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