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