Commit | Line | Data |
---|---|---|
7a8e76a3 SR |
1 | /* |
2 | * Generic ring buffer | |
3 | * | |
4 | * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com> | |
5 | */ | |
6 | #include <linux/ring_buffer.h> | |
14131f2f | 7 | #include <linux/trace_clock.h> |
78d904b4 | 8 | #include <linux/ftrace_irq.h> |
7a8e76a3 SR |
9 | #include <linux/spinlock.h> |
10 | #include <linux/debugfs.h> | |
11 | #include <linux/uaccess.h> | |
a81bd80a | 12 | #include <linux/hardirq.h> |
7a8e76a3 SR |
13 | #include <linux/module.h> |
14 | #include <linux/percpu.h> | |
15 | #include <linux/mutex.h> | |
7a8e76a3 SR |
16 | #include <linux/init.h> |
17 | #include <linux/hash.h> | |
18 | #include <linux/list.h> | |
554f786e | 19 | #include <linux/cpu.h> |
7a8e76a3 SR |
20 | #include <linux/fs.h> |
21 | ||
182e9f5f SR |
22 | #include "trace.h" |
23 | ||
5cc98548 SR |
24 | /* |
25 | * The ring buffer is made up of a list of pages. A separate list of pages is | |
26 | * allocated for each CPU. A writer may only write to a buffer that is | |
27 | * associated with the CPU it is currently executing on. A reader may read | |
28 | * from any per cpu buffer. | |
29 | * | |
30 | * The reader is special. For each per cpu buffer, the reader has its own | |
31 | * reader page. When a reader has read the entire reader page, this reader | |
32 | * page is swapped with another page in the ring buffer. | |
33 | * | |
34 | * Now, as long as the writer is off the reader page, the reader can do what | |
35 | * ever it wants with that page. The writer will never write to that page | |
36 | * again (as long as it is out of the ring buffer). | |
37 | * | |
38 | * Here's some silly ASCII art. | |
39 | * | |
40 | * +------+ | |
41 | * |reader| RING BUFFER | |
42 | * |page | | |
43 | * +------+ +---+ +---+ +---+ | |
44 | * | |-->| |-->| | | |
45 | * +---+ +---+ +---+ | |
46 | * ^ | | |
47 | * | | | |
48 | * +---------------+ | |
49 | * | |
50 | * | |
51 | * +------+ | |
52 | * |reader| RING BUFFER | |
53 | * |page |------------------v | |
54 | * +------+ +---+ +---+ +---+ | |
55 | * | |-->| |-->| | | |
56 | * +---+ +---+ +---+ | |
57 | * ^ | | |
58 | * | | | |
59 | * +---------------+ | |
60 | * | |
61 | * | |
62 | * +------+ | |
63 | * |reader| RING BUFFER | |
64 | * |page |------------------v | |
65 | * +------+ +---+ +---+ +---+ | |
66 | * ^ | |-->| |-->| | | |
67 | * | +---+ +---+ +---+ | |
68 | * | | | |
69 | * | | | |
70 | * +------------------------------+ | |
71 | * | |
72 | * | |
73 | * +------+ | |
74 | * |buffer| RING BUFFER | |
75 | * |page |------------------v | |
76 | * +------+ +---+ +---+ +---+ | |
77 | * ^ | | | |-->| | | |
78 | * | New +---+ +---+ +---+ | |
79 | * | Reader------^ | | |
80 | * | page | | |
81 | * +------------------------------+ | |
82 | * | |
83 | * | |
84 | * After we make this swap, the reader can hand this page off to the splice | |
85 | * code and be done with it. It can even allocate a new page if it needs to | |
86 | * and swap that into the ring buffer. | |
87 | * | |
88 | * We will be using cmpxchg soon to make all this lockless. | |
89 | * | |
90 | */ | |
91 | ||
033601a3 SR |
92 | /* |
93 | * A fast way to enable or disable all ring buffers is to | |
94 | * call tracing_on or tracing_off. Turning off the ring buffers | |
95 | * prevents all ring buffers from being recorded to. | |
96 | * Turning this switch on, makes it OK to write to the | |
97 | * ring buffer, if the ring buffer is enabled itself. | |
98 | * | |
99 | * There's three layers that must be on in order to write | |
100 | * to the ring buffer. | |
101 | * | |
102 | * 1) This global flag must be set. | |
103 | * 2) The ring buffer must be enabled for recording. | |
104 | * 3) The per cpu buffer must be enabled for recording. | |
105 | * | |
106 | * In case of an anomaly, this global flag has a bit set that | |
107 | * will permantly disable all ring buffers. | |
108 | */ | |
109 | ||
110 | /* | |
111 | * Global flag to disable all recording to ring buffers | |
112 | * This has two bits: ON, DISABLED | |
113 | * | |
114 | * ON DISABLED | |
115 | * ---- ---------- | |
116 | * 0 0 : ring buffers are off | |
117 | * 1 0 : ring buffers are on | |
118 | * X 1 : ring buffers are permanently disabled | |
119 | */ | |
120 | ||
121 | enum { | |
122 | RB_BUFFERS_ON_BIT = 0, | |
123 | RB_BUFFERS_DISABLED_BIT = 1, | |
124 | }; | |
125 | ||
126 | enum { | |
127 | RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT, | |
128 | RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT, | |
129 | }; | |
130 | ||
5e39841c | 131 | static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON; |
a3583244 | 132 | |
474d32b6 SR |
133 | #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data) |
134 | ||
a3583244 SR |
135 | /** |
136 | * tracing_on - enable all tracing buffers | |
137 | * | |
138 | * This function enables all tracing buffers that may have been | |
139 | * disabled with tracing_off. | |
140 | */ | |
141 | void tracing_on(void) | |
142 | { | |
033601a3 | 143 | set_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags); |
a3583244 | 144 | } |
c4f50183 | 145 | EXPORT_SYMBOL_GPL(tracing_on); |
a3583244 SR |
146 | |
147 | /** | |
148 | * tracing_off - turn off all tracing buffers | |
149 | * | |
150 | * This function stops all tracing buffers from recording data. | |
151 | * It does not disable any overhead the tracers themselves may | |
152 | * be causing. This function simply causes all recording to | |
153 | * the ring buffers to fail. | |
154 | */ | |
155 | void tracing_off(void) | |
156 | { | |
033601a3 SR |
157 | clear_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags); |
158 | } | |
c4f50183 | 159 | EXPORT_SYMBOL_GPL(tracing_off); |
033601a3 SR |
160 | |
161 | /** | |
162 | * tracing_off_permanent - permanently disable ring buffers | |
163 | * | |
164 | * This function, once called, will disable all ring buffers | |
c3706f00 | 165 | * permanently. |
033601a3 SR |
166 | */ |
167 | void tracing_off_permanent(void) | |
168 | { | |
169 | set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags); | |
a3583244 SR |
170 | } |
171 | ||
988ae9d6 SR |
172 | /** |
173 | * tracing_is_on - show state of ring buffers enabled | |
174 | */ | |
175 | int tracing_is_on(void) | |
176 | { | |
177 | return ring_buffer_flags == RB_BUFFERS_ON; | |
178 | } | |
179 | EXPORT_SYMBOL_GPL(tracing_is_on); | |
180 | ||
d06bbd66 IM |
181 | #include "trace.h" |
182 | ||
e3d6bf0a | 183 | #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array)) |
67d34724 | 184 | #define RB_ALIGNMENT 4U |
7a8e76a3 SR |
185 | #define RB_MAX_SMALL_DATA 28 |
186 | ||
187 | enum { | |
188 | RB_LEN_TIME_EXTEND = 8, | |
189 | RB_LEN_TIME_STAMP = 16, | |
190 | }; | |
191 | ||
192 | /* inline for ring buffer fast paths */ | |
34a148bf | 193 | static unsigned |
7a8e76a3 SR |
194 | rb_event_length(struct ring_buffer_event *event) |
195 | { | |
196 | unsigned length; | |
197 | ||
198 | switch (event->type) { | |
199 | case RINGBUF_TYPE_PADDING: | |
200 | /* undefined */ | |
201 | return -1; | |
202 | ||
203 | case RINGBUF_TYPE_TIME_EXTEND: | |
204 | return RB_LEN_TIME_EXTEND; | |
205 | ||
206 | case RINGBUF_TYPE_TIME_STAMP: | |
207 | return RB_LEN_TIME_STAMP; | |
208 | ||
209 | case RINGBUF_TYPE_DATA: | |
210 | if (event->len) | |
67d34724 | 211 | length = event->len * RB_ALIGNMENT; |
7a8e76a3 SR |
212 | else |
213 | length = event->array[0]; | |
214 | return length + RB_EVNT_HDR_SIZE; | |
215 | default: | |
216 | BUG(); | |
217 | } | |
218 | /* not hit */ | |
219 | return 0; | |
220 | } | |
221 | ||
222 | /** | |
223 | * ring_buffer_event_length - return the length of the event | |
224 | * @event: the event to get the length of | |
225 | */ | |
226 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
227 | { | |
465634ad RR |
228 | unsigned length = rb_event_length(event); |
229 | if (event->type != RINGBUF_TYPE_DATA) | |
230 | return length; | |
231 | length -= RB_EVNT_HDR_SIZE; | |
232 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
233 | length -= sizeof(event->array[0]); | |
234 | return length; | |
7a8e76a3 | 235 | } |
c4f50183 | 236 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
237 | |
238 | /* inline for ring buffer fast paths */ | |
34a148bf | 239 | static void * |
7a8e76a3 SR |
240 | rb_event_data(struct ring_buffer_event *event) |
241 | { | |
242 | BUG_ON(event->type != RINGBUF_TYPE_DATA); | |
243 | /* If length is in len field, then array[0] has the data */ | |
244 | if (event->len) | |
245 | return (void *)&event->array[0]; | |
246 | /* Otherwise length is in array[0] and array[1] has the data */ | |
247 | return (void *)&event->array[1]; | |
248 | } | |
249 | ||
250 | /** | |
251 | * ring_buffer_event_data - return the data of the event | |
252 | * @event: the event to get the data from | |
253 | */ | |
254 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
255 | { | |
256 | return rb_event_data(event); | |
257 | } | |
c4f50183 | 258 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
259 | |
260 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 261 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
262 | |
263 | #define TS_SHIFT 27 | |
264 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
265 | #define TS_DELTA_TEST (~TS_MASK) | |
266 | ||
abc9b56d | 267 | struct buffer_data_page { |
e4c2ce82 | 268 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 269 | local_t commit; /* write committed index */ |
abc9b56d SR |
270 | unsigned char data[]; /* data of buffer page */ |
271 | }; | |
272 | ||
273 | struct buffer_page { | |
274 | local_t write; /* index for next write */ | |
6f807acd | 275 | unsigned read; /* index for next read */ |
e4c2ce82 | 276 | struct list_head list; /* list of free pages */ |
abc9b56d | 277 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
278 | }; |
279 | ||
044fa782 | 280 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 281 | { |
044fa782 | 282 | local_set(&bpage->commit, 0); |
abc9b56d SR |
283 | } |
284 | ||
474d32b6 SR |
285 | /** |
286 | * ring_buffer_page_len - the size of data on the page. | |
287 | * @page: The page to read | |
288 | * | |
289 | * Returns the amount of data on the page, including buffer page header. | |
290 | */ | |
ef7a4a16 SR |
291 | size_t ring_buffer_page_len(void *page) |
292 | { | |
474d32b6 SR |
293 | return local_read(&((struct buffer_data_page *)page)->commit) |
294 | + BUF_PAGE_HDR_SIZE; | |
ef7a4a16 SR |
295 | } |
296 | ||
ed56829c SR |
297 | /* |
298 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
299 | * this issue out. | |
300 | */ | |
34a148bf | 301 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 302 | { |
34a148bf | 303 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 304 | kfree(bpage); |
ed56829c SR |
305 | } |
306 | ||
7a8e76a3 SR |
307 | /* |
308 | * We need to fit the time_stamp delta into 27 bits. | |
309 | */ | |
310 | static inline int test_time_stamp(u64 delta) | |
311 | { | |
312 | if (delta & TS_DELTA_TEST) | |
313 | return 1; | |
314 | return 0; | |
315 | } | |
316 | ||
474d32b6 | 317 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 SR |
318 | |
319 | /* | |
320 | * head_page == tail_page && head == tail then buffer is empty. | |
321 | */ | |
322 | struct ring_buffer_per_cpu { | |
323 | int cpu; | |
324 | struct ring_buffer *buffer; | |
f83c9d0f | 325 | spinlock_t reader_lock; /* serialize readers */ |
3e03fb7f | 326 | raw_spinlock_t lock; |
7a8e76a3 SR |
327 | struct lock_class_key lock_key; |
328 | struct list_head pages; | |
6f807acd SR |
329 | struct buffer_page *head_page; /* read from head */ |
330 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 331 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 332 | struct buffer_page *reader_page; |
7a8e76a3 SR |
333 | unsigned long overrun; |
334 | unsigned long entries; | |
335 | u64 write_stamp; | |
336 | u64 read_stamp; | |
337 | atomic_t record_disabled; | |
338 | }; | |
339 | ||
340 | struct ring_buffer { | |
7a8e76a3 SR |
341 | unsigned pages; |
342 | unsigned flags; | |
343 | int cpus; | |
7a8e76a3 | 344 | atomic_t record_disabled; |
00f62f61 | 345 | cpumask_var_t cpumask; |
7a8e76a3 SR |
346 | |
347 | struct mutex mutex; | |
348 | ||
349 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 350 | |
59222efe | 351 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
352 | struct notifier_block cpu_notify; |
353 | #endif | |
37886f6a | 354 | u64 (*clock)(void); |
7a8e76a3 SR |
355 | }; |
356 | ||
357 | struct ring_buffer_iter { | |
358 | struct ring_buffer_per_cpu *cpu_buffer; | |
359 | unsigned long head; | |
360 | struct buffer_page *head_page; | |
361 | u64 read_stamp; | |
362 | }; | |
363 | ||
f536aafc | 364 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
bf41a158 | 365 | #define RB_WARN_ON(buffer, cond) \ |
3e89c7bb SR |
366 | ({ \ |
367 | int _____ret = unlikely(cond); \ | |
368 | if (_____ret) { \ | |
bf41a158 SR |
369 | atomic_inc(&buffer->record_disabled); \ |
370 | WARN_ON(1); \ | |
371 | } \ | |
3e89c7bb SR |
372 | _____ret; \ |
373 | }) | |
f536aafc | 374 | |
37886f6a SR |
375 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
376 | #define DEBUG_SHIFT 0 | |
377 | ||
378 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) | |
379 | { | |
380 | u64 time; | |
381 | ||
382 | preempt_disable_notrace(); | |
383 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
384 | time = buffer->clock() << DEBUG_SHIFT; | |
385 | preempt_enable_no_resched_notrace(); | |
386 | ||
387 | return time; | |
388 | } | |
389 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
390 | ||
391 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
392 | int cpu, u64 *ts) | |
393 | { | |
394 | /* Just stupid testing the normalize function and deltas */ | |
395 | *ts >>= DEBUG_SHIFT; | |
396 | } | |
397 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
398 | ||
7a8e76a3 SR |
399 | /** |
400 | * check_pages - integrity check of buffer pages | |
401 | * @cpu_buffer: CPU buffer with pages to test | |
402 | * | |
c3706f00 | 403 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
404 | * been corrupted. |
405 | */ | |
406 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
407 | { | |
408 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 409 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 410 | |
3e89c7bb SR |
411 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
412 | return -1; | |
413 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
414 | return -1; | |
7a8e76a3 | 415 | |
044fa782 | 416 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 417 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 418 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
419 | return -1; |
420 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 421 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 422 | return -1; |
7a8e76a3 SR |
423 | } |
424 | ||
425 | return 0; | |
426 | } | |
427 | ||
7a8e76a3 SR |
428 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
429 | unsigned nr_pages) | |
430 | { | |
431 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 432 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
433 | unsigned long addr; |
434 | LIST_HEAD(pages); | |
435 | unsigned i; | |
436 | ||
437 | for (i = 0; i < nr_pages; i++) { | |
044fa782 | 438 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
aa1e0e3b | 439 | GFP_KERNEL, cpu_to_node(cpu_buffer->cpu)); |
044fa782 | 440 | if (!bpage) |
e4c2ce82 | 441 | goto free_pages; |
044fa782 | 442 | list_add(&bpage->list, &pages); |
e4c2ce82 | 443 | |
7a8e76a3 SR |
444 | addr = __get_free_page(GFP_KERNEL); |
445 | if (!addr) | |
446 | goto free_pages; | |
044fa782 SR |
447 | bpage->page = (void *)addr; |
448 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
449 | } |
450 | ||
451 | list_splice(&pages, head); | |
452 | ||
453 | rb_check_pages(cpu_buffer); | |
454 | ||
455 | return 0; | |
456 | ||
457 | free_pages: | |
044fa782 SR |
458 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
459 | list_del_init(&bpage->list); | |
460 | free_buffer_page(bpage); | |
7a8e76a3 SR |
461 | } |
462 | return -ENOMEM; | |
463 | } | |
464 | ||
465 | static struct ring_buffer_per_cpu * | |
466 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) | |
467 | { | |
468 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 469 | struct buffer_page *bpage; |
d769041f | 470 | unsigned long addr; |
7a8e76a3 SR |
471 | int ret; |
472 | ||
473 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
474 | GFP_KERNEL, cpu_to_node(cpu)); | |
475 | if (!cpu_buffer) | |
476 | return NULL; | |
477 | ||
478 | cpu_buffer->cpu = cpu; | |
479 | cpu_buffer->buffer = buffer; | |
f83c9d0f | 480 | spin_lock_init(&cpu_buffer->reader_lock); |
3e03fb7f | 481 | cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; |
7a8e76a3 SR |
482 | INIT_LIST_HEAD(&cpu_buffer->pages); |
483 | ||
044fa782 | 484 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 485 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 486 | if (!bpage) |
e4c2ce82 SR |
487 | goto fail_free_buffer; |
488 | ||
044fa782 | 489 | cpu_buffer->reader_page = bpage; |
d769041f SR |
490 | addr = __get_free_page(GFP_KERNEL); |
491 | if (!addr) | |
e4c2ce82 | 492 | goto fail_free_reader; |
044fa782 SR |
493 | bpage->page = (void *)addr; |
494 | rb_init_page(bpage->page); | |
e4c2ce82 | 495 | |
d769041f | 496 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 497 | |
7a8e76a3 SR |
498 | ret = rb_allocate_pages(cpu_buffer, buffer->pages); |
499 | if (ret < 0) | |
d769041f | 500 | goto fail_free_reader; |
7a8e76a3 SR |
501 | |
502 | cpu_buffer->head_page | |
503 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 504 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 SR |
505 | |
506 | return cpu_buffer; | |
507 | ||
d769041f SR |
508 | fail_free_reader: |
509 | free_buffer_page(cpu_buffer->reader_page); | |
510 | ||
7a8e76a3 SR |
511 | fail_free_buffer: |
512 | kfree(cpu_buffer); | |
513 | return NULL; | |
514 | } | |
515 | ||
516 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
517 | { | |
518 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 519 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 520 | |
d769041f SR |
521 | list_del_init(&cpu_buffer->reader_page->list); |
522 | free_buffer_page(cpu_buffer->reader_page); | |
523 | ||
044fa782 SR |
524 | list_for_each_entry_safe(bpage, tmp, head, list) { |
525 | list_del_init(&bpage->list); | |
526 | free_buffer_page(bpage); | |
7a8e76a3 SR |
527 | } |
528 | kfree(cpu_buffer); | |
529 | } | |
530 | ||
a7b13743 SR |
531 | /* |
532 | * Causes compile errors if the struct buffer_page gets bigger | |
533 | * than the struct page. | |
534 | */ | |
535 | extern int ring_buffer_page_too_big(void); | |
536 | ||
59222efe | 537 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
538 | static int __cpuinit rb_cpu_notify(struct notifier_block *self, |
539 | unsigned long action, void *hcpu); | |
540 | #endif | |
541 | ||
7a8e76a3 SR |
542 | /** |
543 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 544 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
545 | * @flags: attributes to set for the ring buffer. |
546 | * | |
547 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
548 | * flag. This flag means that the buffer will overwrite old data | |
549 | * when the buffer wraps. If this flag is not set, the buffer will | |
550 | * drop data when the tail hits the head. | |
551 | */ | |
552 | struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags) | |
553 | { | |
554 | struct ring_buffer *buffer; | |
555 | int bsize; | |
556 | int cpu; | |
557 | ||
a7b13743 SR |
558 | /* Paranoid! Optimizes out when all is well */ |
559 | if (sizeof(struct buffer_page) > sizeof(struct page)) | |
560 | ring_buffer_page_too_big(); | |
561 | ||
562 | ||
7a8e76a3 SR |
563 | /* keep it in its own cache line */ |
564 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
565 | GFP_KERNEL); | |
566 | if (!buffer) | |
567 | return NULL; | |
568 | ||
9e01c1b7 RR |
569 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
570 | goto fail_free_buffer; | |
571 | ||
7a8e76a3 SR |
572 | buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
573 | buffer->flags = flags; | |
37886f6a | 574 | buffer->clock = trace_clock_local; |
7a8e76a3 SR |
575 | |
576 | /* need at least two pages */ | |
577 | if (buffer->pages == 1) | |
578 | buffer->pages++; | |
579 | ||
3bf832ce FW |
580 | /* |
581 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
582 | * in early initcall, it will not be notified of secondary cpus. | |
583 | * In that off case, we need to allocate for all possible cpus. | |
584 | */ | |
585 | #ifdef CONFIG_HOTPLUG_CPU | |
554f786e SR |
586 | get_online_cpus(); |
587 | cpumask_copy(buffer->cpumask, cpu_online_mask); | |
3bf832ce FW |
588 | #else |
589 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
590 | #endif | |
7a8e76a3 SR |
591 | buffer->cpus = nr_cpu_ids; |
592 | ||
593 | bsize = sizeof(void *) * nr_cpu_ids; | |
594 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
595 | GFP_KERNEL); | |
596 | if (!buffer->buffers) | |
9e01c1b7 | 597 | goto fail_free_cpumask; |
7a8e76a3 SR |
598 | |
599 | for_each_buffer_cpu(buffer, cpu) { | |
600 | buffer->buffers[cpu] = | |
601 | rb_allocate_cpu_buffer(buffer, cpu); | |
602 | if (!buffer->buffers[cpu]) | |
603 | goto fail_free_buffers; | |
604 | } | |
605 | ||
59222efe | 606 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
607 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
608 | buffer->cpu_notify.priority = 0; | |
609 | register_cpu_notifier(&buffer->cpu_notify); | |
610 | #endif | |
611 | ||
612 | put_online_cpus(); | |
7a8e76a3 SR |
613 | mutex_init(&buffer->mutex); |
614 | ||
615 | return buffer; | |
616 | ||
617 | fail_free_buffers: | |
618 | for_each_buffer_cpu(buffer, cpu) { | |
619 | if (buffer->buffers[cpu]) | |
620 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
621 | } | |
622 | kfree(buffer->buffers); | |
623 | ||
9e01c1b7 RR |
624 | fail_free_cpumask: |
625 | free_cpumask_var(buffer->cpumask); | |
554f786e | 626 | put_online_cpus(); |
9e01c1b7 | 627 | |
7a8e76a3 SR |
628 | fail_free_buffer: |
629 | kfree(buffer); | |
630 | return NULL; | |
631 | } | |
c4f50183 | 632 | EXPORT_SYMBOL_GPL(ring_buffer_alloc); |
7a8e76a3 SR |
633 | |
634 | /** | |
635 | * ring_buffer_free - free a ring buffer. | |
636 | * @buffer: the buffer to free. | |
637 | */ | |
638 | void | |
639 | ring_buffer_free(struct ring_buffer *buffer) | |
640 | { | |
641 | int cpu; | |
642 | ||
554f786e SR |
643 | get_online_cpus(); |
644 | ||
59222efe | 645 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
646 | unregister_cpu_notifier(&buffer->cpu_notify); |
647 | #endif | |
648 | ||
7a8e76a3 SR |
649 | for_each_buffer_cpu(buffer, cpu) |
650 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
651 | ||
554f786e SR |
652 | put_online_cpus(); |
653 | ||
9e01c1b7 RR |
654 | free_cpumask_var(buffer->cpumask); |
655 | ||
7a8e76a3 SR |
656 | kfree(buffer); |
657 | } | |
c4f50183 | 658 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 659 | |
37886f6a SR |
660 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
661 | u64 (*clock)(void)) | |
662 | { | |
663 | buffer->clock = clock; | |
664 | } | |
665 | ||
7a8e76a3 SR |
666 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
667 | ||
668 | static void | |
669 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages) | |
670 | { | |
044fa782 | 671 | struct buffer_page *bpage; |
7a8e76a3 SR |
672 | struct list_head *p; |
673 | unsigned i; | |
674 | ||
675 | atomic_inc(&cpu_buffer->record_disabled); | |
676 | synchronize_sched(); | |
677 | ||
678 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
679 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
680 | return; | |
7a8e76a3 | 681 | p = cpu_buffer->pages.next; |
044fa782 SR |
682 | bpage = list_entry(p, struct buffer_page, list); |
683 | list_del_init(&bpage->list); | |
684 | free_buffer_page(bpage); | |
7a8e76a3 | 685 | } |
3e89c7bb SR |
686 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
687 | return; | |
7a8e76a3 SR |
688 | |
689 | rb_reset_cpu(cpu_buffer); | |
690 | ||
691 | rb_check_pages(cpu_buffer); | |
692 | ||
693 | atomic_dec(&cpu_buffer->record_disabled); | |
694 | ||
695 | } | |
696 | ||
697 | static void | |
698 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
699 | struct list_head *pages, unsigned nr_pages) | |
700 | { | |
044fa782 | 701 | struct buffer_page *bpage; |
7a8e76a3 SR |
702 | struct list_head *p; |
703 | unsigned i; | |
704 | ||
705 | atomic_inc(&cpu_buffer->record_disabled); | |
706 | synchronize_sched(); | |
707 | ||
708 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
709 | if (RB_WARN_ON(cpu_buffer, list_empty(pages))) |
710 | return; | |
7a8e76a3 | 711 | p = pages->next; |
044fa782 SR |
712 | bpage = list_entry(p, struct buffer_page, list); |
713 | list_del_init(&bpage->list); | |
714 | list_add_tail(&bpage->list, &cpu_buffer->pages); | |
7a8e76a3 SR |
715 | } |
716 | rb_reset_cpu(cpu_buffer); | |
717 | ||
718 | rb_check_pages(cpu_buffer); | |
719 | ||
720 | atomic_dec(&cpu_buffer->record_disabled); | |
721 | } | |
722 | ||
723 | /** | |
724 | * ring_buffer_resize - resize the ring buffer | |
725 | * @buffer: the buffer to resize. | |
726 | * @size: the new size. | |
727 | * | |
728 | * The tracer is responsible for making sure that the buffer is | |
729 | * not being used while changing the size. | |
730 | * Note: We may be able to change the above requirement by using | |
731 | * RCU synchronizations. | |
732 | * | |
733 | * Minimum size is 2 * BUF_PAGE_SIZE. | |
734 | * | |
735 | * Returns -1 on failure. | |
736 | */ | |
737 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) | |
738 | { | |
739 | struct ring_buffer_per_cpu *cpu_buffer; | |
740 | unsigned nr_pages, rm_pages, new_pages; | |
044fa782 | 741 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
742 | unsigned long buffer_size; |
743 | unsigned long addr; | |
744 | LIST_HEAD(pages); | |
745 | int i, cpu; | |
746 | ||
ee51a1de IM |
747 | /* |
748 | * Always succeed at resizing a non-existent buffer: | |
749 | */ | |
750 | if (!buffer) | |
751 | return size; | |
752 | ||
7a8e76a3 SR |
753 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
754 | size *= BUF_PAGE_SIZE; | |
755 | buffer_size = buffer->pages * BUF_PAGE_SIZE; | |
756 | ||
757 | /* we need a minimum of two pages */ | |
758 | if (size < BUF_PAGE_SIZE * 2) | |
759 | size = BUF_PAGE_SIZE * 2; | |
760 | ||
761 | if (size == buffer_size) | |
762 | return size; | |
763 | ||
764 | mutex_lock(&buffer->mutex); | |
554f786e | 765 | get_online_cpus(); |
7a8e76a3 SR |
766 | |
767 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
768 | ||
769 | if (size < buffer_size) { | |
770 | ||
771 | /* easy case, just free pages */ | |
554f786e SR |
772 | if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) |
773 | goto out_fail; | |
7a8e76a3 SR |
774 | |
775 | rm_pages = buffer->pages - nr_pages; | |
776 | ||
777 | for_each_buffer_cpu(buffer, cpu) { | |
778 | cpu_buffer = buffer->buffers[cpu]; | |
779 | rb_remove_pages(cpu_buffer, rm_pages); | |
780 | } | |
781 | goto out; | |
782 | } | |
783 | ||
784 | /* | |
785 | * This is a bit more difficult. We only want to add pages | |
786 | * when we can allocate enough for all CPUs. We do this | |
787 | * by allocating all the pages and storing them on a local | |
788 | * link list. If we succeed in our allocation, then we | |
789 | * add these pages to the cpu_buffers. Otherwise we just free | |
790 | * them all and return -ENOMEM; | |
791 | */ | |
554f786e SR |
792 | if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) |
793 | goto out_fail; | |
f536aafc | 794 | |
7a8e76a3 SR |
795 | new_pages = nr_pages - buffer->pages; |
796 | ||
797 | for_each_buffer_cpu(buffer, cpu) { | |
798 | for (i = 0; i < new_pages; i++) { | |
044fa782 | 799 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), |
e4c2ce82 SR |
800 | cache_line_size()), |
801 | GFP_KERNEL, cpu_to_node(cpu)); | |
044fa782 | 802 | if (!bpage) |
e4c2ce82 | 803 | goto free_pages; |
044fa782 | 804 | list_add(&bpage->list, &pages); |
7a8e76a3 SR |
805 | addr = __get_free_page(GFP_KERNEL); |
806 | if (!addr) | |
807 | goto free_pages; | |
044fa782 SR |
808 | bpage->page = (void *)addr; |
809 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
810 | } |
811 | } | |
812 | ||
813 | for_each_buffer_cpu(buffer, cpu) { | |
814 | cpu_buffer = buffer->buffers[cpu]; | |
815 | rb_insert_pages(cpu_buffer, &pages, new_pages); | |
816 | } | |
817 | ||
554f786e SR |
818 | if (RB_WARN_ON(buffer, !list_empty(&pages))) |
819 | goto out_fail; | |
7a8e76a3 SR |
820 | |
821 | out: | |
822 | buffer->pages = nr_pages; | |
554f786e | 823 | put_online_cpus(); |
7a8e76a3 SR |
824 | mutex_unlock(&buffer->mutex); |
825 | ||
826 | return size; | |
827 | ||
828 | free_pages: | |
044fa782 SR |
829 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
830 | list_del_init(&bpage->list); | |
831 | free_buffer_page(bpage); | |
7a8e76a3 | 832 | } |
554f786e | 833 | put_online_cpus(); |
641d2f63 | 834 | mutex_unlock(&buffer->mutex); |
7a8e76a3 | 835 | return -ENOMEM; |
554f786e SR |
836 | |
837 | /* | |
838 | * Something went totally wrong, and we are too paranoid | |
839 | * to even clean up the mess. | |
840 | */ | |
841 | out_fail: | |
842 | put_online_cpus(); | |
843 | mutex_unlock(&buffer->mutex); | |
844 | return -1; | |
7a8e76a3 | 845 | } |
c4f50183 | 846 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 847 | |
7a8e76a3 SR |
848 | static inline int rb_null_event(struct ring_buffer_event *event) |
849 | { | |
850 | return event->type == RINGBUF_TYPE_PADDING; | |
851 | } | |
852 | ||
8789a9e7 | 853 | static inline void * |
044fa782 | 854 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 855 | { |
044fa782 | 856 | return bpage->data + index; |
8789a9e7 SR |
857 | } |
858 | ||
044fa782 | 859 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 860 | { |
044fa782 | 861 | return bpage->page->data + index; |
7a8e76a3 SR |
862 | } |
863 | ||
864 | static inline struct ring_buffer_event * | |
d769041f | 865 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 866 | { |
6f807acd SR |
867 | return __rb_page_index(cpu_buffer->reader_page, |
868 | cpu_buffer->reader_page->read); | |
869 | } | |
870 | ||
871 | static inline struct ring_buffer_event * | |
872 | rb_head_event(struct ring_buffer_per_cpu *cpu_buffer) | |
873 | { | |
874 | return __rb_page_index(cpu_buffer->head_page, | |
875 | cpu_buffer->head_page->read); | |
7a8e76a3 SR |
876 | } |
877 | ||
878 | static inline struct ring_buffer_event * | |
879 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
880 | { | |
6f807acd | 881 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
882 | } |
883 | ||
bf41a158 SR |
884 | static inline unsigned rb_page_write(struct buffer_page *bpage) |
885 | { | |
886 | return local_read(&bpage->write); | |
887 | } | |
888 | ||
889 | static inline unsigned rb_page_commit(struct buffer_page *bpage) | |
890 | { | |
abc9b56d | 891 | return local_read(&bpage->page->commit); |
bf41a158 SR |
892 | } |
893 | ||
894 | /* Size is determined by what has been commited */ | |
895 | static inline unsigned rb_page_size(struct buffer_page *bpage) | |
896 | { | |
897 | return rb_page_commit(bpage); | |
898 | } | |
899 | ||
900 | static inline unsigned | |
901 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
902 | { | |
903 | return rb_page_commit(cpu_buffer->commit_page); | |
904 | } | |
905 | ||
906 | static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer) | |
907 | { | |
908 | return rb_page_commit(cpu_buffer->head_page); | |
909 | } | |
910 | ||
7a8e76a3 SR |
911 | /* |
912 | * When the tail hits the head and the buffer is in overwrite mode, | |
913 | * the head jumps to the next page and all content on the previous | |
914 | * page is discarded. But before doing so, we update the overrun | |
915 | * variable of the buffer. | |
916 | */ | |
917 | static void rb_update_overflow(struct ring_buffer_per_cpu *cpu_buffer) | |
918 | { | |
919 | struct ring_buffer_event *event; | |
920 | unsigned long head; | |
921 | ||
922 | for (head = 0; head < rb_head_size(cpu_buffer); | |
923 | head += rb_event_length(event)) { | |
924 | ||
6f807acd | 925 | event = __rb_page_index(cpu_buffer->head_page, head); |
3e89c7bb SR |
926 | if (RB_WARN_ON(cpu_buffer, rb_null_event(event))) |
927 | return; | |
7a8e76a3 SR |
928 | /* Only count data entries */ |
929 | if (event->type != RINGBUF_TYPE_DATA) | |
930 | continue; | |
931 | cpu_buffer->overrun++; | |
932 | cpu_buffer->entries--; | |
933 | } | |
934 | } | |
935 | ||
936 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
044fa782 | 937 | struct buffer_page **bpage) |
7a8e76a3 | 938 | { |
044fa782 | 939 | struct list_head *p = (*bpage)->list.next; |
7a8e76a3 SR |
940 | |
941 | if (p == &cpu_buffer->pages) | |
942 | p = p->next; | |
943 | ||
044fa782 | 944 | *bpage = list_entry(p, struct buffer_page, list); |
7a8e76a3 SR |
945 | } |
946 | ||
bf41a158 SR |
947 | static inline unsigned |
948 | rb_event_index(struct ring_buffer_event *event) | |
949 | { | |
950 | unsigned long addr = (unsigned long)event; | |
951 | ||
952 | return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE); | |
953 | } | |
954 | ||
34a148bf | 955 | static int |
bf41a158 SR |
956 | rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
957 | struct ring_buffer_event *event) | |
958 | { | |
959 | unsigned long addr = (unsigned long)event; | |
960 | unsigned long index; | |
961 | ||
962 | index = rb_event_index(event); | |
963 | addr &= PAGE_MASK; | |
964 | ||
965 | return cpu_buffer->commit_page->page == (void *)addr && | |
966 | rb_commit_index(cpu_buffer) == index; | |
967 | } | |
968 | ||
34a148bf | 969 | static void |
bf41a158 SR |
970 | rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer, |
971 | struct ring_buffer_event *event) | |
7a8e76a3 | 972 | { |
bf41a158 SR |
973 | unsigned long addr = (unsigned long)event; |
974 | unsigned long index; | |
975 | ||
976 | index = rb_event_index(event); | |
977 | addr &= PAGE_MASK; | |
978 | ||
979 | while (cpu_buffer->commit_page->page != (void *)addr) { | |
3e89c7bb SR |
980 | if (RB_WARN_ON(cpu_buffer, |
981 | cpu_buffer->commit_page == cpu_buffer->tail_page)) | |
982 | return; | |
abc9b56d | 983 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
984 | cpu_buffer->commit_page->write; |
985 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
986 | cpu_buffer->write_stamp = |
987 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
988 | } |
989 | ||
990 | /* Now set the commit to the event's index */ | |
abc9b56d | 991 | local_set(&cpu_buffer->commit_page->page->commit, index); |
7a8e76a3 SR |
992 | } |
993 | ||
34a148bf | 994 | static void |
bf41a158 | 995 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 996 | { |
bf41a158 SR |
997 | /* |
998 | * We only race with interrupts and NMIs on this CPU. | |
999 | * If we own the commit event, then we can commit | |
1000 | * all others that interrupted us, since the interruptions | |
1001 | * are in stack format (they finish before they come | |
1002 | * back to us). This allows us to do a simple loop to | |
1003 | * assign the commit to the tail. | |
1004 | */ | |
a8ccf1d6 | 1005 | again: |
bf41a158 | 1006 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
abc9b56d | 1007 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1008 | cpu_buffer->commit_page->write; |
1009 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
1010 | cpu_buffer->write_stamp = |
1011 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1012 | /* add barrier to keep gcc from optimizing too much */ |
1013 | barrier(); | |
1014 | } | |
1015 | while (rb_commit_index(cpu_buffer) != | |
1016 | rb_page_write(cpu_buffer->commit_page)) { | |
abc9b56d | 1017 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1018 | cpu_buffer->commit_page->write; |
1019 | barrier(); | |
1020 | } | |
a8ccf1d6 SR |
1021 | |
1022 | /* again, keep gcc from optimizing */ | |
1023 | barrier(); | |
1024 | ||
1025 | /* | |
1026 | * If an interrupt came in just after the first while loop | |
1027 | * and pushed the tail page forward, we will be left with | |
1028 | * a dangling commit that will never go forward. | |
1029 | */ | |
1030 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1031 | goto again; | |
7a8e76a3 SR |
1032 | } |
1033 | ||
d769041f | 1034 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1035 | { |
abc9b56d | 1036 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1037 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1038 | } |
1039 | ||
34a148bf | 1040 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1041 | { |
1042 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1043 | ||
1044 | /* | |
1045 | * The iterator could be on the reader page (it starts there). | |
1046 | * But the head could have moved, since the reader was | |
1047 | * found. Check for this case and assign the iterator | |
1048 | * to the head page instead of next. | |
1049 | */ | |
1050 | if (iter->head_page == cpu_buffer->reader_page) | |
1051 | iter->head_page = cpu_buffer->head_page; | |
1052 | else | |
1053 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1054 | ||
abc9b56d | 1055 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1056 | iter->head = 0; |
1057 | } | |
1058 | ||
1059 | /** | |
1060 | * ring_buffer_update_event - update event type and data | |
1061 | * @event: the even to update | |
1062 | * @type: the type of event | |
1063 | * @length: the size of the event field in the ring buffer | |
1064 | * | |
1065 | * Update the type and data fields of the event. The length | |
1066 | * is the actual size that is written to the ring buffer, | |
1067 | * and with this, we can determine what to place into the | |
1068 | * data field. | |
1069 | */ | |
34a148bf | 1070 | static void |
7a8e76a3 SR |
1071 | rb_update_event(struct ring_buffer_event *event, |
1072 | unsigned type, unsigned length) | |
1073 | { | |
1074 | event->type = type; | |
1075 | ||
1076 | switch (type) { | |
1077 | ||
1078 | case RINGBUF_TYPE_PADDING: | |
1079 | break; | |
1080 | ||
1081 | case RINGBUF_TYPE_TIME_EXTEND: | |
67d34724 | 1082 | event->len = DIV_ROUND_UP(RB_LEN_TIME_EXTEND, RB_ALIGNMENT); |
7a8e76a3 SR |
1083 | break; |
1084 | ||
1085 | case RINGBUF_TYPE_TIME_STAMP: | |
67d34724 | 1086 | event->len = DIV_ROUND_UP(RB_LEN_TIME_STAMP, RB_ALIGNMENT); |
7a8e76a3 SR |
1087 | break; |
1088 | ||
1089 | case RINGBUF_TYPE_DATA: | |
1090 | length -= RB_EVNT_HDR_SIZE; | |
1091 | if (length > RB_MAX_SMALL_DATA) { | |
1092 | event->len = 0; | |
1093 | event->array[0] = length; | |
1094 | } else | |
67d34724 | 1095 | event->len = DIV_ROUND_UP(length, RB_ALIGNMENT); |
7a8e76a3 SR |
1096 | break; |
1097 | default: | |
1098 | BUG(); | |
1099 | } | |
1100 | } | |
1101 | ||
34a148bf | 1102 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
1103 | { |
1104 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
1105 | ||
1106 | /* zero length can cause confusions */ | |
1107 | if (!length) | |
1108 | length = 1; | |
1109 | ||
1110 | if (length > RB_MAX_SMALL_DATA) | |
1111 | length += sizeof(event.array[0]); | |
1112 | ||
1113 | length += RB_EVNT_HDR_SIZE; | |
1114 | length = ALIGN(length, RB_ALIGNMENT); | |
1115 | ||
1116 | return length; | |
1117 | } | |
1118 | ||
1119 | static struct ring_buffer_event * | |
1120 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
1121 | unsigned type, unsigned long length, u64 *ts) | |
1122 | { | |
98db8df7 | 1123 | struct buffer_page *tail_page, *head_page, *reader_page, *commit_page; |
bf41a158 | 1124 | unsigned long tail, write; |
7a8e76a3 SR |
1125 | struct ring_buffer *buffer = cpu_buffer->buffer; |
1126 | struct ring_buffer_event *event; | |
bf41a158 | 1127 | unsigned long flags; |
78d904b4 | 1128 | bool lock_taken = false; |
7a8e76a3 | 1129 | |
98db8df7 SR |
1130 | commit_page = cpu_buffer->commit_page; |
1131 | /* we just need to protect against interrupts */ | |
1132 | barrier(); | |
7a8e76a3 | 1133 | tail_page = cpu_buffer->tail_page; |
bf41a158 SR |
1134 | write = local_add_return(length, &tail_page->write); |
1135 | tail = write - length; | |
7a8e76a3 | 1136 | |
bf41a158 SR |
1137 | /* See if we shot pass the end of this buffer page */ |
1138 | if (write > BUF_PAGE_SIZE) { | |
7a8e76a3 SR |
1139 | struct buffer_page *next_page = tail_page; |
1140 | ||
3e03fb7f | 1141 | local_irq_save(flags); |
78d904b4 | 1142 | /* |
a81bd80a SR |
1143 | * Since the write to the buffer is still not |
1144 | * fully lockless, we must be careful with NMIs. | |
1145 | * The locks in the writers are taken when a write | |
1146 | * crosses to a new page. The locks protect against | |
1147 | * races with the readers (this will soon be fixed | |
1148 | * with a lockless solution). | |
1149 | * | |
1150 | * Because we can not protect against NMIs, and we | |
1151 | * want to keep traces reentrant, we need to manage | |
1152 | * what happens when we are in an NMI. | |
1153 | * | |
78d904b4 SR |
1154 | * NMIs can happen after we take the lock. |
1155 | * If we are in an NMI, only take the lock | |
1156 | * if it is not already taken. Otherwise | |
1157 | * simply fail. | |
1158 | */ | |
a81bd80a | 1159 | if (unlikely(in_nmi())) { |
78d904b4 | 1160 | if (!__raw_spin_trylock(&cpu_buffer->lock)) |
45141d46 | 1161 | goto out_reset; |
78d904b4 SR |
1162 | } else |
1163 | __raw_spin_lock(&cpu_buffer->lock); | |
1164 | ||
1165 | lock_taken = true; | |
bf41a158 | 1166 | |
7a8e76a3 SR |
1167 | rb_inc_page(cpu_buffer, &next_page); |
1168 | ||
d769041f SR |
1169 | head_page = cpu_buffer->head_page; |
1170 | reader_page = cpu_buffer->reader_page; | |
1171 | ||
1172 | /* we grabbed the lock before incrementing */ | |
3e89c7bb | 1173 | if (RB_WARN_ON(cpu_buffer, next_page == reader_page)) |
45141d46 | 1174 | goto out_reset; |
bf41a158 SR |
1175 | |
1176 | /* | |
1177 | * If for some reason, we had an interrupt storm that made | |
1178 | * it all the way around the buffer, bail, and warn | |
1179 | * about it. | |
1180 | */ | |
98db8df7 | 1181 | if (unlikely(next_page == commit_page)) { |
bf41a158 | 1182 | WARN_ON_ONCE(1); |
45141d46 | 1183 | goto out_reset; |
bf41a158 | 1184 | } |
d769041f | 1185 | |
7a8e76a3 | 1186 | if (next_page == head_page) { |
6f3b3440 | 1187 | if (!(buffer->flags & RB_FL_OVERWRITE)) |
45141d46 | 1188 | goto out_reset; |
7a8e76a3 | 1189 | |
bf41a158 SR |
1190 | /* tail_page has not moved yet? */ |
1191 | if (tail_page == cpu_buffer->tail_page) { | |
1192 | /* count overflows */ | |
1193 | rb_update_overflow(cpu_buffer); | |
1194 | ||
1195 | rb_inc_page(cpu_buffer, &head_page); | |
1196 | cpu_buffer->head_page = head_page; | |
1197 | cpu_buffer->head_page->read = 0; | |
1198 | } | |
1199 | } | |
7a8e76a3 | 1200 | |
bf41a158 SR |
1201 | /* |
1202 | * If the tail page is still the same as what we think | |
1203 | * it is, then it is up to us to update the tail | |
1204 | * pointer. | |
1205 | */ | |
1206 | if (tail_page == cpu_buffer->tail_page) { | |
1207 | local_set(&next_page->write, 0); | |
abc9b56d | 1208 | local_set(&next_page->page->commit, 0); |
bf41a158 SR |
1209 | cpu_buffer->tail_page = next_page; |
1210 | ||
1211 | /* reread the time stamp */ | |
37886f6a | 1212 | *ts = ring_buffer_time_stamp(buffer, cpu_buffer->cpu); |
abc9b56d | 1213 | cpu_buffer->tail_page->page->time_stamp = *ts; |
7a8e76a3 SR |
1214 | } |
1215 | ||
bf41a158 SR |
1216 | /* |
1217 | * The actual tail page has moved forward. | |
1218 | */ | |
1219 | if (tail < BUF_PAGE_SIZE) { | |
1220 | /* Mark the rest of the page with padding */ | |
6f807acd | 1221 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
1222 | event->type = RINGBUF_TYPE_PADDING; |
1223 | } | |
1224 | ||
bf41a158 SR |
1225 | if (tail <= BUF_PAGE_SIZE) |
1226 | /* Set the write back to the previous setting */ | |
1227 | local_set(&tail_page->write, tail); | |
1228 | ||
1229 | /* | |
1230 | * If this was a commit entry that failed, | |
1231 | * increment that too | |
1232 | */ | |
1233 | if (tail_page == cpu_buffer->commit_page && | |
1234 | tail == rb_commit_index(cpu_buffer)) { | |
1235 | rb_set_commit_to_write(cpu_buffer); | |
1236 | } | |
1237 | ||
3e03fb7f SR |
1238 | __raw_spin_unlock(&cpu_buffer->lock); |
1239 | local_irq_restore(flags); | |
bf41a158 SR |
1240 | |
1241 | /* fail and let the caller try again */ | |
1242 | return ERR_PTR(-EAGAIN); | |
7a8e76a3 SR |
1243 | } |
1244 | ||
bf41a158 SR |
1245 | /* We reserved something on the buffer */ |
1246 | ||
3e89c7bb SR |
1247 | if (RB_WARN_ON(cpu_buffer, write > BUF_PAGE_SIZE)) |
1248 | return NULL; | |
7a8e76a3 | 1249 | |
6f807acd | 1250 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
1251 | rb_update_event(event, type, length); |
1252 | ||
bf41a158 SR |
1253 | /* |
1254 | * If this is a commit and the tail is zero, then update | |
1255 | * this page's time stamp. | |
1256 | */ | |
1257 | if (!tail && rb_is_commit(cpu_buffer, event)) | |
abc9b56d | 1258 | cpu_buffer->commit_page->page->time_stamp = *ts; |
bf41a158 | 1259 | |
7a8e76a3 | 1260 | return event; |
bf41a158 | 1261 | |
45141d46 | 1262 | out_reset: |
6f3b3440 LJ |
1263 | /* reset write */ |
1264 | if (tail <= BUF_PAGE_SIZE) | |
1265 | local_set(&tail_page->write, tail); | |
1266 | ||
78d904b4 SR |
1267 | if (likely(lock_taken)) |
1268 | __raw_spin_unlock(&cpu_buffer->lock); | |
3e03fb7f | 1269 | local_irq_restore(flags); |
bf41a158 | 1270 | return NULL; |
7a8e76a3 SR |
1271 | } |
1272 | ||
1273 | static int | |
1274 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1275 | u64 *ts, u64 *delta) | |
1276 | { | |
1277 | struct ring_buffer_event *event; | |
1278 | static int once; | |
bf41a158 | 1279 | int ret; |
7a8e76a3 SR |
1280 | |
1281 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1282 | printk(KERN_WARNING "Delta way too big! %llu" | |
1283 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1284 | (unsigned long long)*delta, |
1285 | (unsigned long long)*ts, | |
1286 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1287 | WARN_ON(1); |
1288 | } | |
1289 | ||
1290 | /* | |
1291 | * The delta is too big, we to add a | |
1292 | * new timestamp. | |
1293 | */ | |
1294 | event = __rb_reserve_next(cpu_buffer, | |
1295 | RINGBUF_TYPE_TIME_EXTEND, | |
1296 | RB_LEN_TIME_EXTEND, | |
1297 | ts); | |
1298 | if (!event) | |
bf41a158 | 1299 | return -EBUSY; |
7a8e76a3 | 1300 | |
bf41a158 SR |
1301 | if (PTR_ERR(event) == -EAGAIN) |
1302 | return -EAGAIN; | |
1303 | ||
1304 | /* Only a commited time event can update the write stamp */ | |
1305 | if (rb_is_commit(cpu_buffer, event)) { | |
1306 | /* | |
1307 | * If this is the first on the page, then we need to | |
1308 | * update the page itself, and just put in a zero. | |
1309 | */ | |
1310 | if (rb_event_index(event)) { | |
1311 | event->time_delta = *delta & TS_MASK; | |
1312 | event->array[0] = *delta >> TS_SHIFT; | |
1313 | } else { | |
abc9b56d | 1314 | cpu_buffer->commit_page->page->time_stamp = *ts; |
bf41a158 SR |
1315 | event->time_delta = 0; |
1316 | event->array[0] = 0; | |
1317 | } | |
7a8e76a3 | 1318 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
1319 | /* let the caller know this was the commit */ |
1320 | ret = 1; | |
1321 | } else { | |
1322 | /* Darn, this is just wasted space */ | |
1323 | event->time_delta = 0; | |
1324 | event->array[0] = 0; | |
1325 | ret = 0; | |
7a8e76a3 SR |
1326 | } |
1327 | ||
bf41a158 SR |
1328 | *delta = 0; |
1329 | ||
1330 | return ret; | |
7a8e76a3 SR |
1331 | } |
1332 | ||
1333 | static struct ring_buffer_event * | |
1334 | rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, | |
1335 | unsigned type, unsigned long length) | |
1336 | { | |
1337 | struct ring_buffer_event *event; | |
1338 | u64 ts, delta; | |
bf41a158 | 1339 | int commit = 0; |
818e3dd3 | 1340 | int nr_loops = 0; |
7a8e76a3 | 1341 | |
bf41a158 | 1342 | again: |
818e3dd3 SR |
1343 | /* |
1344 | * We allow for interrupts to reenter here and do a trace. | |
1345 | * If one does, it will cause this original code to loop | |
1346 | * back here. Even with heavy interrupts happening, this | |
1347 | * should only happen a few times in a row. If this happens | |
1348 | * 1000 times in a row, there must be either an interrupt | |
1349 | * storm or we have something buggy. | |
1350 | * Bail! | |
1351 | */ | |
3e89c7bb | 1352 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
818e3dd3 | 1353 | return NULL; |
818e3dd3 | 1354 | |
37886f6a | 1355 | ts = ring_buffer_time_stamp(cpu_buffer->buffer, cpu_buffer->cpu); |
7a8e76a3 | 1356 | |
bf41a158 SR |
1357 | /* |
1358 | * Only the first commit can update the timestamp. | |
1359 | * Yes there is a race here. If an interrupt comes in | |
1360 | * just after the conditional and it traces too, then it | |
1361 | * will also check the deltas. More than one timestamp may | |
1362 | * also be made. But only the entry that did the actual | |
1363 | * commit will be something other than zero. | |
1364 | */ | |
1365 | if (cpu_buffer->tail_page == cpu_buffer->commit_page && | |
1366 | rb_page_write(cpu_buffer->tail_page) == | |
1367 | rb_commit_index(cpu_buffer)) { | |
1368 | ||
7a8e76a3 SR |
1369 | delta = ts - cpu_buffer->write_stamp; |
1370 | ||
bf41a158 SR |
1371 | /* make sure this delta is calculated here */ |
1372 | barrier(); | |
1373 | ||
1374 | /* Did the write stamp get updated already? */ | |
1375 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
4143c5cb | 1376 | delta = 0; |
bf41a158 | 1377 | |
7a8e76a3 | 1378 | if (test_time_stamp(delta)) { |
7a8e76a3 | 1379 | |
bf41a158 SR |
1380 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
1381 | ||
1382 | if (commit == -EBUSY) | |
7a8e76a3 | 1383 | return NULL; |
bf41a158 SR |
1384 | |
1385 | if (commit == -EAGAIN) | |
1386 | goto again; | |
1387 | ||
1388 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 1389 | } |
bf41a158 SR |
1390 | } else |
1391 | /* Non commits have zero deltas */ | |
7a8e76a3 | 1392 | delta = 0; |
7a8e76a3 SR |
1393 | |
1394 | event = __rb_reserve_next(cpu_buffer, type, length, &ts); | |
bf41a158 SR |
1395 | if (PTR_ERR(event) == -EAGAIN) |
1396 | goto again; | |
1397 | ||
1398 | if (!event) { | |
1399 | if (unlikely(commit)) | |
1400 | /* | |
1401 | * Ouch! We needed a timestamp and it was commited. But | |
1402 | * we didn't get our event reserved. | |
1403 | */ | |
1404 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 | 1405 | return NULL; |
bf41a158 | 1406 | } |
7a8e76a3 | 1407 | |
bf41a158 SR |
1408 | /* |
1409 | * If the timestamp was commited, make the commit our entry | |
1410 | * now so that we will update it when needed. | |
1411 | */ | |
1412 | if (commit) | |
1413 | rb_set_commit_event(cpu_buffer, event); | |
1414 | else if (!rb_is_commit(cpu_buffer, event)) | |
7a8e76a3 SR |
1415 | delta = 0; |
1416 | ||
1417 | event->time_delta = delta; | |
1418 | ||
1419 | return event; | |
1420 | } | |
1421 | ||
bf41a158 SR |
1422 | static DEFINE_PER_CPU(int, rb_need_resched); |
1423 | ||
7a8e76a3 SR |
1424 | /** |
1425 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
1426 | * @buffer: the ring buffer to reserve from | |
1427 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
1428 | * |
1429 | * Returns a reseverd event on the ring buffer to copy directly to. | |
1430 | * The user of this interface will need to get the body to write into | |
1431 | * and can use the ring_buffer_event_data() interface. | |
1432 | * | |
1433 | * The length is the length of the data needed, not the event length | |
1434 | * which also includes the event header. | |
1435 | * | |
1436 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
1437 | * If NULL is returned, then nothing has been allocated or locked. | |
1438 | */ | |
1439 | struct ring_buffer_event * | |
0a987751 | 1440 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
1441 | { |
1442 | struct ring_buffer_per_cpu *cpu_buffer; | |
1443 | struct ring_buffer_event *event; | |
bf41a158 | 1444 | int cpu, resched; |
7a8e76a3 | 1445 | |
033601a3 | 1446 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1447 | return NULL; |
1448 | ||
7a8e76a3 SR |
1449 | if (atomic_read(&buffer->record_disabled)) |
1450 | return NULL; | |
1451 | ||
bf41a158 | 1452 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 1453 | resched = ftrace_preempt_disable(); |
bf41a158 | 1454 | |
7a8e76a3 SR |
1455 | cpu = raw_smp_processor_id(); |
1456 | ||
9e01c1b7 | 1457 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1458 | goto out; |
7a8e76a3 SR |
1459 | |
1460 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1461 | |
1462 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 1463 | goto out; |
7a8e76a3 SR |
1464 | |
1465 | length = rb_calculate_event_length(length); | |
1466 | if (length > BUF_PAGE_SIZE) | |
bf41a158 | 1467 | goto out; |
7a8e76a3 SR |
1468 | |
1469 | event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length); | |
1470 | if (!event) | |
d769041f | 1471 | goto out; |
7a8e76a3 | 1472 | |
bf41a158 SR |
1473 | /* |
1474 | * Need to store resched state on this cpu. | |
1475 | * Only the first needs to. | |
1476 | */ | |
1477 | ||
1478 | if (preempt_count() == 1) | |
1479 | per_cpu(rb_need_resched, cpu) = resched; | |
1480 | ||
7a8e76a3 SR |
1481 | return event; |
1482 | ||
d769041f | 1483 | out: |
182e9f5f | 1484 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1485 | return NULL; |
1486 | } | |
c4f50183 | 1487 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 SR |
1488 | |
1489 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
1490 | struct ring_buffer_event *event) | |
1491 | { | |
7a8e76a3 | 1492 | cpu_buffer->entries++; |
bf41a158 SR |
1493 | |
1494 | /* Only process further if we own the commit */ | |
1495 | if (!rb_is_commit(cpu_buffer, event)) | |
1496 | return; | |
1497 | ||
1498 | cpu_buffer->write_stamp += event->time_delta; | |
1499 | ||
1500 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 SR |
1501 | } |
1502 | ||
1503 | /** | |
1504 | * ring_buffer_unlock_commit - commit a reserved | |
1505 | * @buffer: The buffer to commit to | |
1506 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
1507 | * |
1508 | * This commits the data to the ring buffer, and releases any locks held. | |
1509 | * | |
1510 | * Must be paired with ring_buffer_lock_reserve. | |
1511 | */ | |
1512 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 1513 | struct ring_buffer_event *event) |
7a8e76a3 SR |
1514 | { |
1515 | struct ring_buffer_per_cpu *cpu_buffer; | |
1516 | int cpu = raw_smp_processor_id(); | |
1517 | ||
1518 | cpu_buffer = buffer->buffers[cpu]; | |
1519 | ||
7a8e76a3 SR |
1520 | rb_commit(cpu_buffer, event); |
1521 | ||
bf41a158 SR |
1522 | /* |
1523 | * Only the last preempt count needs to restore preemption. | |
1524 | */ | |
182e9f5f SR |
1525 | if (preempt_count() == 1) |
1526 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1527 | else | |
bf41a158 | 1528 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
1529 | |
1530 | return 0; | |
1531 | } | |
c4f50183 | 1532 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 SR |
1533 | |
1534 | /** | |
1535 | * ring_buffer_write - write data to the buffer without reserving | |
1536 | * @buffer: The ring buffer to write to. | |
1537 | * @length: The length of the data being written (excluding the event header) | |
1538 | * @data: The data to write to the buffer. | |
1539 | * | |
1540 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
1541 | * one function. If you already have the data to write to the buffer, it | |
1542 | * may be easier to simply call this function. | |
1543 | * | |
1544 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
1545 | * and not the length of the event which would hold the header. | |
1546 | */ | |
1547 | int ring_buffer_write(struct ring_buffer *buffer, | |
1548 | unsigned long length, | |
1549 | void *data) | |
1550 | { | |
1551 | struct ring_buffer_per_cpu *cpu_buffer; | |
1552 | struct ring_buffer_event *event; | |
bf41a158 | 1553 | unsigned long event_length; |
7a8e76a3 SR |
1554 | void *body; |
1555 | int ret = -EBUSY; | |
bf41a158 | 1556 | int cpu, resched; |
7a8e76a3 | 1557 | |
033601a3 | 1558 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1559 | return -EBUSY; |
1560 | ||
7a8e76a3 SR |
1561 | if (atomic_read(&buffer->record_disabled)) |
1562 | return -EBUSY; | |
1563 | ||
182e9f5f | 1564 | resched = ftrace_preempt_disable(); |
bf41a158 | 1565 | |
7a8e76a3 SR |
1566 | cpu = raw_smp_processor_id(); |
1567 | ||
9e01c1b7 | 1568 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1569 | goto out; |
7a8e76a3 SR |
1570 | |
1571 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1572 | |
1573 | if (atomic_read(&cpu_buffer->record_disabled)) | |
1574 | goto out; | |
1575 | ||
1576 | event_length = rb_calculate_event_length(length); | |
1577 | event = rb_reserve_next_event(cpu_buffer, | |
1578 | RINGBUF_TYPE_DATA, event_length); | |
1579 | if (!event) | |
1580 | goto out; | |
1581 | ||
1582 | body = rb_event_data(event); | |
1583 | ||
1584 | memcpy(body, data, length); | |
1585 | ||
1586 | rb_commit(cpu_buffer, event); | |
1587 | ||
1588 | ret = 0; | |
1589 | out: | |
182e9f5f | 1590 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1591 | |
1592 | return ret; | |
1593 | } | |
c4f50183 | 1594 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 1595 | |
34a148bf | 1596 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
1597 | { |
1598 | struct buffer_page *reader = cpu_buffer->reader_page; | |
1599 | struct buffer_page *head = cpu_buffer->head_page; | |
1600 | struct buffer_page *commit = cpu_buffer->commit_page; | |
1601 | ||
1602 | return reader->read == rb_page_commit(reader) && | |
1603 | (commit == reader || | |
1604 | (commit == head && | |
1605 | head->read == rb_page_commit(commit))); | |
1606 | } | |
1607 | ||
7a8e76a3 SR |
1608 | /** |
1609 | * ring_buffer_record_disable - stop all writes into the buffer | |
1610 | * @buffer: The ring buffer to stop writes to. | |
1611 | * | |
1612 | * This prevents all writes to the buffer. Any attempt to write | |
1613 | * to the buffer after this will fail and return NULL. | |
1614 | * | |
1615 | * The caller should call synchronize_sched() after this. | |
1616 | */ | |
1617 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
1618 | { | |
1619 | atomic_inc(&buffer->record_disabled); | |
1620 | } | |
c4f50183 | 1621 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
1622 | |
1623 | /** | |
1624 | * ring_buffer_record_enable - enable writes to the buffer | |
1625 | * @buffer: The ring buffer to enable writes | |
1626 | * | |
1627 | * Note, multiple disables will need the same number of enables | |
1628 | * to truely enable the writing (much like preempt_disable). | |
1629 | */ | |
1630 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
1631 | { | |
1632 | atomic_dec(&buffer->record_disabled); | |
1633 | } | |
c4f50183 | 1634 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 SR |
1635 | |
1636 | /** | |
1637 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
1638 | * @buffer: The ring buffer to stop writes to. | |
1639 | * @cpu: The CPU buffer to stop | |
1640 | * | |
1641 | * This prevents all writes to the buffer. Any attempt to write | |
1642 | * to the buffer after this will fail and return NULL. | |
1643 | * | |
1644 | * The caller should call synchronize_sched() after this. | |
1645 | */ | |
1646 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
1647 | { | |
1648 | struct ring_buffer_per_cpu *cpu_buffer; | |
1649 | ||
9e01c1b7 | 1650 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1651 | return; |
7a8e76a3 SR |
1652 | |
1653 | cpu_buffer = buffer->buffers[cpu]; | |
1654 | atomic_inc(&cpu_buffer->record_disabled); | |
1655 | } | |
c4f50183 | 1656 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
1657 | |
1658 | /** | |
1659 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
1660 | * @buffer: The ring buffer to enable writes | |
1661 | * @cpu: The CPU to enable. | |
1662 | * | |
1663 | * Note, multiple disables will need the same number of enables | |
1664 | * to truely enable the writing (much like preempt_disable). | |
1665 | */ | |
1666 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
1667 | { | |
1668 | struct ring_buffer_per_cpu *cpu_buffer; | |
1669 | ||
9e01c1b7 | 1670 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1671 | return; |
7a8e76a3 SR |
1672 | |
1673 | cpu_buffer = buffer->buffers[cpu]; | |
1674 | atomic_dec(&cpu_buffer->record_disabled); | |
1675 | } | |
c4f50183 | 1676 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 SR |
1677 | |
1678 | /** | |
1679 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
1680 | * @buffer: The ring buffer | |
1681 | * @cpu: The per CPU buffer to get the entries from. | |
1682 | */ | |
1683 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
1684 | { | |
1685 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 1686 | unsigned long ret; |
7a8e76a3 | 1687 | |
9e01c1b7 | 1688 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1689 | return 0; |
7a8e76a3 SR |
1690 | |
1691 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 1692 | ret = cpu_buffer->entries; |
554f786e SR |
1693 | |
1694 | return ret; | |
7a8e76a3 | 1695 | } |
c4f50183 | 1696 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
1697 | |
1698 | /** | |
1699 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
1700 | * @buffer: The ring buffer | |
1701 | * @cpu: The per CPU buffer to get the number of overruns from | |
1702 | */ | |
1703 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
1704 | { | |
1705 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 1706 | unsigned long ret; |
7a8e76a3 | 1707 | |
9e01c1b7 | 1708 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1709 | return 0; |
7a8e76a3 SR |
1710 | |
1711 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 1712 | ret = cpu_buffer->overrun; |
554f786e SR |
1713 | |
1714 | return ret; | |
7a8e76a3 | 1715 | } |
c4f50183 | 1716 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 SR |
1717 | |
1718 | /** | |
1719 | * ring_buffer_entries - get the number of entries in a buffer | |
1720 | * @buffer: The ring buffer | |
1721 | * | |
1722 | * Returns the total number of entries in the ring buffer | |
1723 | * (all CPU entries) | |
1724 | */ | |
1725 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
1726 | { | |
1727 | struct ring_buffer_per_cpu *cpu_buffer; | |
1728 | unsigned long entries = 0; | |
1729 | int cpu; | |
1730 | ||
1731 | /* if you care about this being correct, lock the buffer */ | |
1732 | for_each_buffer_cpu(buffer, cpu) { | |
1733 | cpu_buffer = buffer->buffers[cpu]; | |
1734 | entries += cpu_buffer->entries; | |
1735 | } | |
1736 | ||
1737 | return entries; | |
1738 | } | |
c4f50183 | 1739 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
1740 | |
1741 | /** | |
1742 | * ring_buffer_overrun_cpu - get the number of overruns in buffer | |
1743 | * @buffer: The ring buffer | |
1744 | * | |
1745 | * Returns the total number of overruns in the ring buffer | |
1746 | * (all CPU entries) | |
1747 | */ | |
1748 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
1749 | { | |
1750 | struct ring_buffer_per_cpu *cpu_buffer; | |
1751 | unsigned long overruns = 0; | |
1752 | int cpu; | |
1753 | ||
1754 | /* if you care about this being correct, lock the buffer */ | |
1755 | for_each_buffer_cpu(buffer, cpu) { | |
1756 | cpu_buffer = buffer->buffers[cpu]; | |
1757 | overruns += cpu_buffer->overrun; | |
1758 | } | |
1759 | ||
1760 | return overruns; | |
1761 | } | |
c4f50183 | 1762 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 1763 | |
642edba5 | 1764 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
1765 | { |
1766 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1767 | ||
d769041f SR |
1768 | /* Iterator usage is expected to have record disabled */ |
1769 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
1770 | iter->head_page = cpu_buffer->head_page; | |
6f807acd | 1771 | iter->head = cpu_buffer->head_page->read; |
d769041f SR |
1772 | } else { |
1773 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 1774 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
1775 | } |
1776 | if (iter->head) | |
1777 | iter->read_stamp = cpu_buffer->read_stamp; | |
1778 | else | |
abc9b56d | 1779 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 1780 | } |
f83c9d0f | 1781 | |
642edba5 SR |
1782 | /** |
1783 | * ring_buffer_iter_reset - reset an iterator | |
1784 | * @iter: The iterator to reset | |
1785 | * | |
1786 | * Resets the iterator, so that it will start from the beginning | |
1787 | * again. | |
1788 | */ | |
1789 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
1790 | { | |
554f786e | 1791 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
1792 | unsigned long flags; |
1793 | ||
554f786e SR |
1794 | if (!iter) |
1795 | return; | |
1796 | ||
1797 | cpu_buffer = iter->cpu_buffer; | |
1798 | ||
642edba5 SR |
1799 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
1800 | rb_iter_reset(iter); | |
f83c9d0f | 1801 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 1802 | } |
c4f50183 | 1803 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
1804 | |
1805 | /** | |
1806 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
1807 | * @iter: The iterator to check | |
1808 | */ | |
1809 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
1810 | { | |
1811 | struct ring_buffer_per_cpu *cpu_buffer; | |
1812 | ||
1813 | cpu_buffer = iter->cpu_buffer; | |
1814 | ||
bf41a158 SR |
1815 | return iter->head_page == cpu_buffer->commit_page && |
1816 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 1817 | } |
c4f50183 | 1818 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
1819 | |
1820 | static void | |
1821 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1822 | struct ring_buffer_event *event) | |
1823 | { | |
1824 | u64 delta; | |
1825 | ||
1826 | switch (event->type) { | |
1827 | case RINGBUF_TYPE_PADDING: | |
1828 | return; | |
1829 | ||
1830 | case RINGBUF_TYPE_TIME_EXTEND: | |
1831 | delta = event->array[0]; | |
1832 | delta <<= TS_SHIFT; | |
1833 | delta += event->time_delta; | |
1834 | cpu_buffer->read_stamp += delta; | |
1835 | return; | |
1836 | ||
1837 | case RINGBUF_TYPE_TIME_STAMP: | |
1838 | /* FIXME: not implemented */ | |
1839 | return; | |
1840 | ||
1841 | case RINGBUF_TYPE_DATA: | |
1842 | cpu_buffer->read_stamp += event->time_delta; | |
1843 | return; | |
1844 | ||
1845 | default: | |
1846 | BUG(); | |
1847 | } | |
1848 | return; | |
1849 | } | |
1850 | ||
1851 | static void | |
1852 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
1853 | struct ring_buffer_event *event) | |
1854 | { | |
1855 | u64 delta; | |
1856 | ||
1857 | switch (event->type) { | |
1858 | case RINGBUF_TYPE_PADDING: | |
1859 | return; | |
1860 | ||
1861 | case RINGBUF_TYPE_TIME_EXTEND: | |
1862 | delta = event->array[0]; | |
1863 | delta <<= TS_SHIFT; | |
1864 | delta += event->time_delta; | |
1865 | iter->read_stamp += delta; | |
1866 | return; | |
1867 | ||
1868 | case RINGBUF_TYPE_TIME_STAMP: | |
1869 | /* FIXME: not implemented */ | |
1870 | return; | |
1871 | ||
1872 | case RINGBUF_TYPE_DATA: | |
1873 | iter->read_stamp += event->time_delta; | |
1874 | return; | |
1875 | ||
1876 | default: | |
1877 | BUG(); | |
1878 | } | |
1879 | return; | |
1880 | } | |
1881 | ||
d769041f SR |
1882 | static struct buffer_page * |
1883 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 1884 | { |
d769041f SR |
1885 | struct buffer_page *reader = NULL; |
1886 | unsigned long flags; | |
818e3dd3 | 1887 | int nr_loops = 0; |
d769041f | 1888 | |
3e03fb7f SR |
1889 | local_irq_save(flags); |
1890 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
1891 | |
1892 | again: | |
818e3dd3 SR |
1893 | /* |
1894 | * This should normally only loop twice. But because the | |
1895 | * start of the reader inserts an empty page, it causes | |
1896 | * a case where we will loop three times. There should be no | |
1897 | * reason to loop four times (that I know of). | |
1898 | */ | |
3e89c7bb | 1899 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
1900 | reader = NULL; |
1901 | goto out; | |
1902 | } | |
1903 | ||
d769041f SR |
1904 | reader = cpu_buffer->reader_page; |
1905 | ||
1906 | /* If there's more to read, return this page */ | |
bf41a158 | 1907 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
1908 | goto out; |
1909 | ||
1910 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
1911 | if (RB_WARN_ON(cpu_buffer, |
1912 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
1913 | goto out; | |
d769041f SR |
1914 | |
1915 | /* check if we caught up to the tail */ | |
1916 | reader = NULL; | |
bf41a158 | 1917 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 1918 | goto out; |
7a8e76a3 SR |
1919 | |
1920 | /* | |
d769041f SR |
1921 | * Splice the empty reader page into the list around the head. |
1922 | * Reset the reader page to size zero. | |
7a8e76a3 | 1923 | */ |
7a8e76a3 | 1924 | |
d769041f SR |
1925 | reader = cpu_buffer->head_page; |
1926 | cpu_buffer->reader_page->list.next = reader->list.next; | |
1927 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 SR |
1928 | |
1929 | local_set(&cpu_buffer->reader_page->write, 0); | |
abc9b56d | 1930 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
7a8e76a3 | 1931 | |
d769041f SR |
1932 | /* Make the reader page now replace the head */ |
1933 | reader->list.prev->next = &cpu_buffer->reader_page->list; | |
1934 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
7a8e76a3 SR |
1935 | |
1936 | /* | |
d769041f SR |
1937 | * If the tail is on the reader, then we must set the head |
1938 | * to the inserted page, otherwise we set it one before. | |
7a8e76a3 | 1939 | */ |
d769041f | 1940 | cpu_buffer->head_page = cpu_buffer->reader_page; |
7a8e76a3 | 1941 | |
bf41a158 | 1942 | if (cpu_buffer->commit_page != reader) |
d769041f SR |
1943 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
1944 | ||
1945 | /* Finally update the reader page to the new head */ | |
1946 | cpu_buffer->reader_page = reader; | |
1947 | rb_reset_reader_page(cpu_buffer); | |
1948 | ||
1949 | goto again; | |
1950 | ||
1951 | out: | |
3e03fb7f SR |
1952 | __raw_spin_unlock(&cpu_buffer->lock); |
1953 | local_irq_restore(flags); | |
d769041f SR |
1954 | |
1955 | return reader; | |
1956 | } | |
1957 | ||
1958 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
1959 | { | |
1960 | struct ring_buffer_event *event; | |
1961 | struct buffer_page *reader; | |
1962 | unsigned length; | |
1963 | ||
1964 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 1965 | |
d769041f | 1966 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
1967 | if (RB_WARN_ON(cpu_buffer, !reader)) |
1968 | return; | |
7a8e76a3 | 1969 | |
d769041f SR |
1970 | event = rb_reader_event(cpu_buffer); |
1971 | ||
1972 | if (event->type == RINGBUF_TYPE_DATA) | |
1973 | cpu_buffer->entries--; | |
1974 | ||
1975 | rb_update_read_stamp(cpu_buffer, event); | |
1976 | ||
1977 | length = rb_event_length(event); | |
6f807acd | 1978 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
1979 | } |
1980 | ||
1981 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
1982 | { | |
1983 | struct ring_buffer *buffer; | |
1984 | struct ring_buffer_per_cpu *cpu_buffer; | |
1985 | struct ring_buffer_event *event; | |
1986 | unsigned length; | |
1987 | ||
1988 | cpu_buffer = iter->cpu_buffer; | |
1989 | buffer = cpu_buffer->buffer; | |
1990 | ||
1991 | /* | |
1992 | * Check if we are at the end of the buffer. | |
1993 | */ | |
bf41a158 | 1994 | if (iter->head >= rb_page_size(iter->head_page)) { |
3e89c7bb SR |
1995 | if (RB_WARN_ON(buffer, |
1996 | iter->head_page == cpu_buffer->commit_page)) | |
1997 | return; | |
d769041f | 1998 | rb_inc_iter(iter); |
7a8e76a3 SR |
1999 | return; |
2000 | } | |
2001 | ||
2002 | event = rb_iter_head_event(iter); | |
2003 | ||
2004 | length = rb_event_length(event); | |
2005 | ||
2006 | /* | |
2007 | * This should not be called to advance the header if we are | |
2008 | * at the tail of the buffer. | |
2009 | */ | |
3e89c7bb | 2010 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 2011 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
2012 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
2013 | return; | |
7a8e76a3 SR |
2014 | |
2015 | rb_update_iter_read_stamp(iter, event); | |
2016 | ||
2017 | iter->head += length; | |
2018 | ||
2019 | /* check for end of page padding */ | |
bf41a158 SR |
2020 | if ((iter->head >= rb_page_size(iter->head_page)) && |
2021 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
2022 | rb_advance_iter(iter); |
2023 | } | |
2024 | ||
f83c9d0f SR |
2025 | static struct ring_buffer_event * |
2026 | rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
7a8e76a3 SR |
2027 | { |
2028 | struct ring_buffer_per_cpu *cpu_buffer; | |
2029 | struct ring_buffer_event *event; | |
d769041f | 2030 | struct buffer_page *reader; |
818e3dd3 | 2031 | int nr_loops = 0; |
7a8e76a3 | 2032 | |
7a8e76a3 SR |
2033 | cpu_buffer = buffer->buffers[cpu]; |
2034 | ||
2035 | again: | |
818e3dd3 SR |
2036 | /* |
2037 | * We repeat when a timestamp is encountered. It is possible | |
2038 | * to get multiple timestamps from an interrupt entering just | |
2039 | * as one timestamp is about to be written. The max times | |
2040 | * that this can happen is the number of nested interrupts we | |
2041 | * can have. Nesting 10 deep of interrupts is clearly | |
2042 | * an anomaly. | |
2043 | */ | |
3e89c7bb | 2044 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 2045 | return NULL; |
818e3dd3 | 2046 | |
d769041f SR |
2047 | reader = rb_get_reader_page(cpu_buffer); |
2048 | if (!reader) | |
7a8e76a3 SR |
2049 | return NULL; |
2050 | ||
d769041f | 2051 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 SR |
2052 | |
2053 | switch (event->type) { | |
2054 | case RINGBUF_TYPE_PADDING: | |
bf41a158 | 2055 | RB_WARN_ON(cpu_buffer, 1); |
d769041f SR |
2056 | rb_advance_reader(cpu_buffer); |
2057 | return NULL; | |
7a8e76a3 SR |
2058 | |
2059 | case RINGBUF_TYPE_TIME_EXTEND: | |
2060 | /* Internal data, OK to advance */ | |
d769041f | 2061 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2062 | goto again; |
2063 | ||
2064 | case RINGBUF_TYPE_TIME_STAMP: | |
2065 | /* FIXME: not implemented */ | |
d769041f | 2066 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2067 | goto again; |
2068 | ||
2069 | case RINGBUF_TYPE_DATA: | |
2070 | if (ts) { | |
2071 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
37886f6a SR |
2072 | ring_buffer_normalize_time_stamp(buffer, |
2073 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2074 | } |
2075 | return event; | |
2076 | ||
2077 | default: | |
2078 | BUG(); | |
2079 | } | |
2080 | ||
2081 | return NULL; | |
2082 | } | |
c4f50183 | 2083 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 2084 | |
f83c9d0f SR |
2085 | static struct ring_buffer_event * |
2086 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
2087 | { |
2088 | struct ring_buffer *buffer; | |
2089 | struct ring_buffer_per_cpu *cpu_buffer; | |
2090 | struct ring_buffer_event *event; | |
818e3dd3 | 2091 | int nr_loops = 0; |
7a8e76a3 SR |
2092 | |
2093 | if (ring_buffer_iter_empty(iter)) | |
2094 | return NULL; | |
2095 | ||
2096 | cpu_buffer = iter->cpu_buffer; | |
2097 | buffer = cpu_buffer->buffer; | |
2098 | ||
2099 | again: | |
818e3dd3 SR |
2100 | /* |
2101 | * We repeat when a timestamp is encountered. It is possible | |
2102 | * to get multiple timestamps from an interrupt entering just | |
2103 | * as one timestamp is about to be written. The max times | |
2104 | * that this can happen is the number of nested interrupts we | |
2105 | * can have. Nesting 10 deep of interrupts is clearly | |
2106 | * an anomaly. | |
2107 | */ | |
3e89c7bb | 2108 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 2109 | return NULL; |
818e3dd3 | 2110 | |
7a8e76a3 SR |
2111 | if (rb_per_cpu_empty(cpu_buffer)) |
2112 | return NULL; | |
2113 | ||
2114 | event = rb_iter_head_event(iter); | |
2115 | ||
2116 | switch (event->type) { | |
2117 | case RINGBUF_TYPE_PADDING: | |
d769041f | 2118 | rb_inc_iter(iter); |
7a8e76a3 SR |
2119 | goto again; |
2120 | ||
2121 | case RINGBUF_TYPE_TIME_EXTEND: | |
2122 | /* Internal data, OK to advance */ | |
2123 | rb_advance_iter(iter); | |
2124 | goto again; | |
2125 | ||
2126 | case RINGBUF_TYPE_TIME_STAMP: | |
2127 | /* FIXME: not implemented */ | |
2128 | rb_advance_iter(iter); | |
2129 | goto again; | |
2130 | ||
2131 | case RINGBUF_TYPE_DATA: | |
2132 | if (ts) { | |
2133 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
2134 | ring_buffer_normalize_time_stamp(buffer, |
2135 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2136 | } |
2137 | return event; | |
2138 | ||
2139 | default: | |
2140 | BUG(); | |
2141 | } | |
2142 | ||
2143 | return NULL; | |
2144 | } | |
c4f50183 | 2145 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 2146 | |
f83c9d0f SR |
2147 | /** |
2148 | * ring_buffer_peek - peek at the next event to be read | |
2149 | * @buffer: The ring buffer to read | |
2150 | * @cpu: The cpu to peak at | |
2151 | * @ts: The timestamp counter of this event. | |
2152 | * | |
2153 | * This will return the event that will be read next, but does | |
2154 | * not consume the data. | |
2155 | */ | |
2156 | struct ring_buffer_event * | |
2157 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2158 | { | |
2159 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 2160 | struct ring_buffer_event *event; |
f83c9d0f SR |
2161 | unsigned long flags; |
2162 | ||
554f786e | 2163 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2164 | return NULL; |
554f786e | 2165 | |
f83c9d0f SR |
2166 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2167 | event = rb_buffer_peek(buffer, cpu, ts); | |
2168 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2169 | ||
2170 | return event; | |
2171 | } | |
2172 | ||
2173 | /** | |
2174 | * ring_buffer_iter_peek - peek at the next event to be read | |
2175 | * @iter: The ring buffer iterator | |
2176 | * @ts: The timestamp counter of this event. | |
2177 | * | |
2178 | * This will return the event that will be read next, but does | |
2179 | * not increment the iterator. | |
2180 | */ | |
2181 | struct ring_buffer_event * | |
2182 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
2183 | { | |
2184 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2185 | struct ring_buffer_event *event; | |
2186 | unsigned long flags; | |
2187 | ||
2188 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2189 | event = rb_iter_peek(iter, ts); | |
2190 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2191 | ||
2192 | return event; | |
2193 | } | |
2194 | ||
7a8e76a3 SR |
2195 | /** |
2196 | * ring_buffer_consume - return an event and consume it | |
2197 | * @buffer: The ring buffer to get the next event from | |
2198 | * | |
2199 | * Returns the next event in the ring buffer, and that event is consumed. | |
2200 | * Meaning, that sequential reads will keep returning a different event, | |
2201 | * and eventually empty the ring buffer if the producer is slower. | |
2202 | */ | |
2203 | struct ring_buffer_event * | |
2204 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2205 | { | |
554f786e SR |
2206 | struct ring_buffer_per_cpu *cpu_buffer; |
2207 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 2208 | unsigned long flags; |
7a8e76a3 | 2209 | |
554f786e SR |
2210 | /* might be called in atomic */ |
2211 | preempt_disable(); | |
2212 | ||
9e01c1b7 | 2213 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 2214 | goto out; |
7a8e76a3 | 2215 | |
554f786e | 2216 | cpu_buffer = buffer->buffers[cpu]; |
f83c9d0f SR |
2217 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2218 | ||
2219 | event = rb_buffer_peek(buffer, cpu, ts); | |
7a8e76a3 | 2220 | if (!event) |
554f786e | 2221 | goto out_unlock; |
7a8e76a3 | 2222 | |
d769041f | 2223 | rb_advance_reader(cpu_buffer); |
7a8e76a3 | 2224 | |
554f786e | 2225 | out_unlock: |
f83c9d0f SR |
2226 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
2227 | ||
554f786e SR |
2228 | out: |
2229 | preempt_enable(); | |
2230 | ||
7a8e76a3 SR |
2231 | return event; |
2232 | } | |
c4f50183 | 2233 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
2234 | |
2235 | /** | |
2236 | * ring_buffer_read_start - start a non consuming read of the buffer | |
2237 | * @buffer: The ring buffer to read from | |
2238 | * @cpu: The cpu buffer to iterate over | |
2239 | * | |
2240 | * This starts up an iteration through the buffer. It also disables | |
2241 | * the recording to the buffer until the reading is finished. | |
2242 | * This prevents the reading from being corrupted. This is not | |
2243 | * a consuming read, so a producer is not expected. | |
2244 | * | |
2245 | * Must be paired with ring_buffer_finish. | |
2246 | */ | |
2247 | struct ring_buffer_iter * | |
2248 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
2249 | { | |
2250 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2251 | struct ring_buffer_iter *iter; |
d769041f | 2252 | unsigned long flags; |
7a8e76a3 | 2253 | |
9e01c1b7 | 2254 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2255 | return NULL; |
7a8e76a3 SR |
2256 | |
2257 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
2258 | if (!iter) | |
8aabee57 | 2259 | return NULL; |
7a8e76a3 SR |
2260 | |
2261 | cpu_buffer = buffer->buffers[cpu]; | |
2262 | ||
2263 | iter->cpu_buffer = cpu_buffer; | |
2264 | ||
2265 | atomic_inc(&cpu_buffer->record_disabled); | |
2266 | synchronize_sched(); | |
2267 | ||
f83c9d0f | 2268 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 2269 | __raw_spin_lock(&cpu_buffer->lock); |
642edba5 | 2270 | rb_iter_reset(iter); |
3e03fb7f | 2271 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 2272 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
2273 | |
2274 | return iter; | |
2275 | } | |
c4f50183 | 2276 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
2277 | |
2278 | /** | |
2279 | * ring_buffer_finish - finish reading the iterator of the buffer | |
2280 | * @iter: The iterator retrieved by ring_buffer_start | |
2281 | * | |
2282 | * This re-enables the recording to the buffer, and frees the | |
2283 | * iterator. | |
2284 | */ | |
2285 | void | |
2286 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
2287 | { | |
2288 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2289 | ||
2290 | atomic_dec(&cpu_buffer->record_disabled); | |
2291 | kfree(iter); | |
2292 | } | |
c4f50183 | 2293 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
2294 | |
2295 | /** | |
2296 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
2297 | * @iter: The ring buffer iterator | |
2298 | * @ts: The time stamp of the event read. | |
2299 | * | |
2300 | * This reads the next event in the ring buffer and increments the iterator. | |
2301 | */ | |
2302 | struct ring_buffer_event * | |
2303 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
2304 | { | |
2305 | struct ring_buffer_event *event; | |
f83c9d0f SR |
2306 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
2307 | unsigned long flags; | |
7a8e76a3 | 2308 | |
f83c9d0f SR |
2309 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2310 | event = rb_iter_peek(iter, ts); | |
7a8e76a3 | 2311 | if (!event) |
f83c9d0f | 2312 | goto out; |
7a8e76a3 SR |
2313 | |
2314 | rb_advance_iter(iter); | |
f83c9d0f SR |
2315 | out: |
2316 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 SR |
2317 | |
2318 | return event; | |
2319 | } | |
c4f50183 | 2320 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
2321 | |
2322 | /** | |
2323 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
2324 | * @buffer: The ring buffer. | |
2325 | */ | |
2326 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
2327 | { | |
2328 | return BUF_PAGE_SIZE * buffer->pages; | |
2329 | } | |
c4f50183 | 2330 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
2331 | |
2332 | static void | |
2333 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
2334 | { | |
2335 | cpu_buffer->head_page | |
2336 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 2337 | local_set(&cpu_buffer->head_page->write, 0); |
abc9b56d | 2338 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 2339 | |
6f807acd | 2340 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
2341 | |
2342 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
2343 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
2344 | ||
2345 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
2346 | local_set(&cpu_buffer->reader_page->write, 0); | |
abc9b56d | 2347 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 2348 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 2349 | |
7a8e76a3 SR |
2350 | cpu_buffer->overrun = 0; |
2351 | cpu_buffer->entries = 0; | |
69507c06 SR |
2352 | |
2353 | cpu_buffer->write_stamp = 0; | |
2354 | cpu_buffer->read_stamp = 0; | |
7a8e76a3 SR |
2355 | } |
2356 | ||
2357 | /** | |
2358 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
2359 | * @buffer: The ring buffer to reset a per cpu buffer of | |
2360 | * @cpu: The CPU buffer to be reset | |
2361 | */ | |
2362 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
2363 | { | |
2364 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2365 | unsigned long flags; | |
2366 | ||
9e01c1b7 | 2367 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2368 | return; |
7a8e76a3 | 2369 | |
f83c9d0f SR |
2370 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2371 | ||
3e03fb7f | 2372 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
2373 | |
2374 | rb_reset_cpu(cpu_buffer); | |
2375 | ||
3e03fb7f | 2376 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f SR |
2377 | |
2378 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 2379 | } |
c4f50183 | 2380 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
2381 | |
2382 | /** | |
2383 | * ring_buffer_reset - reset a ring buffer | |
2384 | * @buffer: The ring buffer to reset all cpu buffers | |
2385 | */ | |
2386 | void ring_buffer_reset(struct ring_buffer *buffer) | |
2387 | { | |
7a8e76a3 SR |
2388 | int cpu; |
2389 | ||
7a8e76a3 | 2390 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 2391 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 2392 | } |
c4f50183 | 2393 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
2394 | |
2395 | /** | |
2396 | * rind_buffer_empty - is the ring buffer empty? | |
2397 | * @buffer: The ring buffer to test | |
2398 | */ | |
2399 | int ring_buffer_empty(struct ring_buffer *buffer) | |
2400 | { | |
2401 | struct ring_buffer_per_cpu *cpu_buffer; | |
2402 | int cpu; | |
2403 | ||
2404 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
2405 | for_each_buffer_cpu(buffer, cpu) { | |
2406 | cpu_buffer = buffer->buffers[cpu]; | |
2407 | if (!rb_per_cpu_empty(cpu_buffer)) | |
2408 | return 0; | |
2409 | } | |
554f786e | 2410 | |
7a8e76a3 SR |
2411 | return 1; |
2412 | } | |
c4f50183 | 2413 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
2414 | |
2415 | /** | |
2416 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
2417 | * @buffer: The ring buffer | |
2418 | * @cpu: The CPU buffer to test | |
2419 | */ | |
2420 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
2421 | { | |
2422 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2423 | int ret; |
7a8e76a3 | 2424 | |
9e01c1b7 | 2425 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2426 | return 1; |
7a8e76a3 SR |
2427 | |
2428 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e SR |
2429 | ret = rb_per_cpu_empty(cpu_buffer); |
2430 | ||
554f786e SR |
2431 | |
2432 | return ret; | |
7a8e76a3 | 2433 | } |
c4f50183 | 2434 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 SR |
2435 | |
2436 | /** | |
2437 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
2438 | * @buffer_a: One buffer to swap with | |
2439 | * @buffer_b: The other buffer to swap with | |
2440 | * | |
2441 | * This function is useful for tracers that want to take a "snapshot" | |
2442 | * of a CPU buffer and has another back up buffer lying around. | |
2443 | * it is expected that the tracer handles the cpu buffer not being | |
2444 | * used at the moment. | |
2445 | */ | |
2446 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
2447 | struct ring_buffer *buffer_b, int cpu) | |
2448 | { | |
2449 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
2450 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
2451 | int ret = -EINVAL; |
2452 | ||
9e01c1b7 RR |
2453 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
2454 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 2455 | goto out; |
7a8e76a3 SR |
2456 | |
2457 | /* At least make sure the two buffers are somewhat the same */ | |
6d102bc6 | 2458 | if (buffer_a->pages != buffer_b->pages) |
554f786e SR |
2459 | goto out; |
2460 | ||
2461 | ret = -EAGAIN; | |
7a8e76a3 | 2462 | |
97b17efe | 2463 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 2464 | goto out; |
97b17efe SR |
2465 | |
2466 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 2467 | goto out; |
97b17efe SR |
2468 | |
2469 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 2470 | goto out; |
97b17efe | 2471 | |
7a8e76a3 SR |
2472 | cpu_buffer_a = buffer_a->buffers[cpu]; |
2473 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
2474 | ||
97b17efe | 2475 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 2476 | goto out; |
97b17efe SR |
2477 | |
2478 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 2479 | goto out; |
97b17efe | 2480 | |
7a8e76a3 SR |
2481 | /* |
2482 | * We can't do a synchronize_sched here because this | |
2483 | * function can be called in atomic context. | |
2484 | * Normally this will be called from the same CPU as cpu. | |
2485 | * If not it's up to the caller to protect this. | |
2486 | */ | |
2487 | atomic_inc(&cpu_buffer_a->record_disabled); | |
2488 | atomic_inc(&cpu_buffer_b->record_disabled); | |
2489 | ||
2490 | buffer_a->buffers[cpu] = cpu_buffer_b; | |
2491 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
2492 | ||
2493 | cpu_buffer_b->buffer = buffer_a; | |
2494 | cpu_buffer_a->buffer = buffer_b; | |
2495 | ||
2496 | atomic_dec(&cpu_buffer_a->record_disabled); | |
2497 | atomic_dec(&cpu_buffer_b->record_disabled); | |
2498 | ||
554f786e SR |
2499 | ret = 0; |
2500 | out: | |
554f786e | 2501 | return ret; |
7a8e76a3 | 2502 | } |
c4f50183 | 2503 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
7a8e76a3 | 2504 | |
8789a9e7 | 2505 | static void rb_remove_entries(struct ring_buffer_per_cpu *cpu_buffer, |
667d2412 LJ |
2506 | struct buffer_data_page *bpage, |
2507 | unsigned int offset) | |
8789a9e7 SR |
2508 | { |
2509 | struct ring_buffer_event *event; | |
2510 | unsigned long head; | |
2511 | ||
2512 | __raw_spin_lock(&cpu_buffer->lock); | |
667d2412 | 2513 | for (head = offset; head < local_read(&bpage->commit); |
8789a9e7 SR |
2514 | head += rb_event_length(event)) { |
2515 | ||
044fa782 | 2516 | event = __rb_data_page_index(bpage, head); |
8789a9e7 SR |
2517 | if (RB_WARN_ON(cpu_buffer, rb_null_event(event))) |
2518 | return; | |
2519 | /* Only count data entries */ | |
2520 | if (event->type != RINGBUF_TYPE_DATA) | |
2521 | continue; | |
2522 | cpu_buffer->entries--; | |
2523 | } | |
2524 | __raw_spin_unlock(&cpu_buffer->lock); | |
2525 | } | |
2526 | ||
2527 | /** | |
2528 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
2529 | * @buffer: the buffer to allocate for. | |
2530 | * | |
2531 | * This function is used in conjunction with ring_buffer_read_page. | |
2532 | * When reading a full page from the ring buffer, these functions | |
2533 | * can be used to speed up the process. The calling function should | |
2534 | * allocate a few pages first with this function. Then when it | |
2535 | * needs to get pages from the ring buffer, it passes the result | |
2536 | * of this function into ring_buffer_read_page, which will swap | |
2537 | * the page that was allocated, with the read page of the buffer. | |
2538 | * | |
2539 | * Returns: | |
2540 | * The page allocated, or NULL on error. | |
2541 | */ | |
2542 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer) | |
2543 | { | |
044fa782 | 2544 | struct buffer_data_page *bpage; |
ef7a4a16 | 2545 | unsigned long addr; |
8789a9e7 SR |
2546 | |
2547 | addr = __get_free_page(GFP_KERNEL); | |
2548 | if (!addr) | |
2549 | return NULL; | |
2550 | ||
044fa782 | 2551 | bpage = (void *)addr; |
8789a9e7 | 2552 | |
ef7a4a16 SR |
2553 | rb_init_page(bpage); |
2554 | ||
044fa782 | 2555 | return bpage; |
8789a9e7 SR |
2556 | } |
2557 | ||
2558 | /** | |
2559 | * ring_buffer_free_read_page - free an allocated read page | |
2560 | * @buffer: the buffer the page was allocate for | |
2561 | * @data: the page to free | |
2562 | * | |
2563 | * Free a page allocated from ring_buffer_alloc_read_page. | |
2564 | */ | |
2565 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
2566 | { | |
2567 | free_page((unsigned long)data); | |
2568 | } | |
2569 | ||
2570 | /** | |
2571 | * ring_buffer_read_page - extract a page from the ring buffer | |
2572 | * @buffer: buffer to extract from | |
2573 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 2574 | * @len: amount to extract |
8789a9e7 SR |
2575 | * @cpu: the cpu of the buffer to extract |
2576 | * @full: should the extraction only happen when the page is full. | |
2577 | * | |
2578 | * This function will pull out a page from the ring buffer and consume it. | |
2579 | * @data_page must be the address of the variable that was returned | |
2580 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
2581 | * to swap with a page in the ring buffer. | |
2582 | * | |
2583 | * for example: | |
b85fa01e | 2584 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
2585 | * if (!rpage) |
2586 | * return error; | |
ef7a4a16 | 2587 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
2588 | * if (ret >= 0) |
2589 | * process_page(rpage, ret); | |
8789a9e7 SR |
2590 | * |
2591 | * When @full is set, the function will not return true unless | |
2592 | * the writer is off the reader page. | |
2593 | * | |
2594 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
2595 | * The ring buffer can be used anywhere in the kernel and can not | |
2596 | * blindly call wake_up. The layer that uses the ring buffer must be | |
2597 | * responsible for that. | |
2598 | * | |
2599 | * Returns: | |
667d2412 LJ |
2600 | * >=0 if data has been transferred, returns the offset of consumed data. |
2601 | * <0 if no data has been transferred. | |
8789a9e7 SR |
2602 | */ |
2603 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 2604 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
2605 | { |
2606 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2607 | struct ring_buffer_event *event; | |
044fa782 | 2608 | struct buffer_data_page *bpage; |
ef7a4a16 | 2609 | struct buffer_page *reader; |
8789a9e7 | 2610 | unsigned long flags; |
ef7a4a16 | 2611 | unsigned int commit; |
667d2412 | 2612 | unsigned int read; |
4f3640f8 | 2613 | u64 save_timestamp; |
667d2412 | 2614 | int ret = -1; |
8789a9e7 | 2615 | |
554f786e SR |
2616 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
2617 | goto out; | |
2618 | ||
474d32b6 SR |
2619 | /* |
2620 | * If len is not big enough to hold the page header, then | |
2621 | * we can not copy anything. | |
2622 | */ | |
2623 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 2624 | goto out; |
474d32b6 SR |
2625 | |
2626 | len -= BUF_PAGE_HDR_SIZE; | |
2627 | ||
8789a9e7 | 2628 | if (!data_page) |
554f786e | 2629 | goto out; |
8789a9e7 | 2630 | |
044fa782 SR |
2631 | bpage = *data_page; |
2632 | if (!bpage) | |
554f786e | 2633 | goto out; |
8789a9e7 SR |
2634 | |
2635 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2636 | ||
ef7a4a16 SR |
2637 | reader = rb_get_reader_page(cpu_buffer); |
2638 | if (!reader) | |
554f786e | 2639 | goto out_unlock; |
8789a9e7 | 2640 | |
ef7a4a16 SR |
2641 | event = rb_reader_event(cpu_buffer); |
2642 | ||
2643 | read = reader->read; | |
2644 | commit = rb_page_commit(reader); | |
667d2412 | 2645 | |
8789a9e7 | 2646 | /* |
474d32b6 SR |
2647 | * If this page has been partially read or |
2648 | * if len is not big enough to read the rest of the page or | |
2649 | * a writer is still on the page, then | |
2650 | * we must copy the data from the page to the buffer. | |
2651 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 2652 | */ |
474d32b6 | 2653 | if (read || (len < (commit - read)) || |
ef7a4a16 | 2654 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 2655 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
2656 | unsigned int rpos = read; |
2657 | unsigned int pos = 0; | |
ef7a4a16 | 2658 | unsigned int size; |
8789a9e7 SR |
2659 | |
2660 | if (full) | |
554f786e | 2661 | goto out_unlock; |
8789a9e7 | 2662 | |
ef7a4a16 SR |
2663 | if (len > (commit - read)) |
2664 | len = (commit - read); | |
2665 | ||
2666 | size = rb_event_length(event); | |
2667 | ||
2668 | if (len < size) | |
554f786e | 2669 | goto out_unlock; |
ef7a4a16 | 2670 | |
4f3640f8 SR |
2671 | /* save the current timestamp, since the user will need it */ |
2672 | save_timestamp = cpu_buffer->read_stamp; | |
2673 | ||
ef7a4a16 SR |
2674 | /* Need to copy one event at a time */ |
2675 | do { | |
474d32b6 | 2676 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
2677 | |
2678 | len -= size; | |
2679 | ||
2680 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
2681 | rpos = reader->read; |
2682 | pos += size; | |
ef7a4a16 SR |
2683 | |
2684 | event = rb_reader_event(cpu_buffer); | |
2685 | size = rb_event_length(event); | |
2686 | } while (len > size); | |
667d2412 LJ |
2687 | |
2688 | /* update bpage */ | |
ef7a4a16 | 2689 | local_set(&bpage->commit, pos); |
4f3640f8 | 2690 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 2691 | |
474d32b6 SR |
2692 | /* we copied everything to the beginning */ |
2693 | read = 0; | |
8789a9e7 SR |
2694 | } else { |
2695 | /* swap the pages */ | |
044fa782 | 2696 | rb_init_page(bpage); |
ef7a4a16 SR |
2697 | bpage = reader->page; |
2698 | reader->page = *data_page; | |
2699 | local_set(&reader->write, 0); | |
2700 | reader->read = 0; | |
044fa782 | 2701 | *data_page = bpage; |
ef7a4a16 SR |
2702 | |
2703 | /* update the entry counter */ | |
2704 | rb_remove_entries(cpu_buffer, bpage, read); | |
8789a9e7 | 2705 | } |
667d2412 | 2706 | ret = read; |
8789a9e7 | 2707 | |
554f786e | 2708 | out_unlock: |
8789a9e7 SR |
2709 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
2710 | ||
554f786e | 2711 | out: |
8789a9e7 SR |
2712 | return ret; |
2713 | } | |
2714 | ||
a3583244 SR |
2715 | static ssize_t |
2716 | rb_simple_read(struct file *filp, char __user *ubuf, | |
2717 | size_t cnt, loff_t *ppos) | |
2718 | { | |
5e39841c | 2719 | unsigned long *p = filp->private_data; |
a3583244 SR |
2720 | char buf[64]; |
2721 | int r; | |
2722 | ||
033601a3 SR |
2723 | if (test_bit(RB_BUFFERS_DISABLED_BIT, p)) |
2724 | r = sprintf(buf, "permanently disabled\n"); | |
2725 | else | |
2726 | r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p)); | |
a3583244 SR |
2727 | |
2728 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
2729 | } | |
2730 | ||
2731 | static ssize_t | |
2732 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
2733 | size_t cnt, loff_t *ppos) | |
2734 | { | |
5e39841c | 2735 | unsigned long *p = filp->private_data; |
a3583244 | 2736 | char buf[64]; |
5e39841c | 2737 | unsigned long val; |
a3583244 SR |
2738 | int ret; |
2739 | ||
2740 | if (cnt >= sizeof(buf)) | |
2741 | return -EINVAL; | |
2742 | ||
2743 | if (copy_from_user(&buf, ubuf, cnt)) | |
2744 | return -EFAULT; | |
2745 | ||
2746 | buf[cnt] = 0; | |
2747 | ||
2748 | ret = strict_strtoul(buf, 10, &val); | |
2749 | if (ret < 0) | |
2750 | return ret; | |
2751 | ||
033601a3 SR |
2752 | if (val) |
2753 | set_bit(RB_BUFFERS_ON_BIT, p); | |
2754 | else | |
2755 | clear_bit(RB_BUFFERS_ON_BIT, p); | |
a3583244 SR |
2756 | |
2757 | (*ppos)++; | |
2758 | ||
2759 | return cnt; | |
2760 | } | |
2761 | ||
5e2336a0 | 2762 | static const struct file_operations rb_simple_fops = { |
a3583244 SR |
2763 | .open = tracing_open_generic, |
2764 | .read = rb_simple_read, | |
2765 | .write = rb_simple_write, | |
2766 | }; | |
2767 | ||
2768 | ||
2769 | static __init int rb_init_debugfs(void) | |
2770 | { | |
2771 | struct dentry *d_tracer; | |
2772 | struct dentry *entry; | |
2773 | ||
2774 | d_tracer = tracing_init_dentry(); | |
2775 | ||
2776 | entry = debugfs_create_file("tracing_on", 0644, d_tracer, | |
033601a3 | 2777 | &ring_buffer_flags, &rb_simple_fops); |
a3583244 SR |
2778 | if (!entry) |
2779 | pr_warning("Could not create debugfs 'tracing_on' entry\n"); | |
2780 | ||
2781 | return 0; | |
2782 | } | |
2783 | ||
2784 | fs_initcall(rb_init_debugfs); | |
554f786e | 2785 | |
59222efe | 2786 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
2787 | static int __cpuinit rb_cpu_notify(struct notifier_block *self, |
2788 | unsigned long action, void *hcpu) | |
2789 | { | |
2790 | struct ring_buffer *buffer = | |
2791 | container_of(self, struct ring_buffer, cpu_notify); | |
2792 | long cpu = (long)hcpu; | |
2793 | ||
2794 | switch (action) { | |
2795 | case CPU_UP_PREPARE: | |
2796 | case CPU_UP_PREPARE_FROZEN: | |
2797 | if (cpu_isset(cpu, *buffer->cpumask)) | |
2798 | return NOTIFY_OK; | |
2799 | ||
2800 | buffer->buffers[cpu] = | |
2801 | rb_allocate_cpu_buffer(buffer, cpu); | |
2802 | if (!buffer->buffers[cpu]) { | |
2803 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
2804 | cpu); | |
2805 | return NOTIFY_OK; | |
2806 | } | |
2807 | smp_wmb(); | |
2808 | cpu_set(cpu, *buffer->cpumask); | |
2809 | break; | |
2810 | case CPU_DOWN_PREPARE: | |
2811 | case CPU_DOWN_PREPARE_FROZEN: | |
2812 | /* | |
2813 | * Do nothing. | |
2814 | * If we were to free the buffer, then the user would | |
2815 | * lose any trace that was in the buffer. | |
2816 | */ | |
2817 | break; | |
2818 | default: | |
2819 | break; | |
2820 | } | |
2821 | return NOTIFY_OK; | |
2822 | } | |
2823 | #endif |