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)) { |
3554228d SR |
1244 | /* This can easily happen on small ring buffers */ |
1245 | WARN_ON_ONCE(buffer->pages > 2); | |
45141d46 | 1246 | goto out_reset; |
bf41a158 | 1247 | } |
d769041f | 1248 | |
7a8e76a3 | 1249 | if (next_page == head_page) { |
6f3b3440 | 1250 | if (!(buffer->flags & RB_FL_OVERWRITE)) |
45141d46 | 1251 | goto out_reset; |
7a8e76a3 | 1252 | |
bf41a158 SR |
1253 | /* tail_page has not moved yet? */ |
1254 | if (tail_page == cpu_buffer->tail_page) { | |
1255 | /* count overflows */ | |
1256 | rb_update_overflow(cpu_buffer); | |
1257 | ||
1258 | rb_inc_page(cpu_buffer, &head_page); | |
1259 | cpu_buffer->head_page = head_page; | |
1260 | cpu_buffer->head_page->read = 0; | |
1261 | } | |
1262 | } | |
7a8e76a3 | 1263 | |
bf41a158 SR |
1264 | /* |
1265 | * If the tail page is still the same as what we think | |
1266 | * it is, then it is up to us to update the tail | |
1267 | * pointer. | |
1268 | */ | |
1269 | if (tail_page == cpu_buffer->tail_page) { | |
1270 | local_set(&next_page->write, 0); | |
abc9b56d | 1271 | local_set(&next_page->page->commit, 0); |
bf41a158 SR |
1272 | cpu_buffer->tail_page = next_page; |
1273 | ||
1274 | /* reread the time stamp */ | |
37886f6a | 1275 | *ts = ring_buffer_time_stamp(buffer, cpu_buffer->cpu); |
abc9b56d | 1276 | cpu_buffer->tail_page->page->time_stamp = *ts; |
7a8e76a3 SR |
1277 | } |
1278 | ||
bf41a158 SR |
1279 | /* |
1280 | * The actual tail page has moved forward. | |
1281 | */ | |
1282 | if (tail < BUF_PAGE_SIZE) { | |
1283 | /* Mark the rest of the page with padding */ | |
6f807acd | 1284 | event = __rb_page_index(tail_page, tail); |
2d622719 | 1285 | rb_event_set_padding(event); |
7a8e76a3 SR |
1286 | } |
1287 | ||
bf41a158 SR |
1288 | if (tail <= BUF_PAGE_SIZE) |
1289 | /* Set the write back to the previous setting */ | |
1290 | local_set(&tail_page->write, tail); | |
1291 | ||
1292 | /* | |
1293 | * If this was a commit entry that failed, | |
1294 | * increment that too | |
1295 | */ | |
1296 | if (tail_page == cpu_buffer->commit_page && | |
1297 | tail == rb_commit_index(cpu_buffer)) { | |
1298 | rb_set_commit_to_write(cpu_buffer); | |
1299 | } | |
1300 | ||
3e03fb7f SR |
1301 | __raw_spin_unlock(&cpu_buffer->lock); |
1302 | local_irq_restore(flags); | |
bf41a158 SR |
1303 | |
1304 | /* fail and let the caller try again */ | |
1305 | return ERR_PTR(-EAGAIN); | |
7a8e76a3 SR |
1306 | } |
1307 | ||
bf41a158 SR |
1308 | /* We reserved something on the buffer */ |
1309 | ||
3e89c7bb SR |
1310 | if (RB_WARN_ON(cpu_buffer, write > BUF_PAGE_SIZE)) |
1311 | return NULL; | |
7a8e76a3 | 1312 | |
6f807acd | 1313 | event = __rb_page_index(tail_page, tail); |
7a8e76a3 SR |
1314 | rb_update_event(event, type, length); |
1315 | ||
bf41a158 SR |
1316 | /* |
1317 | * If this is a commit and the tail is zero, then update | |
1318 | * this page's time stamp. | |
1319 | */ | |
1320 | if (!tail && rb_is_commit(cpu_buffer, event)) | |
abc9b56d | 1321 | cpu_buffer->commit_page->page->time_stamp = *ts; |
bf41a158 | 1322 | |
7a8e76a3 | 1323 | return event; |
bf41a158 | 1324 | |
45141d46 | 1325 | out_reset: |
6f3b3440 LJ |
1326 | /* reset write */ |
1327 | if (tail <= BUF_PAGE_SIZE) | |
1328 | local_set(&tail_page->write, tail); | |
1329 | ||
78d904b4 SR |
1330 | if (likely(lock_taken)) |
1331 | __raw_spin_unlock(&cpu_buffer->lock); | |
3e03fb7f | 1332 | local_irq_restore(flags); |
bf41a158 | 1333 | return NULL; |
7a8e76a3 SR |
1334 | } |
1335 | ||
1336 | static int | |
1337 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1338 | u64 *ts, u64 *delta) | |
1339 | { | |
1340 | struct ring_buffer_event *event; | |
1341 | static int once; | |
bf41a158 | 1342 | int ret; |
7a8e76a3 SR |
1343 | |
1344 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1345 | printk(KERN_WARNING "Delta way too big! %llu" | |
1346 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1347 | (unsigned long long)*delta, |
1348 | (unsigned long long)*ts, | |
1349 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1350 | WARN_ON(1); |
1351 | } | |
1352 | ||
1353 | /* | |
1354 | * The delta is too big, we to add a | |
1355 | * new timestamp. | |
1356 | */ | |
1357 | event = __rb_reserve_next(cpu_buffer, | |
1358 | RINGBUF_TYPE_TIME_EXTEND, | |
1359 | RB_LEN_TIME_EXTEND, | |
1360 | ts); | |
1361 | if (!event) | |
bf41a158 | 1362 | return -EBUSY; |
7a8e76a3 | 1363 | |
bf41a158 SR |
1364 | if (PTR_ERR(event) == -EAGAIN) |
1365 | return -EAGAIN; | |
1366 | ||
1367 | /* Only a commited time event can update the write stamp */ | |
1368 | if (rb_is_commit(cpu_buffer, event)) { | |
1369 | /* | |
1370 | * If this is the first on the page, then we need to | |
1371 | * update the page itself, and just put in a zero. | |
1372 | */ | |
1373 | if (rb_event_index(event)) { | |
1374 | event->time_delta = *delta & TS_MASK; | |
1375 | event->array[0] = *delta >> TS_SHIFT; | |
1376 | } else { | |
abc9b56d | 1377 | cpu_buffer->commit_page->page->time_stamp = *ts; |
bf41a158 SR |
1378 | event->time_delta = 0; |
1379 | event->array[0] = 0; | |
1380 | } | |
7a8e76a3 | 1381 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
1382 | /* let the caller know this was the commit */ |
1383 | ret = 1; | |
1384 | } else { | |
1385 | /* Darn, this is just wasted space */ | |
1386 | event->time_delta = 0; | |
1387 | event->array[0] = 0; | |
1388 | ret = 0; | |
7a8e76a3 SR |
1389 | } |
1390 | ||
bf41a158 SR |
1391 | *delta = 0; |
1392 | ||
1393 | return ret; | |
7a8e76a3 SR |
1394 | } |
1395 | ||
1396 | static struct ring_buffer_event * | |
1397 | rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, | |
1398 | unsigned type, unsigned long length) | |
1399 | { | |
1400 | struct ring_buffer_event *event; | |
1401 | u64 ts, delta; | |
bf41a158 | 1402 | int commit = 0; |
818e3dd3 | 1403 | int nr_loops = 0; |
7a8e76a3 | 1404 | |
bf41a158 | 1405 | again: |
818e3dd3 SR |
1406 | /* |
1407 | * We allow for interrupts to reenter here and do a trace. | |
1408 | * If one does, it will cause this original code to loop | |
1409 | * back here. Even with heavy interrupts happening, this | |
1410 | * should only happen a few times in a row. If this happens | |
1411 | * 1000 times in a row, there must be either an interrupt | |
1412 | * storm or we have something buggy. | |
1413 | * Bail! | |
1414 | */ | |
3e89c7bb | 1415 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
818e3dd3 | 1416 | return NULL; |
818e3dd3 | 1417 | |
37886f6a | 1418 | ts = ring_buffer_time_stamp(cpu_buffer->buffer, cpu_buffer->cpu); |
7a8e76a3 | 1419 | |
bf41a158 SR |
1420 | /* |
1421 | * Only the first commit can update the timestamp. | |
1422 | * Yes there is a race here. If an interrupt comes in | |
1423 | * just after the conditional and it traces too, then it | |
1424 | * will also check the deltas. More than one timestamp may | |
1425 | * also be made. But only the entry that did the actual | |
1426 | * commit will be something other than zero. | |
1427 | */ | |
1428 | if (cpu_buffer->tail_page == cpu_buffer->commit_page && | |
1429 | rb_page_write(cpu_buffer->tail_page) == | |
1430 | rb_commit_index(cpu_buffer)) { | |
1431 | ||
7a8e76a3 SR |
1432 | delta = ts - cpu_buffer->write_stamp; |
1433 | ||
bf41a158 SR |
1434 | /* make sure this delta is calculated here */ |
1435 | barrier(); | |
1436 | ||
1437 | /* Did the write stamp get updated already? */ | |
1438 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
4143c5cb | 1439 | delta = 0; |
bf41a158 | 1440 | |
7a8e76a3 | 1441 | if (test_time_stamp(delta)) { |
7a8e76a3 | 1442 | |
bf41a158 SR |
1443 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
1444 | ||
1445 | if (commit == -EBUSY) | |
7a8e76a3 | 1446 | return NULL; |
bf41a158 SR |
1447 | |
1448 | if (commit == -EAGAIN) | |
1449 | goto again; | |
1450 | ||
1451 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 1452 | } |
bf41a158 SR |
1453 | } else |
1454 | /* Non commits have zero deltas */ | |
7a8e76a3 | 1455 | delta = 0; |
7a8e76a3 SR |
1456 | |
1457 | event = __rb_reserve_next(cpu_buffer, type, length, &ts); | |
bf41a158 SR |
1458 | if (PTR_ERR(event) == -EAGAIN) |
1459 | goto again; | |
1460 | ||
1461 | if (!event) { | |
1462 | if (unlikely(commit)) | |
1463 | /* | |
1464 | * Ouch! We needed a timestamp and it was commited. But | |
1465 | * we didn't get our event reserved. | |
1466 | */ | |
1467 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 | 1468 | return NULL; |
bf41a158 | 1469 | } |
7a8e76a3 | 1470 | |
bf41a158 SR |
1471 | /* |
1472 | * If the timestamp was commited, make the commit our entry | |
1473 | * now so that we will update it when needed. | |
1474 | */ | |
1475 | if (commit) | |
1476 | rb_set_commit_event(cpu_buffer, event); | |
1477 | else if (!rb_is_commit(cpu_buffer, event)) | |
7a8e76a3 SR |
1478 | delta = 0; |
1479 | ||
1480 | event->time_delta = delta; | |
1481 | ||
1482 | return event; | |
1483 | } | |
1484 | ||
aa18efb2 | 1485 | #define TRACE_RECURSIVE_DEPTH 16 |
261842b7 SR |
1486 | |
1487 | static int trace_recursive_lock(void) | |
1488 | { | |
aa18efb2 | 1489 | current->trace_recursion++; |
261842b7 | 1490 | |
aa18efb2 SR |
1491 | if (likely(current->trace_recursion < TRACE_RECURSIVE_DEPTH)) |
1492 | return 0; | |
e057a5e5 | 1493 | |
aa18efb2 SR |
1494 | /* Disable all tracing before we do anything else */ |
1495 | tracing_off_permanent(); | |
261842b7 | 1496 | |
aa18efb2 SR |
1497 | printk_once(KERN_WARNING "Tracing recursion: depth[%d]:" |
1498 | "HC[%lu]:SC[%lu]:NMI[%lu]\n", | |
1499 | current->trace_recursion, | |
1500 | hardirq_count() >> HARDIRQ_SHIFT, | |
1501 | softirq_count() >> SOFTIRQ_SHIFT, | |
1502 | in_nmi()); | |
261842b7 | 1503 | |
aa18efb2 SR |
1504 | WARN_ON_ONCE(1); |
1505 | return -1; | |
261842b7 SR |
1506 | } |
1507 | ||
1508 | static void trace_recursive_unlock(void) | |
1509 | { | |
aa18efb2 | 1510 | WARN_ON_ONCE(!current->trace_recursion); |
261842b7 | 1511 | |
aa18efb2 | 1512 | current->trace_recursion--; |
261842b7 SR |
1513 | } |
1514 | ||
bf41a158 SR |
1515 | static DEFINE_PER_CPU(int, rb_need_resched); |
1516 | ||
7a8e76a3 SR |
1517 | /** |
1518 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
1519 | * @buffer: the ring buffer to reserve from | |
1520 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
1521 | * |
1522 | * Returns a reseverd event on the ring buffer to copy directly to. | |
1523 | * The user of this interface will need to get the body to write into | |
1524 | * and can use the ring_buffer_event_data() interface. | |
1525 | * | |
1526 | * The length is the length of the data needed, not the event length | |
1527 | * which also includes the event header. | |
1528 | * | |
1529 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
1530 | * If NULL is returned, then nothing has been allocated or locked. | |
1531 | */ | |
1532 | struct ring_buffer_event * | |
0a987751 | 1533 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
1534 | { |
1535 | struct ring_buffer_per_cpu *cpu_buffer; | |
1536 | struct ring_buffer_event *event; | |
bf41a158 | 1537 | int cpu, resched; |
7a8e76a3 | 1538 | |
033601a3 | 1539 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1540 | return NULL; |
1541 | ||
7a8e76a3 SR |
1542 | if (atomic_read(&buffer->record_disabled)) |
1543 | return NULL; | |
1544 | ||
bf41a158 | 1545 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 1546 | resched = ftrace_preempt_disable(); |
bf41a158 | 1547 | |
261842b7 SR |
1548 | if (trace_recursive_lock()) |
1549 | goto out_nocheck; | |
1550 | ||
7a8e76a3 SR |
1551 | cpu = raw_smp_processor_id(); |
1552 | ||
9e01c1b7 | 1553 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1554 | goto out; |
7a8e76a3 SR |
1555 | |
1556 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1557 | |
1558 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 1559 | goto out; |
7a8e76a3 SR |
1560 | |
1561 | length = rb_calculate_event_length(length); | |
1562 | if (length > BUF_PAGE_SIZE) | |
bf41a158 | 1563 | goto out; |
7a8e76a3 SR |
1564 | |
1565 | event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length); | |
1566 | if (!event) | |
d769041f | 1567 | goto out; |
7a8e76a3 | 1568 | |
bf41a158 SR |
1569 | /* |
1570 | * Need to store resched state on this cpu. | |
1571 | * Only the first needs to. | |
1572 | */ | |
1573 | ||
1574 | if (preempt_count() == 1) | |
1575 | per_cpu(rb_need_resched, cpu) = resched; | |
1576 | ||
7a8e76a3 SR |
1577 | return event; |
1578 | ||
d769041f | 1579 | out: |
261842b7 SR |
1580 | trace_recursive_unlock(); |
1581 | ||
1582 | out_nocheck: | |
182e9f5f | 1583 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1584 | return NULL; |
1585 | } | |
c4f50183 | 1586 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 SR |
1587 | |
1588 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
1589 | struct ring_buffer_event *event) | |
1590 | { | |
7a8e76a3 | 1591 | cpu_buffer->entries++; |
bf41a158 SR |
1592 | |
1593 | /* Only process further if we own the commit */ | |
1594 | if (!rb_is_commit(cpu_buffer, event)) | |
1595 | return; | |
1596 | ||
1597 | cpu_buffer->write_stamp += event->time_delta; | |
1598 | ||
1599 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 SR |
1600 | } |
1601 | ||
1602 | /** | |
1603 | * ring_buffer_unlock_commit - commit a reserved | |
1604 | * @buffer: The buffer to commit to | |
1605 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
1606 | * |
1607 | * This commits the data to the ring buffer, and releases any locks held. | |
1608 | * | |
1609 | * Must be paired with ring_buffer_lock_reserve. | |
1610 | */ | |
1611 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 1612 | struct ring_buffer_event *event) |
7a8e76a3 SR |
1613 | { |
1614 | struct ring_buffer_per_cpu *cpu_buffer; | |
1615 | int cpu = raw_smp_processor_id(); | |
1616 | ||
1617 | cpu_buffer = buffer->buffers[cpu]; | |
1618 | ||
7a8e76a3 SR |
1619 | rb_commit(cpu_buffer, event); |
1620 | ||
261842b7 SR |
1621 | trace_recursive_unlock(); |
1622 | ||
bf41a158 SR |
1623 | /* |
1624 | * Only the last preempt count needs to restore preemption. | |
1625 | */ | |
182e9f5f SR |
1626 | if (preempt_count() == 1) |
1627 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1628 | else | |
bf41a158 | 1629 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
1630 | |
1631 | return 0; | |
1632 | } | |
c4f50183 | 1633 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 1634 | |
f3b9aae1 FW |
1635 | static inline void rb_event_discard(struct ring_buffer_event *event) |
1636 | { | |
1637 | event->type = RINGBUF_TYPE_PADDING; | |
1638 | /* time delta must be non zero */ | |
1639 | if (!event->time_delta) | |
1640 | event->time_delta = 1; | |
1641 | } | |
1642 | ||
fa1b47dd SR |
1643 | /** |
1644 | * ring_buffer_event_discard - discard any event in the ring buffer | |
1645 | * @event: the event to discard | |
1646 | * | |
1647 | * Sometimes a event that is in the ring buffer needs to be ignored. | |
1648 | * This function lets the user discard an event in the ring buffer | |
1649 | * and then that event will not be read later. | |
1650 | * | |
1651 | * Note, it is up to the user to be careful with this, and protect | |
1652 | * against races. If the user discards an event that has been consumed | |
1653 | * it is possible that it could corrupt the ring buffer. | |
1654 | */ | |
1655 | void ring_buffer_event_discard(struct ring_buffer_event *event) | |
1656 | { | |
f3b9aae1 | 1657 | rb_event_discard(event); |
fa1b47dd SR |
1658 | } |
1659 | EXPORT_SYMBOL_GPL(ring_buffer_event_discard); | |
1660 | ||
1661 | /** | |
1662 | * ring_buffer_commit_discard - discard an event that has not been committed | |
1663 | * @buffer: the ring buffer | |
1664 | * @event: non committed event to discard | |
1665 | * | |
1666 | * This is similar to ring_buffer_event_discard but must only be | |
1667 | * performed on an event that has not been committed yet. The difference | |
1668 | * is that this will also try to free the event from the ring buffer | |
1669 | * if another event has not been added behind it. | |
1670 | * | |
1671 | * If another event has been added behind it, it will set the event | |
1672 | * up as discarded, and perform the commit. | |
1673 | * | |
1674 | * If this function is called, do not call ring_buffer_unlock_commit on | |
1675 | * the event. | |
1676 | */ | |
1677 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
1678 | struct ring_buffer_event *event) | |
1679 | { | |
1680 | struct ring_buffer_per_cpu *cpu_buffer; | |
1681 | unsigned long new_index, old_index; | |
1682 | struct buffer_page *bpage; | |
1683 | unsigned long index; | |
1684 | unsigned long addr; | |
1685 | int cpu; | |
1686 | ||
1687 | /* The event is discarded regardless */ | |
f3b9aae1 | 1688 | rb_event_discard(event); |
fa1b47dd SR |
1689 | |
1690 | /* | |
1691 | * This must only be called if the event has not been | |
1692 | * committed yet. Thus we can assume that preemption | |
1693 | * is still disabled. | |
1694 | */ | |
1695 | RB_WARN_ON(buffer, !preempt_count()); | |
1696 | ||
1697 | cpu = smp_processor_id(); | |
1698 | cpu_buffer = buffer->buffers[cpu]; | |
1699 | ||
1700 | new_index = rb_event_index(event); | |
1701 | old_index = new_index + rb_event_length(event); | |
1702 | addr = (unsigned long)event; | |
1703 | addr &= PAGE_MASK; | |
1704 | ||
1705 | bpage = cpu_buffer->tail_page; | |
1706 | ||
1707 | if (bpage == (void *)addr && rb_page_write(bpage) == old_index) { | |
1708 | /* | |
1709 | * This is on the tail page. It is possible that | |
1710 | * a write could come in and move the tail page | |
1711 | * and write to the next page. That is fine | |
1712 | * because we just shorten what is on this page. | |
1713 | */ | |
1714 | index = local_cmpxchg(&bpage->write, old_index, new_index); | |
1715 | if (index == old_index) | |
1716 | goto out; | |
1717 | } | |
1718 | ||
1719 | /* | |
1720 | * The commit is still visible by the reader, so we | |
1721 | * must increment entries. | |
1722 | */ | |
1723 | cpu_buffer->entries++; | |
1724 | out: | |
1725 | /* | |
1726 | * If a write came in and pushed the tail page | |
1727 | * we still need to update the commit pointer | |
1728 | * if we were the commit. | |
1729 | */ | |
1730 | if (rb_is_commit(cpu_buffer, event)) | |
1731 | rb_set_commit_to_write(cpu_buffer); | |
1732 | ||
f3b9aae1 FW |
1733 | trace_recursive_unlock(); |
1734 | ||
fa1b47dd SR |
1735 | /* |
1736 | * Only the last preempt count needs to restore preemption. | |
1737 | */ | |
1738 | if (preempt_count() == 1) | |
1739 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1740 | else | |
1741 | preempt_enable_no_resched_notrace(); | |
1742 | ||
1743 | } | |
1744 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
1745 | ||
7a8e76a3 SR |
1746 | /** |
1747 | * ring_buffer_write - write data to the buffer without reserving | |
1748 | * @buffer: The ring buffer to write to. | |
1749 | * @length: The length of the data being written (excluding the event header) | |
1750 | * @data: The data to write to the buffer. | |
1751 | * | |
1752 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
1753 | * one function. If you already have the data to write to the buffer, it | |
1754 | * may be easier to simply call this function. | |
1755 | * | |
1756 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
1757 | * and not the length of the event which would hold the header. | |
1758 | */ | |
1759 | int ring_buffer_write(struct ring_buffer *buffer, | |
1760 | unsigned long length, | |
1761 | void *data) | |
1762 | { | |
1763 | struct ring_buffer_per_cpu *cpu_buffer; | |
1764 | struct ring_buffer_event *event; | |
bf41a158 | 1765 | unsigned long event_length; |
7a8e76a3 SR |
1766 | void *body; |
1767 | int ret = -EBUSY; | |
bf41a158 | 1768 | int cpu, resched; |
7a8e76a3 | 1769 | |
033601a3 | 1770 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1771 | return -EBUSY; |
1772 | ||
7a8e76a3 SR |
1773 | if (atomic_read(&buffer->record_disabled)) |
1774 | return -EBUSY; | |
1775 | ||
182e9f5f | 1776 | resched = ftrace_preempt_disable(); |
bf41a158 | 1777 | |
7a8e76a3 SR |
1778 | cpu = raw_smp_processor_id(); |
1779 | ||
9e01c1b7 | 1780 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1781 | goto out; |
7a8e76a3 SR |
1782 | |
1783 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1784 | |
1785 | if (atomic_read(&cpu_buffer->record_disabled)) | |
1786 | goto out; | |
1787 | ||
1788 | event_length = rb_calculate_event_length(length); | |
1789 | event = rb_reserve_next_event(cpu_buffer, | |
1790 | RINGBUF_TYPE_DATA, event_length); | |
1791 | if (!event) | |
1792 | goto out; | |
1793 | ||
1794 | body = rb_event_data(event); | |
1795 | ||
1796 | memcpy(body, data, length); | |
1797 | ||
1798 | rb_commit(cpu_buffer, event); | |
1799 | ||
1800 | ret = 0; | |
1801 | out: | |
182e9f5f | 1802 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1803 | |
1804 | return ret; | |
1805 | } | |
c4f50183 | 1806 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 1807 | |
34a148bf | 1808 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
1809 | { |
1810 | struct buffer_page *reader = cpu_buffer->reader_page; | |
1811 | struct buffer_page *head = cpu_buffer->head_page; | |
1812 | struct buffer_page *commit = cpu_buffer->commit_page; | |
1813 | ||
1814 | return reader->read == rb_page_commit(reader) && | |
1815 | (commit == reader || | |
1816 | (commit == head && | |
1817 | head->read == rb_page_commit(commit))); | |
1818 | } | |
1819 | ||
7a8e76a3 SR |
1820 | /** |
1821 | * ring_buffer_record_disable - stop all writes into the buffer | |
1822 | * @buffer: The ring buffer to stop writes to. | |
1823 | * | |
1824 | * This prevents all writes to the buffer. Any attempt to write | |
1825 | * to the buffer after this will fail and return NULL. | |
1826 | * | |
1827 | * The caller should call synchronize_sched() after this. | |
1828 | */ | |
1829 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
1830 | { | |
1831 | atomic_inc(&buffer->record_disabled); | |
1832 | } | |
c4f50183 | 1833 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
1834 | |
1835 | /** | |
1836 | * ring_buffer_record_enable - enable writes to the buffer | |
1837 | * @buffer: The ring buffer to enable writes | |
1838 | * | |
1839 | * Note, multiple disables will need the same number of enables | |
1840 | * to truely enable the writing (much like preempt_disable). | |
1841 | */ | |
1842 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
1843 | { | |
1844 | atomic_dec(&buffer->record_disabled); | |
1845 | } | |
c4f50183 | 1846 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 SR |
1847 | |
1848 | /** | |
1849 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
1850 | * @buffer: The ring buffer to stop writes to. | |
1851 | * @cpu: The CPU buffer to stop | |
1852 | * | |
1853 | * This prevents all writes to the buffer. Any attempt to write | |
1854 | * to the buffer after this will fail and return NULL. | |
1855 | * | |
1856 | * The caller should call synchronize_sched() after this. | |
1857 | */ | |
1858 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
1859 | { | |
1860 | struct ring_buffer_per_cpu *cpu_buffer; | |
1861 | ||
9e01c1b7 | 1862 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1863 | return; |
7a8e76a3 SR |
1864 | |
1865 | cpu_buffer = buffer->buffers[cpu]; | |
1866 | atomic_inc(&cpu_buffer->record_disabled); | |
1867 | } | |
c4f50183 | 1868 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
1869 | |
1870 | /** | |
1871 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
1872 | * @buffer: The ring buffer to enable writes | |
1873 | * @cpu: The CPU to enable. | |
1874 | * | |
1875 | * Note, multiple disables will need the same number of enables | |
1876 | * to truely enable the writing (much like preempt_disable). | |
1877 | */ | |
1878 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
1879 | { | |
1880 | struct ring_buffer_per_cpu *cpu_buffer; | |
1881 | ||
9e01c1b7 | 1882 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1883 | return; |
7a8e76a3 SR |
1884 | |
1885 | cpu_buffer = buffer->buffers[cpu]; | |
1886 | atomic_dec(&cpu_buffer->record_disabled); | |
1887 | } | |
c4f50183 | 1888 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 SR |
1889 | |
1890 | /** | |
1891 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
1892 | * @buffer: The ring buffer | |
1893 | * @cpu: The per CPU buffer to get the entries from. | |
1894 | */ | |
1895 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
1896 | { | |
1897 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 1898 | unsigned long ret; |
7a8e76a3 | 1899 | |
9e01c1b7 | 1900 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1901 | return 0; |
7a8e76a3 SR |
1902 | |
1903 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 1904 | ret = cpu_buffer->entries; |
554f786e SR |
1905 | |
1906 | return ret; | |
7a8e76a3 | 1907 | } |
c4f50183 | 1908 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
1909 | |
1910 | /** | |
1911 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
1912 | * @buffer: The ring buffer | |
1913 | * @cpu: The per CPU buffer to get the number of overruns from | |
1914 | */ | |
1915 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
1916 | { | |
1917 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 1918 | unsigned long ret; |
7a8e76a3 | 1919 | |
9e01c1b7 | 1920 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1921 | return 0; |
7a8e76a3 SR |
1922 | |
1923 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 1924 | ret = cpu_buffer->overrun; |
554f786e SR |
1925 | |
1926 | return ret; | |
7a8e76a3 | 1927 | } |
c4f50183 | 1928 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 SR |
1929 | |
1930 | /** | |
1931 | * ring_buffer_entries - get the number of entries in a buffer | |
1932 | * @buffer: The ring buffer | |
1933 | * | |
1934 | * Returns the total number of entries in the ring buffer | |
1935 | * (all CPU entries) | |
1936 | */ | |
1937 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
1938 | { | |
1939 | struct ring_buffer_per_cpu *cpu_buffer; | |
1940 | unsigned long entries = 0; | |
1941 | int cpu; | |
1942 | ||
1943 | /* if you care about this being correct, lock the buffer */ | |
1944 | for_each_buffer_cpu(buffer, cpu) { | |
1945 | cpu_buffer = buffer->buffers[cpu]; | |
1946 | entries += cpu_buffer->entries; | |
1947 | } | |
1948 | ||
1949 | return entries; | |
1950 | } | |
c4f50183 | 1951 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
1952 | |
1953 | /** | |
1954 | * ring_buffer_overrun_cpu - get the number of overruns in buffer | |
1955 | * @buffer: The ring buffer | |
1956 | * | |
1957 | * Returns the total number of overruns in the ring buffer | |
1958 | * (all CPU entries) | |
1959 | */ | |
1960 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
1961 | { | |
1962 | struct ring_buffer_per_cpu *cpu_buffer; | |
1963 | unsigned long overruns = 0; | |
1964 | int cpu; | |
1965 | ||
1966 | /* if you care about this being correct, lock the buffer */ | |
1967 | for_each_buffer_cpu(buffer, cpu) { | |
1968 | cpu_buffer = buffer->buffers[cpu]; | |
1969 | overruns += cpu_buffer->overrun; | |
1970 | } | |
1971 | ||
1972 | return overruns; | |
1973 | } | |
c4f50183 | 1974 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 1975 | |
642edba5 | 1976 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
1977 | { |
1978 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1979 | ||
d769041f SR |
1980 | /* Iterator usage is expected to have record disabled */ |
1981 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
1982 | iter->head_page = cpu_buffer->head_page; | |
6f807acd | 1983 | iter->head = cpu_buffer->head_page->read; |
d769041f SR |
1984 | } else { |
1985 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 1986 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
1987 | } |
1988 | if (iter->head) | |
1989 | iter->read_stamp = cpu_buffer->read_stamp; | |
1990 | else | |
abc9b56d | 1991 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 1992 | } |
f83c9d0f | 1993 | |
642edba5 SR |
1994 | /** |
1995 | * ring_buffer_iter_reset - reset an iterator | |
1996 | * @iter: The iterator to reset | |
1997 | * | |
1998 | * Resets the iterator, so that it will start from the beginning | |
1999 | * again. | |
2000 | */ | |
2001 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
2002 | { | |
554f786e | 2003 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
2004 | unsigned long flags; |
2005 | ||
554f786e SR |
2006 | if (!iter) |
2007 | return; | |
2008 | ||
2009 | cpu_buffer = iter->cpu_buffer; | |
2010 | ||
642edba5 SR |
2011 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2012 | rb_iter_reset(iter); | |
f83c9d0f | 2013 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 2014 | } |
c4f50183 | 2015 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
2016 | |
2017 | /** | |
2018 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
2019 | * @iter: The iterator to check | |
2020 | */ | |
2021 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
2022 | { | |
2023 | struct ring_buffer_per_cpu *cpu_buffer; | |
2024 | ||
2025 | cpu_buffer = iter->cpu_buffer; | |
2026 | ||
bf41a158 SR |
2027 | return iter->head_page == cpu_buffer->commit_page && |
2028 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 2029 | } |
c4f50183 | 2030 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
2031 | |
2032 | static void | |
2033 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
2034 | struct ring_buffer_event *event) | |
2035 | { | |
2036 | u64 delta; | |
2037 | ||
2038 | switch (event->type) { | |
2039 | case RINGBUF_TYPE_PADDING: | |
2040 | return; | |
2041 | ||
2042 | case RINGBUF_TYPE_TIME_EXTEND: | |
2043 | delta = event->array[0]; | |
2044 | delta <<= TS_SHIFT; | |
2045 | delta += event->time_delta; | |
2046 | cpu_buffer->read_stamp += delta; | |
2047 | return; | |
2048 | ||
2049 | case RINGBUF_TYPE_TIME_STAMP: | |
2050 | /* FIXME: not implemented */ | |
2051 | return; | |
2052 | ||
2053 | case RINGBUF_TYPE_DATA: | |
2054 | cpu_buffer->read_stamp += event->time_delta; | |
2055 | return; | |
2056 | ||
2057 | default: | |
2058 | BUG(); | |
2059 | } | |
2060 | return; | |
2061 | } | |
2062 | ||
2063 | static void | |
2064 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
2065 | struct ring_buffer_event *event) | |
2066 | { | |
2067 | u64 delta; | |
2068 | ||
2069 | switch (event->type) { | |
2070 | case RINGBUF_TYPE_PADDING: | |
2071 | return; | |
2072 | ||
2073 | case RINGBUF_TYPE_TIME_EXTEND: | |
2074 | delta = event->array[0]; | |
2075 | delta <<= TS_SHIFT; | |
2076 | delta += event->time_delta; | |
2077 | iter->read_stamp += delta; | |
2078 | return; | |
2079 | ||
2080 | case RINGBUF_TYPE_TIME_STAMP: | |
2081 | /* FIXME: not implemented */ | |
2082 | return; | |
2083 | ||
2084 | case RINGBUF_TYPE_DATA: | |
2085 | iter->read_stamp += event->time_delta; | |
2086 | return; | |
2087 | ||
2088 | default: | |
2089 | BUG(); | |
2090 | } | |
2091 | return; | |
2092 | } | |
2093 | ||
d769041f SR |
2094 | static struct buffer_page * |
2095 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 2096 | { |
d769041f SR |
2097 | struct buffer_page *reader = NULL; |
2098 | unsigned long flags; | |
818e3dd3 | 2099 | int nr_loops = 0; |
d769041f | 2100 | |
3e03fb7f SR |
2101 | local_irq_save(flags); |
2102 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
2103 | |
2104 | again: | |
818e3dd3 SR |
2105 | /* |
2106 | * This should normally only loop twice. But because the | |
2107 | * start of the reader inserts an empty page, it causes | |
2108 | * a case where we will loop three times. There should be no | |
2109 | * reason to loop four times (that I know of). | |
2110 | */ | |
3e89c7bb | 2111 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
2112 | reader = NULL; |
2113 | goto out; | |
2114 | } | |
2115 | ||
d769041f SR |
2116 | reader = cpu_buffer->reader_page; |
2117 | ||
2118 | /* If there's more to read, return this page */ | |
bf41a158 | 2119 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
2120 | goto out; |
2121 | ||
2122 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
2123 | if (RB_WARN_ON(cpu_buffer, |
2124 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
2125 | goto out; | |
d769041f SR |
2126 | |
2127 | /* check if we caught up to the tail */ | |
2128 | reader = NULL; | |
bf41a158 | 2129 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 2130 | goto out; |
7a8e76a3 SR |
2131 | |
2132 | /* | |
d769041f SR |
2133 | * Splice the empty reader page into the list around the head. |
2134 | * Reset the reader page to size zero. | |
7a8e76a3 | 2135 | */ |
7a8e76a3 | 2136 | |
d769041f SR |
2137 | reader = cpu_buffer->head_page; |
2138 | cpu_buffer->reader_page->list.next = reader->list.next; | |
2139 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 SR |
2140 | |
2141 | local_set(&cpu_buffer->reader_page->write, 0); | |
abc9b56d | 2142 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
7a8e76a3 | 2143 | |
d769041f SR |
2144 | /* Make the reader page now replace the head */ |
2145 | reader->list.prev->next = &cpu_buffer->reader_page->list; | |
2146 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
7a8e76a3 SR |
2147 | |
2148 | /* | |
d769041f SR |
2149 | * If the tail is on the reader, then we must set the head |
2150 | * to the inserted page, otherwise we set it one before. | |
7a8e76a3 | 2151 | */ |
d769041f | 2152 | cpu_buffer->head_page = cpu_buffer->reader_page; |
7a8e76a3 | 2153 | |
bf41a158 | 2154 | if (cpu_buffer->commit_page != reader) |
d769041f SR |
2155 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
2156 | ||
2157 | /* Finally update the reader page to the new head */ | |
2158 | cpu_buffer->reader_page = reader; | |
2159 | rb_reset_reader_page(cpu_buffer); | |
2160 | ||
2161 | goto again; | |
2162 | ||
2163 | out: | |
3e03fb7f SR |
2164 | __raw_spin_unlock(&cpu_buffer->lock); |
2165 | local_irq_restore(flags); | |
d769041f SR |
2166 | |
2167 | return reader; | |
2168 | } | |
2169 | ||
2170 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
2171 | { | |
2172 | struct ring_buffer_event *event; | |
2173 | struct buffer_page *reader; | |
2174 | unsigned length; | |
2175 | ||
2176 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 2177 | |
d769041f | 2178 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
2179 | if (RB_WARN_ON(cpu_buffer, !reader)) |
2180 | return; | |
7a8e76a3 | 2181 | |
d769041f SR |
2182 | event = rb_reader_event(cpu_buffer); |
2183 | ||
2d622719 | 2184 | if (event->type == RINGBUF_TYPE_DATA || rb_discarded_event(event)) |
d769041f SR |
2185 | cpu_buffer->entries--; |
2186 | ||
2187 | rb_update_read_stamp(cpu_buffer, event); | |
2188 | ||
2189 | length = rb_event_length(event); | |
6f807acd | 2190 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
2191 | } |
2192 | ||
2193 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
2194 | { | |
2195 | struct ring_buffer *buffer; | |
2196 | struct ring_buffer_per_cpu *cpu_buffer; | |
2197 | struct ring_buffer_event *event; | |
2198 | unsigned length; | |
2199 | ||
2200 | cpu_buffer = iter->cpu_buffer; | |
2201 | buffer = cpu_buffer->buffer; | |
2202 | ||
2203 | /* | |
2204 | * Check if we are at the end of the buffer. | |
2205 | */ | |
bf41a158 | 2206 | if (iter->head >= rb_page_size(iter->head_page)) { |
3e89c7bb SR |
2207 | if (RB_WARN_ON(buffer, |
2208 | iter->head_page == cpu_buffer->commit_page)) | |
2209 | return; | |
d769041f | 2210 | rb_inc_iter(iter); |
7a8e76a3 SR |
2211 | return; |
2212 | } | |
2213 | ||
2214 | event = rb_iter_head_event(iter); | |
2215 | ||
2216 | length = rb_event_length(event); | |
2217 | ||
2218 | /* | |
2219 | * This should not be called to advance the header if we are | |
2220 | * at the tail of the buffer. | |
2221 | */ | |
3e89c7bb | 2222 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 2223 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
2224 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
2225 | return; | |
7a8e76a3 SR |
2226 | |
2227 | rb_update_iter_read_stamp(iter, event); | |
2228 | ||
2229 | iter->head += length; | |
2230 | ||
2231 | /* check for end of page padding */ | |
bf41a158 SR |
2232 | if ((iter->head >= rb_page_size(iter->head_page)) && |
2233 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
2234 | rb_advance_iter(iter); |
2235 | } | |
2236 | ||
f83c9d0f SR |
2237 | static struct ring_buffer_event * |
2238 | rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
7a8e76a3 SR |
2239 | { |
2240 | struct ring_buffer_per_cpu *cpu_buffer; | |
2241 | struct ring_buffer_event *event; | |
d769041f | 2242 | struct buffer_page *reader; |
818e3dd3 | 2243 | int nr_loops = 0; |
7a8e76a3 | 2244 | |
7a8e76a3 SR |
2245 | cpu_buffer = buffer->buffers[cpu]; |
2246 | ||
2247 | again: | |
818e3dd3 SR |
2248 | /* |
2249 | * We repeat when a timestamp is encountered. It is possible | |
2250 | * to get multiple timestamps from an interrupt entering just | |
2251 | * as one timestamp is about to be written. The max times | |
2252 | * that this can happen is the number of nested interrupts we | |
2253 | * can have. Nesting 10 deep of interrupts is clearly | |
2254 | * an anomaly. | |
2255 | */ | |
3e89c7bb | 2256 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 2257 | return NULL; |
818e3dd3 | 2258 | |
d769041f SR |
2259 | reader = rb_get_reader_page(cpu_buffer); |
2260 | if (!reader) | |
7a8e76a3 SR |
2261 | return NULL; |
2262 | ||
d769041f | 2263 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 SR |
2264 | |
2265 | switch (event->type) { | |
2266 | case RINGBUF_TYPE_PADDING: | |
2d622719 TZ |
2267 | if (rb_null_event(event)) |
2268 | RB_WARN_ON(cpu_buffer, 1); | |
2269 | /* | |
2270 | * Because the writer could be discarding every | |
2271 | * event it creates (which would probably be bad) | |
2272 | * if we were to go back to "again" then we may never | |
2273 | * catch up, and will trigger the warn on, or lock | |
2274 | * the box. Return the padding, and we will release | |
2275 | * the current locks, and try again. | |
2276 | */ | |
d769041f | 2277 | rb_advance_reader(cpu_buffer); |
2d622719 | 2278 | return event; |
7a8e76a3 SR |
2279 | |
2280 | case RINGBUF_TYPE_TIME_EXTEND: | |
2281 | /* Internal data, OK to advance */ | |
d769041f | 2282 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2283 | goto again; |
2284 | ||
2285 | case RINGBUF_TYPE_TIME_STAMP: | |
2286 | /* FIXME: not implemented */ | |
d769041f | 2287 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2288 | goto again; |
2289 | ||
2290 | case RINGBUF_TYPE_DATA: | |
2291 | if (ts) { | |
2292 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
37886f6a SR |
2293 | ring_buffer_normalize_time_stamp(buffer, |
2294 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2295 | } |
2296 | return event; | |
2297 | ||
2298 | default: | |
2299 | BUG(); | |
2300 | } | |
2301 | ||
2302 | return NULL; | |
2303 | } | |
c4f50183 | 2304 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 2305 | |
f83c9d0f SR |
2306 | static struct ring_buffer_event * |
2307 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
2308 | { |
2309 | struct ring_buffer *buffer; | |
2310 | struct ring_buffer_per_cpu *cpu_buffer; | |
2311 | struct ring_buffer_event *event; | |
818e3dd3 | 2312 | int nr_loops = 0; |
7a8e76a3 SR |
2313 | |
2314 | if (ring_buffer_iter_empty(iter)) | |
2315 | return NULL; | |
2316 | ||
2317 | cpu_buffer = iter->cpu_buffer; | |
2318 | buffer = cpu_buffer->buffer; | |
2319 | ||
2320 | again: | |
818e3dd3 SR |
2321 | /* |
2322 | * We repeat when a timestamp is encountered. It is possible | |
2323 | * to get multiple timestamps from an interrupt entering just | |
2324 | * as one timestamp is about to be written. The max times | |
2325 | * that this can happen is the number of nested interrupts we | |
2326 | * can have. Nesting 10 deep of interrupts is clearly | |
2327 | * an anomaly. | |
2328 | */ | |
3e89c7bb | 2329 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 2330 | return NULL; |
818e3dd3 | 2331 | |
7a8e76a3 SR |
2332 | if (rb_per_cpu_empty(cpu_buffer)) |
2333 | return NULL; | |
2334 | ||
2335 | event = rb_iter_head_event(iter); | |
2336 | ||
2337 | switch (event->type) { | |
2338 | case RINGBUF_TYPE_PADDING: | |
2d622719 TZ |
2339 | if (rb_null_event(event)) { |
2340 | rb_inc_iter(iter); | |
2341 | goto again; | |
2342 | } | |
2343 | rb_advance_iter(iter); | |
2344 | return event; | |
7a8e76a3 SR |
2345 | |
2346 | case RINGBUF_TYPE_TIME_EXTEND: | |
2347 | /* Internal data, OK to advance */ | |
2348 | rb_advance_iter(iter); | |
2349 | goto again; | |
2350 | ||
2351 | case RINGBUF_TYPE_TIME_STAMP: | |
2352 | /* FIXME: not implemented */ | |
2353 | rb_advance_iter(iter); | |
2354 | goto again; | |
2355 | ||
2356 | case RINGBUF_TYPE_DATA: | |
2357 | if (ts) { | |
2358 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
2359 | ring_buffer_normalize_time_stamp(buffer, |
2360 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2361 | } |
2362 | return event; | |
2363 | ||
2364 | default: | |
2365 | BUG(); | |
2366 | } | |
2367 | ||
2368 | return NULL; | |
2369 | } | |
c4f50183 | 2370 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 2371 | |
f83c9d0f SR |
2372 | /** |
2373 | * ring_buffer_peek - peek at the next event to be read | |
2374 | * @buffer: The ring buffer to read | |
2375 | * @cpu: The cpu to peak at | |
2376 | * @ts: The timestamp counter of this event. | |
2377 | * | |
2378 | * This will return the event that will be read next, but does | |
2379 | * not consume the data. | |
2380 | */ | |
2381 | struct ring_buffer_event * | |
2382 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2383 | { | |
2384 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 2385 | struct ring_buffer_event *event; |
f83c9d0f SR |
2386 | unsigned long flags; |
2387 | ||
554f786e | 2388 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2389 | return NULL; |
554f786e | 2390 | |
2d622719 | 2391 | again: |
f83c9d0f SR |
2392 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2393 | event = rb_buffer_peek(buffer, cpu, ts); | |
2394 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2395 | ||
2d622719 TZ |
2396 | if (event && event->type == RINGBUF_TYPE_PADDING) { |
2397 | cpu_relax(); | |
2398 | goto again; | |
2399 | } | |
2400 | ||
f83c9d0f SR |
2401 | return event; |
2402 | } | |
2403 | ||
2404 | /** | |
2405 | * ring_buffer_iter_peek - peek at the next event to be read | |
2406 | * @iter: The ring buffer iterator | |
2407 | * @ts: The timestamp counter of this event. | |
2408 | * | |
2409 | * This will return the event that will be read next, but does | |
2410 | * not increment the iterator. | |
2411 | */ | |
2412 | struct ring_buffer_event * | |
2413 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
2414 | { | |
2415 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2416 | struct ring_buffer_event *event; | |
2417 | unsigned long flags; | |
2418 | ||
2d622719 | 2419 | again: |
f83c9d0f SR |
2420 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2421 | event = rb_iter_peek(iter, ts); | |
2422 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2423 | ||
2d622719 TZ |
2424 | if (event && event->type == RINGBUF_TYPE_PADDING) { |
2425 | cpu_relax(); | |
2426 | goto again; | |
2427 | } | |
2428 | ||
f83c9d0f SR |
2429 | return event; |
2430 | } | |
2431 | ||
7a8e76a3 SR |
2432 | /** |
2433 | * ring_buffer_consume - return an event and consume it | |
2434 | * @buffer: The ring buffer to get the next event from | |
2435 | * | |
2436 | * Returns the next event in the ring buffer, and that event is consumed. | |
2437 | * Meaning, that sequential reads will keep returning a different event, | |
2438 | * and eventually empty the ring buffer if the producer is slower. | |
2439 | */ | |
2440 | struct ring_buffer_event * | |
2441 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2442 | { | |
554f786e SR |
2443 | struct ring_buffer_per_cpu *cpu_buffer; |
2444 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 2445 | unsigned long flags; |
7a8e76a3 | 2446 | |
2d622719 | 2447 | again: |
554f786e SR |
2448 | /* might be called in atomic */ |
2449 | preempt_disable(); | |
2450 | ||
9e01c1b7 | 2451 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 2452 | goto out; |
7a8e76a3 | 2453 | |
554f786e | 2454 | cpu_buffer = buffer->buffers[cpu]; |
f83c9d0f SR |
2455 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2456 | ||
2457 | event = rb_buffer_peek(buffer, cpu, ts); | |
7a8e76a3 | 2458 | if (!event) |
554f786e | 2459 | goto out_unlock; |
7a8e76a3 | 2460 | |
d769041f | 2461 | rb_advance_reader(cpu_buffer); |
7a8e76a3 | 2462 | |
554f786e | 2463 | out_unlock: |
f83c9d0f SR |
2464 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
2465 | ||
554f786e SR |
2466 | out: |
2467 | preempt_enable(); | |
2468 | ||
2d622719 TZ |
2469 | if (event && event->type == RINGBUF_TYPE_PADDING) { |
2470 | cpu_relax(); | |
2471 | goto again; | |
2472 | } | |
2473 | ||
7a8e76a3 SR |
2474 | return event; |
2475 | } | |
c4f50183 | 2476 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
2477 | |
2478 | /** | |
2479 | * ring_buffer_read_start - start a non consuming read of the buffer | |
2480 | * @buffer: The ring buffer to read from | |
2481 | * @cpu: The cpu buffer to iterate over | |
2482 | * | |
2483 | * This starts up an iteration through the buffer. It also disables | |
2484 | * the recording to the buffer until the reading is finished. | |
2485 | * This prevents the reading from being corrupted. This is not | |
2486 | * a consuming read, so a producer is not expected. | |
2487 | * | |
2488 | * Must be paired with ring_buffer_finish. | |
2489 | */ | |
2490 | struct ring_buffer_iter * | |
2491 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
2492 | { | |
2493 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2494 | struct ring_buffer_iter *iter; |
d769041f | 2495 | unsigned long flags; |
7a8e76a3 | 2496 | |
9e01c1b7 | 2497 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2498 | return NULL; |
7a8e76a3 SR |
2499 | |
2500 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
2501 | if (!iter) | |
8aabee57 | 2502 | return NULL; |
7a8e76a3 SR |
2503 | |
2504 | cpu_buffer = buffer->buffers[cpu]; | |
2505 | ||
2506 | iter->cpu_buffer = cpu_buffer; | |
2507 | ||
2508 | atomic_inc(&cpu_buffer->record_disabled); | |
2509 | synchronize_sched(); | |
2510 | ||
f83c9d0f | 2511 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 2512 | __raw_spin_lock(&cpu_buffer->lock); |
642edba5 | 2513 | rb_iter_reset(iter); |
3e03fb7f | 2514 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 2515 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
2516 | |
2517 | return iter; | |
2518 | } | |
c4f50183 | 2519 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
2520 | |
2521 | /** | |
2522 | * ring_buffer_finish - finish reading the iterator of the buffer | |
2523 | * @iter: The iterator retrieved by ring_buffer_start | |
2524 | * | |
2525 | * This re-enables the recording to the buffer, and frees the | |
2526 | * iterator. | |
2527 | */ | |
2528 | void | |
2529 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
2530 | { | |
2531 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2532 | ||
2533 | atomic_dec(&cpu_buffer->record_disabled); | |
2534 | kfree(iter); | |
2535 | } | |
c4f50183 | 2536 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
2537 | |
2538 | /** | |
2539 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
2540 | * @iter: The ring buffer iterator | |
2541 | * @ts: The time stamp of the event read. | |
2542 | * | |
2543 | * This reads the next event in the ring buffer and increments the iterator. | |
2544 | */ | |
2545 | struct ring_buffer_event * | |
2546 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
2547 | { | |
2548 | struct ring_buffer_event *event; | |
f83c9d0f SR |
2549 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
2550 | unsigned long flags; | |
7a8e76a3 | 2551 | |
2d622719 | 2552 | again: |
f83c9d0f SR |
2553 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2554 | event = rb_iter_peek(iter, ts); | |
7a8e76a3 | 2555 | if (!event) |
f83c9d0f | 2556 | goto out; |
7a8e76a3 SR |
2557 | |
2558 | rb_advance_iter(iter); | |
f83c9d0f SR |
2559 | out: |
2560 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 2561 | |
2d622719 TZ |
2562 | if (event && event->type == RINGBUF_TYPE_PADDING) { |
2563 | cpu_relax(); | |
2564 | goto again; | |
2565 | } | |
2566 | ||
7a8e76a3 SR |
2567 | return event; |
2568 | } | |
c4f50183 | 2569 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
2570 | |
2571 | /** | |
2572 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
2573 | * @buffer: The ring buffer. | |
2574 | */ | |
2575 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
2576 | { | |
2577 | return BUF_PAGE_SIZE * buffer->pages; | |
2578 | } | |
c4f50183 | 2579 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
2580 | |
2581 | static void | |
2582 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
2583 | { | |
2584 | cpu_buffer->head_page | |
2585 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 2586 | local_set(&cpu_buffer->head_page->write, 0); |
abc9b56d | 2587 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 2588 | |
6f807acd | 2589 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
2590 | |
2591 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
2592 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
2593 | ||
2594 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
2595 | local_set(&cpu_buffer->reader_page->write, 0); | |
abc9b56d | 2596 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 2597 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 2598 | |
7a8e76a3 SR |
2599 | cpu_buffer->overrun = 0; |
2600 | cpu_buffer->entries = 0; | |
69507c06 SR |
2601 | |
2602 | cpu_buffer->write_stamp = 0; | |
2603 | cpu_buffer->read_stamp = 0; | |
7a8e76a3 SR |
2604 | } |
2605 | ||
2606 | /** | |
2607 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
2608 | * @buffer: The ring buffer to reset a per cpu buffer of | |
2609 | * @cpu: The CPU buffer to be reset | |
2610 | */ | |
2611 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
2612 | { | |
2613 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2614 | unsigned long flags; | |
2615 | ||
9e01c1b7 | 2616 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2617 | return; |
7a8e76a3 | 2618 | |
f83c9d0f SR |
2619 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2620 | ||
3e03fb7f | 2621 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
2622 | |
2623 | rb_reset_cpu(cpu_buffer); | |
2624 | ||
3e03fb7f | 2625 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f SR |
2626 | |
2627 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 2628 | } |
c4f50183 | 2629 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
2630 | |
2631 | /** | |
2632 | * ring_buffer_reset - reset a ring buffer | |
2633 | * @buffer: The ring buffer to reset all cpu buffers | |
2634 | */ | |
2635 | void ring_buffer_reset(struct ring_buffer *buffer) | |
2636 | { | |
7a8e76a3 SR |
2637 | int cpu; |
2638 | ||
7a8e76a3 | 2639 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 2640 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 2641 | } |
c4f50183 | 2642 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
2643 | |
2644 | /** | |
2645 | * rind_buffer_empty - is the ring buffer empty? | |
2646 | * @buffer: The ring buffer to test | |
2647 | */ | |
2648 | int ring_buffer_empty(struct ring_buffer *buffer) | |
2649 | { | |
2650 | struct ring_buffer_per_cpu *cpu_buffer; | |
2651 | int cpu; | |
2652 | ||
2653 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
2654 | for_each_buffer_cpu(buffer, cpu) { | |
2655 | cpu_buffer = buffer->buffers[cpu]; | |
2656 | if (!rb_per_cpu_empty(cpu_buffer)) | |
2657 | return 0; | |
2658 | } | |
554f786e | 2659 | |
7a8e76a3 SR |
2660 | return 1; |
2661 | } | |
c4f50183 | 2662 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
2663 | |
2664 | /** | |
2665 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
2666 | * @buffer: The ring buffer | |
2667 | * @cpu: The CPU buffer to test | |
2668 | */ | |
2669 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
2670 | { | |
2671 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2672 | int ret; |
7a8e76a3 | 2673 | |
9e01c1b7 | 2674 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2675 | return 1; |
7a8e76a3 SR |
2676 | |
2677 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e SR |
2678 | ret = rb_per_cpu_empty(cpu_buffer); |
2679 | ||
554f786e SR |
2680 | |
2681 | return ret; | |
7a8e76a3 | 2682 | } |
c4f50183 | 2683 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 SR |
2684 | |
2685 | /** | |
2686 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
2687 | * @buffer_a: One buffer to swap with | |
2688 | * @buffer_b: The other buffer to swap with | |
2689 | * | |
2690 | * This function is useful for tracers that want to take a "snapshot" | |
2691 | * of a CPU buffer and has another back up buffer lying around. | |
2692 | * it is expected that the tracer handles the cpu buffer not being | |
2693 | * used at the moment. | |
2694 | */ | |
2695 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
2696 | struct ring_buffer *buffer_b, int cpu) | |
2697 | { | |
2698 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
2699 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
2700 | int ret = -EINVAL; |
2701 | ||
9e01c1b7 RR |
2702 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
2703 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 2704 | goto out; |
7a8e76a3 SR |
2705 | |
2706 | /* At least make sure the two buffers are somewhat the same */ | |
6d102bc6 | 2707 | if (buffer_a->pages != buffer_b->pages) |
554f786e SR |
2708 | goto out; |
2709 | ||
2710 | ret = -EAGAIN; | |
7a8e76a3 | 2711 | |
97b17efe | 2712 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 2713 | goto out; |
97b17efe SR |
2714 | |
2715 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 2716 | goto out; |
97b17efe SR |
2717 | |
2718 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 2719 | goto out; |
97b17efe | 2720 | |
7a8e76a3 SR |
2721 | cpu_buffer_a = buffer_a->buffers[cpu]; |
2722 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
2723 | ||
97b17efe | 2724 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 2725 | goto out; |
97b17efe SR |
2726 | |
2727 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 2728 | goto out; |
97b17efe | 2729 | |
7a8e76a3 SR |
2730 | /* |
2731 | * We can't do a synchronize_sched here because this | |
2732 | * function can be called in atomic context. | |
2733 | * Normally this will be called from the same CPU as cpu. | |
2734 | * If not it's up to the caller to protect this. | |
2735 | */ | |
2736 | atomic_inc(&cpu_buffer_a->record_disabled); | |
2737 | atomic_inc(&cpu_buffer_b->record_disabled); | |
2738 | ||
2739 | buffer_a->buffers[cpu] = cpu_buffer_b; | |
2740 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
2741 | ||
2742 | cpu_buffer_b->buffer = buffer_a; | |
2743 | cpu_buffer_a->buffer = buffer_b; | |
2744 | ||
2745 | atomic_dec(&cpu_buffer_a->record_disabled); | |
2746 | atomic_dec(&cpu_buffer_b->record_disabled); | |
2747 | ||
554f786e SR |
2748 | ret = 0; |
2749 | out: | |
554f786e | 2750 | return ret; |
7a8e76a3 | 2751 | } |
c4f50183 | 2752 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
7a8e76a3 | 2753 | |
8789a9e7 | 2754 | static void rb_remove_entries(struct ring_buffer_per_cpu *cpu_buffer, |
667d2412 LJ |
2755 | struct buffer_data_page *bpage, |
2756 | unsigned int offset) | |
8789a9e7 SR |
2757 | { |
2758 | struct ring_buffer_event *event; | |
2759 | unsigned long head; | |
2760 | ||
2761 | __raw_spin_lock(&cpu_buffer->lock); | |
667d2412 | 2762 | for (head = offset; head < local_read(&bpage->commit); |
8789a9e7 SR |
2763 | head += rb_event_length(event)) { |
2764 | ||
044fa782 | 2765 | event = __rb_data_page_index(bpage, head); |
8789a9e7 SR |
2766 | if (RB_WARN_ON(cpu_buffer, rb_null_event(event))) |
2767 | return; | |
2768 | /* Only count data entries */ | |
2769 | if (event->type != RINGBUF_TYPE_DATA) | |
2770 | continue; | |
2771 | cpu_buffer->entries--; | |
2772 | } | |
2773 | __raw_spin_unlock(&cpu_buffer->lock); | |
2774 | } | |
2775 | ||
2776 | /** | |
2777 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
2778 | * @buffer: the buffer to allocate for. | |
2779 | * | |
2780 | * This function is used in conjunction with ring_buffer_read_page. | |
2781 | * When reading a full page from the ring buffer, these functions | |
2782 | * can be used to speed up the process. The calling function should | |
2783 | * allocate a few pages first with this function. Then when it | |
2784 | * needs to get pages from the ring buffer, it passes the result | |
2785 | * of this function into ring_buffer_read_page, which will swap | |
2786 | * the page that was allocated, with the read page of the buffer. | |
2787 | * | |
2788 | * Returns: | |
2789 | * The page allocated, or NULL on error. | |
2790 | */ | |
2791 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer) | |
2792 | { | |
044fa782 | 2793 | struct buffer_data_page *bpage; |
ef7a4a16 | 2794 | unsigned long addr; |
8789a9e7 SR |
2795 | |
2796 | addr = __get_free_page(GFP_KERNEL); | |
2797 | if (!addr) | |
2798 | return NULL; | |
2799 | ||
044fa782 | 2800 | bpage = (void *)addr; |
8789a9e7 | 2801 | |
ef7a4a16 SR |
2802 | rb_init_page(bpage); |
2803 | ||
044fa782 | 2804 | return bpage; |
8789a9e7 SR |
2805 | } |
2806 | ||
2807 | /** | |
2808 | * ring_buffer_free_read_page - free an allocated read page | |
2809 | * @buffer: the buffer the page was allocate for | |
2810 | * @data: the page to free | |
2811 | * | |
2812 | * Free a page allocated from ring_buffer_alloc_read_page. | |
2813 | */ | |
2814 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
2815 | { | |
2816 | free_page((unsigned long)data); | |
2817 | } | |
2818 | ||
2819 | /** | |
2820 | * ring_buffer_read_page - extract a page from the ring buffer | |
2821 | * @buffer: buffer to extract from | |
2822 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 2823 | * @len: amount to extract |
8789a9e7 SR |
2824 | * @cpu: the cpu of the buffer to extract |
2825 | * @full: should the extraction only happen when the page is full. | |
2826 | * | |
2827 | * This function will pull out a page from the ring buffer and consume it. | |
2828 | * @data_page must be the address of the variable that was returned | |
2829 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
2830 | * to swap with a page in the ring buffer. | |
2831 | * | |
2832 | * for example: | |
b85fa01e | 2833 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
2834 | * if (!rpage) |
2835 | * return error; | |
ef7a4a16 | 2836 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
2837 | * if (ret >= 0) |
2838 | * process_page(rpage, ret); | |
8789a9e7 SR |
2839 | * |
2840 | * When @full is set, the function will not return true unless | |
2841 | * the writer is off the reader page. | |
2842 | * | |
2843 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
2844 | * The ring buffer can be used anywhere in the kernel and can not | |
2845 | * blindly call wake_up. The layer that uses the ring buffer must be | |
2846 | * responsible for that. | |
2847 | * | |
2848 | * Returns: | |
667d2412 LJ |
2849 | * >=0 if data has been transferred, returns the offset of consumed data. |
2850 | * <0 if no data has been transferred. | |
8789a9e7 SR |
2851 | */ |
2852 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 2853 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
2854 | { |
2855 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2856 | struct ring_buffer_event *event; | |
044fa782 | 2857 | struct buffer_data_page *bpage; |
ef7a4a16 | 2858 | struct buffer_page *reader; |
8789a9e7 | 2859 | unsigned long flags; |
ef7a4a16 | 2860 | unsigned int commit; |
667d2412 | 2861 | unsigned int read; |
4f3640f8 | 2862 | u64 save_timestamp; |
667d2412 | 2863 | int ret = -1; |
8789a9e7 | 2864 | |
554f786e SR |
2865 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
2866 | goto out; | |
2867 | ||
474d32b6 SR |
2868 | /* |
2869 | * If len is not big enough to hold the page header, then | |
2870 | * we can not copy anything. | |
2871 | */ | |
2872 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 2873 | goto out; |
474d32b6 SR |
2874 | |
2875 | len -= BUF_PAGE_HDR_SIZE; | |
2876 | ||
8789a9e7 | 2877 | if (!data_page) |
554f786e | 2878 | goto out; |
8789a9e7 | 2879 | |
044fa782 SR |
2880 | bpage = *data_page; |
2881 | if (!bpage) | |
554f786e | 2882 | goto out; |
8789a9e7 SR |
2883 | |
2884 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2885 | ||
ef7a4a16 SR |
2886 | reader = rb_get_reader_page(cpu_buffer); |
2887 | if (!reader) | |
554f786e | 2888 | goto out_unlock; |
8789a9e7 | 2889 | |
ef7a4a16 SR |
2890 | event = rb_reader_event(cpu_buffer); |
2891 | ||
2892 | read = reader->read; | |
2893 | commit = rb_page_commit(reader); | |
667d2412 | 2894 | |
8789a9e7 | 2895 | /* |
474d32b6 SR |
2896 | * If this page has been partially read or |
2897 | * if len is not big enough to read the rest of the page or | |
2898 | * a writer is still on the page, then | |
2899 | * we must copy the data from the page to the buffer. | |
2900 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 2901 | */ |
474d32b6 | 2902 | if (read || (len < (commit - read)) || |
ef7a4a16 | 2903 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 2904 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
2905 | unsigned int rpos = read; |
2906 | unsigned int pos = 0; | |
ef7a4a16 | 2907 | unsigned int size; |
8789a9e7 SR |
2908 | |
2909 | if (full) | |
554f786e | 2910 | goto out_unlock; |
8789a9e7 | 2911 | |
ef7a4a16 SR |
2912 | if (len > (commit - read)) |
2913 | len = (commit - read); | |
2914 | ||
2915 | size = rb_event_length(event); | |
2916 | ||
2917 | if (len < size) | |
554f786e | 2918 | goto out_unlock; |
ef7a4a16 | 2919 | |
4f3640f8 SR |
2920 | /* save the current timestamp, since the user will need it */ |
2921 | save_timestamp = cpu_buffer->read_stamp; | |
2922 | ||
ef7a4a16 SR |
2923 | /* Need to copy one event at a time */ |
2924 | do { | |
474d32b6 | 2925 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
2926 | |
2927 | len -= size; | |
2928 | ||
2929 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
2930 | rpos = reader->read; |
2931 | pos += size; | |
ef7a4a16 SR |
2932 | |
2933 | event = rb_reader_event(cpu_buffer); | |
2934 | size = rb_event_length(event); | |
2935 | } while (len > size); | |
667d2412 LJ |
2936 | |
2937 | /* update bpage */ | |
ef7a4a16 | 2938 | local_set(&bpage->commit, pos); |
4f3640f8 | 2939 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 2940 | |
474d32b6 SR |
2941 | /* we copied everything to the beginning */ |
2942 | read = 0; | |
8789a9e7 SR |
2943 | } else { |
2944 | /* swap the pages */ | |
044fa782 | 2945 | rb_init_page(bpage); |
ef7a4a16 SR |
2946 | bpage = reader->page; |
2947 | reader->page = *data_page; | |
2948 | local_set(&reader->write, 0); | |
2949 | reader->read = 0; | |
044fa782 | 2950 | *data_page = bpage; |
ef7a4a16 SR |
2951 | |
2952 | /* update the entry counter */ | |
2953 | rb_remove_entries(cpu_buffer, bpage, read); | |
8789a9e7 | 2954 | } |
667d2412 | 2955 | ret = read; |
8789a9e7 | 2956 | |
554f786e | 2957 | out_unlock: |
8789a9e7 SR |
2958 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
2959 | ||
554f786e | 2960 | out: |
8789a9e7 SR |
2961 | return ret; |
2962 | } | |
2963 | ||
a3583244 SR |
2964 | static ssize_t |
2965 | rb_simple_read(struct file *filp, char __user *ubuf, | |
2966 | size_t cnt, loff_t *ppos) | |
2967 | { | |
5e39841c | 2968 | unsigned long *p = filp->private_data; |
a3583244 SR |
2969 | char buf[64]; |
2970 | int r; | |
2971 | ||
033601a3 SR |
2972 | if (test_bit(RB_BUFFERS_DISABLED_BIT, p)) |
2973 | r = sprintf(buf, "permanently disabled\n"); | |
2974 | else | |
2975 | r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p)); | |
a3583244 SR |
2976 | |
2977 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
2978 | } | |
2979 | ||
2980 | static ssize_t | |
2981 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
2982 | size_t cnt, loff_t *ppos) | |
2983 | { | |
5e39841c | 2984 | unsigned long *p = filp->private_data; |
a3583244 | 2985 | char buf[64]; |
5e39841c | 2986 | unsigned long val; |
a3583244 SR |
2987 | int ret; |
2988 | ||
2989 | if (cnt >= sizeof(buf)) | |
2990 | return -EINVAL; | |
2991 | ||
2992 | if (copy_from_user(&buf, ubuf, cnt)) | |
2993 | return -EFAULT; | |
2994 | ||
2995 | buf[cnt] = 0; | |
2996 | ||
2997 | ret = strict_strtoul(buf, 10, &val); | |
2998 | if (ret < 0) | |
2999 | return ret; | |
3000 | ||
033601a3 SR |
3001 | if (val) |
3002 | set_bit(RB_BUFFERS_ON_BIT, p); | |
3003 | else | |
3004 | clear_bit(RB_BUFFERS_ON_BIT, p); | |
a3583244 SR |
3005 | |
3006 | (*ppos)++; | |
3007 | ||
3008 | return cnt; | |
3009 | } | |
3010 | ||
5e2336a0 | 3011 | static const struct file_operations rb_simple_fops = { |
a3583244 SR |
3012 | .open = tracing_open_generic, |
3013 | .read = rb_simple_read, | |
3014 | .write = rb_simple_write, | |
3015 | }; | |
3016 | ||
3017 | ||
3018 | static __init int rb_init_debugfs(void) | |
3019 | { | |
3020 | struct dentry *d_tracer; | |
a3583244 SR |
3021 | |
3022 | d_tracer = tracing_init_dentry(); | |
3023 | ||
5452af66 FW |
3024 | trace_create_file("tracing_on", 0644, d_tracer, |
3025 | &ring_buffer_flags, &rb_simple_fops); | |
a3583244 SR |
3026 | |
3027 | return 0; | |
3028 | } | |
3029 | ||
3030 | fs_initcall(rb_init_debugfs); | |
554f786e | 3031 | |
59222efe | 3032 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
3033 | static int rb_cpu_notify(struct notifier_block *self, |
3034 | unsigned long action, void *hcpu) | |
554f786e SR |
3035 | { |
3036 | struct ring_buffer *buffer = | |
3037 | container_of(self, struct ring_buffer, cpu_notify); | |
3038 | long cpu = (long)hcpu; | |
3039 | ||
3040 | switch (action) { | |
3041 | case CPU_UP_PREPARE: | |
3042 | case CPU_UP_PREPARE_FROZEN: | |
3043 | if (cpu_isset(cpu, *buffer->cpumask)) | |
3044 | return NOTIFY_OK; | |
3045 | ||
3046 | buffer->buffers[cpu] = | |
3047 | rb_allocate_cpu_buffer(buffer, cpu); | |
3048 | if (!buffer->buffers[cpu]) { | |
3049 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
3050 | cpu); | |
3051 | return NOTIFY_OK; | |
3052 | } | |
3053 | smp_wmb(); | |
3054 | cpu_set(cpu, *buffer->cpumask); | |
3055 | break; | |
3056 | case CPU_DOWN_PREPARE: | |
3057 | case CPU_DOWN_PREPARE_FROZEN: | |
3058 | /* | |
3059 | * Do nothing. | |
3060 | * If we were to free the buffer, then the user would | |
3061 | * lose any trace that was in the buffer. | |
3062 | */ | |
3063 | break; | |
3064 | default: | |
3065 | break; | |
3066 | } | |
3067 | return NOTIFY_OK; | |
3068 | } | |
3069 | #endif |