int rb_null_event(struct ring_buffer_event *event) |
716 | { | |
717 | return event->type == RINGBUF_TYPE_PADDING; | |
718 | } | |
719 | ||
8789a9e7 | 720 | static inline void * |
044fa782 | 721 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 722 | { |
044fa782 | 723 | return bpage->data + index; |
8789a9e7 SR |
724 | } |
725 | ||
044fa782 | 726 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 727 | { |
044fa782 | 728 | return bpage->page->data + index; |
7a8e76a3 SR |
729 | } |
730 | ||
731 | static inline struct ring_buffer_event * | |
d769041f | 732 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 733 | { |
6f807acd SR |
734 | return __rb_page_index(cpu_buffer->reader_page, |
735 | cpu_buffer->reader_page->read); | |
736 | } | |
737 | ||
738 | static inline struct ring_buffer_event * | |
739 | rb_head_event(struct ring_buffer_per_cpu *cpu_buffer) | |
740 | { | |
741 | return __rb_page_index(cpu_buffer->head_page, | |
742 | cpu_buffer->head_page->read); | |
7a8e76a3 SR |
743 | } |
744 | ||
745 | static inline struct ring_buffer_event * | |
746 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
747 | { | |
6f807acd | 748 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
749 | } |
750 | ||
bf41a158 SR |
751 | static inline unsigned rb_page_write(struct buffer_page *bpage) |
752 | { | |
753 | return local_read(&bpage->write); | |
754 | } | |
755 | ||
756 | static inline unsigned rb_page_commit(struct buffer_page *bpage) | |
757 | { | |
abc9b56d | 758 | return local_read(&bpage->page->commit); |
bf41a158 SR |
759 | } |
760 | ||
761 | /* Size is determined by what has been commited */ | |
762 | static inline unsigned rb_page_size(struct buffer_page *bpage) | |
763 | { | |
764 | return rb_page_commit(bpage); | |
765 | } | |
766 | ||
767 | static inline unsigned | |
768 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
769 | { | |
770 | return rb_page_commit(cpu_buffer->commit_page); | |
771 | } | |
772 | ||
773 | static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer) | |
774 | { | |
775 | return rb_page_commit(cpu_buffer->head_page); | |
776 | } | |
777 | ||
7a8e76a3 SR |
778 | /* |
779 | * When the tail hits the head and the buffer is in overwrite mode, | |
780 | * the head jumps to the next page and all content on the previous | |
781 | * page is discarded. But before doing so, we update the overrun | |
782 | * variable of the buffer. | |
783 | */ | |
784 | static void rb_update_overflow(struct ring_buffer_per_cpu *cpu_buffer) | |
785 | { | |
786 | struct ring_buffer_event *event; | |
787 | unsigned long head; | |
788 | ||
789 | for (head = 0; head < rb_head_size(cpu_buffer); | |
790 | head += rb_event_length(event)) { | |
791 | ||
6f807acd | 792 | event = __rb_page_index(cpu_buffer->head_page, head); |
3e89c7bb SR |
793 | if (RB_WARN_ON(cpu_buffer, rb_null_event(event))) |
794 | return; | |
7a8e76a3 SR |
795 | /* Only count data entries */ |
796 | if (event->type != RINGBUF_TYPE_DATA) | |
797 | continue; | |
798 | cpu_buffer->overrun++; | |
799 | cpu_buffer->entries--; | |
800 | } | |
801 | } | |
802 | ||
803 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
044fa782 | 804 | struct buffer_page **bpage) |
7a8e76a3 | 805 | { |
044fa782 | 806 | struct list_head *p = (*bpage)->list.next; |
7a8e76a3 SR |
807 | |
808 | if (p == &cpu_buffer->pages) | |
809 | p = p->next; | |
810 | ||
044fa782 | 811 | *bpage = list_entry(p, struct buffer_page, list); |
7a8e76a3 SR |
812 | } |
813 | ||
bf41a158 SR |
814 | static inline unsigned |
815 | rb_event_index(struct ring_buffer_event *event) | |
816 | { | |
817 | unsigned long addr = (unsigned long)event; | |
818 | ||
819 | return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE); | |
820 | } | |
821 | ||
34a148bf | 822 | static int |
bf41a158 SR |
823 | rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
824 | struct ring_buffer_event *event) | |
825 | { | |
826 | unsigned long addr = (unsigned long)event; | |
827 | unsigned long index; | |
828 | ||
829 | index = rb_event_index(event); | |
830 | addr &= PAGE_MASK; | |
831 | ||
832 | return cpu_buffer->commit_page->page == (void *)addr && | |
833 | rb_commit_index(cpu_buffer) == index; | |
834 | } | |
835 | ||
34a148bf | 836 | static void |
bf41a158 SR |
837 | rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer, |
838 | struct ring_buffer_event *event) | |
7a8e76a3 | 839 | { |
bf41a158 SR |
840 | unsigned long addr = (unsigned long)event; |
841 | unsigned long index; | |
842 | ||
843 | index = rb_event_index(event); | |
844 | addr &= PAGE_MASK; | |
845 | ||
846 | while (cpu_buffer->commit_page->page != (void *)addr) { | |
3e89c7bb SR |
847 | if (RB_WARN_ON(cpu_buffer, |
848 | cpu_buffer->commit_page == cpu_buffer->tail_page)) | |
849 | return; | |
abc9b56d | 850 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
851 | cpu_buffer->commit_page->write; |
852 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
853 | cpu_buffer->write_stamp = |
854 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
855 | } |
856 | ||
857 | /* Now set the commit to the event's index */ | |
abc9b56d | 858 | local_set(&cpu_buffer->commit_page->page->commit, index); |
7a8e76a3 SR |
859 | } |
860 | ||
34a148bf | 861 | static void |
bf41a158 | 862 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 863 | { |
bf41a158 SR |
864 | /* |
865 | * We only race with interrupts and NMIs on this CPU. | |
866 | * If we own the commit event, then we can commit | |
867 | * all others that interrupted us, since the interruptions | |
868 | * are in stack format (they finish before they come | |
869 | * back to us). This allows us to do a simple loop to | |
870 | * assign the commit to the tail. | |
871 | */ | |
a8ccf1d6 | 872 | again: |
bf41a158 | 873 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
abc9b56d | 874 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
875 | cpu_buffer->commit_page->write; |
876 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
877 | cpu_buffer->write_stamp = |
878 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
879 | /* add barrier to keep gcc from optimizing too much */ |
880 | barrier(); | |
881 | } | |
882 | while (rb_commit_index(cpu_buffer) != | |
883 | rb_page_write(cpu_buffer->commit_page)) { | |
abc9b56d | 884 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
885 | cpu_buffer->commit_page->write; |
886 | barrier(); | |
887 | } | |
a8ccf1d6 SR |
888 | |
889 | /* again, keep gcc from optimizing */ | |
890 | barrier(); | |
891 | ||
892 | /* | |
893 | * If an interrupt came in just after the first while loop | |
894 | * and pushed the tail page forward, we will be left with | |
895 | * a dangling commit that will never go forward. | |
896 | */ | |
897 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
898 | goto again; | |
7a8e76a3 SR |
899 | } |
900 | ||
d769041f | 901 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 902 | { |
abc9b56d | 903 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 904 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
905 | } |
906 | ||
34a148bf | 907 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
908 | { |
909 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
910 | ||
911 | /* | |
912 | * The iterator could be on the reader page (it starts there). | |
913 | * But the head could have moved, since the reader was | |
914 | * found. Check for this case and assign the iterator | |
915 | * to the head page instead of next. | |
916 | */ | |
917 | if (iter->head_page == cpu_buffer->reader_page) | |
918 | iter->head_page = cpu_buffer->head_page; | |
919 | else | |
920 | rb_inc_page(cpu_buffer, &iter->head_page); | |
921 | ||
abc9b56d | 922 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
923 | iter->head = 0; |
924 | } | |
925 | ||
926 | /** | |
927 | * ring_buffer_update_event - update event type and data | |
928 | * @event: the even to update | |
929 | * @type: the type of event | |
930 | * @length: the size of the event field in the ring buffer | |
931 | * | |
932 | * Update the type and data fields of the event. The length | |
933 | * is the actual size that is written to the ring buffer, | |
934 | * and with this, we can determine what to place into the | |
935 | * data field. | |
936 | */ | |
34a148bf | 937 | static void |
7a8e76a3 SR |
938 | rb_update_event(struct ring_buffer_event *event, |
939 | unsigned type, unsigned length) | |
940 | { | |
941 | event->type = type; | |
942 | ||
943 | switch (type) { | |
944 | ||
945 | case RINGBUF_TYPE_PADDING: | |
946 | break; | |
947 | ||
948 | case RINGBUF_TYPE_TIME_EXTEND: | |
67d34724 | 949 | event->len = DIV_ROUND_UP(RB_LEN_TIME_EXTEND, RB_ALIGNMENT); |
7a8e76a3 SR |
950 | break; |
951 | ||
952 | case RINGBUF_TYPE_TIME_STAMP: | |
67d34724 | 953 | event->len = DIV_ROUND_UP(RB_LEN_TIME_STAMP, RB_ALIGNMENT); |
7a8e76a3 SR |
954 | break; |
955 | ||
956 | case RINGBUF_TYPE_DATA: | |
957 | length -= RB_EVNT_HDR_SIZE; | |
958 | if (length > RB_MAX_SMALL_DATA) { | |
959 | event->len = 0; | |
960 | event->array[0] = length; | |
961 | } else | |
67d34724 | 962 | event->len = DIV_ROUND_UP(length, RB_ALIGNMENT); |
7a8e76a3 SR |
963 | break; |
964 | default: | |
965 | BUG(); | |
966 | } | |
967 | } | |
968 | ||
34a148bf | 969 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
970 | { |
971 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
972 | ||
973 | /* zero length can cause confusions */ | |
974 | if (!length) | |
975 | length = 1; | |
976 | ||
977 | if (length > RB_MAX_SMALL_DATA) | |
978 | length += sizeof(event.array[0]); | |
979 | ||
980 | length += RB_EVNT_HDR_SIZE; | |
981 | length = ALIGN(length, RB_ALIGNMENT); | |
982 | ||
983 | return length; | |
984 | } | |
985 | ||
986 | static struct ring_buffer_event * | |
987 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
988 | unsigned type, unsigned long length, u64 *ts) | |
989 | { | |
98db8df7 | 990 | struct buffer_page *tail_page, *head_page, *reader_page, *commit_page; |
bf41a158 | 991 | unsigned long tail, write; |
7a8e76a3 SR |
992 | struct ring_buffer *buffer = cpu_buffer->buffer; |
993 | struct ring_buffer_event *event; | |
bf41a158 | 994 | unsigned long flags; |
78d904b4 | 995 | bool lock_taken = false; |
7a8e76a3 | 996 | |
98db8df7 SR |
997 | commit_page = cpu_buffer->commit_page; |
998 | /* we just need to protect against interrupts */ | |
999 | barrier(); | |
7a8e76a3 | 1000 | tail_page = cpu_buffer->tail_page; |
bf41a158 SR |
1001 | write = local_add_return(length, &tail_page->write); |
1002 | tail = write - length; | |
7a8e76a3 | 1003 | |
bf41a158 SR |
1004 | /* See if we shot pass the end of this buffer page */ |
1005 | if (write > BUF_PAGE_SIZE) { | |
7a8e76a3 SR |
1006 | struct buffer_page *next_page = tail_page; |
1007 | ||
3e03fb7f | 1008 | local_irq_save(flags); |
78d904b4 | 1009 | /* |
a81bd80a SR |
1010 | * Since the write to the buffer is still not |
1011 | * fully lockless, we must be careful with NMIs. | |
1012 | * The locks in the writers are taken when a write | |
1013 | * crosses to a new page. The locks protect against | |
1014 | * races with the readers (this will soon be fixed | |
1015 | * with a lockless solution). | |
1016 | * | |
1017 | * Because we can not protect against NMIs, and we | |
1018 | * want to keep traces reentrant, we need to manage | |
1019 | * what happens when we are in an NMI. | |
1020 | * | |
78d904b4 SR |
1021 | * NMIs can happen after we take the lock. |
1022 | * If we are in an NMI, only take the lock | |
1023 | * if it is not already taken. Otherwise | |
1024 | * simply fail. | |
1025 | */ | |
a81bd80a | 1026 | if (unlikely(in_nmi())) { |
78d904b4 | 1027 | if (!__raw_spin_trylock(&cpu_buffer->lock)) |
45141d46 | 1028 | goto out_reset; |
78d904b4 SR |
1029 | } else |
1030 | __raw_spin_lock(&cpu_buffer->lock); | |
1031 | ||
1032 | lock_taken = true; | |
bf41a158 | 1033 | |
7a8e76a3 SR |
1034 | rb_inc_page(cpu_buffer, &next_page); |
1035 | ||
d769041f SR |
1036 | head_page = cpu_buffer->head_page; |
1037 | reader_page = cpu_buffer->reader_page; | |
1038 | ||
1039 | /* we grabbed the lock before incrementing */ | |
3e89c7bb | 1040 | if (RB_WARN_ON(cpu_buffer, next_page == reader_page)) |
45141d46 | 1041 | goto out_reset; |
bf41a158 SR |
1042 | |
1043 | /* | |
1044 | * If for some reason, we had an interrupt storm that made | |
1045 | * it all the way around the buffer, bail, and warn | |
1046 | * about it. | |
1047 | */ | |
98db8df7 | 1048 | if (unlikely(next_page == commit_page)) { |
bf41a158 | 1049 | WARN_ON_ONCE(1); |
45141d46 | 1050 | goto out_reset; |
bf41a158 | 1051 | } |
d769041f | 1052 | |
7a8e76a3 | 1053 | if (next_page == head_page) { |
6f3b3440 | 1054 | if (!(buffer->flags & RB_FL_OVERWRITE)) |
45141d46 | 1055 | goto out_reset; |
7a8e76a3 | 1056 | |
bf41a158 SR |
1057 | /* tail_page has not moved yet? */ |
1058 | if (tail_page == cpu_buffer->tail_page) { | |
1059 | /* count overflows */ | |
1060 | rb_update_overflow(cpu_buffer); | |
1061 | ||
1062 | rb_inc_page(cpu_buffer, &head_page); | |
1063 | cpu_buffer->head_page = head_page; | |
1064 | cpu_buffer->head_page->read = 0; | |
1065 | } | |
1066 | } | |
7a8e76a3 | 1067 | |
bf41a158 SR |
1068 | /* |
1069 | * If the tail page is still the same as what we think | |
1070 | * it is, then it is up to us to update the tail | |
1071 | * pointer. | |
1072 | */ | |
1073 | if (tail_page == cpu_buffer->tail_page) { | |
1074 | local_set(&next_page->write, 0); | |
abc9b56d | 1075 | local_set(&next_page->page->commit, 0); |
bf41a158 SR |
1076 | cpu_buffer->tail_page = next_page; |
1077 | ||
1078 | /* reread the time stamp */ | |
1079 | *ts = ring_buffer_time_stamp(cpu_buffer->cpu); | |
abc9b56d | 1080 | cpu_buffer->tail_page->page->time_stamp = *ts; |
7a8e76a3 SR |
1081 | } |
1082 | ||
bf41a158 SR |
1083 | /* |
1084 | * The actual tail page has moved forward. | |
1085 | */ | |
1086 | if (tail < BUF_PAGE_SIZE) { | |
1087 | /* Mark the rest of the page with padding */ | |
6f807acd | 1088 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
1089 | event->type = RINGBUF_TYPE_PADDING; |
1090 | } | |
1091 | ||
bf41a158 SR |
1092 | if (tail <= BUF_PAGE_SIZE) |
1093 | /* Set the write back to the previous setting */ | |
1094 | local_set(&tail_page->write, tail); | |
1095 | ||
1096 | /* | |
1097 | * If this was a commit entry that failed, | |
1098 | * increment that too | |
1099 | */ | |
1100 | if (tail_page == cpu_buffer->commit_page && | |
1101 | tail == rb_commit_index(cpu_buffer)) { | |
1102 | rb_set_commit_to_write(cpu_buffer); | |
1103 | } | |
1104 | ||
3e03fb7f SR |
1105 | __raw_spin_unlock(&cpu_buffer->lock); |
1106 | local_irq_restore(flags); | |
bf41a158 SR |
1107 | |
1108 | /* fail and let the caller try again */ | |
1109 | return ERR_PTR(-EAGAIN); | |
7a8e76a3 SR |
1110 | } |
1111 | ||
bf41a158 SR |
1112 | /* We reserved something on the buffer */ |
1113 | ||
3e89c7bb SR |
1114 | if (RB_WARN_ON(cpu_buffer, write > BUF_PAGE_SIZE)) |
1115 | return NULL; | |
7a8e76a3 | 1116 | |
6f807acd | 1117 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
1118 | rb_update_event(event, type, length); |
1119 | ||
bf41a158 SR |
1120 | /* |
1121 | * If this is a commit and the tail is zero, then update | |
1122 | * this page's time stamp. | |
1123 | */ | |
1124 | if (!tail && rb_is_commit(cpu_buffer, event)) | |
abc9b56d | 1125 | cpu_buffer->commit_page->page->time_stamp = *ts; |
bf41a158 | 1126 | |
7a8e76a3 | 1127 | return event; |
bf41a158 | 1128 | |
45141d46 | 1129 | out_reset: |
6f3b3440 LJ |
1130 | /* reset write */ |
1131 | if (tail <= BUF_PAGE_SIZE) | |
1132 | local_set(&tail_page->write, tail); | |
1133 | ||
78d904b4 SR |
1134 | if (likely(lock_taken)) |
1135 | __raw_spin_unlock(&cpu_buffer->lock); | |
3e03fb7f | 1136 | local_irq_restore(flags); |
bf41a158 | 1137 | return NULL; |
7a8e76a3 SR |
1138 | } |
1139 | ||
1140 | static int | |
1141 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1142 | u64 *ts, u64 *delta) | |
1143 | { | |
1144 | struct ring_buffer_event *event; | |
1145 | static int once; | |
bf41a158 | 1146 | int ret; |
7a8e76a3 SR |
1147 | |
1148 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1149 | printk(KERN_WARNING "Delta way too big! %llu" | |
1150 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1151 | (unsigned long long)*delta, |
1152 | (unsigned long long)*ts, | |
1153 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1154 | WARN_ON(1); |
1155 | } | |
1156 | ||
1157 | /* | |
1158 | * The delta is too big, we to add a | |
1159 | * new timestamp. | |
1160 | */ | |
1161 | event = __rb_reserve_next(cpu_buffer, | |
1162 | RINGBUF_TYPE_TIME_EXTEND, | |
1163 | RB_LEN_TIME_EXTEND, | |
1164 | ts); | |
1165 | if (!event) | |
bf41a158 | 1166 | return -EBUSY; |
7a8e76a3 | 1167 | |
bf41a158 SR |
1168 | if (PTR_ERR(event) == -EAGAIN) |
1169 | return -EAGAIN; | |
1170 | ||
1171 | /* Only a commited time event can update the write stamp */ | |
1172 | if (rb_is_commit(cpu_buffer, event)) { | |
1173 | /* | |
1174 | * If this is the first on the page, then we need to | |
1175 | * update the page itself, and just put in a zero. | |
1176 | */ | |
1177 | if (rb_event_index(event)) { | |
1178 | event->time_delta = *delta & TS_MASK; | |
1179 | event->array[0] = *delta >> TS_SHIFT; | |
1180 | } else { | |
abc9b56d | 1181 | cpu_buffer->commit_page->page->time_stamp = *ts; |
bf41a158 SR |
1182 | event->time_delta = 0; |
1183 | event->array[0] = 0; | |
1184 | } | |
7a8e76a3 | 1185 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
1186 | /* let the caller know this was the commit */ |
1187 | ret = 1; | |
1188 | } else { | |
1189 | /* Darn, this is just wasted space */ | |
1190 | event->time_delta = 0; | |
1191 | event->array[0] = 0; | |
1192 | ret = 0; | |
7a8e76a3 SR |
1193 | } |
1194 | ||
bf41a158 SR |
1195 | *delta = 0; |
1196 | ||
1197 | return ret; | |
7a8e76a3 SR |
1198 | } |
1199 | ||
1200 | static struct ring_buffer_event * | |
1201 | rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, | |
1202 | unsigned type, unsigned long length) | |
1203 | { | |
1204 | struct ring_buffer_event *event; | |
1205 | u64 ts, delta; | |
bf41a158 | 1206 | int commit = 0; |
818e3dd3 | 1207 | int nr_loops = 0; |
7a8e76a3 | 1208 | |
bf41a158 | 1209 | again: |
818e3dd3 SR |
1210 | /* |
1211 | * We allow for interrupts to reenter here and do a trace. | |
1212 | * If one does, it will cause this original code to loop | |
1213 | * back here. Even with heavy interrupts happening, this | |
1214 | * should only happen a few times in a row. If this happens | |
1215 | * 1000 times in a row, there must be either an interrupt | |
1216 | * storm or we have something buggy. | |
1217 | * Bail! | |
1218 | */ | |
3e89c7bb | 1219 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
818e3dd3 | 1220 | return NULL; |
818e3dd3 | 1221 | |
7a8e76a3 SR |
1222 | ts = ring_buffer_time_stamp(cpu_buffer->cpu); |
1223 | ||
bf41a158 SR |
1224 | /* |
1225 | * Only the first commit can update the timestamp. | |
1226 | * Yes there is a race here. If an interrupt comes in | |
1227 | * just after the conditional and it traces too, then it | |
1228 | * will also check the deltas. More than one timestamp may | |
1229 | * also be made. But only the entry that did the actual | |
1230 | * commit will be something other than zero. | |
1231 | */ | |
1232 | if (cpu_buffer->tail_page == cpu_buffer->commit_page && | |
1233 | rb_page_write(cpu_buffer->tail_page) == | |
1234 | rb_commit_index(cpu_buffer)) { | |
1235 | ||
7a8e76a3 SR |
1236 | delta = ts - cpu_buffer->write_stamp; |
1237 | ||
bf41a158 SR |
1238 | /* make sure this delta is calculated here */ |
1239 | barrier(); | |
1240 | ||
1241 | /* Did the write stamp get updated already? */ | |
1242 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
4143c5cb | 1243 | delta = 0; |
bf41a158 | 1244 | |
7a8e76a3 | 1245 | if (test_time_stamp(delta)) { |
7a8e76a3 | 1246 | |
bf41a158 SR |
1247 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
1248 | ||
1249 | if (commit == -EBUSY) | |
7a8e76a3 | 1250 | return NULL; |
bf41a158 SR |
1251 | |
1252 | if (commit == -EAGAIN) | |
1253 | goto again; | |
1254 | ||
1255 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 1256 | } |
bf41a158 SR |
1257 | } else |
1258 | /* Non commits have zero deltas */ | |
7a8e76a3 | 1259 | delta = 0; |
7a8e76a3 SR |
1260 | |
1261 | event = __rb_reserve_next(cpu_buffer, type, length, &ts); | |
bf41a158 SR |
1262 | if (PTR_ERR(event) == -EAGAIN) |
1263 | goto again; | |
1264 | ||
1265 | if (!event) { | |
1266 | if (unlikely(commit)) | |
1267 | /* | |
1268 | * Ouch! We needed a timestamp and it was commited. But | |
1269 | * we didn't get our event reserved. | |
1270 | */ | |
1271 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 | 1272 | return NULL; |
bf41a158 | 1273 | } |
7a8e76a3 | 1274 | |
bf41a158 SR |
1275 | /* |
1276 | * If the timestamp was commited, make the commit our entry | |
1277 | * now so that we will update it when needed. | |
1278 | */ | |
1279 | if (commit) | |
1280 | rb_set_commit_event(cpu_buffer, event); | |
1281 | else if (!rb_is_commit(cpu_buffer, event)) | |
7a8e76a3 SR |
1282 | delta = 0; |
1283 | ||
1284 | event->time_delta = delta; | |
1285 | ||
1286 | return event; | |
1287 | } | |
1288 | ||
bf41a158 SR |
1289 | static DEFINE_PER_CPU(int, rb_need_resched); |
1290 | ||
7a8e76a3 SR |
1291 | /** |
1292 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
1293 | * @buffer: the ring buffer to reserve from | |
1294 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
1295 | * |
1296 | * Returns a reseverd event on the ring buffer to copy directly to. | |
1297 | * The user of this interface will need to get the body to write into | |
1298 | * and can use the ring_buffer_event_data() interface. | |
1299 | * | |
1300 | * The length is the length of the data needed, not the event length | |
1301 | * which also includes the event header. | |
1302 | * | |
1303 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
1304 | * If NULL is returned, then nothing has been allocated or locked. | |
1305 | */ | |
1306 | struct ring_buffer_event * | |
0a987751 | 1307 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
1308 | { |
1309 | struct ring_buffer_per_cpu *cpu_buffer; | |
1310 | struct ring_buffer_event *event; | |
bf41a158 | 1311 | int cpu, resched; |
7a8e76a3 | 1312 | |
033601a3 | 1313 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1314 | return NULL; |
1315 | ||
7a8e76a3 SR |
1316 | if (atomic_read(&buffer->record_disabled)) |
1317 | return NULL; | |
1318 | ||
bf41a158 | 1319 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 1320 | resched = ftrace_preempt_disable(); |
bf41a158 | 1321 | |
7a8e76a3 SR |
1322 | cpu = raw_smp_processor_id(); |
1323 | ||
9e01c1b7 | 1324 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1325 | goto out; |
7a8e76a3 SR |
1326 | |
1327 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1328 | |
1329 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 1330 | goto out; |
7a8e76a3 SR |
1331 | |
1332 | length = rb_calculate_event_length(length); | |
1333 | if (length > BUF_PAGE_SIZE) | |
bf41a158 | 1334 | goto out; |
7a8e76a3 SR |
1335 | |
1336 | event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length); | |
1337 | if (!event) | |
d769041f | 1338 | goto out; |
7a8e76a3 | 1339 | |
bf41a158 SR |
1340 | /* |
1341 | * Need to store resched state on this cpu. | |
1342 | * Only the first needs to. | |
1343 | */ | |
1344 | ||
1345 | if (preempt_count() == 1) | |
1346 | per_cpu(rb_need_resched, cpu) = resched; | |
1347 | ||
7a8e76a3 SR |
1348 | return event; |
1349 | ||
d769041f | 1350 | out: |
182e9f5f | 1351 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1352 | return NULL; |
1353 | } | |
c4f50183 | 1354 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 SR |
1355 | |
1356 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
1357 | struct ring_buffer_event *event) | |
1358 | { | |
7a8e76a3 | 1359 | cpu_buffer->entries++; |
bf41a158 SR |
1360 | |
1361 | /* Only process further if we own the commit */ | |
1362 | if (!rb_is_commit(cpu_buffer, event)) | |
1363 | return; | |
1364 | ||
1365 | cpu_buffer->write_stamp += event->time_delta; | |
1366 | ||
1367 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 SR |
1368 | } |
1369 | ||
1370 | /** | |
1371 | * ring_buffer_unlock_commit - commit a reserved | |
1372 | * @buffer: The buffer to commit to | |
1373 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
1374 | * |
1375 | * This commits the data to the ring buffer, and releases any locks held. | |
1376 | * | |
1377 | * Must be paired with ring_buffer_lock_reserve. | |
1378 | */ | |
1379 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 1380 | struct ring_buffer_event *event) |
7a8e76a3 SR |
1381 | { |
1382 | struct ring_buffer_per_cpu *cpu_buffer; | |
1383 | int cpu = raw_smp_processor_id(); | |
1384 | ||
1385 | cpu_buffer = buffer->buffers[cpu]; | |
1386 | ||
7a8e76a3 SR |
1387 | rb_commit(cpu_buffer, event); |
1388 | ||
bf41a158 SR |
1389 | /* |
1390 | * Only the last preempt count needs to restore preemption. | |
1391 | */ | |
182e9f5f SR |
1392 | if (preempt_count() == 1) |
1393 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1394 | else | |
bf41a158 | 1395 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
1396 | |
1397 | return 0; | |
1398 | } | |
c4f50183 | 1399 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 SR |
1400 | |
1401 | /** | |
1402 | * ring_buffer_write - write data to the buffer without reserving | |
1403 | * @buffer: The ring buffer to write to. | |
1404 | * @length: The length of the data being written (excluding the event header) | |
1405 | * @data: The data to write to the buffer. | |
1406 | * | |
1407 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
1408 | * one function. If you already have the data to write to the buffer, it | |
1409 | * may be easier to simply call this function. | |
1410 | * | |
1411 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
1412 | * and not the length of the event which would hold the header. | |
1413 | */ | |
1414 | int ring_buffer_write(struct ring_buffer *buffer, | |
1415 | unsigned long length, | |
1416 | void *data) | |
1417 | { | |
1418 | struct ring_buffer_per_cpu *cpu_buffer; | |
1419 | struct ring_buffer_event *event; | |
bf41a158 | 1420 | unsigned long event_length; |
7a8e76a3 SR |
1421 | void *body; |
1422 | int ret = -EBUSY; | |
bf41a158 | 1423 | int cpu, resched; |
7a8e76a3 | 1424 | |
033601a3 | 1425 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1426 | return -EBUSY; |
1427 | ||
7a8e76a3 SR |
1428 | if (atomic_read(&buffer->record_disabled)) |
1429 | return -EBUSY; | |
1430 | ||
182e9f5f | 1431 | resched = ftrace_preempt_disable(); |
bf41a158 | 1432 | |
7a8e76a3 SR |
1433 | cpu = raw_smp_processor_id(); |
1434 | ||
9e01c1b7 | 1435 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1436 | goto out; |
7a8e76a3 SR |
1437 | |
1438 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1439 | |
1440 | if (atomic_read(&cpu_buffer->record_disabled)) | |
1441 | goto out; | |
1442 | ||
1443 | event_length = rb_calculate_event_length(length); | |
1444 | event = rb_reserve_next_event(cpu_buffer, | |
1445 | RINGBUF_TYPE_DATA, event_length); | |
1446 | if (!event) | |
1447 | goto out; | |
1448 | ||
1449 | body = rb_event_data(event); | |
1450 | ||
1451 | memcpy(body, data, length); | |
1452 | ||
1453 | rb_commit(cpu_buffer, event); | |
1454 | ||
1455 | ret = 0; | |
1456 | out: | |
182e9f5f | 1457 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1458 | |
1459 | return ret; | |
1460 | } | |
c4f50183 | 1461 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 1462 | |
34a148bf | 1463 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
1464 | { |
1465 | struct buffer_page *reader = cpu_buffer->reader_page; | |
1466 | struct buffer_page *head = cpu_buffer->head_page; | |
1467 | struct buffer_page *commit = cpu_buffer->commit_page; | |
1468 | ||
1469 | return reader->read == rb_page_commit(reader) && | |
1470 | (commit == reader || | |
1471 | (commit == head && | |
1472 | head->read == rb_page_commit(commit))); | |
1473 | } | |
1474 | ||
7a8e76a3 SR |
1475 | /** |
1476 | * ring_buffer_record_disable - stop all writes into the buffer | |
1477 | * @buffer: The ring buffer to stop writes to. | |
1478 | * | |
1479 | * This prevents all writes to the buffer. Any attempt to write | |
1480 | * to the buffer after this will fail and return NULL. | |
1481 | * | |
1482 | * The caller should call synchronize_sched() after this. | |
1483 | */ | |
1484 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
1485 | { | |
1486 | atomic_inc(&buffer->record_disabled); | |
1487 | } | |
c4f50183 | 1488 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
1489 | |
1490 | /** | |
1491 | * ring_buffer_record_enable - enable writes to the buffer | |
1492 | * @buffer: The ring buffer to enable writes | |
1493 | * | |
1494 | * Note, multiple disables will need the same number of enables | |
1495 | * to truely enable the writing (much like preempt_disable). | |
1496 | */ | |
1497 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
1498 | { | |
1499 | atomic_dec(&buffer->record_disabled); | |
1500 | } | |
c4f50183 | 1501 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 SR |
1502 | |
1503 | /** | |
1504 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
1505 | * @buffer: The ring buffer to stop writes to. | |
1506 | * @cpu: The CPU buffer to stop | |
1507 | * | |
1508 | * This prevents all writes to the buffer. Any attempt to write | |
1509 | * to the buffer after this will fail and return NULL. | |
1510 | * | |
1511 | * The caller should call synchronize_sched() after this. | |
1512 | */ | |
1513 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
1514 | { | |
1515 | struct ring_buffer_per_cpu *cpu_buffer; | |
1516 | ||
9e01c1b7 | 1517 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
1518 | return; |
1519 | ||
1520 | cpu_buffer = buffer->buffers[cpu]; | |
1521 | atomic_inc(&cpu_buffer->record_disabled); | |
1522 | } | |
c4f50183 | 1523 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
1524 | |
1525 | /** | |
1526 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
1527 | * @buffer: The ring buffer to enable writes | |
1528 | * @cpu: The CPU to enable. | |
1529 | * | |
1530 | * Note, multiple disables will need the same number of enables | |
1531 | * to truely enable the writing (much like preempt_disable). | |
1532 | */ | |
1533 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
1534 | { | |
1535 | struct ring_buffer_per_cpu *cpu_buffer; | |
1536 | ||
9e01c1b7 | 1537 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
1538 | return; |
1539 | ||
1540 | cpu_buffer = buffer->buffers[cpu]; | |
1541 | atomic_dec(&cpu_buffer->record_disabled); | |
1542 | } | |
c4f50183 | 1543 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 SR |
1544 | |
1545 | /** | |
1546 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
1547 | * @buffer: The ring buffer | |
1548 | * @cpu: The per CPU buffer to get the entries from. | |
1549 | */ | |
1550 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
1551 | { | |
1552 | struct ring_buffer_per_cpu *cpu_buffer; | |
1553 | ||
9e01c1b7 | 1554 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
1555 | return 0; |
1556 | ||
1557 | cpu_buffer = buffer->buffers[cpu]; | |
1558 | return cpu_buffer->entries; | |
1559 | } | |
c4f50183 | 1560 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
1561 | |
1562 | /** | |
1563 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
1564 | * @buffer: The ring buffer | |
1565 | * @cpu: The per CPU buffer to get the number of overruns from | |
1566 | */ | |
1567 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
1568 | { | |
1569 | struct ring_buffer_per_cpu *cpu_buffer; | |
1570 | ||
9e01c1b7 | 1571 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
1572 | return 0; |
1573 | ||
1574 | cpu_buffer = buffer->buffers[cpu]; | |
1575 | return cpu_buffer->overrun; | |
1576 | } | |
c4f50183 | 1577 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 SR |
1578 | |
1579 | /** | |
1580 | * ring_buffer_entries - get the number of entries in a buffer | |
1581 | * @buffer: The ring buffer | |
1582 | * | |
1583 | * Returns the total number of entries in the ring buffer | |
1584 | * (all CPU entries) | |
1585 | */ | |
1586 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
1587 | { | |
1588 | struct ring_buffer_per_cpu *cpu_buffer; | |
1589 | unsigned long entries = 0; | |
1590 | int cpu; | |
1591 | ||
1592 | /* if you care about this being correct, lock the buffer */ | |
1593 | for_each_buffer_cpu(buffer, cpu) { | |
1594 | cpu_buffer = buffer->buffers[cpu]; | |
1595 | entries += cpu_buffer->entries; | |
1596 | } | |
1597 | ||
1598 | return entries; | |
1599 | } | |
c4f50183 | 1600 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
1601 | |
1602 | /** | |
1603 | * ring_buffer_overrun_cpu - get the number of overruns in buffer | |
1604 | * @buffer: The ring buffer | |
1605 | * | |
1606 | * Returns the total number of overruns in the ring buffer | |
1607 | * (all CPU entries) | |
1608 | */ | |
1609 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
1610 | { | |
1611 | struct ring_buffer_per_cpu *cpu_buffer; | |
1612 | unsigned long overruns = 0; | |
1613 | int cpu; | |
1614 | ||
1615 | /* if you care about this being correct, lock the buffer */ | |
1616 | for_each_buffer_cpu(buffer, cpu) { | |
1617 | cpu_buffer = buffer->buffers[cpu]; | |
1618 | overruns += cpu_buffer->overrun; | |
1619 | } | |
1620 | ||
1621 | return overruns; | |
1622 | } | |
c4f50183 | 1623 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 1624 | |
642edba5 | 1625 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
1626 | { |
1627 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1628 | ||
d769041f SR |
1629 | /* Iterator usage is expected to have record disabled */ |
1630 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
1631 | iter->head_page = cpu_buffer->head_page; | |
6f807acd | 1632 | iter->head = cpu_buffer->head_page->read; |
d769041f SR |
1633 | } else { |
1634 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 1635 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
1636 | } |
1637 | if (iter->head) | |
1638 | iter->read_stamp = cpu_buffer->read_stamp; | |
1639 | else | |
abc9b56d | 1640 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 1641 | } |
f83c9d0f | 1642 | |
642edba5 SR |
1643 | /** |
1644 | * ring_buffer_iter_reset - reset an iterator | |
1645 | * @iter: The iterator to reset | |
1646 | * | |
1647 | * Resets the iterator, so that it will start from the beginning | |
1648 | * again. | |
1649 | */ | |
1650 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
1651 | { | |
1652 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1653 | unsigned long flags; | |
1654 | ||
1655 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
1656 | rb_iter_reset(iter); | |
f83c9d0f | 1657 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 1658 | } |
c4f50183 | 1659 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
1660 | |
1661 | /** | |
1662 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
1663 | * @iter: The iterator to check | |
1664 | */ | |
1665 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
1666 | { | |
1667 | struct ring_buffer_per_cpu *cpu_buffer; | |
1668 | ||
1669 | cpu_buffer = iter->cpu_buffer; | |
1670 | ||
bf41a158 SR |
1671 | return iter->head_page == cpu_buffer->commit_page && |
1672 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 1673 | } |
c4f50183 | 1674 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
1675 | |
1676 | static void | |
1677 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1678 | struct ring_buffer_event *event) | |
1679 | { | |
1680 | u64 delta; | |
1681 | ||
1682 | switch (event->type) { | |
1683 | case RINGBUF_TYPE_PADDING: | |
1684 | return; | |
1685 | ||
1686 | case RINGBUF_TYPE_TIME_EXTEND: | |
1687 | delta = event->array[0]; | |
1688 | delta <<= TS_SHIFT; | |
1689 | delta += event->time_delta; | |
1690 | cpu_buffer->read_stamp += delta; | |
1691 | return; | |
1692 | ||
1693 | case RINGBUF_TYPE_TIME_STAMP: | |
1694 | /* FIXME: not implemented */ | |
1695 | return; | |
1696 | ||
1697 | case RINGBUF_TYPE_DATA: | |
1698 | cpu_buffer->read_stamp += event->time_delta; | |
1699 | return; | |
1700 | ||
1701 | default: | |
1702 | BUG(); | |
1703 | } | |
1704 | return; | |
1705 | } | |
1706 | ||
1707 | static void | |
1708 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
1709 | struct ring_buffer_event *event) | |
1710 | { | |
1711 | u64 delta; | |
1712 | ||
1713 | switch (event->type) { | |
1714 | case RINGBUF_TYPE_PADDING: | |
1715 | return; | |
1716 | ||
1717 | case RINGBUF_TYPE_TIME_EXTEND: | |
1718 | delta = event->array[0]; | |
1719 | delta <<= TS_SHIFT; | |
1720 | delta += event->time_delta; | |
1721 | iter->read_stamp += delta; | |
1722 | return; | |
1723 | ||
1724 | case RINGBUF_TYPE_TIME_STAMP: | |
1725 | /* FIXME: not implemented */ | |
1726 | return; | |
1727 | ||
1728 | case RINGBUF_TYPE_DATA: | |
1729 | iter->read_stamp += event->time_delta; | |
1730 | return; | |
1731 | ||
1732 | default: | |
1733 | BUG(); | |
1734 | } | |
1735 | return; | |
1736 | } | |
1737 | ||
d769041f SR |
1738 | static struct buffer_page * |
1739 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 1740 | { |
d769041f SR |
1741 | struct buffer_page *reader = NULL; |
1742 | unsigned long flags; | |
818e3dd3 | 1743 | int nr_loops = 0; |
d769041f | 1744 | |
3e03fb7f SR |
1745 | local_irq_save(flags); |
1746 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
1747 | |
1748 | again: | |
818e3dd3 SR |
1749 | /* |
1750 | * This should normally only loop twice. But because the | |
1751 | * start of the reader inserts an empty page, it causes | |
1752 | * a case where we will loop three times. There should be no | |
1753 | * reason to loop four times (that I know of). | |
1754 | */ | |
3e89c7bb | 1755 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
1756 | reader = NULL; |
1757 | goto out; | |
1758 | } | |
1759 | ||
d769041f SR |
1760 | reader = cpu_buffer->reader_page; |
1761 | ||
1762 | /* If there's more to read, return this page */ | |
bf41a158 | 1763 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
1764 | goto out; |
1765 | ||
1766 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
1767 | if (RB_WARN_ON(cpu_buffer, |
1768 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
1769 | goto out; | |
d769041f SR |
1770 | |
1771 | /* check if we caught up to the tail */ | |
1772 | reader = NULL; | |
bf41a158 | 1773 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 1774 | goto out; |
7a8e76a3 SR |
1775 | |
1776 | /* | |
d769041f SR |
1777 | * Splice the empty reader page into the list around the head. |
1778 | * Reset the reader page to size zero. | |
7a8e76a3 | 1779 | */ |
7a8e76a3 | 1780 | |
d769041f SR |
1781 | reader = cpu_buffer->head_page; |
1782 | cpu_buffer->reader_page->list.next = reader->list.next; | |
1783 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 SR |
1784 | |
1785 | local_set(&cpu_buffer->reader_page->write, 0); | |
abc9b56d | 1786 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
7a8e76a3 | 1787 | |
d769041f SR |
1788 | /* Make the reader page now replace the head */ |
1789 | reader->list.prev->next = &cpu_buffer->reader_page->list; | |
1790 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
7a8e76a3 SR |
1791 | |
1792 | /* | |
d769041f SR |
1793 | * If the tail is on the reader, then we must set the head |
1794 | * to the inserted page, otherwise we set it one before. | |
7a8e76a3 | 1795 | */ |
d769041f | 1796 | cpu_buffer->head_page = cpu_buffer->reader_page; |
7a8e76a3 | 1797 | |
bf41a158 | 1798 | if (cpu_buffer->commit_page != reader) |
d769041f SR |
1799 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
1800 | ||
1801 | /* Finally update the reader page to the new head */ | |
1802 | cpu_buffer->reader_page = reader; | |
1803 | rb_reset_reader_page(cpu_buffer); | |
1804 | ||
1805 | goto again; | |
1806 | ||
1807 | out: | |
3e03fb7f SR |
1808 | __raw_spin_unlock(&cpu_buffer->lock); |
1809 | local_irq_restore(flags); | |
d769041f SR |
1810 | |
1811 | return reader; | |
1812 | } | |
1813 | ||
1814 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
1815 | { | |
1816 | struct ring_buffer_event *event; | |
1817 | struct buffer_page *reader; | |
1818 | unsigned length; | |
1819 | ||
1820 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 1821 | |
d769041f | 1822 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
1823 | if (RB_WARN_ON(cpu_buffer, !reader)) |
1824 | return; | |
7a8e76a3 | 1825 | |
d769041f SR |
1826 | event = rb_reader_event(cpu_buffer); |
1827 | ||
1828 | if (event->type == RINGBUF_TYPE_DATA) | |
1829 | cpu_buffer->entries--; | |
1830 | ||
1831 | rb_update_read_stamp(cpu_buffer, event); | |
1832 | ||
1833 | length = rb_event_length(event); | |
6f807acd | 1834 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
1835 | } |
1836 | ||
1837 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
1838 | { | |
1839 | struct ring_buffer *buffer; | |
1840 | struct ring_buffer_per_cpu *cpu_buffer; | |
1841 | struct ring_buffer_event *event; | |
1842 | unsigned length; | |
1843 | ||
1844 | cpu_buffer = iter->cpu_buffer; | |
1845 | buffer = cpu_buffer->buffer; | |
1846 | ||
1847 | /* | |
1848 | * Check if we are at the end of the buffer. | |
1849 | */ | |
bf41a158 | 1850 | if (iter->head >= rb_page_size(iter->head_page)) { |
3e89c7bb SR |
1851 | if (RB_WARN_ON(buffer, |
1852 | iter->head_page == cpu_buffer->commit_page)) | |
1853 | return; | |
d769041f | 1854 | rb_inc_iter(iter); |
7a8e76a3 SR |
1855 | return; |
1856 | } | |
1857 | ||
1858 | event = rb_iter_head_event(iter); | |
1859 | ||
1860 | length = rb_event_length(event); | |
1861 | ||
1862 | /* | |
1863 | * This should not be called to advance the header if we are | |
1864 | * at the tail of the buffer. | |
1865 | */ | |
3e89c7bb | 1866 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 1867 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
1868 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
1869 | return; | |
7a8e76a3 SR |
1870 | |
1871 | rb_update_iter_read_stamp(iter, event); | |
1872 | ||
1873 | iter->head += length; | |
1874 | ||
1875 | /* check for end of page padding */ | |
bf41a158 SR |
1876 | if ((iter->head >= rb_page_size(iter->head_page)) && |
1877 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
1878 | rb_advance_iter(iter); |
1879 | } | |
1880 | ||
f83c9d0f SR |
1881 | static struct ring_buffer_event * |
1882 | rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
7a8e76a3 SR |
1883 | { |
1884 | struct ring_buffer_per_cpu *cpu_buffer; | |
1885 | struct ring_buffer_event *event; | |
d769041f | 1886 | struct buffer_page *reader; |
818e3dd3 | 1887 | int nr_loops = 0; |
7a8e76a3 | 1888 | |
9e01c1b7 | 1889 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
1890 | return NULL; |
1891 | ||
1892 | cpu_buffer = buffer->buffers[cpu]; | |
1893 | ||
1894 | again: | |
818e3dd3 SR |
1895 | /* |
1896 | * We repeat when a timestamp is encountered. It is possible | |
1897 | * to get multiple timestamps from an interrupt entering just | |
1898 | * as one timestamp is about to be written. The max times | |
1899 | * that this can happen is the number of nested interrupts we | |
1900 | * can have. Nesting 10 deep of interrupts is clearly | |
1901 | * an anomaly. | |
1902 | */ | |
3e89c7bb | 1903 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 1904 | return NULL; |
818e3dd3 | 1905 | |
d769041f SR |
1906 | reader = rb_get_reader_page(cpu_buffer); |
1907 | if (!reader) | |
7a8e76a3 SR |
1908 | return NULL; |
1909 | ||
d769041f | 1910 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 SR |
1911 | |
1912 | switch (event->type) { | |
1913 | case RINGBUF_TYPE_PADDING: | |
bf41a158 | 1914 | RB_WARN_ON(cpu_buffer, 1); |
d769041f SR |
1915 | rb_advance_reader(cpu_buffer); |
1916 | return NULL; | |
7a8e76a3 SR |
1917 | |
1918 | case RINGBUF_TYPE_TIME_EXTEND: | |
1919 | /* Internal data, OK to advance */ | |
d769041f | 1920 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
1921 | goto again; |
1922 | ||
1923 | case RINGBUF_TYPE_TIME_STAMP: | |
1924 | /* FIXME: not implemented */ | |
d769041f | 1925 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
1926 | goto again; |
1927 | ||
1928 | case RINGBUF_TYPE_DATA: | |
1929 | if (ts) { | |
1930 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
1931 | ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts); | |
1932 | } | |
1933 | return event; | |
1934 | ||
1935 | default: | |
1936 | BUG(); | |
1937 | } | |
1938 | ||
1939 | return NULL; | |
1940 | } | |
c4f50183 | 1941 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 1942 | |
f83c9d0f SR |
1943 | static struct ring_buffer_event * |
1944 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
1945 | { |
1946 | struct ring_buffer *buffer; | |
1947 | struct ring_buffer_per_cpu *cpu_buffer; | |
1948 | struct ring_buffer_event *event; | |
818e3dd3 | 1949 | int nr_loops = 0; |
7a8e76a3 SR |
1950 | |
1951 | if (ring_buffer_iter_empty(iter)) | |
1952 | return NULL; | |
1953 | ||
1954 | cpu_buffer = iter->cpu_buffer; | |
1955 | buffer = cpu_buffer->buffer; | |
1956 | ||
1957 | again: | |
818e3dd3 SR |
1958 | /* |
1959 | * We repeat when a timestamp is encountered. It is possible | |
1960 | * to get multiple timestamps from an interrupt entering just | |
1961 | * as one timestamp is about to be written. The max times | |
1962 | * that this can happen is the number of nested interrupts we | |
1963 | * can have. Nesting 10 deep of interrupts is clearly | |
1964 | * an anomaly. | |
1965 | */ | |
3e89c7bb | 1966 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 1967 | return NULL; |
818e3dd3 | 1968 | |
7a8e76a3 SR |
1969 | if (rb_per_cpu_empty(cpu_buffer)) |
1970 | return NULL; | |
1971 | ||
1972 | event = rb_iter_head_event(iter); | |
1973 | ||
1974 | switch (event->type) { | |
1975 | case RINGBUF_TYPE_PADDING: | |
d769041f | 1976 | rb_inc_iter(iter); |
7a8e76a3 SR |
1977 | goto again; |
1978 | ||
1979 | case RINGBUF_TYPE_TIME_EXTEND: | |
1980 | /* Internal data, OK to advance */ | |
1981 | rb_advance_iter(iter); | |
1982 | goto again; | |
1983 | ||
1984 | case RINGBUF_TYPE_TIME_STAMP: | |
1985 | /* FIXME: not implemented */ | |
1986 | rb_advance_iter(iter); | |
1987 | goto again; | |
1988 | ||
1989 | case RINGBUF_TYPE_DATA: | |
1990 | if (ts) { | |
1991 | *ts = iter->read_stamp + event->time_delta; | |
1992 | ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts); | |
1993 | } | |
1994 | return event; | |
1995 | ||
1996 | default: | |
1997 | BUG(); | |
1998 | } | |
1999 | ||
2000 | return NULL; | |
2001 | } | |
c4f50183 | 2002 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 2003 | |
f83c9d0f SR |
2004 | /** |
2005 | * ring_buffer_peek - peek at the next event to be read | |
2006 | * @buffer: The ring buffer to read | |
2007 | * @cpu: The cpu to peak at | |
2008 | * @ts: The timestamp counter of this event. | |
2009 | * | |
2010 | * This will return the event that will be read next, but does | |
2011 | * not consume the data. | |
2012 | */ | |
2013 | struct ring_buffer_event * | |
2014 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2015 | { | |
2016 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2017 | struct ring_buffer_event *event; | |
2018 | unsigned long flags; | |
2019 | ||
2020 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2021 | event = rb_buffer_peek(buffer, cpu, ts); | |
2022 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2023 | ||
2024 | return event; | |
2025 | } | |
2026 | ||
2027 | /** | |
2028 | * ring_buffer_iter_peek - peek at the next event to be read | |
2029 | * @iter: The ring buffer iterator | |
2030 | * @ts: The timestamp counter of this event. | |
2031 | * | |
2032 | * This will return the event that will be read next, but does | |
2033 | * not increment the iterator. | |
2034 | */ | |
2035 | struct ring_buffer_event * | |
2036 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
2037 | { | |
2038 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2039 | struct ring_buffer_event *event; | |
2040 | unsigned long flags; | |
2041 | ||
2042 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2043 | event = rb_iter_peek(iter, ts); | |
2044 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2045 | ||
2046 | return event; | |
2047 | } | |
2048 | ||
7a8e76a3 SR |
2049 | /** |
2050 | * ring_buffer_consume - return an event and consume it | |
2051 | * @buffer: The ring buffer to get the next event from | |
2052 | * | |
2053 | * Returns the next event in the ring buffer, and that event is consumed. | |
2054 | * Meaning, that sequential reads will keep returning a different event, | |
2055 | * and eventually empty the ring buffer if the producer is slower. | |
2056 | */ | |
2057 | struct ring_buffer_event * | |
2058 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2059 | { | |
f83c9d0f | 2060 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; |
7a8e76a3 | 2061 | struct ring_buffer_event *event; |
f83c9d0f | 2062 | unsigned long flags; |
7a8e76a3 | 2063 | |
9e01c1b7 | 2064 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
2065 | return NULL; |
2066 | ||
f83c9d0f SR |
2067 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2068 | ||
2069 | event = rb_buffer_peek(buffer, cpu, ts); | |
7a8e76a3 | 2070 | if (!event) |
f83c9d0f | 2071 | goto out; |
7a8e76a3 | 2072 | |
d769041f | 2073 | rb_advance_reader(cpu_buffer); |
7a8e76a3 | 2074 | |
f83c9d0f SR |
2075 | out: |
2076 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2077 | ||
7a8e76a3 SR |
2078 | return event; |
2079 | } | |
c4f50183 | 2080 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
2081 | |
2082 | /** | |
2083 | * ring_buffer_read_start - start a non consuming read of the buffer | |
2084 | * @buffer: The ring buffer to read from | |
2085 | * @cpu: The cpu buffer to iterate over | |
2086 | * | |
2087 | * This starts up an iteration through the buffer. It also disables | |
2088 | * the recording to the buffer until the reading is finished. | |
2089 | * This prevents the reading from being corrupted. This is not | |
2090 | * a consuming read, so a producer is not expected. | |
2091 | * | |
2092 | * Must be paired with ring_buffer_finish. | |
2093 | */ | |
2094 | struct ring_buffer_iter * | |
2095 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
2096 | { | |
2097 | struct ring_buffer_per_cpu *cpu_buffer; | |
2098 | struct ring_buffer_iter *iter; | |
d769041f | 2099 | unsigned long flags; |
7a8e76a3 | 2100 | |
9e01c1b7 | 2101 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
2102 | return NULL; |
2103 | ||
2104 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
2105 | if (!iter) | |
2106 | return NULL; | |
2107 | ||
2108 | cpu_buffer = buffer->buffers[cpu]; | |
2109 | ||
2110 | iter->cpu_buffer = cpu_buffer; | |
2111 | ||
2112 | atomic_inc(&cpu_buffer->record_disabled); | |
2113 | synchronize_sched(); | |
2114 | ||
f83c9d0f | 2115 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 2116 | __raw_spin_lock(&cpu_buffer->lock); |
642edba5 | 2117 | rb_iter_reset(iter); |
3e03fb7f | 2118 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 2119 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
2120 | |
2121 | return iter; | |
2122 | } | |
c4f50183 | 2123 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
2124 | |
2125 | /** | |
2126 | * ring_buffer_finish - finish reading the iterator of the buffer | |
2127 | * @iter: The iterator retrieved by ring_buffer_start | |
2128 | * | |
2129 | * This re-enables the recording to the buffer, and frees the | |
2130 | * iterator. | |
2131 | */ | |
2132 | void | |
2133 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
2134 | { | |
2135 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2136 | ||
2137 | atomic_dec(&cpu_buffer->record_disabled); | |
2138 | kfree(iter); | |
2139 | } | |
c4f50183 | 2140 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
2141 | |
2142 | /** | |
2143 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
2144 | * @iter: The ring buffer iterator | |
2145 | * @ts: The time stamp of the event read. | |
2146 | * | |
2147 | * This reads the next event in the ring buffer and increments the iterator. | |
2148 | */ | |
2149 | struct ring_buffer_event * | |
2150 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
2151 | { | |
2152 | struct ring_buffer_event *event; | |
f83c9d0f SR |
2153 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
2154 | unsigned long flags; | |
7a8e76a3 | 2155 | |
f83c9d0f SR |
2156 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2157 | event = rb_iter_peek(iter, ts); | |
7a8e76a3 | 2158 | if (!event) |
f83c9d0f | 2159 | goto out; |
7a8e76a3 SR |
2160 | |
2161 | rb_advance_iter(iter); | |
f83c9d0f SR |
2162 | out: |
2163 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 SR |
2164 | |
2165 | return event; | |
2166 | } | |
c4f50183 | 2167 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
2168 | |
2169 | /** | |
2170 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
2171 | * @buffer: The ring buffer. | |
2172 | */ | |
2173 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
2174 | { | |
2175 | return BUF_PAGE_SIZE * buffer->pages; | |
2176 | } | |
c4f50183 | 2177 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
2178 | |
2179 | static void | |
2180 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
2181 | { | |
2182 | cpu_buffer->head_page | |
2183 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 2184 | local_set(&cpu_buffer->head_page->write, 0); |
abc9b56d | 2185 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 2186 | |
6f807acd | 2187 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
2188 | |
2189 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
2190 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
2191 | ||
2192 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
2193 | local_set(&cpu_buffer->reader_page->write, 0); | |
abc9b56d | 2194 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 2195 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 2196 | |
7a8e76a3 SR |
2197 | cpu_buffer->overrun = 0; |
2198 | cpu_buffer->entries = 0; | |
69507c06 SR |
2199 | |
2200 | cpu_buffer->write_stamp = 0; | |
2201 | cpu_buffer->read_stamp = 0; | |
7a8e76a3 SR |
2202 | } |
2203 | ||
2204 | /** | |
2205 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
2206 | * @buffer: The ring buffer to reset a per cpu buffer of | |
2207 | * @cpu: The CPU buffer to be reset | |
2208 | */ | |
2209 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
2210 | { | |
2211 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2212 | unsigned long flags; | |
2213 | ||
9e01c1b7 | 2214 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
2215 | return; |
2216 | ||
f83c9d0f SR |
2217 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2218 | ||
3e03fb7f | 2219 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
2220 | |
2221 | rb_reset_cpu(cpu_buffer); | |
2222 | ||
3e03fb7f | 2223 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f SR |
2224 | |
2225 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 2226 | } |
c4f50183 | 2227 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
2228 | |
2229 | /** | |
2230 | * ring_buffer_reset - reset a ring buffer | |
2231 | * @buffer: The ring buffer to reset all cpu buffers | |
2232 | */ | |
2233 | void ring_buffer_reset(struct ring_buffer *buffer) | |
2234 | { | |
7a8e76a3 SR |
2235 | int cpu; |
2236 | ||
7a8e76a3 | 2237 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 2238 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 2239 | } |
c4f50183 | 2240 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
2241 | |
2242 | /** | |
2243 | * rind_buffer_empty - is the ring buffer empty? | |
2244 | * @buffer: The ring buffer to test | |
2245 | */ | |
2246 | int ring_buffer_empty(struct ring_buffer *buffer) | |
2247 | { | |
2248 | struct ring_buffer_per_cpu *cpu_buffer; | |
2249 | int cpu; | |
2250 | ||
2251 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
2252 | for_each_buffer_cpu(buffer, cpu) { | |
2253 | cpu_buffer = buffer->buffers[cpu]; | |
2254 | if (!rb_per_cpu_empty(cpu_buffer)) | |
2255 | return 0; | |
2256 | } | |
2257 | return 1; | |
2258 | } | |
c4f50183 | 2259 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
2260 | |
2261 | /** | |
2262 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
2263 | * @buffer: The ring buffer | |
2264 | * @cpu: The CPU buffer to test | |
2265 | */ | |
2266 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
2267 | { | |
2268 | struct ring_buffer_per_cpu *cpu_buffer; | |
2269 | ||
9e01c1b7 | 2270 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
7a8e76a3 SR |
2271 | return 1; |
2272 | ||
2273 | cpu_buffer = buffer->buffers[cpu]; | |
2274 | return rb_per_cpu_empty(cpu_buffer); | |
2275 | } | |
c4f50183 | 2276 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 SR |
2277 | |
2278 | /** | |
2279 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
2280 | * @buffer_a: One buffer to swap with | |
2281 | * @buffer_b: The other buffer to swap with | |
2282 | * | |
2283 | * This function is useful for tracers that want to take a "snapshot" | |
2284 | * of a CPU buffer and has another back up buffer lying around. | |
2285 | * it is expected that the tracer handles the cpu buffer not being | |
2286 | * used at the moment. | |
2287 | */ | |
2288 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
2289 | struct ring_buffer *buffer_b, int cpu) | |
2290 | { | |
2291 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
2292 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
2293 | ||
9e01c1b7 RR |
2294 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
2295 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
7a8e76a3 SR |
2296 | return -EINVAL; |
2297 | ||
2298 | /* At least make sure the two buffers are somewhat the same */ | |
6d102bc6 | 2299 | if (buffer_a->pages != buffer_b->pages) |
7a8e76a3 SR |
2300 | return -EINVAL; |
2301 | ||
97b17efe SR |
2302 | if (ring_buffer_flags != RB_BUFFERS_ON) |
2303 | return -EAGAIN; | |
2304 | ||
2305 | if (atomic_read(&buffer_a->record_disabled)) | |
2306 | return -EAGAIN; | |
2307 | ||
2308 | if (atomic_read(&buffer_b->record_disabled)) | |
2309 | return -EAGAIN; | |
2310 | ||
7a8e76a3 SR |
2311 | cpu_buffer_a = buffer_a->buffers[cpu]; |
2312 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
2313 | ||
97b17efe SR |
2314 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
2315 | return -EAGAIN; | |
2316 | ||
2317 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
2318 | return -EAGAIN; | |
2319 | ||
7a8e76a3 SR |
2320 | /* |
2321 | * We can't do a synchronize_sched here because this | |
2322 | * function can be called in atomic context. | |
2323 | * Normally this will be called from the same CPU as cpu. | |
2324 | * If not it's up to the caller to protect this. | |
2325 | */ | |
2326 | atomic_inc(&cpu_buffer_a->record_disabled); | |
2327 | atomic_inc(&cpu_buffer_b->record_disabled); | |
2328 | ||
2329 | buffer_a->buffers[cpu] = cpu_buffer_b; | |
2330 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
2331 | ||
2332 | cpu_buffer_b->buffer = buffer_a; | |
2333 | cpu_buffer_a->buffer = buffer_b; | |
2334 | ||
2335 | atomic_dec(&cpu_buffer_a->record_disabled); | |
2336 | atomic_dec(&cpu_buffer_b->record_disabled); | |
2337 | ||
2338 | return 0; | |
2339 | } | |
c4f50183 | 2340 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
7a8e76a3 | 2341 | |
8789a9e7 | 2342 | static void rb_remove_entries(struct ring_buffer_per_cpu *cpu_buffer, |
667d2412 LJ |
2343 | struct buffer_data_page *bpage, |
2344 | unsigned int offset) | |
8789a9e7 SR |
2345 | { |
2346 | struct ring_buffer_event *event; | |
2347 | unsigned long head; | |
2348 | ||
2349 | __raw_spin_lock(&cpu_buffer->lock); | |
667d2412 | 2350 | for (head = offset; head < local_read(&bpage->commit); |
8789a9e7 SR |
2351 | head += rb_event_length(event)) { |
2352 | ||
044fa782 | 2353 | event = __rb_data_page_index(bpage, head); |
8789a9e7 SR |
2354 | if (RB_WARN_ON(cpu_buffer, rb_null_event(event))) |
2355 | return; | |
2356 | /* Only count data entries */ | |
2357 | if (event->type != RINGBUF_TYPE_DATA) | |
2358 | continue; | |
2359 | cpu_buffer->entries--; | |
2360 | } | |
2361 | __raw_spin_unlock(&cpu_buffer->lock); | |
2362 | } | |
2363 | ||
2364 | /** | |
2365 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
2366 | * @buffer: the buffer to allocate for. | |
2367 | * | |
2368 | * This function is used in conjunction with ring_buffer_read_page. | |
2369 | * When reading a full page from the ring buffer, these functions | |
2370 | * can be used to speed up the process. The calling function should | |
2371 | * allocate a few pages first with this function. Then when it | |
2372 | * needs to get pages from the ring buffer, it passes the result | |
2373 | * of this function into ring_buffer_read_page, which will swap | |
2374 | * the page that was allocated, with the read page of the buffer. | |
2375 | * | |
2376 | * Returns: | |
2377 | * The page allocated, or NULL on error. | |
2378 | */ | |
2379 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer) | |
2380 | { | |
2381 | unsigned long addr; | |
044fa782 | 2382 | struct buffer_data_page *bpage; |
8789a9e7 SR |
2383 | |
2384 | addr = __get_free_page(GFP_KERNEL); | |
2385 | if (!addr) | |
2386 | return NULL; | |
2387 | ||
044fa782 | 2388 | bpage = (void *)addr; |
8789a9e7 | 2389 | |
044fa782 | 2390 | return bpage; |
8789a9e7 SR |
2391 | } |
2392 | ||
2393 | /** | |
2394 | * ring_buffer_free_read_page - free an allocated read page | |
2395 | * @buffer: the buffer the page was allocate for | |
2396 | * @data: the page to free | |
2397 | * | |
2398 | * Free a page allocated from ring_buffer_alloc_read_page. | |
2399 | */ | |
2400 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
2401 | { | |
2402 | free_page((unsigned long)data); | |
2403 | } | |
2404 | ||
2405 | /** | |
2406 | * ring_buffer_read_page - extract a page from the ring buffer | |
2407 | * @buffer: buffer to extract from | |
2408 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
2409 | * @cpu: the cpu of the buffer to extract | |
2410 | * @full: should the extraction only happen when the page is full. | |
2411 | * | |
2412 | * This function will pull out a page from the ring buffer and consume it. | |
2413 | * @data_page must be the address of the variable that was returned | |
2414 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
2415 | * to swap with a page in the ring buffer. | |
2416 | * | |
2417 | * for example: | |
b85fa01e | 2418 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
2419 | * if (!rpage) |
2420 | * return error; | |
2421 | * ret = ring_buffer_read_page(buffer, &rpage, cpu, 0); | |
667d2412 LJ |
2422 | * if (ret >= 0) |
2423 | * process_page(rpage, ret); | |
8789a9e7 SR |
2424 | * |
2425 | * When @full is set, the function will not return true unless | |
2426 | * the writer is off the reader page. | |
2427 | * | |
2428 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
2429 | * The ring buffer can be used anywhere in the kernel and can not | |
2430 | * blindly call wake_up. The layer that uses the ring buffer must be | |
2431 | * responsible for that. | |
2432 | * | |
2433 | * Returns: | |
667d2412 LJ |
2434 | * >=0 if data has been transferred, returns the offset of consumed data. |
2435 | * <0 if no data has been transferred. | |
8789a9e7 SR |
2436 | */ |
2437 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
2438 | void **data_page, int cpu, int full) | |
2439 | { | |
2440 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2441 | struct ring_buffer_event *event; | |
044fa782 | 2442 | struct buffer_data_page *bpage; |
8789a9e7 | 2443 | unsigned long flags; |
667d2412 LJ |
2444 | unsigned int read; |
2445 | int ret = -1; | |
8789a9e7 SR |
2446 | |
2447 | if (!data_page) | |
2448 | return 0; | |
2449 | ||
044fa782 SR |
2450 | bpage = *data_page; |
2451 | if (!bpage) | |
8789a9e7 SR |
2452 | return 0; |
2453 | ||
2454 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2455 | ||
2456 | /* | |
2457 | * rb_buffer_peek will get the next ring buffer if | |
2458 | * the current reader page is empty. | |
2459 | */ | |
2460 | event = rb_buffer_peek(buffer, cpu, NULL); | |
2461 | if (!event) | |
2462 | goto out; | |
2463 | ||
2464 | /* check for data */ | |
2465 | if (!local_read(&cpu_buffer->reader_page->page->commit)) | |
2466 | goto out; | |
667d2412 LJ |
2467 | |
2468 | read = cpu_buffer->reader_page->read; | |
8789a9e7 SR |
2469 | /* |
2470 | * If the writer is already off of the read page, then simply | |
2471 | * switch the read page with the given page. Otherwise | |
2472 | * we need to copy the data from the reader to the writer. | |
2473 | */ | |
2474 | if (cpu_buffer->reader_page == cpu_buffer->commit_page) { | |
b85fa01e | 2475 | unsigned int commit = rb_page_commit(cpu_buffer->reader_page); |
667d2412 | 2476 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
8789a9e7 SR |
2477 | |
2478 | if (full) | |
2479 | goto out; | |
2480 | /* The writer is still on the reader page, we must copy */ | |
667d2412 | 2481 | memcpy(bpage->data + read, rpage->data + read, commit - read); |
8789a9e7 SR |
2482 | |
2483 | /* consume what was read */ | |
b85fa01e | 2484 | cpu_buffer->reader_page->read = commit; |
667d2412 LJ |
2485 | |
2486 | /* update bpage */ | |
2487 | local_set(&bpage->commit, commit); | |
2488 | if (!read) | |
2489 | bpage->time_stamp = rpage->time_stamp; | |
8789a9e7 SR |
2490 | } else { |
2491 | /* swap the pages */ | |
044fa782 SR |
2492 | rb_init_page(bpage); |
2493 | bpage = cpu_buffer->reader_page->page; | |
8789a9e7 | 2494 | cpu_buffer->reader_page->page = *data_page; |
41be4da4 | 2495 | local_set(&cpu_buffer->reader_page->write, 0); |
8789a9e7 | 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); |