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 | ||
d1b182a8 SR |
24 | /* |
25 | * The ring buffer header is special. We must manually up keep it. | |
26 | */ | |
27 | int ring_buffer_print_entry_header(struct trace_seq *s) | |
28 | { | |
29 | int ret; | |
30 | ||
31 | ret = trace_seq_printf(s, "\ttype : 2 bits\n"); | |
32 | ret = trace_seq_printf(s, "\tlen : 3 bits\n"); | |
33 | ret = trace_seq_printf(s, "\ttime_delta : 27 bits\n"); | |
34 | ret = trace_seq_printf(s, "\tarray : 32 bits\n"); | |
35 | ret = trace_seq_printf(s, "\n"); | |
36 | ret = trace_seq_printf(s, "\tpadding : type == %d\n", | |
37 | RINGBUF_TYPE_PADDING); | |
38 | ret = trace_seq_printf(s, "\ttime_extend : type == %d\n", | |
39 | RINGBUF_TYPE_TIME_EXTEND); | |
40 | ret = trace_seq_printf(s, "\tdata : type == %d\n", | |
41 | RINGBUF_TYPE_DATA); | |
42 | ||
43 | return ret; | |
44 | } | |
45 | ||
5cc98548 SR |
46 | /* |
47 | * The ring buffer is made up of a list of pages. A separate list of pages is | |
48 | * allocated for each CPU. A writer may only write to a buffer that is | |
49 | * associated with the CPU it is currently executing on. A reader may read | |
50 | * from any per cpu buffer. | |
51 | * | |
52 | * The reader is special. For each per cpu buffer, the reader has its own | |
53 | * reader page. When a reader has read the entire reader page, this reader | |
54 | * page is swapped with another page in the ring buffer. | |
55 | * | |
56 | * Now, as long as the writer is off the reader page, the reader can do what | |
57 | * ever it wants with that page. The writer will never write to that page | |
58 | * again (as long as it is out of the ring buffer). | |
59 | * | |
60 | * Here's some silly ASCII art. | |
61 | * | |
62 | * +------+ | |
63 | * |reader| RING BUFFER | |
64 | * |page | | |
65 | * +------+ +---+ +---+ +---+ | |
66 | * | |-->| |-->| | | |
67 | * +---+ +---+ +---+ | |
68 | * ^ | | |
69 | * | | | |
70 | * +---------------+ | |
71 | * | |
72 | * | |
73 | * +------+ | |
74 | * |reader| RING BUFFER | |
75 | * |page |------------------v | |
76 | * +------+ +---+ +---+ +---+ | |
77 | * | |-->| |-->| | | |
78 | * +---+ +---+ +---+ | |
79 | * ^ | | |
80 | * | | | |
81 | * +---------------+ | |
82 | * | |
83 | * | |
84 | * +------+ | |
85 | * |reader| RING BUFFER | |
86 | * |page |------------------v | |
87 | * +------+ +---+ +---+ +---+ | |
88 | * ^ | |-->| |-->| | | |
89 | * | +---+ +---+ +---+ | |
90 | * | | | |
91 | * | | | |
92 | * +------------------------------+ | |
93 | * | |
94 | * | |
95 | * +------+ | |
96 | * |buffer| RING BUFFER | |
97 | * |page |------------------v | |
98 | * +------+ +---+ +---+ +---+ | |
99 | * ^ | | | |-->| | | |
100 | * | New +---+ +---+ +---+ | |
101 | * | Reader------^ | | |
102 | * | page | | |
103 | * +------------------------------+ | |
104 | * | |
105 | * | |
106 | * After we make this swap, the reader can hand this page off to the splice | |
107 | * code and be done with it. It can even allocate a new page if it needs to | |
108 | * and swap that into the ring buffer. | |
109 | * | |
110 | * We will be using cmpxchg soon to make all this lockless. | |
111 | * | |
112 | */ | |
113 | ||
033601a3 SR |
114 | /* |
115 | * A fast way to enable or disable all ring buffers is to | |
116 | * call tracing_on or tracing_off. Turning off the ring buffers | |
117 | * prevents all ring buffers from being recorded to. | |
118 | * Turning this switch on, makes it OK to write to the | |
119 | * ring buffer, if the ring buffer is enabled itself. | |
120 | * | |
121 | * There's three layers that must be on in order to write | |
122 | * to the ring buffer. | |
123 | * | |
124 | * 1) This global flag must be set. | |
125 | * 2) The ring buffer must be enabled for recording. | |
126 | * 3) The per cpu buffer must be enabled for recording. | |
127 | * | |
128 | * In case of an anomaly, this global flag has a bit set that | |
129 | * will permantly disable all ring buffers. | |
130 | */ | |
131 | ||
132 | /* | |
133 | * Global flag to disable all recording to ring buffers | |
134 | * This has two bits: ON, DISABLED | |
135 | * | |
136 | * ON DISABLED | |
137 | * ---- ---------- | |
138 | * 0 0 : ring buffers are off | |
139 | * 1 0 : ring buffers are on | |
140 | * X 1 : ring buffers are permanently disabled | |
141 | */ | |
142 | ||
143 | enum { | |
144 | RB_BUFFERS_ON_BIT = 0, | |
145 | RB_BUFFERS_DISABLED_BIT = 1, | |
146 | }; | |
147 | ||
148 | enum { | |
149 | RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT, | |
150 | RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT, | |
151 | }; | |
152 | ||
5e39841c | 153 | static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON; |
a3583244 | 154 | |
474d32b6 SR |
155 | #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data) |
156 | ||
a3583244 SR |
157 | /** |
158 | * tracing_on - enable all tracing buffers | |
159 | * | |
160 | * This function enables all tracing buffers that may have been | |
161 | * disabled with tracing_off. | |
162 | */ | |
163 | void tracing_on(void) | |
164 | { | |
033601a3 | 165 | set_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags); |
a3583244 | 166 | } |
c4f50183 | 167 | EXPORT_SYMBOL_GPL(tracing_on); |
a3583244 SR |
168 | |
169 | /** | |
170 | * tracing_off - turn off all tracing buffers | |
171 | * | |
172 | * This function stops all tracing buffers from recording data. | |
173 | * It does not disable any overhead the tracers themselves may | |
174 | * be causing. This function simply causes all recording to | |
175 | * the ring buffers to fail. | |
176 | */ | |
177 | void tracing_off(void) | |
178 | { | |
033601a3 SR |
179 | clear_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags); |
180 | } | |
c4f50183 | 181 | EXPORT_SYMBOL_GPL(tracing_off); |
033601a3 SR |
182 | |
183 | /** | |
184 | * tracing_off_permanent - permanently disable ring buffers | |
185 | * | |
186 | * This function, once called, will disable all ring buffers | |
c3706f00 | 187 | * permanently. |
033601a3 SR |
188 | */ |
189 | void tracing_off_permanent(void) | |
190 | { | |
191 | set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags); | |
a3583244 SR |
192 | } |
193 | ||
988ae9d6 SR |
194 | /** |
195 | * tracing_is_on - show state of ring buffers enabled | |
196 | */ | |
197 | int tracing_is_on(void) | |
198 | { | |
199 | return ring_buffer_flags == RB_BUFFERS_ON; | |
200 | } | |
201 | EXPORT_SYMBOL_GPL(tracing_is_on); | |
202 | ||
d06bbd66 IM |
203 | #include "trace.h" |
204 | ||
e3d6bf0a | 205 | #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array)) |
67d34724 | 206 | #define RB_ALIGNMENT 4U |
7a8e76a3 SR |
207 | #define RB_MAX_SMALL_DATA 28 |
208 | ||
209 | enum { | |
210 | RB_LEN_TIME_EXTEND = 8, | |
211 | RB_LEN_TIME_STAMP = 16, | |
212 | }; | |
213 | ||
2d622719 TZ |
214 | static inline int rb_null_event(struct ring_buffer_event *event) |
215 | { | |
216 | return event->type == RINGBUF_TYPE_PADDING && event->time_delta == 0; | |
217 | } | |
218 | ||
219 | static inline int rb_discarded_event(struct ring_buffer_event *event) | |
220 | { | |
221 | return event->type == RINGBUF_TYPE_PADDING && event->time_delta; | |
222 | } | |
223 | ||
224 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
225 | { | |
226 | event->type = RINGBUF_TYPE_PADDING; | |
227 | event->time_delta = 0; | |
228 | } | |
229 | ||
34a148bf | 230 | static unsigned |
2d622719 | 231 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
232 | { |
233 | unsigned length; | |
234 | ||
2d622719 TZ |
235 | if (event->len) |
236 | length = event->len * RB_ALIGNMENT; | |
237 | else | |
238 | length = event->array[0]; | |
239 | return length + RB_EVNT_HDR_SIZE; | |
240 | } | |
241 | ||
242 | /* inline for ring buffer fast paths */ | |
243 | static unsigned | |
244 | rb_event_length(struct ring_buffer_event *event) | |
245 | { | |
7a8e76a3 SR |
246 | switch (event->type) { |
247 | case RINGBUF_TYPE_PADDING: | |
2d622719 TZ |
248 | if (rb_null_event(event)) |
249 | /* undefined */ | |
250 | return -1; | |
251 | return rb_event_data_length(event); | |
7a8e76a3 SR |
252 | |
253 | case RINGBUF_TYPE_TIME_EXTEND: | |
254 | return RB_LEN_TIME_EXTEND; | |
255 | ||
256 | case RINGBUF_TYPE_TIME_STAMP: | |
257 | return RB_LEN_TIME_STAMP; | |
258 | ||
259 | case RINGBUF_TYPE_DATA: | |
2d622719 | 260 | return rb_event_data_length(event); |
7a8e76a3 SR |
261 | default: |
262 | BUG(); | |
263 | } | |
264 | /* not hit */ | |
265 | return 0; | |
266 | } | |
267 | ||
268 | /** | |
269 | * ring_buffer_event_length - return the length of the event | |
270 | * @event: the event to get the length of | |
271 | */ | |
272 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
273 | { | |
465634ad RR |
274 | unsigned length = rb_event_length(event); |
275 | if (event->type != RINGBUF_TYPE_DATA) | |
276 | return length; | |
277 | length -= RB_EVNT_HDR_SIZE; | |
278 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
279 | length -= sizeof(event->array[0]); | |
280 | return length; | |
7a8e76a3 | 281 | } |
c4f50183 | 282 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
283 | |
284 | /* inline for ring buffer fast paths */ | |
34a148bf | 285 | static void * |
7a8e76a3 SR |
286 | rb_event_data(struct ring_buffer_event *event) |
287 | { | |
288 | BUG_ON(event->type != RINGBUF_TYPE_DATA); | |
289 | /* If length is in len field, then array[0] has the data */ | |
290 | if (event->len) | |
291 | return (void *)&event->array[0]; | |
292 | /* Otherwise length is in array[0] and array[1] has the data */ | |
293 | return (void *)&event->array[1]; | |
294 | } | |
295 | ||
296 | /** | |
297 | * ring_buffer_event_data - return the data of the event | |
298 | * @event: the event to get the data from | |
299 | */ | |
300 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
301 | { | |
302 | return rb_event_data(event); | |
303 | } | |
c4f50183 | 304 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
305 | |
306 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 307 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
308 | |
309 | #define TS_SHIFT 27 | |
310 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
311 | #define TS_DELTA_TEST (~TS_MASK) | |
312 | ||
abc9b56d | 313 | struct buffer_data_page { |
e4c2ce82 | 314 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 315 | local_t commit; /* write committed index */ |
abc9b56d SR |
316 | unsigned char data[]; /* data of buffer page */ |
317 | }; | |
318 | ||
319 | struct buffer_page { | |
320 | local_t write; /* index for next write */ | |
6f807acd | 321 | unsigned read; /* index for next read */ |
e4c2ce82 | 322 | struct list_head list; /* list of free pages */ |
abc9b56d | 323 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
324 | }; |
325 | ||
044fa782 | 326 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 327 | { |
044fa782 | 328 | local_set(&bpage->commit, 0); |
abc9b56d SR |
329 | } |
330 | ||
474d32b6 SR |
331 | /** |
332 | * ring_buffer_page_len - the size of data on the page. | |
333 | * @page: The page to read | |
334 | * | |
335 | * Returns the amount of data on the page, including buffer page header. | |
336 | */ | |
ef7a4a16 SR |
337 | size_t ring_buffer_page_len(void *page) |
338 | { | |
474d32b6 SR |
339 | return local_read(&((struct buffer_data_page *)page)->commit) |
340 | + BUF_PAGE_HDR_SIZE; | |
ef7a4a16 SR |
341 | } |
342 | ||
ed56829c SR |
343 | /* |
344 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
345 | * this issue out. | |
346 | */ | |
34a148bf | 347 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 348 | { |
34a148bf | 349 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 350 | kfree(bpage); |
ed56829c SR |
351 | } |
352 | ||
7a8e76a3 SR |
353 | /* |
354 | * We need to fit the time_stamp delta into 27 bits. | |
355 | */ | |
356 | static inline int test_time_stamp(u64 delta) | |
357 | { | |
358 | if (delta & TS_DELTA_TEST) | |
359 | return 1; | |
360 | return 0; | |
361 | } | |
362 | ||
474d32b6 | 363 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 | 364 | |
d1b182a8 SR |
365 | int ring_buffer_print_page_header(struct trace_seq *s) |
366 | { | |
367 | struct buffer_data_page field; | |
368 | int ret; | |
369 | ||
370 | ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t" | |
371 | "offset:0;\tsize:%u;\n", | |
372 | (unsigned int)sizeof(field.time_stamp)); | |
373 | ||
374 | ret = trace_seq_printf(s, "\tfield: local_t commit;\t" | |
375 | "offset:%u;\tsize:%u;\n", | |
376 | (unsigned int)offsetof(typeof(field), commit), | |
377 | (unsigned int)sizeof(field.commit)); | |
378 | ||
379 | ret = trace_seq_printf(s, "\tfield: char data;\t" | |
380 | "offset:%u;\tsize:%u;\n", | |
381 | (unsigned int)offsetof(typeof(field), data), | |
382 | (unsigned int)BUF_PAGE_SIZE); | |
383 | ||
384 | return ret; | |
385 | } | |
386 | ||
7a8e76a3 SR |
387 | /* |
388 | * head_page == tail_page && head == tail then buffer is empty. | |
389 | */ | |
390 | struct ring_buffer_per_cpu { | |
391 | int cpu; | |
392 | struct ring_buffer *buffer; | |
f83c9d0f | 393 | spinlock_t reader_lock; /* serialize readers */ |
3e03fb7f | 394 | raw_spinlock_t lock; |
7a8e76a3 SR |
395 | struct lock_class_key lock_key; |
396 | struct list_head pages; | |
6f807acd SR |
397 | struct buffer_page *head_page; /* read from head */ |
398 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 399 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 400 | struct buffer_page *reader_page; |
7a8e76a3 SR |
401 | unsigned long overrun; |
402 | unsigned long entries; | |
403 | u64 write_stamp; | |
404 | u64 read_stamp; | |
405 | atomic_t record_disabled; | |
406 | }; | |
407 | ||
408 | struct ring_buffer { | |
7a8e76a3 SR |
409 | unsigned pages; |
410 | unsigned flags; | |
411 | int cpus; | |
7a8e76a3 | 412 | atomic_t record_disabled; |
00f62f61 | 413 | cpumask_var_t cpumask; |
7a8e76a3 SR |
414 | |
415 | struct mutex mutex; | |
416 | ||
417 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 418 | |
59222efe | 419 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
420 | struct notifier_block cpu_notify; |
421 | #endif | |
37886f6a | 422 | u64 (*clock)(void); |
7a8e76a3 SR |
423 | }; |
424 | ||
425 | struct ring_buffer_iter { | |
426 | struct ring_buffer_per_cpu *cpu_buffer; | |
427 | unsigned long head; | |
428 | struct buffer_page *head_page; | |
429 | u64 read_stamp; | |
430 | }; | |
431 | ||
f536aafc | 432 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
bf41a158 | 433 | #define RB_WARN_ON(buffer, cond) \ |
3e89c7bb SR |
434 | ({ \ |
435 | int _____ret = unlikely(cond); \ | |
436 | if (_____ret) { \ | |
bf41a158 SR |
437 | atomic_inc(&buffer->record_disabled); \ |
438 | WARN_ON(1); \ | |
439 | } \ | |
3e89c7bb SR |
440 | _____ret; \ |
441 | }) | |
f536aafc | 442 | |
37886f6a SR |
443 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
444 | #define DEBUG_SHIFT 0 | |
445 | ||
446 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) | |
447 | { | |
448 | u64 time; | |
449 | ||
450 | preempt_disable_notrace(); | |
451 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
452 | time = buffer->clock() << DEBUG_SHIFT; | |
453 | preempt_enable_no_resched_notrace(); | |
454 | ||
455 | return time; | |
456 | } | |
457 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
458 | ||
459 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
460 | int cpu, u64 *ts) | |
461 | { | |
462 | /* Just stupid testing the normalize function and deltas */ | |
463 | *ts >>= DEBUG_SHIFT; | |
464 | } | |
465 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
466 | ||
7a8e76a3 SR |
467 | /** |
468 | * check_pages - integrity check of buffer pages | |
469 | * @cpu_buffer: CPU buffer with pages to test | |
470 | * | |
c3706f00 | 471 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
472 | * been corrupted. |
473 | */ | |
474 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
475 | { | |
476 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 477 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 478 | |
3e89c7bb SR |
479 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
480 | return -1; | |
481 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
482 | return -1; | |
7a8e76a3 | 483 | |
044fa782 | 484 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 485 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 486 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
487 | return -1; |
488 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 489 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 490 | return -1; |
7a8e76a3 SR |
491 | } |
492 | ||
493 | return 0; | |
494 | } | |
495 | ||
7a8e76a3 SR |
496 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
497 | unsigned nr_pages) | |
498 | { | |
499 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 500 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
501 | unsigned long addr; |
502 | LIST_HEAD(pages); | |
503 | unsigned i; | |
504 | ||
505 | for (i = 0; i < nr_pages; i++) { | |
044fa782 | 506 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
aa1e0e3b | 507 | GFP_KERNEL, cpu_to_node(cpu_buffer->cpu)); |
044fa782 | 508 | if (!bpage) |
e4c2ce82 | 509 | goto free_pages; |
044fa782 | 510 | list_add(&bpage->list, &pages); |
e4c2ce82 | 511 | |
7a8e76a3 SR |
512 | addr = __get_free_page(GFP_KERNEL); |
513 | if (!addr) | |
514 | goto free_pages; | |
044fa782 SR |
515 | bpage->page = (void *)addr; |
516 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
517 | } |
518 | ||
519 | list_splice(&pages, head); | |
520 | ||
521 | rb_check_pages(cpu_buffer); | |
522 | ||
523 | return 0; | |
524 | ||
525 | free_pages: | |
044fa782 SR |
526 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
527 | list_del_init(&bpage->list); | |
528 | free_buffer_page(bpage); | |
7a8e76a3 SR |
529 | } |
530 | return -ENOMEM; | |
531 | } | |
532 | ||
533 | static struct ring_buffer_per_cpu * | |
534 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) | |
535 | { | |
536 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 537 | struct buffer_page *bpage; |
d769041f | 538 | unsigned long addr; |
7a8e76a3 SR |
539 | int ret; |
540 | ||
541 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
542 | GFP_KERNEL, cpu_to_node(cpu)); | |
543 | if (!cpu_buffer) | |
544 | return NULL; | |
545 | ||
546 | cpu_buffer->cpu = cpu; | |
547 | cpu_buffer->buffer = buffer; | |
f83c9d0f | 548 | spin_lock_init(&cpu_buffer->reader_lock); |
3e03fb7f | 549 | cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; |
7a8e76a3 SR |
550 | INIT_LIST_HEAD(&cpu_buffer->pages); |
551 | ||
044fa782 | 552 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 553 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 554 | if (!bpage) |
e4c2ce82 SR |
555 | goto fail_free_buffer; |
556 | ||
044fa782 | 557 | cpu_buffer->reader_page = bpage; |
d769041f SR |
558 | addr = __get_free_page(GFP_KERNEL); |
559 | if (!addr) | |
e4c2ce82 | 560 | goto fail_free_reader; |
044fa782 SR |
561 | bpage->page = (void *)addr; |
562 | rb_init_page(bpage->page); | |
e4c2ce82 | 563 | |
d769041f | 564 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 565 | |
7a8e76a3 SR |
566 | ret = rb_allocate_pages(cpu_buffer, buffer->pages); |
567 | if (ret < 0) | |
d769041f | 568 | goto fail_free_reader; |
7a8e76a3 SR |
569 | |
570 | cpu_buffer->head_page | |
571 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 572 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 SR |
573 | |
574 | return cpu_buffer; | |
575 | ||
d769041f SR |
576 | fail_free_reader: |
577 | free_buffer_page(cpu_buffer->reader_page); | |
578 | ||
7a8e76a3 SR |
579 | fail_free_buffer: |
580 | kfree(cpu_buffer); | |
581 | return NULL; | |
582 | } | |
583 | ||
584 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
585 | { | |
586 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 587 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 588 | |
d769041f SR |
589 | free_buffer_page(cpu_buffer->reader_page); |
590 | ||
044fa782 SR |
591 | list_for_each_entry_safe(bpage, tmp, head, list) { |
592 | list_del_init(&bpage->list); | |
593 | free_buffer_page(bpage); | |
7a8e76a3 SR |
594 | } |
595 | kfree(cpu_buffer); | |
596 | } | |
597 | ||
a7b13743 SR |
598 | /* |
599 | * Causes compile errors if the struct buffer_page gets bigger | |
600 | * than the struct page. | |
601 | */ | |
602 | extern int ring_buffer_page_too_big(void); | |
603 | ||
59222efe | 604 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
605 | static int rb_cpu_notify(struct notifier_block *self, |
606 | unsigned long action, void *hcpu); | |
554f786e SR |
607 | #endif |
608 | ||
7a8e76a3 SR |
609 | /** |
610 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 611 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
612 | * @flags: attributes to set for the ring buffer. |
613 | * | |
614 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
615 | * flag. This flag means that the buffer will overwrite old data | |
616 | * when the buffer wraps. If this flag is not set, the buffer will | |
617 | * drop data when the tail hits the head. | |
618 | */ | |
619 | struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags) | |
620 | { | |
621 | struct ring_buffer *buffer; | |
622 | int bsize; | |
623 | int cpu; | |
624 | ||
a7b13743 SR |
625 | /* Paranoid! Optimizes out when all is well */ |
626 | if (sizeof(struct buffer_page) > sizeof(struct page)) | |
627 | ring_buffer_page_too_big(); | |
628 | ||
629 | ||
7a8e76a3 SR |
630 | /* keep it in its own cache line */ |
631 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
632 | GFP_KERNEL); | |
633 | if (!buffer) | |
634 | return NULL; | |
635 | ||
9e01c1b7 RR |
636 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
637 | goto fail_free_buffer; | |
638 | ||
7a8e76a3 SR |
639 | buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
640 | buffer->flags = flags; | |
37886f6a | 641 | buffer->clock = trace_clock_local; |
7a8e76a3 SR |
642 | |
643 | /* need at least two pages */ | |
644 | if (buffer->pages == 1) | |
645 | buffer->pages++; | |
646 | ||
3bf832ce FW |
647 | /* |
648 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
649 | * in early initcall, it will not be notified of secondary cpus. | |
650 | * In that off case, we need to allocate for all possible cpus. | |
651 | */ | |
652 | #ifdef CONFIG_HOTPLUG_CPU | |
554f786e SR |
653 | get_online_cpus(); |
654 | cpumask_copy(buffer->cpumask, cpu_online_mask); | |
3bf832ce FW |
655 | #else |
656 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
657 | #endif | |
7a8e76a3 SR |
658 | buffer->cpus = nr_cpu_ids; |
659 | ||
660 | bsize = sizeof(void *) * nr_cpu_ids; | |
661 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
662 | GFP_KERNEL); | |
663 | if (!buffer->buffers) | |
9e01c1b7 | 664 | goto fail_free_cpumask; |
7a8e76a3 SR |
665 | |
666 | for_each_buffer_cpu(buffer, cpu) { | |
667 | buffer->buffers[cpu] = | |
668 | rb_allocate_cpu_buffer(buffer, cpu); | |
669 | if (!buffer->buffers[cpu]) | |
670 | goto fail_free_buffers; | |
671 | } | |
672 | ||
59222efe | 673 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
674 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
675 | buffer->cpu_notify.priority = 0; | |
676 | register_cpu_notifier(&buffer->cpu_notify); | |
677 | #endif | |
678 | ||
679 | put_online_cpus(); | |
7a8e76a3 SR |
680 | mutex_init(&buffer->mutex); |
681 | ||
682 | return buffer; | |
683 | ||
684 | fail_free_buffers: | |
685 | for_each_buffer_cpu(buffer, cpu) { | |
686 | if (buffer->buffers[cpu]) | |
687 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
688 | } | |
689 | kfree(buffer->buffers); | |
690 | ||
9e01c1b7 RR |
691 | fail_free_cpumask: |
692 | free_cpumask_var(buffer->cpumask); | |
554f786e | 693 | put_online_cpus(); |
9e01c1b7 | 694 | |
7a8e76a3 SR |
695 | fail_free_buffer: |
696 | kfree(buffer); | |
697 | return NULL; | |
698 | } | |
c4f50183 | 699 | EXPORT_SYMBOL_GPL(ring_buffer_alloc); |
7a8e76a3 SR |
700 | |
701 | /** | |
702 | * ring_buffer_free - free a ring buffer. | |
703 | * @buffer: the buffer to free. | |
704 | */ | |
705 | void | |
706 | ring_buffer_free(struct ring_buffer *buffer) | |
707 | { | |
708 | int cpu; | |
709 | ||
554f786e SR |
710 | get_online_cpus(); |
711 | ||
59222efe | 712 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
713 | unregister_cpu_notifier(&buffer->cpu_notify); |
714 | #endif | |
715 | ||
7a8e76a3 SR |
716 | for_each_buffer_cpu(buffer, cpu) |
717 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
718 | ||
554f786e SR |
719 | put_online_cpus(); |
720 | ||
9e01c1b7 RR |
721 | free_cpumask_var(buffer->cpumask); |
722 | ||
7a8e76a3 SR |
723 | kfree(buffer); |
724 | } | |
c4f50183 | 725 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 726 | |
37886f6a SR |
727 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
728 | u64 (*clock)(void)) | |
729 | { | |
730 | buffer->clock = clock; | |
731 | } | |
732 | ||
7a8e76a3 SR |
733 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
734 | ||
735 | static void | |
736 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages) | |
737 | { | |
044fa782 | 738 | struct buffer_page *bpage; |
7a8e76a3 SR |
739 | struct list_head *p; |
740 | unsigned i; | |
741 | ||
742 | atomic_inc(&cpu_buffer->record_disabled); | |
743 | synchronize_sched(); | |
744 | ||
745 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
746 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
747 | return; | |
7a8e76a3 | 748 | p = cpu_buffer->pages.next; |
044fa782 SR |
749 | bpage = list_entry(p, struct buffer_page, list); |
750 | list_del_init(&bpage->list); | |
751 | free_buffer_page(bpage); | |
7a8e76a3 | 752 | } |
3e89c7bb SR |
753 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
754 | return; | |
7a8e76a3 SR |
755 | |
756 | rb_reset_cpu(cpu_buffer); | |
757 | ||
758 | rb_check_pages(cpu_buffer); | |
759 | ||
760 | atomic_dec(&cpu_buffer->record_disabled); | |
761 | ||
762 | } | |
763 | ||
764 | static void | |
765 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
766 | struct list_head *pages, unsigned nr_pages) | |
767 | { | |
044fa782 | 768 | struct buffer_page *bpage; |
7a8e76a3 SR |
769 | struct list_head *p; |
770 | unsigned i; | |
771 | ||
772 | atomic_inc(&cpu_buffer->record_disabled); | |
773 | synchronize_sched(); | |
774 | ||
775 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
776 | if (RB_WARN_ON(cpu_buffer, list_empty(pages))) |
777 | return; | |
7a8e76a3 | 778 | p = pages->next; |
044fa782 SR |
779 | bpage = list_entry(p, struct buffer_page, list); |
780 | list_del_init(&bpage->list); | |
781 | list_add_tail(&bpage->list, &cpu_buffer->pages); | |
7a8e76a3 SR |
782 | } |
783 | rb_reset_cpu(cpu_buffer); | |
784 | ||
785 | rb_check_pages(cpu_buffer); | |
786 | ||
787 | atomic_dec(&cpu_buffer->record_disabled); | |
788 | } | |
789 | ||
790 | /** | |
791 | * ring_buffer_resize - resize the ring buffer | |
792 | * @buffer: the buffer to resize. | |
793 | * @size: the new size. | |
794 | * | |
795 | * The tracer is responsible for making sure that the buffer is | |
796 | * not being used while changing the size. | |
797 | * Note: We may be able to change the above requirement by using | |
798 | * RCU synchronizations. | |
799 | * | |
800 | * Minimum size is 2 * BUF_PAGE_SIZE. | |
801 | * | |
802 | * Returns -1 on failure. | |
803 | */ | |
804 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) | |
805 | { | |
806 | struct ring_buffer_per_cpu *cpu_buffer; | |
807 | unsigned nr_pages, rm_pages, new_pages; | |
044fa782 | 808 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
809 | unsigned long buffer_size; |
810 | unsigned long addr; | |
811 | LIST_HEAD(pages); | |
812 | int i, cpu; | |
813 | ||
ee51a1de IM |
814 | /* |
815 | * Always succeed at resizing a non-existent buffer: | |
816 | */ | |
817 | if (!buffer) | |
818 | return size; | |
819 | ||
7a8e76a3 SR |
820 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
821 | size *= BUF_PAGE_SIZE; | |
822 | buffer_size = buffer->pages * BUF_PAGE_SIZE; | |
823 | ||
824 | /* we need a minimum of two pages */ | |
825 | if (size < BUF_PAGE_SIZE * 2) | |
826 | size = BUF_PAGE_SIZE * 2; | |
827 | ||
828 | if (size == buffer_size) | |
829 | return size; | |
830 | ||
831 | mutex_lock(&buffer->mutex); | |
554f786e | 832 | get_online_cpus(); |
7a8e76a3 SR |
833 | |
834 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
835 | ||
836 | if (size < buffer_size) { | |
837 | ||
838 | /* easy case, just free pages */ | |
554f786e SR |
839 | if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) |
840 | goto out_fail; | |
7a8e76a3 SR |
841 | |
842 | rm_pages = buffer->pages - nr_pages; | |
843 | ||
844 | for_each_buffer_cpu(buffer, cpu) { | |
845 | cpu_buffer = buffer->buffers[cpu]; | |
846 | rb_remove_pages(cpu_buffer, rm_pages); | |
847 | } | |
848 | goto out; | |
849 | } | |
850 | ||
851 | /* | |
852 | * This is a bit more difficult. We only want to add pages | |
853 | * when we can allocate enough for all CPUs. We do this | |
854 | * by allocating all the pages and storing them on a local | |
855 | * link list. If we succeed in our allocation, then we | |
856 | * add these pages to the cpu_buffers. Otherwise we just free | |
857 | * them all and return -ENOMEM; | |
858 | */ | |
554f786e SR |
859 | if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) |
860 | goto out_fail; | |
f536aafc | 861 | |
7a8e76a3 SR |
862 | new_pages = nr_pages - buffer->pages; |
863 | ||
864 | for_each_buffer_cpu(buffer, cpu) { | |
865 | for (i = 0; i < new_pages; i++) { | |
044fa782 | 866 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), |
e4c2ce82 SR |
867 | cache_line_size()), |
868 | GFP_KERNEL, cpu_to_node(cpu)); | |
044fa782 | 869 | if (!bpage) |
e4c2ce82 | 870 | goto free_pages; |
044fa782 | 871 | list_add(&bpage->list, &pages); |
7a8e76a3 SR |
872 | addr = __get_free_page(GFP_KERNEL); |
873 | if (!addr) | |
874 | goto free_pages; | |
044fa782 SR |
875 | bpage->page = (void *)addr; |
876 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
877 | } |
878 | } | |
879 | ||
880 | for_each_buffer_cpu(buffer, cpu) { | |
881 | cpu_buffer = buffer->buffers[cpu]; | |
882 | rb_insert_pages(cpu_buffer, &pages, new_pages); | |
883 | } | |
884 | ||
554f786e SR |
885 | if (RB_WARN_ON(buffer, !list_empty(&pages))) |
886 | goto out_fail; | |
7a8e76a3 SR |
887 | |
888 | out: | |
889 | buffer->pages = nr_pages; | |
554f786e | 890 | put_online_cpus(); |
7a8e76a3 SR |
891 | mutex_unlock(&buffer->mutex); |
892 | ||
893 | return size; | |
894 | ||
895 | free_pages: | |
044fa782 SR |
896 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
897 | list_del_init(&bpage->list); | |
898 | free_buffer_page(bpage); | |
7a8e76a3 | 899 | } |
554f786e | 900 | put_online_cpus(); |
641d2f63 | 901 | mutex_unlock(&buffer->mutex); |
7a8e76a3 | 902 | return -ENOMEM; |
554f786e SR |
903 | |
904 | /* | |
905 | * Something went totally wrong, and we are too paranoid | |
906 | * to even clean up the mess. | |
907 | */ | |
908 | out_fail: | |
909 | put_online_cpus(); | |
910 | mutex_unlock(&buffer->mutex); | |
911 | return -1; | |
7a8e76a3 | 912 | } |
c4f50183 | 913 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 914 | |
8789a9e7 | 915 | static inline void * |
044fa782 | 916 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 917 | { |
044fa782 | 918 | return bpage->data + index; |
8789a9e7 SR |
919 | } |
920 | ||
044fa782 | 921 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 922 | { |
044fa782 | 923 | return bpage->page->data + index; |
7a8e76a3 SR |
924 | } |
925 | ||
926 | static inline struct ring_buffer_event * | |
d769041f | 927 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 928 | { |
6f807acd SR |
929 | return __rb_page_index(cpu_buffer->reader_page, |
930 | cpu_buffer->reader_page->read); | |
931 | } | |
932 | ||
933 | static inline struct ring_buffer_event * | |
934 | rb_head_event(struct ring_buffer_per_cpu *cpu_buffer) | |
935 | { | |
936 | return __rb_page_index(cpu_buffer->head_page, | |
937 | cpu_buffer->head_page->read); | |
7a8e76a3 SR |
938 | } |
939 | ||
940 | static inline struct ring_buffer_event * | |
941 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
942 | { | |
6f807acd | 943 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
944 | } |
945 | ||
bf41a158 SR |
946 | static inline unsigned rb_page_write(struct buffer_page *bpage) |
947 | { | |
948 | return local_read(&bpage->write); | |
949 | } | |
950 | ||
951 | static inline unsigned rb_page_commit(struct buffer_page *bpage) | |
952 | { | |
abc9b56d | 953 | return local_read(&bpage->page->commit); |
bf41a158 SR |
954 | } |
955 | ||
956 | /* Size is determined by what has been commited */ | |
957 | static inline unsigned rb_page_size(struct buffer_page *bpage) | |
958 | { | |
959 | return rb_page_commit(bpage); | |
960 | } | |
961 | ||
962 | static inline unsigned | |
963 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
964 | { | |
965 | return rb_page_commit(cpu_buffer->commit_page); | |
966 | } | |
967 | ||
968 | static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer) | |
969 | { | |
970 | return rb_page_commit(cpu_buffer->head_page); | |
971 | } | |
972 | ||
7a8e76a3 SR |
973 | /* |
974 | * When the tail hits the head and the buffer is in overwrite mode, | |
975 | * the head jumps to the next page and all content on the previous | |
976 | * page is discarded. But before doing so, we update the overrun | |
977 | * variable of the buffer. | |
978 | */ | |
979 | static void rb_update_overflow(struct ring_buffer_per_cpu *cpu_buffer) | |
980 | { | |
981 | struct ring_buffer_event *event; | |
982 | unsigned long head; | |
983 | ||
984 | for (head = 0; head < rb_head_size(cpu_buffer); | |
985 | head += rb_event_length(event)) { | |
986 | ||
6f807acd | 987 | event = __rb_page_index(cpu_buffer->head_page, head); |
3e89c7bb SR |
988 | if (RB_WARN_ON(cpu_buffer, rb_null_event(event))) |
989 | return; | |
7a8e76a3 SR |
990 | /* Only count data entries */ |
991 | if (event->type != RINGBUF_TYPE_DATA) | |
992 | continue; | |
993 | cpu_buffer->overrun++; | |
994 | cpu_buffer->entries--; | |
995 | } | |
996 | } | |
997 | ||
998 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
044fa782 | 999 | struct buffer_page **bpage) |
7a8e76a3 | 1000 | { |
044fa782 | 1001 | struct list_head *p = (*bpage)->list.next; |
7a8e76a3 SR |
1002 | |
1003 | if (p == &cpu_buffer->pages) | |
1004 | p = p->next; | |
1005 | ||
044fa782 | 1006 | *bpage = list_entry(p, struct buffer_page, list); |
7a8e76a3 SR |
1007 | } |
1008 | ||
bf41a158 SR |
1009 | static inline unsigned |
1010 | rb_event_index(struct ring_buffer_event *event) | |
1011 | { | |
1012 | unsigned long addr = (unsigned long)event; | |
1013 | ||
1014 | return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE); | |
1015 | } | |
1016 | ||
34a148bf | 1017 | static int |
bf41a158 SR |
1018 | rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1019 | struct ring_buffer_event *event) | |
1020 | { | |
1021 | unsigned long addr = (unsigned long)event; | |
1022 | unsigned long index; | |
1023 | ||
1024 | index = rb_event_index(event); | |
1025 | addr &= PAGE_MASK; | |
1026 | ||
1027 | return cpu_buffer->commit_page->page == (void *)addr && | |
1028 | rb_commit_index(cpu_buffer) == index; | |
1029 | } | |
1030 | ||
34a148bf | 1031 | static void |
bf41a158 SR |
1032 | rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer, |
1033 | struct ring_buffer_event *event) | |
7a8e76a3 | 1034 | { |
bf41a158 SR |
1035 | unsigned long addr = (unsigned long)event; |
1036 | unsigned long index; | |
1037 | ||
1038 | index = rb_event_index(event); | |
1039 | addr &= PAGE_MASK; | |
1040 | ||
1041 | while (cpu_buffer->commit_page->page != (void *)addr) { | |
3e89c7bb SR |
1042 | if (RB_WARN_ON(cpu_buffer, |
1043 | cpu_buffer->commit_page == cpu_buffer->tail_page)) | |
1044 | return; | |
abc9b56d | 1045 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1046 | cpu_buffer->commit_page->write; |
1047 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
1048 | cpu_buffer->write_stamp = |
1049 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1050 | } |
1051 | ||
1052 | /* Now set the commit to the event's index */ | |
abc9b56d | 1053 | local_set(&cpu_buffer->commit_page->page->commit, index); |
7a8e76a3 SR |
1054 | } |
1055 | ||
34a148bf | 1056 | static void |
bf41a158 | 1057 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1058 | { |
bf41a158 SR |
1059 | /* |
1060 | * We only race with interrupts and NMIs on this CPU. | |
1061 | * If we own the commit event, then we can commit | |
1062 | * all others that interrupted us, since the interruptions | |
1063 | * are in stack format (they finish before they come | |
1064 | * back to us). This allows us to do a simple loop to | |
1065 | * assign the commit to the tail. | |
1066 | */ | |
a8ccf1d6 | 1067 | again: |
bf41a158 | 1068 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
abc9b56d | 1069 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1070 | cpu_buffer->commit_page->write; |
1071 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
1072 | cpu_buffer->write_stamp = |
1073 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1074 | /* add barrier to keep gcc from optimizing too much */ |
1075 | barrier(); | |
1076 | } | |
1077 | while (rb_commit_index(cpu_buffer) != | |
1078 | rb_page_write(cpu_buffer->commit_page)) { | |
abc9b56d | 1079 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1080 | cpu_buffer->commit_page->write; |
1081 | barrier(); | |
1082 | } | |
a8ccf1d6 SR |
1083 | |
1084 | /* again, keep gcc from optimizing */ | |
1085 | barrier(); | |
1086 | ||
1087 | /* | |
1088 | * If an interrupt came in just after the first while loop | |
1089 | * and pushed the tail page forward, we will be left with | |
1090 | * a dangling commit that will never go forward. | |
1091 | */ | |
1092 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1093 | goto again; | |
7a8e76a3 SR |
1094 | } |
1095 | ||
d769041f | 1096 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1097 | { |
abc9b56d | 1098 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1099 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1100 | } |
1101 | ||
34a148bf | 1102 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1103 | { |
1104 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1105 | ||
1106 | /* | |
1107 | * The iterator could be on the reader page (it starts there). | |
1108 | * But the head could have moved, since the reader was | |
1109 | * found. Check for this case and assign the iterator | |
1110 | * to the head page instead of next. | |
1111 | */ | |
1112 | if (iter->head_page == cpu_buffer->reader_page) | |
1113 | iter->head_page = cpu_buffer->head_page; | |
1114 | else | |
1115 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1116 | ||
abc9b56d | 1117 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1118 | iter->head = 0; |
1119 | } | |
1120 | ||
1121 | /** | |
1122 | * ring_buffer_update_event - update event type and data | |
1123 | * @event: the even to update | |
1124 | * @type: the type of event | |
1125 | * @length: the size of the event field in the ring buffer | |
1126 | * | |
1127 | * Update the type and data fields of the event. The length | |
1128 | * is the actual size that is written to the ring buffer, | |
1129 | * and with this, we can determine what to place into the | |
1130 | * data field. | |
1131 | */ | |
34a148bf | 1132 | static void |
7a8e76a3 SR |
1133 | rb_update_event(struct ring_buffer_event *event, |
1134 | unsigned type, unsigned length) | |
1135 | { | |
1136 | event->type = type; | |
1137 | ||
1138 | switch (type) { | |
1139 | ||
1140 | case RINGBUF_TYPE_PADDING: | |
1141 | break; | |
1142 | ||
1143 | case RINGBUF_TYPE_TIME_EXTEND: | |
67d34724 | 1144 | event->len = DIV_ROUND_UP(RB_LEN_TIME_EXTEND, RB_ALIGNMENT); |
7a8e76a3 SR |
1145 | break; |
1146 | ||
1147 | case RINGBUF_TYPE_TIME_STAMP: | |
67d34724 | 1148 | event->len = DIV_ROUND_UP(RB_LEN_TIME_STAMP, RB_ALIGNMENT); |
7a8e76a3 SR |
1149 | break; |
1150 | ||
1151 | case RINGBUF_TYPE_DATA: | |
1152 | length -= RB_EVNT_HDR_SIZE; | |
1153 | if (length > RB_MAX_SMALL_DATA) { | |
1154 | event->len = 0; | |
1155 | event->array[0] = length; | |
1156 | } else | |
67d34724 | 1157 | event->len = DIV_ROUND_UP(length, RB_ALIGNMENT); |
7a8e76a3 SR |
1158 | break; |
1159 | default: | |
1160 | BUG(); | |
1161 | } | |
1162 | } | |
1163 | ||
34a148bf | 1164 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
1165 | { |
1166 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
1167 | ||
1168 | /* zero length can cause confusions */ | |
1169 | if (!length) | |
1170 | length = 1; | |
1171 | ||
1172 | if (length > RB_MAX_SMALL_DATA) | |
1173 | length += sizeof(event.array[0]); | |
1174 | ||
1175 | length += RB_EVNT_HDR_SIZE; | |
1176 | length = ALIGN(length, RB_ALIGNMENT); | |
1177 | ||
1178 | return length; | |
1179 | } | |
1180 | ||
1181 | static struct ring_buffer_event * | |
1182 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
1183 | unsigned type, unsigned long length, u64 *ts) | |
1184 | { | |
98db8df7 | 1185 | struct buffer_page *tail_page, *head_page, *reader_page, *commit_page; |
bf41a158 | 1186 | unsigned long tail, write; |
7a8e76a3 SR |
1187 | struct ring_buffer *buffer = cpu_buffer->buffer; |
1188 | struct ring_buffer_event *event; | |
bf41a158 | 1189 | unsigned long flags; |
78d904b4 | 1190 | bool lock_taken = false; |
7a8e76a3 | 1191 | |
98db8df7 SR |
1192 | commit_page = cpu_buffer->commit_page; |
1193 | /* we just need to protect against interrupts */ | |
1194 | barrier(); | |
7a8e76a3 | 1195 | tail_page = cpu_buffer->tail_page; |
bf41a158 SR |
1196 | write = local_add_return(length, &tail_page->write); |
1197 | tail = write - length; | |
7a8e76a3 | 1198 | |
bf41a158 SR |
1199 | /* See if we shot pass the end of this buffer page */ |
1200 | if (write > BUF_PAGE_SIZE) { | |
7a8e76a3 SR |
1201 | struct buffer_page *next_page = tail_page; |
1202 | ||
3e03fb7f | 1203 | local_irq_save(flags); |
78d904b4 | 1204 | /* |
a81bd80a SR |
1205 | * Since the write to the buffer is still not |
1206 | * fully lockless, we must be careful with NMIs. | |
1207 | * The locks in the writers are taken when a write | |
1208 | * crosses to a new page. The locks protect against | |
1209 | * races with the readers (this will soon be fixed | |
1210 | * with a lockless solution). | |
1211 | * | |
1212 | * Because we can not protect against NMIs, and we | |
1213 | * want to keep traces reentrant, we need to manage | |
1214 | * what happens when we are in an NMI. | |
1215 | * | |
78d904b4 SR |
1216 | * NMIs can happen after we take the lock. |
1217 | * If we are in an NMI, only take the lock | |
1218 | * if it is not already taken. Otherwise | |
1219 | * simply fail. | |
1220 | */ | |
a81bd80a | 1221 | if (unlikely(in_nmi())) { |
78d904b4 | 1222 | if (!__raw_spin_trylock(&cpu_buffer->lock)) |
45141d46 | 1223 | goto out_reset; |
78d904b4 SR |
1224 | } else |
1225 | __raw_spin_lock(&cpu_buffer->lock); | |
1226 | ||
1227 | lock_taken = true; | |
bf41a158 | 1228 | |
7a8e76a3 SR |
1229 | rb_inc_page(cpu_buffer, &next_page); |
1230 | ||
d769041f SR |
1231 | head_page = cpu_buffer->head_page; |
1232 | reader_page = cpu_buffer->reader_page; | |
1233 | ||
1234 | /* we grabbed the lock before incrementing */ | |
3e89c7bb | 1235 | if (RB_WARN_ON(cpu_buffer, next_page == reader_page)) |
45141d46 | 1236 | goto out_reset; |
bf41a158 SR |
1237 | |
1238 | /* | |
1239 | * If for some reason, we had an interrupt storm that made | |
1240 | * it all the way around the buffer, bail, and warn | |
1241 | * about it. | |
1242 | */ | |
98db8df7 | 1243 | if (unlikely(next_page == commit_page)) { |
bf41a158 | 1244 | WARN_ON_ONCE(1); |
45141d46 | 1245 | goto out_reset; |
bf41a158 | 1246 | } |
d769041f | 1247 | |
7a8e76a3 | 1248 | if (next_page == head_page) { |
6f3b3440 | 1249 | if (!(buffer->flags & RB_FL_OVERWRITE)) |
45141d46 | 1250 | goto out_reset; |
7a8e76a3 | 1251 | |
bf41a158 SR |
1252 | /* tail_page has not moved yet? */ |
1253 | if (tail_page == cpu_buffer->tail_page) { | |
1254 | /* count overflows */ | |
1255 | rb_update_overflow(cpu_buffer); | |
1256 | ||
1257 | rb_inc_page(cpu_buffer, &head_page); | |
1258 | cpu_buffer->head_page = head_page; | |
1259 | cpu_buffer->head_page->read = 0; | |
1260 | } | |
1261 | } | |
7a8e76a3 | 1262 | |
bf41a158 SR |
1263 | /* |
1264 | * If the tail page is still the same as what we think | |
1265 | * it is, then it is up to us to update the tail | |
1266 | * pointer. | |
1267 | */ | |
1268 | if (tail_page == cpu_buffer->tail_page) { | |
1269 | local_set(&next_page->write, 0); | |
abc9b56d | 1270 | local_set(&next_page->page->commit, 0); |
bf41a158 SR |
1271 | cpu_buffer->tail_page = next_page; |
1272 | ||
1273 | /* reread the time stamp */ | |
37886f6a | 1274 | *ts = ring_buffer_time_stamp(buffer, cpu_buffer->cpu); |
abc9b56d | 1275 | cpu_buffer->tail_page->page->time_stamp = *ts; |
7a8e76a3 SR |
1276 | } |
1277 | ||
bf41a158 SR |
1278 | /* |
1279 | * The actual tail page has moved forward. | |
1280 | */ | |
1281 | if (tail < BUF_PAGE_SIZE) { | |
1282 | /* Mark the rest of the page with padding */ | |
6f807acd | 1283 | event = __rb_page_index(tail_page, tail); |
2d622719 | 1284 | rb_event_set_padding(event); |
7a8e76a3 SR |
1285 | } |
1286 | ||
bf41a158 SR |
1287 | if (tail <= BUF_PAGE_SIZE) |
1288 | /* Set the write back to the previous setting */ | |
1289 | local_set(&tail_page->write, tail); | |
1290 | ||
1291 | /* | |
1292 | * If this was a commit entry that failed, | |
1293 | * increment that too | |
1294 | */ | |
1295 | if (tail_page == cpu_buffer->commit_page && | |
1296 | tail == rb_commit_index(cpu_buffer)) { | |
1297 | rb_set_commit_to_write(cpu_buffer); | |
1298 | } | |
1299 | ||
3e03fb7f SR |
1300 | __raw_spin_unlock(&cpu_buffer->lock); |
1301 | local_irq_restore(flags); | |
bf41a158 SR |
1302 | |
1303 | /* fail and let the caller try again */ | |
1304 | return ERR_PTR(-EAGAIN); | |
7a8e76a3 SR |
1305 | } |
1306 | ||
bf41a158 SR |
1307 | /* We reserved something on the buffer */ |
1308 | ||
3e89c7bb SR |
1309 | if (RB_WARN_ON(cpu_buffer, write > BUF_PAGE_SIZE)) |
1310 | return NULL; | |
7a8e76a3 | 1311 | |
6f807acd | 1312 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
1313 | rb_update_event(event, type, length); |
1314 | ||
bf41a158 SR |
1315 | /* |
1316 | * If this is a commit and the tail is zero, then update | |
1317 | * this page's time stamp. | |
1318 | */ | |
1319 | if (!tail && rb_is_commit(cpu_buffer, event)) | |
abc9b56d | 1320 | cpu_buffer->commit_page->page->time_stamp = *ts; |
bf41a158 | 1321 | |
7a8e76a3 | 1322 | return event; |
bf41a158 | 1323 | |
45141d46 | 1324 | out_reset: |
6f3b3440 LJ |
1325 | /* reset write */ |
1326 | if (tail <= BUF_PAGE_SIZE) | |
1327 | local_set(&tail_page->write, tail); | |
1328 | ||
78d904b4 SR |
1329 | if (likely(lock_taken)) |
1330 | __raw_spin_unlock(&cpu_buffer->lock); | |
3e03fb7f | 1331 | local_irq_restore(flags); |
bf41a158 | 1332 | return NULL; |
7a8e76a3 SR |
1333 | } |
1334 | ||
1335 | static int | |
1336 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1337 | u64 *ts, u64 *delta) | |
1338 | { | |
1339 | struct ring_buffer_event *event; | |
1340 | static int once; | |
bf41a158 | 1341 | int ret; |
7a8e76a3 SR |
1342 | |
1343 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1344 | printk(KERN_WARNING "Delta way too big! %llu" | |
1345 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1346 | (unsigned long long)*delta, |
1347 | (unsigned long long)*ts, | |
1348 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1349 | WARN_ON(1); |
1350 | } | |
1351 | ||
1352 | /* | |
1353 | * The delta is too big, we to add a | |
1354 | * new timestamp. | |
1355 | */ | |
1356 | event = __rb_reserve_next(cpu_buffer, | |
1357 | RINGBUF_TYPE_TIME_EXTEND, | |
1358 | RB_LEN_TIME_EXTEND, | |
1359 | ts); | |
1360 | if (!event) | |
bf41a158 | 1361 | return -EBUSY; |
7a8e76a3 | 1362 | |
bf41a158 SR |
1363 | if (PTR_ERR(event) == -EAGAIN) |
1364 | return -EAGAIN; | |
1365 | ||
1366 | /* Only a commited time event can update the write stamp */ | |
1367 | if (rb_is_commit(cpu_buffer, event)) { | |
1368 | /* | |
1369 | * If this is the first on the page, then we need to | |
1370 | * update the page itself, and just put in a zero. | |
1371 | */ | |
1372 | if (rb_event_index(event)) { | |
1373 | event->time_delta = *delta & TS_MASK; | |
1374 | event->array[0] = *delta >> TS_SHIFT; | |
1375 | } else { | |
abc9b56d | 1376 | cpu_buffer->commit_page->page->time_stamp = *ts; |
bf41a158 SR |
1377 | event->time_delta = 0; |
1378 | event->array[0] = 0; | |
1379 | } | |
7a8e76a3 | 1380 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
1381 | /* let the caller know this was the commit */ |
1382 | ret = 1; | |
1383 | } else { | |
1384 | /* Darn, this is just wasted space */ | |
1385 | event->time_delta = 0; | |
1386 | event->array[0] = 0; | |
1387 | ret = 0; | |
7a8e76a3 SR |
1388 | } |
1389 | ||
bf41a158 SR |
1390 | *delta = 0; |
1391 | ||
1392 | return ret; | |
7a8e76a3 SR |
1393 | } |
1394 | ||
1395 | static struct ring_buffer_event * | |
1396 | rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, | |
1397 | unsigned type, unsigned long length) | |
1398 | { | |
1399 | struct ring_buffer_event *event; | |
1400 | u64 ts, delta; | |
bf41a158 | 1401 | int commit = 0; |
818e3dd3 | 1402 | int nr_loops = 0; |
7a8e76a3 | 1403 | |
bf41a158 | 1404 | again: |
818e3dd3 SR |
1405 | /* |
1406 | * We allow for interrupts to reenter here and do a trace. | |
1407 | * If one does, it will cause this original code to loop | |
1408 | * back here. Even with heavy interrupts happening, this | |
1409 | * should only happen a few times in a row. If this happens | |
1410 | * 1000 times in a row, there must be either an interrupt | |
1411 | * storm or we have something buggy. | |
1412 | * Bail! | |
1413 | */ | |
3e89c7bb | 1414 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
818e3dd3 | 1415 | return NULL; |
818e3dd3 | 1416 | |
37886f6a | 1417 | ts = ring_buffer_time_stamp(cpu_buffer->buffer, cpu_buffer->cpu); |
7a8e76a3 | 1418 | |
bf41a158 SR |
1419 | /* |
1420 | * Only the first commit can update the timestamp. | |
1421 | * Yes there is a race here. If an interrupt comes in | |
1422 | * just after the conditional and it traces too, then it | |
1423 | * will also check the deltas. More than one timestamp may | |
1424 | * also be made. But only the entry that did the actual | |
1425 | * commit will be something other than zero. | |
1426 | */ | |
1427 | if (cpu_buffer->tail_page == cpu_buffer->commit_page && | |
1428 | rb_page_write(cpu_buffer->tail_page) == | |
1429 | rb_commit_index(cpu_buffer)) { | |
1430 | ||
7a8e76a3 SR |
1431 | delta = ts - cpu_buffer->write_stamp; |
1432 | ||
bf41a158 SR |
1433 | /* make sure this delta is calculated here */ |
1434 | barrier(); | |
1435 | ||
1436 | /* Did the write stamp get updated already? */ | |
1437 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
4143c5cb | 1438 | delta = 0; |
bf41a158 | 1439 | |
7a8e76a3 | 1440 | if (test_time_stamp(delta)) { |
7a8e76a3 | 1441 | |
bf41a158 SR |
1442 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
1443 | ||
1444 | if (commit == -EBUSY) | |
7a8e76a3 | 1445 | return NULL; |
bf41a158 SR |
1446 | |
1447 | if (commit == -EAGAIN) | |
1448 | goto again; | |
1449 | ||
1450 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 1451 | } |
bf41a158 SR |
1452 | } else |
1453 | /* Non commits have zero deltas */ | |
7a8e76a3 | 1454 | delta = 0; |
7a8e76a3 SR |
1455 | |
1456 | event = __rb_reserve_next(cpu_buffer, type, length, &ts); | |
bf41a158 SR |
1457 | if (PTR_ERR(event) == -EAGAIN) |
1458 | goto again; | |
1459 | ||
1460 | if (!event) { | |
1461 | if (unlikely(commit)) | |
1462 | /* | |
1463 | * Ouch! We needed a timestamp and it was commited. But | |
1464 | * we didn't get our event reserved. | |
1465 | */ | |
1466 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 | 1467 | return NULL; |
bf41a158 | 1468 | } |
7a8e76a3 | 1469 | |
bf41a158 SR |
1470 | /* |
1471 | * If the timestamp was commited, make the commit our entry | |
1472 | * now so that we will update it when needed. | |
1473 | */ | |
1474 | if (commit) | |
1475 | rb_set_commit_event(cpu_buffer, event); | |
1476 | else if (!rb_is_commit(cpu_buffer, event)) | |
7a8e76a3 SR |
1477 | delta = 0; |
1478 | ||
1479 | event->time_delta = delta; | |
1480 | ||
1481 | return event; | |
1482 | } | |
1483 | ||
bf41a158 SR |
1484 | static DEFINE_PER_CPU(int, rb_need_resched); |
1485 | ||
7a8e76a3 SR |
1486 | /** |
1487 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
1488 | * @buffer: the ring buffer to reserve from | |
1489 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
1490 | * |
1491 | * Returns a reseverd event on the ring buffer to copy directly to. | |
1492 | * The user of this interface will need to get the body to write into | |
1493 | * and can use the ring_buffer_event_data() interface. | |
1494 | * | |
1495 | * The length is the length of the data needed, not the event length | |
1496 | * which also includes the event header. | |
1497 | * | |
1498 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
1499 | * If NULL is returned, then nothing has been allocated or locked. | |
1500 | */ | |
1501 | struct ring_buffer_event * | |
0a987751 | 1502 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
1503 | { |
1504 | struct ring_buffer_per_cpu *cpu_buffer; | |
1505 | struct ring_buffer_event *event; | |
bf41a158 | 1506 | int cpu, resched; |
7a8e76a3 | 1507 | |
033601a3 | 1508 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1509 | return NULL; |
1510 | ||
7a8e76a3 SR |
1511 | if (atomic_read(&buffer->record_disabled)) |
1512 | return NULL; | |
1513 | ||
bf41a158 | 1514 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 1515 | resched = ftrace_preempt_disable(); |
bf41a158 | 1516 | |
7a8e76a3 SR |
1517 | cpu = raw_smp_processor_id(); |
1518 | ||
9e01c1b7 | 1519 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1520 | goto out; |
7a8e76a3 SR |
1521 | |
1522 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1523 | |
1524 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 1525 | goto out; |
7a8e76a3 SR |
1526 | |
1527 | length = rb_calculate_event_length(length); | |
1528 | if (length > BUF_PAGE_SIZE) | |
bf41a158 | 1529 | goto out; |
7a8e76a3 SR |
1530 | |
1531 | event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length); | |
1532 | if (!event) | |
d769041f | 1533 | goto out; |
7a8e76a3 | 1534 | |
bf41a158 SR |
1535 | /* |
1536 | * Need to store resched state on this cpu. | |
1537 | * Only the first needs to. | |
1538 | */ | |
1539 | ||
1540 | if (preempt_count() == 1) | |
1541 | per_cpu(rb_need_resched, cpu) = resched; | |
1542 | ||
7a8e76a3 SR |
1543 | return event; |
1544 | ||
d769041f | 1545 | out: |
182e9f5f | 1546 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1547 | return NULL; |
1548 | } | |
c4f50183 | 1549 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 SR |
1550 | |
1551 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
1552 | struct ring_buffer_event *event) | |
1553 | { | |
7a8e76a3 | 1554 | cpu_buffer->entries++; |
bf41a158 SR |
1555 | |
1556 | /* Only process further if we own the commit */ | |
1557 | if (!rb_is_commit(cpu_buffer, event)) | |
1558 | return; | |
1559 | ||
1560 | cpu_buffer->write_stamp += event->time_delta; | |
1561 | ||
1562 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 SR |
1563 | } |
1564 | ||
1565 | /** | |
1566 | * ring_buffer_unlock_commit - commit a reserved | |
1567 | * @buffer: The buffer to commit to | |
1568 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
1569 | * |
1570 | * This commits the data to the ring buffer, and releases any locks held. | |
1571 | * | |
1572 | * Must be paired with ring_buffer_lock_reserve. | |
1573 | */ | |
1574 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 1575 | struct ring_buffer_event *event) |
7a8e76a3 SR |
1576 | { |
1577 | struct ring_buffer_per_cpu *cpu_buffer; | |
1578 | int cpu = raw_smp_processor_id(); | |
1579 | ||
1580 | cpu_buffer = buffer->buffers[cpu]; | |
1581 | ||
7a8e76a3 SR |
1582 | rb_commit(cpu_buffer, event); |
1583 | ||
bf41a158 SR |
1584 | /* |
1585 | * Only the last preempt count needs to restore preemption. | |
1586 | */ | |
182e9f5f SR |
1587 | if (preempt_count() == 1) |
1588 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1589 | else | |
bf41a158 | 1590 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
1591 | |
1592 | return 0; | |
1593 | } | |
c4f50183 | 1594 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 1595 | |
fa1b47dd SR |
1596 | /** |
1597 | * ring_buffer_event_discard - discard any event in the ring buffer | |
1598 | * @event: the event to discard | |
1599 | * | |
1600 | * Sometimes a event that is in the ring buffer needs to be ignored. | |
1601 | * This function lets the user discard an event in the ring buffer | |
1602 | * and then that event will not be read later. | |
1603 | * | |
1604 | * Note, it is up to the user to be careful with this, and protect | |
1605 | * against races. If the user discards an event that has been consumed | |
1606 | * it is possible that it could corrupt the ring buffer. | |
1607 | */ | |
1608 | void ring_buffer_event_discard(struct ring_buffer_event *event) | |
1609 | { | |
1610 | event->type = RINGBUF_TYPE_PADDING; | |
1611 | /* time delta must be non zero */ | |
1612 | if (!event->time_delta) | |
1613 | event->time_delta = 1; | |
1614 | } | |
1615 | EXPORT_SYMBOL_GPL(ring_buffer_event_discard); | |
1616 | ||
1617 | /** | |
1618 | * ring_buffer_commit_discard - discard an event that has not been committed | |
1619 | * @buffer: the ring buffer | |
1620 | * @event: non committed event to discard | |
1621 | * | |
1622 | * This is similar to ring_buffer_event_discard but must only be | |
1623 | * performed on an event that has not been committed yet. The difference | |
1624 | * is that this will also try to free the event from the ring buffer | |
1625 | * if another event has not been added behind it. | |
1626 | * | |
1627 | * If another event has been added behind it, it will set the event | |
1628 | * up as discarded, and perform the commit. | |
1629 | * | |
1630 | * If this function is called, do not call ring_buffer_unlock_commit on | |
1631 | * the event. | |
1632 | */ | |
1633 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
1634 | struct ring_buffer_event *event) | |
1635 | { | |
1636 | struct ring_buffer_per_cpu *cpu_buffer; | |
1637 | unsigned long new_index, old_index; | |
1638 | struct buffer_page *bpage; | |
1639 | unsigned long index; | |
1640 | unsigned long addr; | |
1641 | int cpu; | |
1642 | ||
1643 | /* The event is discarded regardless */ | |
1644 | ring_buffer_event_discard(event); | |
1645 | ||
1646 | /* | |
1647 | * This must only be called if the event has not been | |
1648 | * committed yet. Thus we can assume that preemption | |
1649 | * is still disabled. | |
1650 | */ | |
1651 | RB_WARN_ON(buffer, !preempt_count()); | |
1652 | ||
1653 | cpu = smp_processor_id(); | |
1654 | cpu_buffer = buffer->buffers[cpu]; | |
1655 | ||
1656 | new_index = rb_event_index(event); | |
1657 | old_index = new_index + rb_event_length(event); | |
1658 | addr = (unsigned long)event; | |
1659 | addr &= PAGE_MASK; | |
1660 | ||
1661 | bpage = cpu_buffer->tail_page; | |
1662 | ||
1663 | if (bpage == (void *)addr && rb_page_write(bpage) == old_index) { | |
1664 | /* | |
1665 | * This is on the tail page. It is possible that | |
1666 | * a write could come in and move the tail page | |
1667 | * and write to the next page. That is fine | |
1668 | * because we just shorten what is on this page. | |
1669 | */ | |
1670 | index = local_cmpxchg(&bpage->write, old_index, new_index); | |
1671 | if (index == old_index) | |
1672 | goto out; | |
1673 | } | |
1674 | ||
1675 | /* | |
1676 | * The commit is still visible by the reader, so we | |
1677 | * must increment entries. | |
1678 | */ | |
1679 | cpu_buffer->entries++; | |
1680 | out: | |
1681 | /* | |
1682 | * If a write came in and pushed the tail page | |
1683 | * we still need to update the commit pointer | |
1684 | * if we were the commit. | |
1685 | */ | |
1686 | if (rb_is_commit(cpu_buffer, event)) | |
1687 | rb_set_commit_to_write(cpu_buffer); | |
1688 | ||
1689 | /* | |
1690 | * Only the last preempt count needs to restore preemption. | |
1691 | */ | |
1692 | if (preempt_count() == 1) | |
1693 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1694 | else | |
1695 | preempt_enable_no_resched_notrace(); | |
1696 | ||
1697 | } | |
1698 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
1699 | ||
7a8e76a3 SR |
1700 | /** |
1701 | * ring_buffer_write - write data to the buffer without reserving | |
1702 | * @buffer: The ring buffer to write to. | |
1703 | * @length: The length of the data being written (excluding the event header) | |
1704 | * @data: The data to write to the buffer. | |
1705 | * | |
1706 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
1707 | * one function. If you already have the data to write to the buffer, it | |
1708 | * may be easier to simply call this function. | |
1709 | * | |
1710 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
1711 | * and not the length of the event which would hold the header. | |
1712 | */ | |
1713 | int ring_buffer_write(struct ring_buffer *buffer, | |
1714 | unsigned long length, | |
1715 | void *data) | |
1716 | { | |
1717 | struct ring_buffer_per_cpu *cpu_buffer; | |
1718 | struct ring_buffer_event *event; | |
bf41a158 | 1719 | unsigned long event_length; |
7a8e76a3 SR |
1720 | void *body; |
1721 | int ret = -EBUSY; | |
bf41a158 | 1722 | int cpu, resched; |
7a8e76a3 | 1723 | |
033601a3 | 1724 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1725 | return -EBUSY; |
1726 | ||
7a8e76a3 SR |
1727 | if (atomic_read(&buffer->record_disabled)) |
1728 | return -EBUSY; | |
1729 | ||
182e9f5f | 1730 | resched = ftrace_preempt_disable(); |
bf41a158 | 1731 | |
7a8e76a3 SR |
1732 | cpu = raw_smp_processor_id(); |
1733 | ||
9e01c1b7 | 1734 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1735 | goto out; |
7a8e76a3 SR |
1736 | |
1737 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1738 | |
1739 | if (atomic_read(&cpu_buffer->record_disabled)) | |
1740 | goto out; | |
1741 | ||
1742 | event_length = rb_calculate_event_length(length); | |
1743 | event = rb_reserve_next_event(cpu_buffer, | |
1744 | RINGBUF_TYPE_DATA, event_length); | |
1745 | if (!event) | |
1746 | goto out; | |
1747 | ||
1748 | body = rb_event_data(event); | |
1749 | ||
1750 | memcpy(body, data, length); | |
1751 | ||
1752 | rb_commit(cpu_buffer, event); | |
1753 | ||
1754 | ret = 0; | |
1755 | out: | |
182e9f5f | 1756 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1757 | |
1758 | return ret; | |
1759 | } | |
c4f50183 | 1760 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 1761 | |
34a148bf | 1762 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
1763 | { |
1764 | struct buffer_page *reader = cpu_buffer->reader_page; | |
1765 | struct buffer_page *head = cpu_buffer->head_page; | |
1766 | struct buffer_page *commit = cpu_buffer->commit_page; | |
1767 | ||
1768 | return reader->read == rb_page_commit(reader) && | |
1769 | (commit == reader || | |
1770 | (commit == head && | |
1771 | head->read == rb_page_commit(commit))); | |
1772 | } | |
1773 | ||
7a8e76a3 SR |
1774 | /** |
1775 | * ring_buffer_record_disable - stop all writes into the buffer | |
1776 | * @buffer: The ring buffer to stop writes to. | |
1777 | * | |
1778 | * This prevents all writes to the buffer. Any attempt to write | |
1779 | * to the buffer after this will fail and return NULL. | |
1780 | * | |
1781 | * The caller should call synchronize_sched() after this. | |
1782 | */ | |
1783 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
1784 | { | |
1785 | atomic_inc(&buffer->record_disabled); | |
1786 | } | |
c4f50183 | 1787 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
1788 | |
1789 | /** | |
1790 | * ring_buffer_record_enable - enable writes to the buffer | |
1791 | * @buffer: The ring buffer to enable writes | |
1792 | * | |
1793 | * Note, multiple disables will need the same number of enables | |
1794 | * to truely enable the writing (much like preempt_disable). | |
1795 | */ | |
1796 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
1797 | { | |
1798 | atomic_dec(&buffer->record_disabled); | |
1799 | } | |
c4f50183 | 1800 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 SR |
1801 | |
1802 | /** | |
1803 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
1804 | * @buffer: The ring buffer to stop writes to. | |
1805 | * @cpu: The CPU buffer to stop | |
1806 | * | |
1807 | * This prevents all writes to the buffer. Any attempt to write | |
1808 | * to the buffer after this will fail and return NULL. | |
1809 | * | |
1810 | * The caller should call synchronize_sched() after this. | |
1811 | */ | |
1812 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
1813 | { | |
1814 | struct ring_buffer_per_cpu *cpu_buffer; | |
1815 | ||
9e01c1b7 | 1816 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1817 | return; |
7a8e76a3 SR |
1818 | |
1819 | cpu_buffer = buffer->buffers[cpu]; | |
1820 | atomic_inc(&cpu_buffer->record_disabled); | |
1821 | } | |
c4f50183 | 1822 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
1823 | |
1824 | /** | |
1825 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
1826 | * @buffer: The ring buffer to enable writes | |
1827 | * @cpu: The CPU to enable. | |
1828 | * | |
1829 | * Note, multiple disables will need the same number of enables | |
1830 | * to truely enable the writing (much like preempt_disable). | |
1831 | */ | |
1832 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
1833 | { | |
1834 | struct ring_buffer_per_cpu *cpu_buffer; | |
1835 | ||
9e01c1b7 | 1836 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1837 | return; |
7a8e76a3 SR |
1838 | |
1839 | cpu_buffer = buffer->buffers[cpu]; | |
1840 | atomic_dec(&cpu_buffer->record_disabled); | |
1841 | } | |
c4f50183 | 1842 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 SR |
1843 | |
1844 | /** | |
1845 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
1846 | * @buffer: The ring buffer | |
1847 | * @cpu: The per CPU buffer to get the entries from. | |
1848 | */ | |
1849 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
1850 | { | |
1851 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 1852 | unsigned long ret; |
7a8e76a3 | 1853 | |
9e01c1b7 | 1854 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1855 | return 0; |
7a8e76a3 SR |
1856 | |
1857 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 1858 | ret = cpu_buffer->entries; |
554f786e SR |
1859 | |
1860 | return ret; | |
7a8e76a3 | 1861 | } |
c4f50183 | 1862 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
1863 | |
1864 | /** | |
1865 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
1866 | * @buffer: The ring buffer | |
1867 | * @cpu: The per CPU buffer to get the number of overruns from | |
1868 | */ | |
1869 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
1870 | { | |
1871 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 1872 | unsigned long ret; |
7a8e76a3 | 1873 | |
9e01c1b7 | 1874 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1875 | return 0; |
7a8e76a3 SR |
1876 | |
1877 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 1878 | ret = cpu_buffer->overrun; |
554f786e SR |
1879 | |
1880 | return ret; | |
7a8e76a3 | 1881 | } |
c4f50183 | 1882 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 SR |
1883 | |
1884 | /** | |
1885 | * ring_buffer_entries - get the number of entries in a buffer | |
1886 | * @buffer: The ring buffer | |
1887 | * | |
1888 | * Returns the total number of entries in the ring buffer | |
1889 | * (all CPU entries) | |
1890 | */ | |
1891 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
1892 | { | |
1893 | struct ring_buffer_per_cpu *cpu_buffer; | |
1894 | unsigned long entries = 0; | |
1895 | int cpu; | |
1896 | ||
1897 | /* if you care about this being correct, lock the buffer */ | |
1898 | for_each_buffer_cpu(buffer, cpu) { | |
1899 | cpu_buffer = buffer->buffers[cpu]; | |
1900 | entries += cpu_buffer->entries; | |
1901 | } | |
1902 | ||
1903 | return entries; | |
1904 | } | |
c4f50183 | 1905 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
1906 | |
1907 | /** | |
1908 | * ring_buffer_overrun_cpu - get the number of overruns in buffer | |
1909 | * @buffer: The ring buffer | |
1910 | * | |
1911 | * Returns the total number of overruns in the ring buffer | |
1912 | * (all CPU entries) | |
1913 | */ | |
1914 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
1915 | { | |
1916 | struct ring_buffer_per_cpu *cpu_buffer; | |
1917 | unsigned long overruns = 0; | |
1918 | int cpu; | |
1919 | ||
1920 | /* if you care about this being correct, lock the buffer */ | |
1921 | for_each_buffer_cpu(buffer, cpu) { | |
1922 | cpu_buffer = buffer->buffers[cpu]; | |
1923 | overruns += cpu_buffer->overrun; | |
1924 | } | |
1925 | ||
1926 | return overruns; | |
1927 | } | |
c4f50183 | 1928 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 1929 | |
642edba5 | 1930 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
1931 | { |
1932 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1933 | ||
d769041f SR |
1934 | /* Iterator usage is expected to have record disabled */ |
1935 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
1936 | iter->head_page = cpu_buffer->head_page; | |
6f807acd | 1937 | iter->head = cpu_buffer->head_page->read; |
d769041f SR |
1938 | } else { |
1939 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 1940 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
1941 | } |
1942 | if (iter->head) | |
1943 | iter->read_stamp = cpu_buffer->read_stamp; | |
1944 | else | |
abc9b56d | 1945 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 1946 | } |
f83c9d0f | 1947 | |
642edba5 SR |
1948 | /** |
1949 | * ring_buffer_iter_reset - reset an iterator | |
1950 | * @iter: The iterator to reset | |
1951 | * | |
1952 | * Resets the iterator, so that it will start from the beginning | |
1953 | * again. | |
1954 | */ | |
1955 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
1956 | { | |
554f786e | 1957 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
1958 | unsigned long flags; |
1959 | ||
554f786e SR |
1960 | if (!iter) |
1961 | return; | |
1962 | ||
1963 | cpu_buffer = iter->cpu_buffer; | |
1964 | ||
642edba5 SR |
1965 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
1966 | rb_iter_reset(iter); | |
f83c9d0f | 1967 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 1968 | } |
c4f50183 | 1969 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
1970 | |
1971 | /** | |
1972 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
1973 | * @iter: The iterator to check | |
1974 | */ | |
1975 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
1976 | { | |
1977 | struct ring_buffer_per_cpu *cpu_buffer; | |
1978 | ||
1979 | cpu_buffer = iter->cpu_buffer; | |
1980 | ||
bf41a158 SR |
1981 | return iter->head_page == cpu_buffer->commit_page && |
1982 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 1983 | } |
c4f50183 | 1984 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
1985 | |
1986 | static void | |
1987 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1988 | struct ring_buffer_event *event) | |
1989 | { | |
1990 | u64 delta; | |
1991 | ||
1992 | switch (event->type) { | |
1993 | case RINGBUF_TYPE_PADDING: | |
1994 | return; | |
1995 | ||
1996 | case RINGBUF_TYPE_TIME_EXTEND: | |
1997 | delta = event->array[0]; | |
1998 | delta <<= TS_SHIFT; | |
1999 | delta += event->time_delta; | |
2000 | cpu_buffer->read_stamp += delta; | |
2001 | return; | |
2002 | ||
2003 | case RINGBUF_TYPE_TIME_STAMP: | |
2004 | /* FIXME: not implemented */ | |
2005 | return; | |
2006 | ||
2007 | case RINGBUF_TYPE_DATA: | |
2008 | cpu_buffer->read_stamp += event->time_delta; | |
2009 | return; | |
2010 | ||
2011 | default: | |
2012 | BUG(); | |
2013 | } | |
2014 | return; | |
2015 | } | |
2016 | ||
2017 | static void | |
2018 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
2019 | struct ring_buffer_event *event) | |
2020 | { | |
2021 | u64 delta; | |
2022 | ||
2023 | switch (event->type) { | |
2024 | case RINGBUF_TYPE_PADDING: | |
2025 | return; | |
2026 | ||
2027 | case RINGBUF_TYPE_TIME_EXTEND: | |
2028 | delta = event->array[0]; | |
2029 | delta <<= TS_SHIFT; | |
2030 | delta += event->time_delta; | |
2031 | iter->read_stamp += delta; | |
2032 | return; | |
2033 | ||
2034 | case RINGBUF_TYPE_TIME_STAMP: | |
2035 | /* FIXME: not implemented */ | |
2036 | return; | |
2037 | ||
2038 | case RINGBUF_TYPE_DATA: | |
2039 | iter->read_stamp += event->time_delta; | |
2040 | return; | |
2041 | ||
2042 | default: | |
2043 | BUG(); | |
2044 | } | |
2045 | return; | |
2046 | } | |
2047 | ||
d769041f SR |
2048 | static struct buffer_page * |
2049 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 2050 | { |
d769041f SR |
2051 | struct buffer_page *reader = NULL; |
2052 | unsigned long flags; | |
818e3dd3 | 2053 | int nr_loops = 0; |
d769041f | 2054 | |
3e03fb7f SR |
2055 | local_irq_save(flags); |
2056 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
2057 | |
2058 | again: | |
818e3dd3 SR |
2059 | /* |
2060 | * This should normally only loop twice. But because the | |
2061 | * start of the reader inserts an empty page, it causes | |
2062 | * a case where we will loop three times. There should be no | |
2063 | * reason to loop four times (that I know of). | |
2064 | */ | |
3e89c7bb | 2065 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
2066 | reader = NULL; |
2067 | goto out; | |
2068 | } | |
2069 | ||
d769041f SR |
2070 | reader = cpu_buffer->reader_page; |
2071 | ||
2072 | /* If there's more to read, return this page */ | |
bf41a158 | 2073 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
2074 | goto out; |
2075 | ||
2076 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
2077 | if (RB_WARN_ON(cpu_buffer, |
2078 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
2079 | goto out; | |
d769041f SR |
2080 | |
2081 | /* check if we caught up to the tail */ | |
2082 | reader = NULL; | |
bf41a158 | 2083 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 2084 | goto out; |
7a8e76a3 SR |
2085 | |
2086 | /* | |
d769041f SR |
2087 | * Splice the empty reader page into the list around the head. |
2088 | * Reset the reader page to size zero. | |
7a8e76a3 | 2089 | */ |
7a8e76a3 | 2090 | |
d769041f SR |
2091 | reader = cpu_buffer->head_page; |
2092 | cpu_buffer->reader_page->list.next = reader->list.next; | |
2093 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 SR |
2094 | |
2095 | local_set(&cpu_buffer->reader_page->write, 0); | |
abc9b56d | 2096 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
7a8e76a3 | 2097 | |
d769041f SR |
2098 | /* Make the reader page now replace the head */ |
2099 | reader->list.prev->next = &cpu_buffer->reader_page->list; | |
2100 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
7a8e76a3 SR |
2101 | |
2102 | /* | |
d769041f SR |
2103 | * If the tail is on the reader, then we must set the head |
2104 | * to the inserted page, otherwise we set it one before. | |
7a8e76a3 | 2105 | */ |
d769041f | 2106 | cpu_buffer->head_page = cpu_buffer->reader_page; |
7a8e76a3 | 2107 | |
bf41a158 | 2108 | if (cpu_buffer->commit_page != reader) |
d769041f SR |
2109 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
2110 | ||
2111 | /* Finally update the reader page to the new head */ | |
2112 | cpu_buffer->reader_page = reader; | |
2113 | rb_reset_reader_page(cpu_buffer); | |
2114 | ||
2115 | goto again; | |
2116 | ||
2117 | out: | |
3e03fb7f SR |
2118 | __raw_spin_unlock(&cpu_buffer->lock); |
2119 | local_irq_restore(flags); | |
d769041f SR |
2120 | |
2121 | return reader; | |
2122 | } | |
2123 | ||
2124 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
2125 | { | |
2126 | struct ring_buffer_event *event; | |
2127 | struct buffer_page *reader; | |
2128 | unsigned length; | |
2129 | ||
2130 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 2131 | |
d769041f | 2132 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
2133 | if (RB_WARN_ON(cpu_buffer, !reader)) |
2134 | return; | |
7a8e76a3 | 2135 | |
d769041f SR |
2136 | event = rb_reader_event(cpu_buffer); |
2137 | ||
2d622719 | 2138 | if (event->type == RINGBUF_TYPE_DATA || rb_discarded_event(event)) |
d769041f SR |
2139 | cpu_buffer->entries--; |
2140 | ||
2141 | rb_update_read_stamp(cpu_buffer, event); | |
2142 | ||
2143 | length = rb_event_length(event); | |
6f807acd | 2144 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
2145 | } |
2146 | ||
2147 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
2148 | { | |
2149 | struct ring_buffer *buffer; | |
2150 | struct ring_buffer_per_cpu *cpu_buffer; | |
2151 | struct ring_buffer_event *event; | |
2152 | unsigned length; | |
2153 | ||
2154 | cpu_buffer = iter->cpu_buffer; | |
2155 | buffer = cpu_buffer->buffer; | |
2156 | ||
2157 | /* | |
2158 | * Check if we are at the end of the buffer. | |
2159 | */ | |
bf41a158 | 2160 | if (iter->head >= rb_page_size(iter->head_page)) { |
3e89c7bb SR |
2161 | if (RB_WARN_ON(buffer, |
2162 | iter->head_page == cpu_buffer->commit_page)) | |
2163 | return; | |
d769041f | 2164 | rb_inc_iter(iter); |
7a8e76a3 SR |
2165 | return; |
2166 | } | |
2167 | ||
2168 | event = rb_iter_head_event(iter); | |
2169 | ||
2170 | length = rb_event_length(event); | |
2171 | ||
2172 | /* | |
2173 | * This should not be called to advance the header if we are | |
2174 | * at the tail of the buffer. | |
2175 | */ | |
3e89c7bb | 2176 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 2177 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
2178 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
2179 | return; | |
7a8e76a3 SR |
2180 | |
2181 | rb_update_iter_read_stamp(iter, event); | |
2182 | ||
2183 | iter->head += length; | |
2184 | ||
2185 | /* check for end of page padding */ | |
bf41a158 SR |
2186 | if ((iter->head >= rb_page_size(iter->head_page)) && |
2187 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
2188 | rb_advance_iter(iter); |
2189 | } | |
2190 | ||
f83c9d0f SR |
2191 | static struct ring_buffer_event * |
2192 | rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
7a8e76a3 SR |
2193 | { |
2194 | struct ring_buffer_per_cpu *cpu_buffer; | |
2195 | struct ring_buffer_event *event; | |
d769041f | 2196 | struct buffer_page *reader; |
818e3dd3 | 2197 | int nr_loops = 0; |
7a8e76a3 | 2198 | |
7a8e76a3 SR |
2199 | cpu_buffer = buffer->buffers[cpu]; |
2200 | ||
2201 | again: | |
818e3dd3 SR |
2202 | /* |
2203 | * We repeat when a timestamp is encountered. It is possible | |
2204 | * to get multiple timestamps from an interrupt entering just | |
2205 | * as one timestamp is about to be written. The max times | |
2206 | * that this can happen is the number of nested interrupts we | |
2207 | * can have. Nesting 10 deep of interrupts is clearly | |
2208 | * an anomaly. | |
2209 | */ | |
3e89c7bb | 2210 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 2211 | return NULL; |
818e3dd3 | 2212 | |
d769041f SR |
2213 | reader = rb_get_reader_page(cpu_buffer); |
2214 | if (!reader) | |
7a8e76a3 SR |
2215 | return NULL; |
2216 | ||
d769041f | 2217 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 SR |
2218 | |
2219 | switch (event->type) { | |
2220 | case RINGBUF_TYPE_PADDING: | |
2d622719 TZ |
2221 | if (rb_null_event(event)) |
2222 | RB_WARN_ON(cpu_buffer, 1); | |
2223 | /* | |
2224 | * Because the writer could be discarding every | |
2225 | * event it creates (which would probably be bad) | |
2226 | * if we were to go back to "again" then we may never | |
2227 | * catch up, and will trigger the warn on, or lock | |
2228 | * the box. Return the padding, and we will release | |
2229 | * the current locks, and try again. | |
2230 | */ | |
d769041f | 2231 | rb_advance_reader(cpu_buffer); |
2d622719 | 2232 | return event; |
7a8e76a3 SR |
2233 | |
2234 | case RINGBUF_TYPE_TIME_EXTEND: | |
2235 | /* Internal data, OK to advance */ | |
d769041f | 2236 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2237 | goto again; |
2238 | ||
2239 | case RINGBUF_TYPE_TIME_STAMP: | |
2240 | /* FIXME: not implemented */ | |
d769041f | 2241 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2242 | goto again; |
2243 | ||
2244 | case RINGBUF_TYPE_DATA: | |
2245 | if (ts) { | |
2246 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
37886f6a SR |
2247 | ring_buffer_normalize_time_stamp(buffer, |
2248 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2249 | } |
2250 | return event; | |
2251 | ||
2252 | default: | |
2253 | BUG(); | |
2254 | } | |
2255 | ||
2256 | return NULL; | |
2257 | } | |
c4f50183 | 2258 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 2259 | |
f83c9d0f SR |
2260 | static struct ring_buffer_event * |
2261 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
2262 | { |
2263 | struct ring_buffer *buffer; | |
2264 | struct ring_buffer_per_cpu *cpu_buffer; | |
2265 | struct ring_buffer_event *event; | |
818e3dd3 | 2266 | int nr_loops = 0; |
7a8e76a3 SR |
2267 | |
2268 | if (ring_buffer_iter_empty(iter)) | |
2269 | return NULL; | |
2270 | ||
2271 | cpu_buffer = iter->cpu_buffer; | |
2272 | buffer = cpu_buffer->buffer; | |
2273 | ||
2274 | again: | |
818e3dd3 SR |
2275 | /* |
2276 | * We repeat when a timestamp is encountered. It is possible | |
2277 | * to get multiple timestamps from an interrupt entering just | |
2278 | * as one timestamp is about to be written. The max times | |
2279 | * that this can happen is the number of nested interrupts we | |
2280 | * can have. Nesting 10 deep of interrupts is clearly | |
2281 | * an anomaly. | |
2282 | */ | |
3e89c7bb | 2283 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 2284 | return NULL; |
818e3dd3 | 2285 | |
7a8e76a3 SR |
2286 | if (rb_per_cpu_empty(cpu_buffer)) |
2287 | return NULL; | |
2288 | ||
2289 | event = rb_iter_head_event(iter); | |
2290 | ||
2291 | switch (event->type) { | |
2292 | case RINGBUF_TYPE_PADDING: | |
2d622719 TZ |
2293 | if (rb_null_event(event)) { |
2294 | rb_inc_iter(iter); | |
2295 | goto again; | |
2296 | } | |
2297 | rb_advance_iter(iter); | |
2298 | return event; | |
7a8e76a3 SR |
2299 | |
2300 | case RINGBUF_TYPE_TIME_EXTEND: | |
2301 | /* Internal data, OK to advance */ | |
2302 | rb_advance_iter(iter); | |
2303 | goto again; | |
2304 | ||
2305 | case RINGBUF_TYPE_TIME_STAMP: | |
2306 | /* FIXME: not implemented */ | |
2307 | rb_advance_iter(iter); | |
2308 | goto again; | |
2309 | ||
2310 | case RINGBUF_TYPE_DATA: | |
2311 | if (ts) { | |
2312 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
2313 | ring_buffer_normalize_time_stamp(buffer, |
2314 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2315 | } |
2316 | return event; | |
2317 | ||
2318 | default: | |
2319 | BUG(); | |
2320 | } | |
2321 | ||
2322 | return NULL; | |
2323 | } | |
c4f50183 | 2324 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 2325 | |
f83c9d0f SR |
2326 | /** |
2327 | * ring_buffer_peek - peek at the next event to be read | |
2328 | * @buffer: The ring buffer to read | |
2329 | * @cpu: The cpu to peak at | |
2330 | * @ts: The timestamp counter of this event. | |
2331 | * | |
2332 | * This will return the event that will be read next, but does | |
2333 | * not consume the data. | |
2334 | */ | |
2335 | struct ring_buffer_event * | |
2336 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2337 | { | |
2338 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 2339 | struct ring_buffer_event *event; |
f83c9d0f SR |
2340 | unsigned long flags; |
2341 | ||
554f786e | 2342 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2343 | return NULL; |
554f786e | 2344 | |
2d622719 | 2345 | again: |
f83c9d0f SR |
2346 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2347 | event = rb_buffer_peek(buffer, cpu, ts); | |
2348 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2349 | ||
2d622719 TZ |
2350 | if (event && event->type == RINGBUF_TYPE_PADDING) { |
2351 | cpu_relax(); | |
2352 | goto again; | |
2353 | } | |
2354 | ||
f83c9d0f SR |
2355 | return event; |
2356 | } | |
2357 | ||
2358 | /** | |
2359 | * ring_buffer_iter_peek - peek at the next event to be read | |
2360 | * @iter: The ring buffer iterator | |
2361 | * @ts: The timestamp counter of this event. | |
2362 | * | |
2363 | * This will return the event that will be read next, but does | |
2364 | * not increment the iterator. | |
2365 | */ | |
2366 | struct ring_buffer_event * | |
2367 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
2368 | { | |
2369 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2370 | struct ring_buffer_event *event; | |
2371 | unsigned long flags; | |
2372 | ||
2d622719 | 2373 | again: |
f83c9d0f SR |
2374 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2375 | event = rb_iter_peek(iter, ts); | |
2376 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2377 | ||
2d622719 TZ |
2378 | if (event && event->type == RINGBUF_TYPE_PADDING) { |
2379 | cpu_relax(); | |
2380 | goto again; | |
2381 | } | |
2382 | ||
f83c9d0f SR |
2383 | return event; |
2384 | } | |
2385 | ||
7a8e76a3 SR |
2386 | /** |
2387 | * ring_buffer_consume - return an event and consume it | |
2388 | * @buffer: The ring buffer to get the next event from | |
2389 | * | |
2390 | * Returns the next event in the ring buffer, and that event is consumed. | |
2391 | * Meaning, that sequential reads will keep returning a different event, | |
2392 | * and eventually empty the ring buffer if the producer is slower. | |
2393 | */ | |
2394 | struct ring_buffer_event * | |
2395 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2396 | { | |
554f786e SR |
2397 | struct ring_buffer_per_cpu *cpu_buffer; |
2398 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 2399 | unsigned long flags; |
7a8e76a3 | 2400 | |
2d622719 | 2401 | again: |
554f786e SR |
2402 | /* might be called in atomic */ |
2403 | preempt_disable(); | |
2404 | ||
9e01c1b7 | 2405 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 2406 | goto out; |
7a8e76a3 | 2407 | |
554f786e | 2408 | cpu_buffer = buffer->buffers[cpu]; |
f83c9d0f SR |
2409 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2410 | ||
2411 | event = rb_buffer_peek(buffer, cpu, ts); | |
7a8e76a3 | 2412 | if (!event) |
554f786e | 2413 | goto out_unlock; |
7a8e76a3 | 2414 | |
d769041f | 2415 | rb_advance_reader(cpu_buffer); |
7a8e76a3 | 2416 | |
554f786e | 2417 | out_unlock: |
f83c9d0f SR |
2418 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
2419 | ||
554f786e SR |
2420 | out: |
2421 | preempt_enable(); | |
2422 | ||
2d622719 TZ |
2423 | if (event && event->type == RINGBUF_TYPE_PADDING) { |
2424 | cpu_relax(); | |
2425 | goto again; | |
2426 | } | |
2427 | ||
7a8e76a3 SR |
2428 | return event; |
2429 | } | |
c4f50183 | 2430 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
2431 | |
2432 | /** | |
2433 | * ring_buffer_read_start - start a non consuming read of the buffer | |
2434 | * @buffer: The ring buffer to read from | |
2435 | * @cpu: The cpu buffer to iterate over | |
2436 | * | |
2437 | * This starts up an iteration through the buffer. It also disables | |
2438 | * the recording to the buffer until the reading is finished. | |
2439 | * This prevents the reading from being corrupted. This is not | |
2440 | * a consuming read, so a producer is not expected. | |
2441 | * | |
2442 | * Must be paired with ring_buffer_finish. | |
2443 | */ | |
2444 | struct ring_buffer_iter * | |
2445 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
2446 | { | |
2447 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2448 | struct ring_buffer_iter *iter; |
d769041f | 2449 | unsigned long flags; |
7a8e76a3 | 2450 | |
9e01c1b7 | 2451 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2452 | return NULL; |
7a8e76a3 SR |
2453 | |
2454 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
2455 | if (!iter) | |
8aabee57 | 2456 | return NULL; |
7a8e76a3 SR |
2457 | |
2458 | cpu_buffer = buffer->buffers[cpu]; | |
2459 | ||
2460 | iter->cpu_buffer = cpu_buffer; | |
2461 | ||
2462 | atomic_inc(&cpu_buffer->record_disabled); | |
2463 | synchronize_sched(); | |
2464 | ||
f83c9d0f | 2465 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 2466 | __raw_spin_lock(&cpu_buffer->lock); |
642edba5 | 2467 | rb_iter_reset(iter); |
3e03fb7f | 2468 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 2469 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
2470 | |
2471 | return iter; | |
2472 | } | |
c4f50183 | 2473 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
2474 | |
2475 | /** | |
2476 | * ring_buffer_finish - finish reading the iterator of the buffer | |
2477 | * @iter: The iterator retrieved by ring_buffer_start | |
2478 | * | |
2479 | * This re-enables the recording to the buffer, and frees the | |
2480 | * iterator. | |
2481 | */ | |
2482 | void | |
2483 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
2484 | { | |
2485 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2486 | ||
2487 | atomic_dec(&cpu_buffer->record_disabled); | |
2488 | kfree(iter); | |
2489 | } | |
c4f50183 | 2490 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
2491 | |
2492 | /** | |
2493 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
2494 | * @iter: The ring buffer iterator | |
2495 | * @ts: The time stamp of the event read. | |
2496 | * | |
2497 | * This reads the next event in the ring buffer and increments the iterator. | |
2498 | */ | |
2499 | struct ring_buffer_event * | |
2500 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
2501 | { | |
2502 | struct ring_buffer_event *event; | |
f83c9d0f SR |
2503 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
2504 | unsigned long flags; | |
7a8e76a3 | 2505 | |
2d622719 | 2506 | again: |
f83c9d0f SR |
2507 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2508 | event = rb_iter_peek(iter, ts); | |
7a8e76a3 | 2509 | if (!event) |
f83c9d0f | 2510 | goto out; |
7a8e76a3 SR |
2511 | |
2512 | rb_advance_iter(iter); | |
f83c9d0f SR |
2513 | out: |
2514 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 2515 | |
2d622719 TZ |
2516 | if (event && event->type == RINGBUF_TYPE_PADDING) { |
2517 | cpu_relax(); | |
2518 | goto again; | |
2519 | } | |
2520 | ||
7a8e76a3 SR |
2521 | return event; |
2522 | } | |
c4f50183 | 2523 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
2524 | |
2525 | /** | |
2526 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
2527 | * @buffer: The ring buffer. | |
2528 | */ | |
2529 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
2530 | { | |
2531 | return BUF_PAGE_SIZE * buffer->pages; | |
2532 | } | |
c4f50183 | 2533 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
2534 | |
2535 | static void | |
2536 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
2537 | { | |
2538 | cpu_buffer->head_page | |
2539 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 2540 | local_set(&cpu_buffer->head_page->write, 0); |
abc9b56d | 2541 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 2542 | |
6f807acd | 2543 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
2544 | |
2545 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
2546 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
2547 | ||
2548 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
2549 | local_set(&cpu_buffer->reader_page->write, 0); | |
abc9b56d | 2550 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 2551 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 2552 | |
7a8e76a3 SR |
2553 | cpu_buffer->overrun = 0; |
2554 | cpu_buffer->entries = 0; | |
69507c06 SR |
2555 | |
2556 | cpu_buffer->write_stamp = 0; | |
2557 | cpu_buffer->read_stamp = 0; | |
7a8e76a3 SR |
2558 | } |
2559 | ||
2560 | /** | |
2561 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
2562 | * @buffer: The ring buffer to reset a per cpu buffer of | |
2563 | * @cpu: The CPU buffer to be reset | |
2564 | */ | |
2565 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
2566 | { | |
2567 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2568 | unsigned long flags; | |
2569 | ||
9e01c1b7 | 2570 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2571 | return; |
7a8e76a3 | 2572 | |
f83c9d0f SR |
2573 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2574 | ||
3e03fb7f | 2575 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
2576 | |
2577 | rb_reset_cpu(cpu_buffer); | |
2578 | ||
3e03fb7f | 2579 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f SR |
2580 | |
2581 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 2582 | } |
c4f50183 | 2583 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
2584 | |
2585 | /** | |
2586 | * ring_buffer_reset - reset a ring buffer | |
2587 | * @buffer: The ring buffer to reset all cpu buffers | |
2588 | */ | |
2589 | void ring_buffer_reset(struct ring_buffer *buffer) | |
2590 | { | |
7a8e76a3 SR |
2591 | int cpu; |
2592 | ||
7a8e76a3 | 2593 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 2594 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 2595 | } |
c4f50183 | 2596 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
2597 | |
2598 | /** | |
2599 | * rind_buffer_empty - is the ring buffer empty? | |
2600 | * @buffer: The ring buffer to test | |
2601 | */ | |
2602 | int ring_buffer_empty(struct ring_buffer *buffer) | |
2603 | { | |
2604 | struct ring_buffer_per_cpu *cpu_buffer; | |
2605 | int cpu; | |
2606 | ||
2607 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
2608 | for_each_buffer_cpu(buffer, cpu) { | |
2609 | cpu_buffer = buffer->buffers[cpu]; | |
2610 | if (!rb_per_cpu_empty(cpu_buffer)) | |
2611 | return 0; | |
2612 | } | |
554f786e | 2613 | |
7a8e76a3 SR |
2614 | return 1; |
2615 | } | |
c4f50183 | 2616 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
2617 | |
2618 | /** | |
2619 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
2620 | * @buffer: The ring buffer | |
2621 | * @cpu: The CPU buffer to test | |
2622 | */ | |
2623 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
2624 | { | |
2625 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2626 | int ret; |
7a8e76a3 | 2627 | |
9e01c1b7 | 2628 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2629 | return 1; |
7a8e76a3 SR |
2630 | |
2631 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e SR |
2632 | ret = rb_per_cpu_empty(cpu_buffer); |
2633 | ||
554f786e SR |
2634 | |
2635 | return ret; | |
7a8e76a3 | 2636 | } |
c4f50183 | 2637 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 SR |
2638 | |
2639 | /** | |
2640 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
2641 | * @buffer_a: One buffer to swap with | |
2642 | * @buffer_b: The other buffer to swap with | |
2643 | * | |
2644 | * This function is useful for tracers that want to take a "snapshot" | |
2645 | * of a CPU buffer and has another back up buffer lying around. | |
2646 | * it is expected that the tracer handles the cpu buffer not being | |
2647 | * used at the moment. | |
2648 | */ | |
2649 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
2650 | struct ring_buffer *buffer_b, int cpu) | |
2651 | { | |
2652 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
2653 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
2654 | int ret = -EINVAL; |
2655 | ||
9e01c1b7 RR |
2656 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
2657 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 2658 | goto out; |
7a8e76a3 SR |
2659 | |
2660 | /* At least make sure the two buffers are somewhat the same */ | |
6d102bc6 | 2661 | if (buffer_a->pages != buffer_b->pages) |
554f786e SR |
2662 | goto out; |
2663 | ||
2664 | ret = -EAGAIN; | |
7a8e76a3 | 2665 | |
97b17efe | 2666 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 2667 | goto out; |
97b17efe SR |
2668 | |
2669 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 2670 | goto out; |
97b17efe SR |
2671 | |
2672 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 2673 | goto out; |
97b17efe | 2674 | |
7a8e76a3 SR |
2675 | cpu_buffer_a = buffer_a->buffers[cpu]; |
2676 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
2677 | ||
97b17efe | 2678 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 2679 | goto out; |
97b17efe SR |
2680 | |
2681 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 2682 | goto out; |
97b17efe | 2683 | |
7a8e76a3 SR |
2684 | /* |
2685 | * We can't do a synchronize_sched here because this | |
2686 | * function can be called in atomic context. | |
2687 | * Normally this will be called from the same CPU as cpu. | |
2688 | * If not it's up to the caller to protect this. | |
2689 | */ | |
2690 | atomic_inc(&cpu_buffer_a->record_disabled); | |
2691 | atomic_inc(&cpu_buffer_b->record_disabled); | |
2692 | ||
2693 | buffer_a->buffers[cpu] = cpu_buffer_b; | |
2694 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
2695 | ||
2696 | cpu_buffer_b->buffer = buffer_a; | |
2697 | cpu_buffer_a->buffer = buffer_b; | |
2698 | ||
2699 | atomic_dec(&cpu_buffer_a->record_disabled); | |
2700 | atomic_dec(&cpu_buffer_b->record_disabled); | |
2701 | ||
554f786e SR |
2702 | ret = 0; |
2703 | out: | |
554f786e | 2704 | return ret; |
7a8e76a3 | 2705 | } |
c4f50183 | 2706 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
7a8e76a3 | 2707 | |
8789a9e7 | 2708 | static void rb_remove_entries(struct ring_buffer_per_cpu *cpu_buffer, |
667d2412 LJ |
2709 | struct buffer_data_page *bpage, |
2710 | unsigned int offset) | |
8789a9e7 SR |
2711 | { |
2712 | struct ring_buffer_event *event; | |
2713 | unsigned long head; | |
2714 | ||
2715 | __raw_spin_lock(&cpu_buffer->lock); | |
667d2412 | 2716 | for (head = offset; head < local_read(&bpage->commit); |
8789a9e7 SR |
2717 | head += rb_event_length(event)) { |
2718 | ||
044fa782 | 2719 | event = __rb_data_page_index(bpage, head); |
8789a9e7 SR |
2720 | if (RB_WARN_ON(cpu_buffer, rb_null_event(event))) |
2721 | return; | |
2722 | /* Only count data entries */ | |
2723 | if (event->type != RINGBUF_TYPE_DATA) | |
2724 | continue; | |
2725 | cpu_buffer->entries--; | |
2726 | } | |
2727 | __raw_spin_unlock(&cpu_buffer->lock); | |
2728 | } | |
2729 | ||
2730 | /** | |
2731 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
2732 | * @buffer: the buffer to allocate for. | |
2733 | * | |
2734 | * This function is used in conjunction with ring_buffer_read_page. | |
2735 | * When reading a full page from the ring buffer, these functions | |
2736 | * can be used to speed up the process. The calling function should | |
2737 | * allocate a few pages first with this function. Then when it | |
2738 | * needs to get pages from the ring buffer, it passes the result | |
2739 | * of this function into ring_buffer_read_page, which will swap | |
2740 | * the page that was allocated, with the read page of the buffer. | |
2741 | * | |
2742 | * Returns: | |
2743 | * The page allocated, or NULL on error. | |
2744 | */ | |
2745 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer) | |
2746 | { | |
044fa782 | 2747 | struct buffer_data_page *bpage; |
ef7a4a16 | 2748 | unsigned long addr; |
8789a9e7 SR |
2749 | |
2750 | addr = __get_free_page(GFP_KERNEL); | |
2751 | if (!addr) | |
2752 | return NULL; | |
2753 | ||
044fa782 | 2754 | bpage = (void *)addr; |
8789a9e7 | 2755 | |
ef7a4a16 SR |
2756 | rb_init_page(bpage); |
2757 | ||
044fa782 | 2758 | return bpage; |
8789a9e7 SR |
2759 | } |
2760 | ||
2761 | /** | |
2762 | * ring_buffer_free_read_page - free an allocated read page | |
2763 | * @buffer: the buffer the page was allocate for | |
2764 | * @data: the page to free | |
2765 | * | |
2766 | * Free a page allocated from ring_buffer_alloc_read_page. | |
2767 | */ | |
2768 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
2769 | { | |
2770 | free_page((unsigned long)data); | |
2771 | } | |
2772 | ||
2773 | /** | |
2774 | * ring_buffer_read_page - extract a page from the ring buffer | |
2775 | * @buffer: buffer to extract from | |
2776 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 2777 | * @len: amount to extract |
8789a9e7 SR |
2778 | * @cpu: the cpu of the buffer to extract |
2779 | * @full: should the extraction only happen when the page is full. | |
2780 | * | |
2781 | * This function will pull out a page from the ring buffer and consume it. | |
2782 | * @data_page must be the address of the variable that was returned | |
2783 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
2784 | * to swap with a page in the ring buffer. | |
2785 | * | |
2786 | * for example: | |
b85fa01e | 2787 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
2788 | * if (!rpage) |
2789 | * return error; | |
ef7a4a16 | 2790 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
2791 | * if (ret >= 0) |
2792 | * process_page(rpage, ret); | |
8789a9e7 SR |
2793 | * |
2794 | * When @full is set, the function will not return true unless | |
2795 | * the writer is off the reader page. | |
2796 | * | |
2797 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
2798 | * The ring buffer can be used anywhere in the kernel and can not | |
2799 | * blindly call wake_up. The layer that uses the ring buffer must be | |
2800 | * responsible for that. | |
2801 | * | |
2802 | * Returns: | |
667d2412 LJ |
2803 | * >=0 if data has been transferred, returns the offset of consumed data. |
2804 | * <0 if no data has been transferred. | |
8789a9e7 SR |
2805 | */ |
2806 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 2807 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
2808 | { |
2809 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2810 | struct ring_buffer_event *event; | |
044fa782 | 2811 | struct buffer_data_page *bpage; |
ef7a4a16 | 2812 | struct buffer_page *reader; |
8789a9e7 | 2813 | unsigned long flags; |
ef7a4a16 | 2814 | unsigned int commit; |
667d2412 | 2815 | unsigned int read; |
4f3640f8 | 2816 | u64 save_timestamp; |
667d2412 | 2817 | int ret = -1; |
8789a9e7 | 2818 | |
554f786e SR |
2819 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
2820 | goto out; | |
2821 | ||
474d32b6 SR |
2822 | /* |
2823 | * If len is not big enough to hold the page header, then | |
2824 | * we can not copy anything. | |
2825 | */ | |
2826 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 2827 | goto out; |
474d32b6 SR |
2828 | |
2829 | len -= BUF_PAGE_HDR_SIZE; | |
2830 | ||
8789a9e7 | 2831 | if (!data_page) |
554f786e | 2832 | goto out; |
8789a9e7 | 2833 | |
044fa782 SR |
2834 | bpage = *data_page; |
2835 | if (!bpage) | |
554f786e | 2836 | goto out; |
8789a9e7 SR |
2837 | |
2838 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2839 | ||
ef7a4a16 SR |
2840 | reader = rb_get_reader_page(cpu_buffer); |
2841 | if (!reader) | |
554f786e | 2842 | goto out_unlock; |
8789a9e7 | 2843 | |
ef7a4a16 SR |
2844 | event = rb_reader_event(cpu_buffer); |
2845 | ||
2846 | read = reader->read; | |
2847 | commit = rb_page_commit(reader); | |
667d2412 | 2848 | |
8789a9e7 | 2849 | /* |
474d32b6 SR |
2850 | * If this page has been partially read or |
2851 | * if len is not big enough to read the rest of the page or | |
2852 | * a writer is still on the page, then | |
2853 | * we must copy the data from the page to the buffer. | |
2854 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 2855 | */ |
474d32b6 | 2856 | if (read || (len < (commit - read)) || |
ef7a4a16 | 2857 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 2858 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
2859 | unsigned int rpos = read; |
2860 | unsigned int pos = 0; | |
ef7a4a16 | 2861 | unsigned int size; |
8789a9e7 SR |
2862 | |
2863 | if (full) | |
554f786e | 2864 | goto out_unlock; |
8789a9e7 | 2865 | |
ef7a4a16 SR |
2866 | if (len > (commit - read)) |
2867 | len = (commit - read); | |
2868 | ||
2869 | size = rb_event_length(event); | |
2870 | ||
2871 | if (len < size) | |
554f786e | 2872 | goto out_unlock; |
ef7a4a16 | 2873 | |
4f3640f8 SR |
2874 | /* save the current timestamp, since the user will need it */ |
2875 | save_timestamp = cpu_buffer->read_stamp; | |
2876 | ||
ef7a4a16 SR |
2877 | /* Need to copy one event at a time */ |
2878 | do { | |
474d32b6 | 2879 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
2880 | |
2881 | len -= size; | |
2882 | ||
2883 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
2884 | rpos = reader->read; |
2885 | pos += size; | |
ef7a4a16 SR |
2886 | |
2887 | event = rb_reader_event(cpu_buffer); | |
2888 | size = rb_event_length(event); | |
2889 | } while (len > size); | |
667d2412 LJ |
2890 | |
2891 | /* update bpage */ | |
ef7a4a16 | 2892 | local_set(&bpage->commit, pos); |
4f3640f8 | 2893 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 2894 | |
474d32b6 SR |
2895 | /* we copied everything to the beginning */ |
2896 | read = 0; | |
8789a9e7 SR |
2897 | } else { |
2898 | /* swap the pages */ | |
044fa782 | 2899 | rb_init_page(bpage); |
ef7a4a16 SR |
2900 | bpage = reader->page; |
2901 | reader->page = *data_page; | |
2902 | local_set(&reader->write, 0); | |
2903 | reader->read = 0; | |
044fa782 | 2904 | *data_page = bpage; |
ef7a4a16 SR |
2905 | |
2906 | /* update the entry counter */ | |
2907 | rb_remove_entries(cpu_buffer, bpage, read); | |
8789a9e7 | 2908 | } |
667d2412 | 2909 | ret = read; |
8789a9e7 | 2910 | |
554f786e | 2911 | out_unlock: |
8789a9e7 SR |
2912 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
2913 | ||
554f786e | 2914 | out: |
8789a9e7 SR |
2915 | return ret; |
2916 | } | |
2917 | ||
a3583244 SR |
2918 | static ssize_t |
2919 | rb_simple_read(struct file *filp, char __user *ubuf, | |
2920 | size_t cnt, loff_t *ppos) | |
2921 | { | |
5e39841c | 2922 | unsigned long *p = filp->private_data; |
a3583244 SR |
2923 | char buf[64]; |
2924 | int r; | |
2925 | ||
033601a3 SR |
2926 | if (test_bit(RB_BUFFERS_DISABLED_BIT, p)) |
2927 | r = sprintf(buf, "permanently disabled\n"); | |
2928 | else | |
2929 | r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p)); | |
a3583244 SR |
2930 | |
2931 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
2932 | } | |
2933 | ||
2934 | static ssize_t | |
2935 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
2936 | size_t cnt, loff_t *ppos) | |
2937 | { | |
5e39841c | 2938 | unsigned long *p = filp->private_data; |
a3583244 | 2939 | char buf[64]; |
5e39841c | 2940 | unsigned long val; |
a3583244 SR |
2941 | int ret; |
2942 | ||
2943 | if (cnt >= sizeof(buf)) | |
2944 | return -EINVAL; | |
2945 | ||
2946 | if (copy_from_user(&buf, ubuf, cnt)) | |
2947 | return -EFAULT; | |
2948 | ||
2949 | buf[cnt] = 0; | |
2950 | ||
2951 | ret = strict_strtoul(buf, 10, &val); | |
2952 | if (ret < 0) | |
2953 | return ret; | |
2954 | ||
033601a3 SR |
2955 | if (val) |
2956 | set_bit(RB_BUFFERS_ON_BIT, p); | |
2957 | else | |
2958 | clear_bit(RB_BUFFERS_ON_BIT, p); | |
a3583244 SR |
2959 | |
2960 | (*ppos)++; | |
2961 | ||
2962 | return cnt; | |
2963 | } | |
2964 | ||
5e2336a0 | 2965 | static const struct file_operations rb_simple_fops = { |
a3583244 SR |
2966 | .open = tracing_open_generic, |
2967 | .read = rb_simple_read, | |
2968 | .write = rb_simple_write, | |
2969 | }; | |
2970 | ||
2971 | ||
2972 | static __init int rb_init_debugfs(void) | |
2973 | { | |
2974 | struct dentry *d_tracer; | |
a3583244 SR |
2975 | |
2976 | d_tracer = tracing_init_dentry(); | |
2977 | ||
5452af66 FW |
2978 | trace_create_file("tracing_on", 0644, d_tracer, |
2979 | &ring_buffer_flags, &rb_simple_fops); | |
a3583244 SR |
2980 | |
2981 | return 0; | |
2982 | } | |
2983 | ||
2984 | fs_initcall(rb_init_debugfs); | |
554f786e | 2985 | |
59222efe | 2986 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
2987 | static int rb_cpu_notify(struct notifier_block *self, |
2988 | unsigned long action, void *hcpu) | |
554f786e SR |
2989 | { |
2990 | struct ring_buffer *buffer = | |
2991 | container_of(self, struct ring_buffer, cpu_notify); | |
2992 | long cpu = (long)hcpu; | |
2993 | ||
2994 | switch (action) { | |
2995 | case CPU_UP_PREPARE: | |
2996 | case CPU_UP_PREPARE_FROZEN: | |
2997 | if (cpu_isset(cpu, *buffer->cpumask)) | |
2998 | return NOTIFY_OK; | |
2999 | ||
3000 | buffer->buffers[cpu] = | |
3001 | rb_allocate_cpu_buffer(buffer, cpu); | |
3002 | if (!buffer->buffers[cpu]) { | |
3003 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
3004 | cpu); | |
3005 | return NOTIFY_OK; | |
3006 | } | |
3007 | smp_wmb(); | |
3008 | cpu_set(cpu, *buffer->cpumask); | |
3009 | break; | |
3010 | case CPU_DOWN_PREPARE: | |
3011 | case CPU_DOWN_PREPARE_FROZEN: | |
3012 | /* | |
3013 | * Do nothing. | |
3014 | * If we were to free the buffer, then the user would | |
3015 | * lose any trace that was in the buffer. | |
3016 | */ | |
3017 | break; | |
3018 | default: | |
3019 | break; | |
3020 | } | |
3021 | return NOTIFY_OK; | |
3022 | } | |
3023 | #endif |