]>
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 | ||
334d4169 LJ |
31 | ret = trace_seq_printf(s, "# compressed entry header\n"); |
32 | ret = trace_seq_printf(s, "\ttype_len : 5 bits\n"); | |
d1b182a8 SR |
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); | |
334d4169 LJ |
40 | ret = trace_seq_printf(s, "\tdata max type_len == %d\n", |
41 | RINGBUF_TYPE_DATA_TYPE_LEN_MAX); | |
d1b182a8 SR |
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 |
334d4169 LJ |
207 | #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
208 | ||
209 | /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */ | |
210 | #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX | |
7a8e76a3 SR |
211 | |
212 | enum { | |
213 | RB_LEN_TIME_EXTEND = 8, | |
214 | RB_LEN_TIME_STAMP = 16, | |
215 | }; | |
216 | ||
2d622719 TZ |
217 | static inline int rb_null_event(struct ring_buffer_event *event) |
218 | { | |
334d4169 LJ |
219 | return event->type_len == RINGBUF_TYPE_PADDING |
220 | && event->time_delta == 0; | |
2d622719 TZ |
221 | } |
222 | ||
223 | static inline int rb_discarded_event(struct ring_buffer_event *event) | |
224 | { | |
334d4169 | 225 | return event->type_len == RINGBUF_TYPE_PADDING && event->time_delta; |
2d622719 TZ |
226 | } |
227 | ||
228 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
229 | { | |
334d4169 | 230 | event->type_len = RINGBUF_TYPE_PADDING; |
2d622719 TZ |
231 | event->time_delta = 0; |
232 | } | |
233 | ||
34a148bf | 234 | static unsigned |
2d622719 | 235 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
236 | { |
237 | unsigned length; | |
238 | ||
334d4169 LJ |
239 | if (event->type_len) |
240 | length = event->type_len * RB_ALIGNMENT; | |
2d622719 TZ |
241 | else |
242 | length = event->array[0]; | |
243 | return length + RB_EVNT_HDR_SIZE; | |
244 | } | |
245 | ||
246 | /* inline for ring buffer fast paths */ | |
247 | static unsigned | |
248 | rb_event_length(struct ring_buffer_event *event) | |
249 | { | |
334d4169 | 250 | switch (event->type_len) { |
7a8e76a3 | 251 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
252 | if (rb_null_event(event)) |
253 | /* undefined */ | |
254 | return -1; | |
334d4169 | 255 | return event->array[0] + RB_EVNT_HDR_SIZE; |
7a8e76a3 SR |
256 | |
257 | case RINGBUF_TYPE_TIME_EXTEND: | |
258 | return RB_LEN_TIME_EXTEND; | |
259 | ||
260 | case RINGBUF_TYPE_TIME_STAMP: | |
261 | return RB_LEN_TIME_STAMP; | |
262 | ||
263 | case RINGBUF_TYPE_DATA: | |
2d622719 | 264 | return rb_event_data_length(event); |
7a8e76a3 SR |
265 | default: |
266 | BUG(); | |
267 | } | |
268 | /* not hit */ | |
269 | return 0; | |
270 | } | |
271 | ||
272 | /** | |
273 | * ring_buffer_event_length - return the length of the event | |
274 | * @event: the event to get the length of | |
275 | */ | |
276 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
277 | { | |
465634ad | 278 | unsigned length = rb_event_length(event); |
334d4169 | 279 | if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
465634ad RR |
280 | return length; |
281 | length -= RB_EVNT_HDR_SIZE; | |
282 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
283 | length -= sizeof(event->array[0]); | |
284 | return length; | |
7a8e76a3 | 285 | } |
c4f50183 | 286 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
287 | |
288 | /* inline for ring buffer fast paths */ | |
34a148bf | 289 | static void * |
7a8e76a3 SR |
290 | rb_event_data(struct ring_buffer_event *event) |
291 | { | |
334d4169 | 292 | BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX); |
7a8e76a3 | 293 | /* If length is in len field, then array[0] has the data */ |
334d4169 | 294 | if (event->type_len) |
7a8e76a3 SR |
295 | return (void *)&event->array[0]; |
296 | /* Otherwise length is in array[0] and array[1] has the data */ | |
297 | return (void *)&event->array[1]; | |
298 | } | |
299 | ||
300 | /** | |
301 | * ring_buffer_event_data - return the data of the event | |
302 | * @event: the event to get the data from | |
303 | */ | |
304 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
305 | { | |
306 | return rb_event_data(event); | |
307 | } | |
c4f50183 | 308 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
309 | |
310 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 311 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
312 | |
313 | #define TS_SHIFT 27 | |
314 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
315 | #define TS_DELTA_TEST (~TS_MASK) | |
316 | ||
abc9b56d | 317 | struct buffer_data_page { |
e4c2ce82 | 318 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 319 | local_t commit; /* write committed index */ |
abc9b56d SR |
320 | unsigned char data[]; /* data of buffer page */ |
321 | }; | |
322 | ||
323 | struct buffer_page { | |
778c55d4 | 324 | struct list_head list; /* list of buffer pages */ |
abc9b56d | 325 | local_t write; /* index for next write */ |
6f807acd | 326 | unsigned read; /* index for next read */ |
778c55d4 | 327 | local_t entries; /* entries on this page */ |
abc9b56d | 328 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
329 | }; |
330 | ||
044fa782 | 331 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 332 | { |
044fa782 | 333 | local_set(&bpage->commit, 0); |
abc9b56d SR |
334 | } |
335 | ||
474d32b6 SR |
336 | /** |
337 | * ring_buffer_page_len - the size of data on the page. | |
338 | * @page: The page to read | |
339 | * | |
340 | * Returns the amount of data on the page, including buffer page header. | |
341 | */ | |
ef7a4a16 SR |
342 | size_t ring_buffer_page_len(void *page) |
343 | { | |
474d32b6 SR |
344 | return local_read(&((struct buffer_data_page *)page)->commit) |
345 | + BUF_PAGE_HDR_SIZE; | |
ef7a4a16 SR |
346 | } |
347 | ||
ed56829c SR |
348 | /* |
349 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
350 | * this issue out. | |
351 | */ | |
34a148bf | 352 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 353 | { |
34a148bf | 354 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 355 | kfree(bpage); |
ed56829c SR |
356 | } |
357 | ||
7a8e76a3 SR |
358 | /* |
359 | * We need to fit the time_stamp delta into 27 bits. | |
360 | */ | |
361 | static inline int test_time_stamp(u64 delta) | |
362 | { | |
363 | if (delta & TS_DELTA_TEST) | |
364 | return 1; | |
365 | return 0; | |
366 | } | |
367 | ||
474d32b6 | 368 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 | 369 | |
be957c44 SR |
370 | /* Max payload is BUF_PAGE_SIZE - header (8bytes) */ |
371 | #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) | |
372 | ||
d1b182a8 SR |
373 | int ring_buffer_print_page_header(struct trace_seq *s) |
374 | { | |
375 | struct buffer_data_page field; | |
376 | int ret; | |
377 | ||
378 | ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t" | |
379 | "offset:0;\tsize:%u;\n", | |
380 | (unsigned int)sizeof(field.time_stamp)); | |
381 | ||
382 | ret = trace_seq_printf(s, "\tfield: local_t commit;\t" | |
383 | "offset:%u;\tsize:%u;\n", | |
384 | (unsigned int)offsetof(typeof(field), commit), | |
385 | (unsigned int)sizeof(field.commit)); | |
386 | ||
387 | ret = trace_seq_printf(s, "\tfield: char data;\t" | |
388 | "offset:%u;\tsize:%u;\n", | |
389 | (unsigned int)offsetof(typeof(field), data), | |
390 | (unsigned int)BUF_PAGE_SIZE); | |
391 | ||
392 | return ret; | |
393 | } | |
394 | ||
7a8e76a3 SR |
395 | /* |
396 | * head_page == tail_page && head == tail then buffer is empty. | |
397 | */ | |
398 | struct ring_buffer_per_cpu { | |
399 | int cpu; | |
400 | struct ring_buffer *buffer; | |
f83c9d0f | 401 | spinlock_t reader_lock; /* serialize readers */ |
3e03fb7f | 402 | raw_spinlock_t lock; |
7a8e76a3 SR |
403 | struct lock_class_key lock_key; |
404 | struct list_head pages; | |
6f807acd SR |
405 | struct buffer_page *head_page; /* read from head */ |
406 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 407 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 408 | struct buffer_page *reader_page; |
f0d2c681 SR |
409 | unsigned long nmi_dropped; |
410 | unsigned long commit_overrun; | |
7a8e76a3 | 411 | unsigned long overrun; |
e4906eff SR |
412 | unsigned long read; |
413 | local_t entries; | |
7a8e76a3 SR |
414 | u64 write_stamp; |
415 | u64 read_stamp; | |
416 | atomic_t record_disabled; | |
417 | }; | |
418 | ||
419 | struct ring_buffer { | |
7a8e76a3 SR |
420 | unsigned pages; |
421 | unsigned flags; | |
422 | int cpus; | |
7a8e76a3 | 423 | atomic_t record_disabled; |
00f62f61 | 424 | cpumask_var_t cpumask; |
7a8e76a3 SR |
425 | |
426 | struct mutex mutex; | |
427 | ||
428 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 429 | |
59222efe | 430 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
431 | struct notifier_block cpu_notify; |
432 | #endif | |
37886f6a | 433 | u64 (*clock)(void); |
7a8e76a3 SR |
434 | }; |
435 | ||
436 | struct ring_buffer_iter { | |
437 | struct ring_buffer_per_cpu *cpu_buffer; | |
438 | unsigned long head; | |
439 | struct buffer_page *head_page; | |
440 | u64 read_stamp; | |
441 | }; | |
442 | ||
f536aafc | 443 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
bf41a158 | 444 | #define RB_WARN_ON(buffer, cond) \ |
3e89c7bb SR |
445 | ({ \ |
446 | int _____ret = unlikely(cond); \ | |
447 | if (_____ret) { \ | |
bf41a158 SR |
448 | atomic_inc(&buffer->record_disabled); \ |
449 | WARN_ON(1); \ | |
450 | } \ | |
3e89c7bb SR |
451 | _____ret; \ |
452 | }) | |
f536aafc | 453 | |
37886f6a SR |
454 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
455 | #define DEBUG_SHIFT 0 | |
456 | ||
457 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) | |
458 | { | |
459 | u64 time; | |
460 | ||
461 | preempt_disable_notrace(); | |
462 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
463 | time = buffer->clock() << DEBUG_SHIFT; | |
464 | preempt_enable_no_resched_notrace(); | |
465 | ||
466 | return time; | |
467 | } | |
468 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
469 | ||
470 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
471 | int cpu, u64 *ts) | |
472 | { | |
473 | /* Just stupid testing the normalize function and deltas */ | |
474 | *ts >>= DEBUG_SHIFT; | |
475 | } | |
476 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
477 | ||
7a8e76a3 SR |
478 | /** |
479 | * check_pages - integrity check of buffer pages | |
480 | * @cpu_buffer: CPU buffer with pages to test | |
481 | * | |
c3706f00 | 482 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
483 | * been corrupted. |
484 | */ | |
485 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
486 | { | |
487 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 488 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 489 | |
3e89c7bb SR |
490 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
491 | return -1; | |
492 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
493 | return -1; | |
7a8e76a3 | 494 | |
044fa782 | 495 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 496 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 497 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
498 | return -1; |
499 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 500 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 501 | return -1; |
7a8e76a3 SR |
502 | } |
503 | ||
504 | return 0; | |
505 | } | |
506 | ||
7a8e76a3 SR |
507 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
508 | unsigned nr_pages) | |
509 | { | |
510 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 511 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
512 | unsigned long addr; |
513 | LIST_HEAD(pages); | |
514 | unsigned i; | |
515 | ||
516 | for (i = 0; i < nr_pages; i++) { | |
044fa782 | 517 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
aa1e0e3b | 518 | GFP_KERNEL, cpu_to_node(cpu_buffer->cpu)); |
044fa782 | 519 | if (!bpage) |
e4c2ce82 | 520 | goto free_pages; |
044fa782 | 521 | list_add(&bpage->list, &pages); |
e4c2ce82 | 522 | |
7a8e76a3 SR |
523 | addr = __get_free_page(GFP_KERNEL); |
524 | if (!addr) | |
525 | goto free_pages; | |
044fa782 SR |
526 | bpage->page = (void *)addr; |
527 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
528 | } |
529 | ||
530 | list_splice(&pages, head); | |
531 | ||
532 | rb_check_pages(cpu_buffer); | |
533 | ||
534 | return 0; | |
535 | ||
536 | free_pages: | |
044fa782 SR |
537 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
538 | list_del_init(&bpage->list); | |
539 | free_buffer_page(bpage); | |
7a8e76a3 SR |
540 | } |
541 | return -ENOMEM; | |
542 | } | |
543 | ||
544 | static struct ring_buffer_per_cpu * | |
545 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) | |
546 | { | |
547 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 548 | struct buffer_page *bpage; |
d769041f | 549 | unsigned long addr; |
7a8e76a3 SR |
550 | int ret; |
551 | ||
552 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
553 | GFP_KERNEL, cpu_to_node(cpu)); | |
554 | if (!cpu_buffer) | |
555 | return NULL; | |
556 | ||
557 | cpu_buffer->cpu = cpu; | |
558 | cpu_buffer->buffer = buffer; | |
f83c9d0f | 559 | spin_lock_init(&cpu_buffer->reader_lock); |
3e03fb7f | 560 | cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; |
7a8e76a3 SR |
561 | INIT_LIST_HEAD(&cpu_buffer->pages); |
562 | ||
044fa782 | 563 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 564 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 565 | if (!bpage) |
e4c2ce82 SR |
566 | goto fail_free_buffer; |
567 | ||
044fa782 | 568 | cpu_buffer->reader_page = bpage; |
d769041f SR |
569 | addr = __get_free_page(GFP_KERNEL); |
570 | if (!addr) | |
e4c2ce82 | 571 | goto fail_free_reader; |
044fa782 SR |
572 | bpage->page = (void *)addr; |
573 | rb_init_page(bpage->page); | |
e4c2ce82 | 574 | |
d769041f | 575 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 576 | |
7a8e76a3 SR |
577 | ret = rb_allocate_pages(cpu_buffer, buffer->pages); |
578 | if (ret < 0) | |
d769041f | 579 | goto fail_free_reader; |
7a8e76a3 SR |
580 | |
581 | cpu_buffer->head_page | |
582 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 583 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 SR |
584 | |
585 | return cpu_buffer; | |
586 | ||
d769041f SR |
587 | fail_free_reader: |
588 | free_buffer_page(cpu_buffer->reader_page); | |
589 | ||
7a8e76a3 SR |
590 | fail_free_buffer: |
591 | kfree(cpu_buffer); | |
592 | return NULL; | |
593 | } | |
594 | ||
595 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
596 | { | |
597 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 598 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 599 | |
d769041f SR |
600 | free_buffer_page(cpu_buffer->reader_page); |
601 | ||
044fa782 SR |
602 | list_for_each_entry_safe(bpage, tmp, head, list) { |
603 | list_del_init(&bpage->list); | |
604 | free_buffer_page(bpage); | |
7a8e76a3 SR |
605 | } |
606 | kfree(cpu_buffer); | |
607 | } | |
608 | ||
a7b13743 SR |
609 | /* |
610 | * Causes compile errors if the struct buffer_page gets bigger | |
611 | * than the struct page. | |
612 | */ | |
613 | extern int ring_buffer_page_too_big(void); | |
614 | ||
59222efe | 615 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
616 | static int rb_cpu_notify(struct notifier_block *self, |
617 | unsigned long action, void *hcpu); | |
554f786e SR |
618 | #endif |
619 | ||
7a8e76a3 SR |
620 | /** |
621 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 622 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
623 | * @flags: attributes to set for the ring buffer. |
624 | * | |
625 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
626 | * flag. This flag means that the buffer will overwrite old data | |
627 | * when the buffer wraps. If this flag is not set, the buffer will | |
628 | * drop data when the tail hits the head. | |
629 | */ | |
630 | struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags) | |
631 | { | |
632 | struct ring_buffer *buffer; | |
633 | int bsize; | |
634 | int cpu; | |
635 | ||
a7b13743 SR |
636 | /* Paranoid! Optimizes out when all is well */ |
637 | if (sizeof(struct buffer_page) > sizeof(struct page)) | |
638 | ring_buffer_page_too_big(); | |
639 | ||
640 | ||
7a8e76a3 SR |
641 | /* keep it in its own cache line */ |
642 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
643 | GFP_KERNEL); | |
644 | if (!buffer) | |
645 | return NULL; | |
646 | ||
9e01c1b7 RR |
647 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
648 | goto fail_free_buffer; | |
649 | ||
7a8e76a3 SR |
650 | buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
651 | buffer->flags = flags; | |
37886f6a | 652 | buffer->clock = trace_clock_local; |
7a8e76a3 SR |
653 | |
654 | /* need at least two pages */ | |
655 | if (buffer->pages == 1) | |
656 | buffer->pages++; | |
657 | ||
3bf832ce FW |
658 | /* |
659 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
660 | * in early initcall, it will not be notified of secondary cpus. | |
661 | * In that off case, we need to allocate for all possible cpus. | |
662 | */ | |
663 | #ifdef CONFIG_HOTPLUG_CPU | |
554f786e SR |
664 | get_online_cpus(); |
665 | cpumask_copy(buffer->cpumask, cpu_online_mask); | |
3bf832ce FW |
666 | #else |
667 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
668 | #endif | |
7a8e76a3 SR |
669 | buffer->cpus = nr_cpu_ids; |
670 | ||
671 | bsize = sizeof(void *) * nr_cpu_ids; | |
672 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
673 | GFP_KERNEL); | |
674 | if (!buffer->buffers) | |
9e01c1b7 | 675 | goto fail_free_cpumask; |
7a8e76a3 SR |
676 | |
677 | for_each_buffer_cpu(buffer, cpu) { | |
678 | buffer->buffers[cpu] = | |
679 | rb_allocate_cpu_buffer(buffer, cpu); | |
680 | if (!buffer->buffers[cpu]) | |
681 | goto fail_free_buffers; | |
682 | } | |
683 | ||
59222efe | 684 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
685 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
686 | buffer->cpu_notify.priority = 0; | |
687 | register_cpu_notifier(&buffer->cpu_notify); | |
688 | #endif | |
689 | ||
690 | put_online_cpus(); | |
7a8e76a3 SR |
691 | mutex_init(&buffer->mutex); |
692 | ||
693 | return buffer; | |
694 | ||
695 | fail_free_buffers: | |
696 | for_each_buffer_cpu(buffer, cpu) { | |
697 | if (buffer->buffers[cpu]) | |
698 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
699 | } | |
700 | kfree(buffer->buffers); | |
701 | ||
9e01c1b7 RR |
702 | fail_free_cpumask: |
703 | free_cpumask_var(buffer->cpumask); | |
554f786e | 704 | put_online_cpus(); |
9e01c1b7 | 705 | |
7a8e76a3 SR |
706 | fail_free_buffer: |
707 | kfree(buffer); | |
708 | return NULL; | |
709 | } | |
c4f50183 | 710 | EXPORT_SYMBOL_GPL(ring_buffer_alloc); |
7a8e76a3 SR |
711 | |
712 | /** | |
713 | * ring_buffer_free - free a ring buffer. | |
714 | * @buffer: the buffer to free. | |
715 | */ | |
716 | void | |
717 | ring_buffer_free(struct ring_buffer *buffer) | |
718 | { | |
719 | int cpu; | |
720 | ||
554f786e SR |
721 | get_online_cpus(); |
722 | ||
59222efe | 723 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
724 | unregister_cpu_notifier(&buffer->cpu_notify); |
725 | #endif | |
726 | ||
7a8e76a3 SR |
727 | for_each_buffer_cpu(buffer, cpu) |
728 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
729 | ||
554f786e SR |
730 | put_online_cpus(); |
731 | ||
9e01c1b7 RR |
732 | free_cpumask_var(buffer->cpumask); |
733 | ||
7a8e76a3 SR |
734 | kfree(buffer); |
735 | } | |
c4f50183 | 736 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 737 | |
37886f6a SR |
738 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
739 | u64 (*clock)(void)) | |
740 | { | |
741 | buffer->clock = clock; | |
742 | } | |
743 | ||
7a8e76a3 SR |
744 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
745 | ||
746 | static void | |
747 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages) | |
748 | { | |
044fa782 | 749 | struct buffer_page *bpage; |
7a8e76a3 SR |
750 | struct list_head *p; |
751 | unsigned i; | |
752 | ||
753 | atomic_inc(&cpu_buffer->record_disabled); | |
754 | synchronize_sched(); | |
755 | ||
756 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
757 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
758 | return; | |
7a8e76a3 | 759 | p = cpu_buffer->pages.next; |
044fa782 SR |
760 | bpage = list_entry(p, struct buffer_page, list); |
761 | list_del_init(&bpage->list); | |
762 | free_buffer_page(bpage); | |
7a8e76a3 | 763 | } |
3e89c7bb SR |
764 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
765 | return; | |
7a8e76a3 SR |
766 | |
767 | rb_reset_cpu(cpu_buffer); | |
768 | ||
769 | rb_check_pages(cpu_buffer); | |
770 | ||
771 | atomic_dec(&cpu_buffer->record_disabled); | |
772 | ||
773 | } | |
774 | ||
775 | static void | |
776 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
777 | struct list_head *pages, unsigned nr_pages) | |
778 | { | |
044fa782 | 779 | struct buffer_page *bpage; |
7a8e76a3 SR |
780 | struct list_head *p; |
781 | unsigned i; | |
782 | ||
783 | atomic_inc(&cpu_buffer->record_disabled); | |
784 | synchronize_sched(); | |
785 | ||
786 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
787 | if (RB_WARN_ON(cpu_buffer, list_empty(pages))) |
788 | return; | |
7a8e76a3 | 789 | p = pages->next; |
044fa782 SR |
790 | bpage = list_entry(p, struct buffer_page, list); |
791 | list_del_init(&bpage->list); | |
792 | list_add_tail(&bpage->list, &cpu_buffer->pages); | |
7a8e76a3 SR |
793 | } |
794 | rb_reset_cpu(cpu_buffer); | |
795 | ||
796 | rb_check_pages(cpu_buffer); | |
797 | ||
798 | atomic_dec(&cpu_buffer->record_disabled); | |
799 | } | |
800 | ||
801 | /** | |
802 | * ring_buffer_resize - resize the ring buffer | |
803 | * @buffer: the buffer to resize. | |
804 | * @size: the new size. | |
805 | * | |
806 | * The tracer is responsible for making sure that the buffer is | |
807 | * not being used while changing the size. | |
808 | * Note: We may be able to change the above requirement by using | |
809 | * RCU synchronizations. | |
810 | * | |
811 | * Minimum size is 2 * BUF_PAGE_SIZE. | |
812 | * | |
813 | * Returns -1 on failure. | |
814 | */ | |
815 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) | |
816 | { | |
817 | struct ring_buffer_per_cpu *cpu_buffer; | |
818 | unsigned nr_pages, rm_pages, new_pages; | |
044fa782 | 819 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
820 | unsigned long buffer_size; |
821 | unsigned long addr; | |
822 | LIST_HEAD(pages); | |
823 | int i, cpu; | |
824 | ||
ee51a1de IM |
825 | /* |
826 | * Always succeed at resizing a non-existent buffer: | |
827 | */ | |
828 | if (!buffer) | |
829 | return size; | |
830 | ||
7a8e76a3 SR |
831 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
832 | size *= BUF_PAGE_SIZE; | |
833 | buffer_size = buffer->pages * BUF_PAGE_SIZE; | |
834 | ||
835 | /* we need a minimum of two pages */ | |
836 | if (size < BUF_PAGE_SIZE * 2) | |
837 | size = BUF_PAGE_SIZE * 2; | |
838 | ||
839 | if (size == buffer_size) | |
840 | return size; | |
841 | ||
842 | mutex_lock(&buffer->mutex); | |
554f786e | 843 | get_online_cpus(); |
7a8e76a3 SR |
844 | |
845 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
846 | ||
847 | if (size < buffer_size) { | |
848 | ||
849 | /* easy case, just free pages */ | |
554f786e SR |
850 | if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) |
851 | goto out_fail; | |
7a8e76a3 SR |
852 | |
853 | rm_pages = buffer->pages - nr_pages; | |
854 | ||
855 | for_each_buffer_cpu(buffer, cpu) { | |
856 | cpu_buffer = buffer->buffers[cpu]; | |
857 | rb_remove_pages(cpu_buffer, rm_pages); | |
858 | } | |
859 | goto out; | |
860 | } | |
861 | ||
862 | /* | |
863 | * This is a bit more difficult. We only want to add pages | |
864 | * when we can allocate enough for all CPUs. We do this | |
865 | * by allocating all the pages and storing them on a local | |
866 | * link list. If we succeed in our allocation, then we | |
867 | * add these pages to the cpu_buffers. Otherwise we just free | |
868 | * them all and return -ENOMEM; | |
869 | */ | |
554f786e SR |
870 | if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) |
871 | goto out_fail; | |
f536aafc | 872 | |
7a8e76a3 SR |
873 | new_pages = nr_pages - buffer->pages; |
874 | ||
875 | for_each_buffer_cpu(buffer, cpu) { | |
876 | for (i = 0; i < new_pages; i++) { | |
044fa782 | 877 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), |
e4c2ce82 SR |
878 | cache_line_size()), |
879 | GFP_KERNEL, cpu_to_node(cpu)); | |
044fa782 | 880 | if (!bpage) |
e4c2ce82 | 881 | goto free_pages; |
044fa782 | 882 | list_add(&bpage->list, &pages); |
7a8e76a3 SR |
883 | addr = __get_free_page(GFP_KERNEL); |
884 | if (!addr) | |
885 | goto free_pages; | |
044fa782 SR |
886 | bpage->page = (void *)addr; |
887 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
888 | } |
889 | } | |
890 | ||
891 | for_each_buffer_cpu(buffer, cpu) { | |
892 | cpu_buffer = buffer->buffers[cpu]; | |
893 | rb_insert_pages(cpu_buffer, &pages, new_pages); | |
894 | } | |
895 | ||
554f786e SR |
896 | if (RB_WARN_ON(buffer, !list_empty(&pages))) |
897 | goto out_fail; | |
7a8e76a3 SR |
898 | |
899 | out: | |
900 | buffer->pages = nr_pages; | |
554f786e | 901 | put_online_cpus(); |
7a8e76a3 SR |
902 | mutex_unlock(&buffer->mutex); |
903 | ||
904 | return size; | |
905 | ||
906 | free_pages: | |
044fa782 SR |
907 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
908 | list_del_init(&bpage->list); | |
909 | free_buffer_page(bpage); | |
7a8e76a3 | 910 | } |
554f786e | 911 | put_online_cpus(); |
641d2f63 | 912 | mutex_unlock(&buffer->mutex); |
7a8e76a3 | 913 | return -ENOMEM; |
554f786e SR |
914 | |
915 | /* | |
916 | * Something went totally wrong, and we are too paranoid | |
917 | * to even clean up the mess. | |
918 | */ | |
919 | out_fail: | |
920 | put_online_cpus(); | |
921 | mutex_unlock(&buffer->mutex); | |
922 | return -1; | |
7a8e76a3 | 923 | } |
c4f50183 | 924 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 925 | |
8789a9e7 | 926 | static inline void * |
044fa782 | 927 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 928 | { |
044fa782 | 929 | return bpage->data + index; |
8789a9e7 SR |
930 | } |
931 | ||
044fa782 | 932 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 933 | { |
044fa782 | 934 | return bpage->page->data + index; |
7a8e76a3 SR |
935 | } |
936 | ||
937 | static inline struct ring_buffer_event * | |
d769041f | 938 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 939 | { |
6f807acd SR |
940 | return __rb_page_index(cpu_buffer->reader_page, |
941 | cpu_buffer->reader_page->read); | |
942 | } | |
943 | ||
944 | static inline struct ring_buffer_event * | |
945 | rb_head_event(struct ring_buffer_per_cpu *cpu_buffer) | |
946 | { | |
947 | return __rb_page_index(cpu_buffer->head_page, | |
948 | cpu_buffer->head_page->read); | |
7a8e76a3 SR |
949 | } |
950 | ||
951 | static inline struct ring_buffer_event * | |
952 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
953 | { | |
6f807acd | 954 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
955 | } |
956 | ||
bf41a158 SR |
957 | static inline unsigned rb_page_write(struct buffer_page *bpage) |
958 | { | |
959 | return local_read(&bpage->write); | |
960 | } | |
961 | ||
962 | static inline unsigned rb_page_commit(struct buffer_page *bpage) | |
963 | { | |
abc9b56d | 964 | return local_read(&bpage->page->commit); |
bf41a158 SR |
965 | } |
966 | ||
967 | /* Size is determined by what has been commited */ | |
968 | static inline unsigned rb_page_size(struct buffer_page *bpage) | |
969 | { | |
970 | return rb_page_commit(bpage); | |
971 | } | |
972 | ||
973 | static inline unsigned | |
974 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
975 | { | |
976 | return rb_page_commit(cpu_buffer->commit_page); | |
977 | } | |
978 | ||
979 | static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer) | |
980 | { | |
981 | return rb_page_commit(cpu_buffer->head_page); | |
982 | } | |
983 | ||
7a8e76a3 | 984 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, |
044fa782 | 985 | struct buffer_page **bpage) |
7a8e76a3 | 986 | { |
044fa782 | 987 | struct list_head *p = (*bpage)->list.next; |
7a8e76a3 SR |
988 | |
989 | if (p == &cpu_buffer->pages) | |
990 | p = p->next; | |
991 | ||
044fa782 | 992 | *bpage = list_entry(p, struct buffer_page, list); |
7a8e76a3 SR |
993 | } |
994 | ||
bf41a158 SR |
995 | static inline unsigned |
996 | rb_event_index(struct ring_buffer_event *event) | |
997 | { | |
998 | unsigned long addr = (unsigned long)event; | |
999 | ||
1000 | return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE); | |
1001 | } | |
1002 | ||
34a148bf | 1003 | static int |
bf41a158 SR |
1004 | rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1005 | struct ring_buffer_event *event) | |
1006 | { | |
1007 | unsigned long addr = (unsigned long)event; | |
1008 | unsigned long index; | |
1009 | ||
1010 | index = rb_event_index(event); | |
1011 | addr &= PAGE_MASK; | |
1012 | ||
1013 | return cpu_buffer->commit_page->page == (void *)addr && | |
1014 | rb_commit_index(cpu_buffer) == index; | |
1015 | } | |
1016 | ||
34a148bf | 1017 | static void |
bf41a158 SR |
1018 | rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer, |
1019 | struct ring_buffer_event *event) | |
7a8e76a3 | 1020 | { |
bf41a158 SR |
1021 | unsigned long addr = (unsigned long)event; |
1022 | unsigned long index; | |
1023 | ||
1024 | index = rb_event_index(event); | |
1025 | addr &= PAGE_MASK; | |
1026 | ||
1027 | while (cpu_buffer->commit_page->page != (void *)addr) { | |
3e89c7bb SR |
1028 | if (RB_WARN_ON(cpu_buffer, |
1029 | cpu_buffer->commit_page == cpu_buffer->tail_page)) | |
1030 | return; | |
abc9b56d | 1031 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1032 | cpu_buffer->commit_page->write; |
1033 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
1034 | cpu_buffer->write_stamp = |
1035 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1036 | } |
1037 | ||
1038 | /* Now set the commit to the event's index */ | |
abc9b56d | 1039 | local_set(&cpu_buffer->commit_page->page->commit, index); |
7a8e76a3 SR |
1040 | } |
1041 | ||
34a148bf | 1042 | static void |
bf41a158 | 1043 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1044 | { |
bf41a158 SR |
1045 | /* |
1046 | * We only race with interrupts and NMIs on this CPU. | |
1047 | * If we own the commit event, then we can commit | |
1048 | * all others that interrupted us, since the interruptions | |
1049 | * are in stack format (they finish before they come | |
1050 | * back to us). This allows us to do a simple loop to | |
1051 | * assign the commit to the tail. | |
1052 | */ | |
a8ccf1d6 | 1053 | again: |
bf41a158 | 1054 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
abc9b56d | 1055 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1056 | cpu_buffer->commit_page->write; |
1057 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
1058 | cpu_buffer->write_stamp = |
1059 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1060 | /* add barrier to keep gcc from optimizing too much */ |
1061 | barrier(); | |
1062 | } | |
1063 | while (rb_commit_index(cpu_buffer) != | |
1064 | rb_page_write(cpu_buffer->commit_page)) { | |
abc9b56d | 1065 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1066 | cpu_buffer->commit_page->write; |
1067 | barrier(); | |
1068 | } | |
a8ccf1d6 SR |
1069 | |
1070 | /* again, keep gcc from optimizing */ | |
1071 | barrier(); | |
1072 | ||
1073 | /* | |
1074 | * If an interrupt came in just after the first while loop | |
1075 | * and pushed the tail page forward, we will be left with | |
1076 | * a dangling commit that will never go forward. | |
1077 | */ | |
1078 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1079 | goto again; | |
7a8e76a3 SR |
1080 | } |
1081 | ||
d769041f | 1082 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1083 | { |
abc9b56d | 1084 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1085 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1086 | } |
1087 | ||
34a148bf | 1088 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1089 | { |
1090 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1091 | ||
1092 | /* | |
1093 | * The iterator could be on the reader page (it starts there). | |
1094 | * But the head could have moved, since the reader was | |
1095 | * found. Check for this case and assign the iterator | |
1096 | * to the head page instead of next. | |
1097 | */ | |
1098 | if (iter->head_page == cpu_buffer->reader_page) | |
1099 | iter->head_page = cpu_buffer->head_page; | |
1100 | else | |
1101 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1102 | ||
abc9b56d | 1103 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1104 | iter->head = 0; |
1105 | } | |
1106 | ||
1107 | /** | |
1108 | * ring_buffer_update_event - update event type and data | |
1109 | * @event: the even to update | |
1110 | * @type: the type of event | |
1111 | * @length: the size of the event field in the ring buffer | |
1112 | * | |
1113 | * Update the type and data fields of the event. The length | |
1114 | * is the actual size that is written to the ring buffer, | |
1115 | * and with this, we can determine what to place into the | |
1116 | * data field. | |
1117 | */ | |
34a148bf | 1118 | static void |
7a8e76a3 SR |
1119 | rb_update_event(struct ring_buffer_event *event, |
1120 | unsigned type, unsigned length) | |
1121 | { | |
334d4169 | 1122 | event->type_len = type; |
7a8e76a3 SR |
1123 | |
1124 | switch (type) { | |
1125 | ||
1126 | case RINGBUF_TYPE_PADDING: | |
7a8e76a3 | 1127 | case RINGBUF_TYPE_TIME_EXTEND: |
7a8e76a3 | 1128 | case RINGBUF_TYPE_TIME_STAMP: |
7a8e76a3 SR |
1129 | break; |
1130 | ||
334d4169 | 1131 | case 0: |
7a8e76a3 | 1132 | length -= RB_EVNT_HDR_SIZE; |
334d4169 | 1133 | if (length > RB_MAX_SMALL_DATA) |
7a8e76a3 | 1134 | event->array[0] = length; |
334d4169 LJ |
1135 | else |
1136 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
7a8e76a3 SR |
1137 | break; |
1138 | default: | |
1139 | BUG(); | |
1140 | } | |
1141 | } | |
1142 | ||
34a148bf | 1143 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
1144 | { |
1145 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
1146 | ||
1147 | /* zero length can cause confusions */ | |
1148 | if (!length) | |
1149 | length = 1; | |
1150 | ||
1151 | if (length > RB_MAX_SMALL_DATA) | |
1152 | length += sizeof(event.array[0]); | |
1153 | ||
1154 | length += RB_EVNT_HDR_SIZE; | |
1155 | length = ALIGN(length, RB_ALIGNMENT); | |
1156 | ||
1157 | return length; | |
1158 | } | |
1159 | ||
6634ff26 | 1160 | |
7a8e76a3 | 1161 | static struct ring_buffer_event * |
6634ff26 SR |
1162 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
1163 | unsigned long length, unsigned long tail, | |
1164 | struct buffer_page *commit_page, | |
1165 | struct buffer_page *tail_page, u64 *ts) | |
7a8e76a3 | 1166 | { |
6634ff26 | 1167 | struct buffer_page *next_page, *head_page, *reader_page; |
7a8e76a3 SR |
1168 | struct ring_buffer *buffer = cpu_buffer->buffer; |
1169 | struct ring_buffer_event *event; | |
78d904b4 | 1170 | bool lock_taken = false; |
6634ff26 | 1171 | unsigned long flags; |
aa20ae84 SR |
1172 | |
1173 | next_page = tail_page; | |
1174 | ||
1175 | local_irq_save(flags); | |
1176 | /* | |
1177 | * Since the write to the buffer is still not | |
1178 | * fully lockless, we must be careful with NMIs. | |
1179 | * The locks in the writers are taken when a write | |
1180 | * crosses to a new page. The locks protect against | |
1181 | * races with the readers (this will soon be fixed | |
1182 | * with a lockless solution). | |
1183 | * | |
1184 | * Because we can not protect against NMIs, and we | |
1185 | * want to keep traces reentrant, we need to manage | |
1186 | * what happens when we are in an NMI. | |
1187 | * | |
1188 | * NMIs can happen after we take the lock. | |
1189 | * If we are in an NMI, only take the lock | |
1190 | * if it is not already taken. Otherwise | |
1191 | * simply fail. | |
1192 | */ | |
1193 | if (unlikely(in_nmi())) { | |
1194 | if (!__raw_spin_trylock(&cpu_buffer->lock)) { | |
1195 | cpu_buffer->nmi_dropped++; | |
1196 | goto out_reset; | |
1197 | } | |
1198 | } else | |
1199 | __raw_spin_lock(&cpu_buffer->lock); | |
1200 | ||
1201 | lock_taken = true; | |
1202 | ||
1203 | rb_inc_page(cpu_buffer, &next_page); | |
1204 | ||
1205 | head_page = cpu_buffer->head_page; | |
1206 | reader_page = cpu_buffer->reader_page; | |
1207 | ||
1208 | /* we grabbed the lock before incrementing */ | |
1209 | if (RB_WARN_ON(cpu_buffer, next_page == reader_page)) | |
1210 | goto out_reset; | |
1211 | ||
1212 | /* | |
1213 | * If for some reason, we had an interrupt storm that made | |
1214 | * it all the way around the buffer, bail, and warn | |
1215 | * about it. | |
1216 | */ | |
1217 | if (unlikely(next_page == commit_page)) { | |
1218 | cpu_buffer->commit_overrun++; | |
1219 | goto out_reset; | |
1220 | } | |
1221 | ||
1222 | if (next_page == head_page) { | |
1223 | if (!(buffer->flags & RB_FL_OVERWRITE)) | |
1224 | goto out_reset; | |
1225 | ||
1226 | /* tail_page has not moved yet? */ | |
1227 | if (tail_page == cpu_buffer->tail_page) { | |
1228 | /* count overflows */ | |
1229 | cpu_buffer->overrun += | |
1230 | local_read(&head_page->entries); | |
1231 | ||
1232 | rb_inc_page(cpu_buffer, &head_page); | |
1233 | cpu_buffer->head_page = head_page; | |
1234 | cpu_buffer->head_page->read = 0; | |
1235 | } | |
1236 | } | |
1237 | ||
1238 | /* | |
1239 | * If the tail page is still the same as what we think | |
1240 | * it is, then it is up to us to update the tail | |
1241 | * pointer. | |
1242 | */ | |
1243 | if (tail_page == cpu_buffer->tail_page) { | |
1244 | local_set(&next_page->write, 0); | |
1245 | local_set(&next_page->entries, 0); | |
1246 | local_set(&next_page->page->commit, 0); | |
1247 | cpu_buffer->tail_page = next_page; | |
1248 | ||
1249 | /* reread the time stamp */ | |
1250 | *ts = ring_buffer_time_stamp(buffer, cpu_buffer->cpu); | |
1251 | cpu_buffer->tail_page->page->time_stamp = *ts; | |
1252 | } | |
1253 | ||
1254 | /* | |
1255 | * The actual tail page has moved forward. | |
1256 | */ | |
1257 | if (tail < BUF_PAGE_SIZE) { | |
1258 | /* Mark the rest of the page with padding */ | |
1259 | event = __rb_page_index(tail_page, tail); | |
1260 | rb_event_set_padding(event); | |
1261 | } | |
1262 | ||
8e7abf1c SR |
1263 | /* Set the write back to the previous setting */ |
1264 | local_sub(length, &tail_page->write); | |
aa20ae84 SR |
1265 | |
1266 | /* | |
1267 | * If this was a commit entry that failed, | |
1268 | * increment that too | |
1269 | */ | |
1270 | if (tail_page == cpu_buffer->commit_page && | |
1271 | tail == rb_commit_index(cpu_buffer)) { | |
1272 | rb_set_commit_to_write(cpu_buffer); | |
1273 | } | |
1274 | ||
1275 | __raw_spin_unlock(&cpu_buffer->lock); | |
1276 | local_irq_restore(flags); | |
1277 | ||
1278 | /* fail and let the caller try again */ | |
1279 | return ERR_PTR(-EAGAIN); | |
1280 | ||
45141d46 | 1281 | out_reset: |
6f3b3440 | 1282 | /* reset write */ |
8e7abf1c | 1283 | local_sub(length, &tail_page->write); |
6f3b3440 | 1284 | |
78d904b4 SR |
1285 | if (likely(lock_taken)) |
1286 | __raw_spin_unlock(&cpu_buffer->lock); | |
3e03fb7f | 1287 | local_irq_restore(flags); |
bf41a158 | 1288 | return NULL; |
7a8e76a3 SR |
1289 | } |
1290 | ||
6634ff26 SR |
1291 | static struct ring_buffer_event * |
1292 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
1293 | unsigned type, unsigned long length, u64 *ts) | |
1294 | { | |
1295 | struct buffer_page *tail_page, *commit_page; | |
1296 | struct ring_buffer_event *event; | |
1297 | unsigned long tail, write; | |
1298 | ||
1299 | commit_page = cpu_buffer->commit_page; | |
1300 | /* we just need to protect against interrupts */ | |
1301 | barrier(); | |
1302 | tail_page = cpu_buffer->tail_page; | |
1303 | write = local_add_return(length, &tail_page->write); | |
1304 | tail = write - length; | |
1305 | ||
1306 | /* See if we shot pass the end of this buffer page */ | |
1307 | if (write > BUF_PAGE_SIZE) | |
1308 | return rb_move_tail(cpu_buffer, length, tail, | |
1309 | commit_page, tail_page, ts); | |
1310 | ||
1311 | /* We reserved something on the buffer */ | |
1312 | ||
1313 | if (RB_WARN_ON(cpu_buffer, write > BUF_PAGE_SIZE)) | |
1314 | return NULL; | |
1315 | ||
1316 | event = __rb_page_index(tail_page, tail); | |
1317 | rb_update_event(event, type, length); | |
1318 | ||
1319 | /* The passed in type is zero for DATA */ | |
1320 | if (likely(!type)) | |
1321 | local_inc(&tail_page->entries); | |
1322 | ||
1323 | /* | |
1324 | * If this is a commit and the tail is zero, then update | |
1325 | * this page's time stamp. | |
1326 | */ | |
1327 | if (!tail && rb_is_commit(cpu_buffer, event)) | |
1328 | cpu_buffer->commit_page->page->time_stamp = *ts; | |
1329 | ||
1330 | return event; | |
1331 | } | |
1332 | ||
7a8e76a3 SR |
1333 | static int |
1334 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1335 | u64 *ts, u64 *delta) | |
1336 | { | |
1337 | struct ring_buffer_event *event; | |
1338 | static int once; | |
bf41a158 | 1339 | int ret; |
7a8e76a3 SR |
1340 | |
1341 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1342 | printk(KERN_WARNING "Delta way too big! %llu" | |
1343 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1344 | (unsigned long long)*delta, |
1345 | (unsigned long long)*ts, | |
1346 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1347 | WARN_ON(1); |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * The delta is too big, we to add a | |
1352 | * new timestamp. | |
1353 | */ | |
1354 | event = __rb_reserve_next(cpu_buffer, | |
1355 | RINGBUF_TYPE_TIME_EXTEND, | |
1356 | RB_LEN_TIME_EXTEND, | |
1357 | ts); | |
1358 | if (!event) | |
bf41a158 | 1359 | return -EBUSY; |
7a8e76a3 | 1360 | |
bf41a158 SR |
1361 | if (PTR_ERR(event) == -EAGAIN) |
1362 | return -EAGAIN; | |
1363 | ||
1364 | /* Only a commited time event can update the write stamp */ | |
1365 | if (rb_is_commit(cpu_buffer, event)) { | |
1366 | /* | |
1367 | * If this is the first on the page, then we need to | |
1368 | * update the page itself, and just put in a zero. | |
1369 | */ | |
1370 | if (rb_event_index(event)) { | |
1371 | event->time_delta = *delta & TS_MASK; | |
1372 | event->array[0] = *delta >> TS_SHIFT; | |
1373 | } else { | |
abc9b56d | 1374 | cpu_buffer->commit_page->page->time_stamp = *ts; |
bf41a158 SR |
1375 | event->time_delta = 0; |
1376 | event->array[0] = 0; | |
1377 | } | |
7a8e76a3 | 1378 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
1379 | /* let the caller know this was the commit */ |
1380 | ret = 1; | |
1381 | } else { | |
1382 | /* Darn, this is just wasted space */ | |
1383 | event->time_delta = 0; | |
1384 | event->array[0] = 0; | |
1385 | ret = 0; | |
7a8e76a3 SR |
1386 | } |
1387 | ||
bf41a158 SR |
1388 | *delta = 0; |
1389 | ||
1390 | return ret; | |
7a8e76a3 SR |
1391 | } |
1392 | ||
1393 | static struct ring_buffer_event * | |
1394 | rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 1395 | unsigned long length) |
7a8e76a3 SR |
1396 | { |
1397 | struct ring_buffer_event *event; | |
1398 | u64 ts, delta; | |
bf41a158 | 1399 | int commit = 0; |
818e3dd3 | 1400 | int nr_loops = 0; |
7a8e76a3 | 1401 | |
be957c44 | 1402 | length = rb_calculate_event_length(length); |
bf41a158 | 1403 | again: |
818e3dd3 SR |
1404 | /* |
1405 | * We allow for interrupts to reenter here and do a trace. | |
1406 | * If one does, it will cause this original code to loop | |
1407 | * back here. Even with heavy interrupts happening, this | |
1408 | * should only happen a few times in a row. If this happens | |
1409 | * 1000 times in a row, there must be either an interrupt | |
1410 | * storm or we have something buggy. | |
1411 | * Bail! | |
1412 | */ | |
3e89c7bb | 1413 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
818e3dd3 | 1414 | return NULL; |
818e3dd3 | 1415 | |
37886f6a | 1416 | ts = ring_buffer_time_stamp(cpu_buffer->buffer, cpu_buffer->cpu); |
7a8e76a3 | 1417 | |
bf41a158 SR |
1418 | /* |
1419 | * Only the first commit can update the timestamp. | |
1420 | * Yes there is a race here. If an interrupt comes in | |
1421 | * just after the conditional and it traces too, then it | |
1422 | * will also check the deltas. More than one timestamp may | |
1423 | * also be made. But only the entry that did the actual | |
1424 | * commit will be something other than zero. | |
1425 | */ | |
1426 | if (cpu_buffer->tail_page == cpu_buffer->commit_page && | |
1427 | rb_page_write(cpu_buffer->tail_page) == | |
1428 | rb_commit_index(cpu_buffer)) { | |
1429 | ||
7a8e76a3 SR |
1430 | delta = ts - cpu_buffer->write_stamp; |
1431 | ||
bf41a158 SR |
1432 | /* make sure this delta is calculated here */ |
1433 | barrier(); | |
1434 | ||
1435 | /* Did the write stamp get updated already? */ | |
1436 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
4143c5cb | 1437 | delta = 0; |
bf41a158 | 1438 | |
7a8e76a3 | 1439 | if (test_time_stamp(delta)) { |
7a8e76a3 | 1440 | |
bf41a158 SR |
1441 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
1442 | ||
1443 | if (commit == -EBUSY) | |
7a8e76a3 | 1444 | return NULL; |
bf41a158 SR |
1445 | |
1446 | if (commit == -EAGAIN) | |
1447 | goto again; | |
1448 | ||
1449 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 1450 | } |
bf41a158 SR |
1451 | } else |
1452 | /* Non commits have zero deltas */ | |
7a8e76a3 | 1453 | delta = 0; |
7a8e76a3 | 1454 | |
1cd8d735 | 1455 | event = __rb_reserve_next(cpu_buffer, 0, length, &ts); |
bf41a158 SR |
1456 | if (PTR_ERR(event) == -EAGAIN) |
1457 | goto again; | |
1458 | ||
1459 | if (!event) { | |
1460 | if (unlikely(commit)) | |
1461 | /* | |
1462 | * Ouch! We needed a timestamp and it was commited. But | |
1463 | * we didn't get our event reserved. | |
1464 | */ | |
1465 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 | 1466 | return NULL; |
bf41a158 | 1467 | } |
7a8e76a3 | 1468 | |
bf41a158 SR |
1469 | /* |
1470 | * If the timestamp was commited, make the commit our entry | |
1471 | * now so that we will update it when needed. | |
1472 | */ | |
1473 | if (commit) | |
1474 | rb_set_commit_event(cpu_buffer, event); | |
1475 | else if (!rb_is_commit(cpu_buffer, event)) | |
7a8e76a3 SR |
1476 | delta = 0; |
1477 | ||
1478 | event->time_delta = delta; | |
1479 | ||
1480 | return event; | |
1481 | } | |
1482 | ||
aa18efb2 | 1483 | #define TRACE_RECURSIVE_DEPTH 16 |
261842b7 SR |
1484 | |
1485 | static int trace_recursive_lock(void) | |
1486 | { | |
aa18efb2 | 1487 | current->trace_recursion++; |
261842b7 | 1488 | |
aa18efb2 SR |
1489 | if (likely(current->trace_recursion < TRACE_RECURSIVE_DEPTH)) |
1490 | return 0; | |
e057a5e5 | 1491 | |
aa18efb2 SR |
1492 | /* Disable all tracing before we do anything else */ |
1493 | tracing_off_permanent(); | |
261842b7 | 1494 | |
7d7d2b80 | 1495 | printk_once(KERN_WARNING "Tracing recursion: depth[%ld]:" |
aa18efb2 SR |
1496 | "HC[%lu]:SC[%lu]:NMI[%lu]\n", |
1497 | current->trace_recursion, | |
1498 | hardirq_count() >> HARDIRQ_SHIFT, | |
1499 | softirq_count() >> SOFTIRQ_SHIFT, | |
1500 | in_nmi()); | |
261842b7 | 1501 | |
aa18efb2 SR |
1502 | WARN_ON_ONCE(1); |
1503 | return -1; | |
261842b7 SR |
1504 | } |
1505 | ||
1506 | static void trace_recursive_unlock(void) | |
1507 | { | |
aa18efb2 | 1508 | WARN_ON_ONCE(!current->trace_recursion); |
261842b7 | 1509 | |
aa18efb2 | 1510 | current->trace_recursion--; |
261842b7 SR |
1511 | } |
1512 | ||
bf41a158 SR |
1513 | static DEFINE_PER_CPU(int, rb_need_resched); |
1514 | ||
7a8e76a3 SR |
1515 | /** |
1516 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
1517 | * @buffer: the ring buffer to reserve from | |
1518 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
1519 | * |
1520 | * Returns a reseverd event on the ring buffer to copy directly to. | |
1521 | * The user of this interface will need to get the body to write into | |
1522 | * and can use the ring_buffer_event_data() interface. | |
1523 | * | |
1524 | * The length is the length of the data needed, not the event length | |
1525 | * which also includes the event header. | |
1526 | * | |
1527 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
1528 | * If NULL is returned, then nothing has been allocated or locked. | |
1529 | */ | |
1530 | struct ring_buffer_event * | |
0a987751 | 1531 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
1532 | { |
1533 | struct ring_buffer_per_cpu *cpu_buffer; | |
1534 | struct ring_buffer_event *event; | |
bf41a158 | 1535 | int cpu, resched; |
7a8e76a3 | 1536 | |
033601a3 | 1537 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1538 | return NULL; |
1539 | ||
7a8e76a3 SR |
1540 | if (atomic_read(&buffer->record_disabled)) |
1541 | return NULL; | |
1542 | ||
bf41a158 | 1543 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 1544 | resched = ftrace_preempt_disable(); |
bf41a158 | 1545 | |
261842b7 SR |
1546 | if (trace_recursive_lock()) |
1547 | goto out_nocheck; | |
1548 | ||
7a8e76a3 SR |
1549 | cpu = raw_smp_processor_id(); |
1550 | ||
9e01c1b7 | 1551 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1552 | goto out; |
7a8e76a3 SR |
1553 | |
1554 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1555 | |
1556 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 1557 | goto out; |
7a8e76a3 | 1558 | |
be957c44 | 1559 | if (length > BUF_MAX_DATA_SIZE) |
bf41a158 | 1560 | goto out; |
7a8e76a3 | 1561 | |
1cd8d735 | 1562 | event = rb_reserve_next_event(cpu_buffer, length); |
7a8e76a3 | 1563 | if (!event) |
d769041f | 1564 | goto out; |
7a8e76a3 | 1565 | |
bf41a158 SR |
1566 | /* |
1567 | * Need to store resched state on this cpu. | |
1568 | * Only the first needs to. | |
1569 | */ | |
1570 | ||
1571 | if (preempt_count() == 1) | |
1572 | per_cpu(rb_need_resched, cpu) = resched; | |
1573 | ||
7a8e76a3 SR |
1574 | return event; |
1575 | ||
d769041f | 1576 | out: |
261842b7 SR |
1577 | trace_recursive_unlock(); |
1578 | ||
1579 | out_nocheck: | |
182e9f5f | 1580 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1581 | return NULL; |
1582 | } | |
c4f50183 | 1583 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 SR |
1584 | |
1585 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
1586 | struct ring_buffer_event *event) | |
1587 | { | |
e4906eff | 1588 | local_inc(&cpu_buffer->entries); |
bf41a158 SR |
1589 | |
1590 | /* Only process further if we own the commit */ | |
1591 | if (!rb_is_commit(cpu_buffer, event)) | |
1592 | return; | |
1593 | ||
1594 | cpu_buffer->write_stamp += event->time_delta; | |
1595 | ||
1596 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 SR |
1597 | } |
1598 | ||
1599 | /** | |
1600 | * ring_buffer_unlock_commit - commit a reserved | |
1601 | * @buffer: The buffer to commit to | |
1602 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
1603 | * |
1604 | * This commits the data to the ring buffer, and releases any locks held. | |
1605 | * | |
1606 | * Must be paired with ring_buffer_lock_reserve. | |
1607 | */ | |
1608 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 1609 | struct ring_buffer_event *event) |
7a8e76a3 SR |
1610 | { |
1611 | struct ring_buffer_per_cpu *cpu_buffer; | |
1612 | int cpu = raw_smp_processor_id(); | |
1613 | ||
1614 | cpu_buffer = buffer->buffers[cpu]; | |
1615 | ||
7a8e76a3 SR |
1616 | rb_commit(cpu_buffer, event); |
1617 | ||
261842b7 SR |
1618 | trace_recursive_unlock(); |
1619 | ||
bf41a158 SR |
1620 | /* |
1621 | * Only the last preempt count needs to restore preemption. | |
1622 | */ | |
182e9f5f SR |
1623 | if (preempt_count() == 1) |
1624 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1625 | else | |
bf41a158 | 1626 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
1627 | |
1628 | return 0; | |
1629 | } | |
c4f50183 | 1630 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 1631 | |
f3b9aae1 FW |
1632 | static inline void rb_event_discard(struct ring_buffer_event *event) |
1633 | { | |
334d4169 LJ |
1634 | /* array[0] holds the actual length for the discarded event */ |
1635 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
1636 | event->type_len = RINGBUF_TYPE_PADDING; | |
f3b9aae1 FW |
1637 | /* time delta must be non zero */ |
1638 | if (!event->time_delta) | |
1639 | event->time_delta = 1; | |
1640 | } | |
1641 | ||
fa1b47dd SR |
1642 | /** |
1643 | * ring_buffer_event_discard - discard any event in the ring buffer | |
1644 | * @event: the event to discard | |
1645 | * | |
1646 | * Sometimes a event that is in the ring buffer needs to be ignored. | |
1647 | * This function lets the user discard an event in the ring buffer | |
1648 | * and then that event will not be read later. | |
1649 | * | |
1650 | * Note, it is up to the user to be careful with this, and protect | |
1651 | * against races. If the user discards an event that has been consumed | |
1652 | * it is possible that it could corrupt the ring buffer. | |
1653 | */ | |
1654 | void ring_buffer_event_discard(struct ring_buffer_event *event) | |
1655 | { | |
f3b9aae1 | 1656 | rb_event_discard(event); |
fa1b47dd SR |
1657 | } |
1658 | EXPORT_SYMBOL_GPL(ring_buffer_event_discard); | |
1659 | ||
1660 | /** | |
1661 | * ring_buffer_commit_discard - discard an event that has not been committed | |
1662 | * @buffer: the ring buffer | |
1663 | * @event: non committed event to discard | |
1664 | * | |
1665 | * This is similar to ring_buffer_event_discard but must only be | |
1666 | * performed on an event that has not been committed yet. The difference | |
1667 | * is that this will also try to free the event from the ring buffer | |
1668 | * if another event has not been added behind it. | |
1669 | * | |
1670 | * If another event has been added behind it, it will set the event | |
1671 | * up as discarded, and perform the commit. | |
1672 | * | |
1673 | * If this function is called, do not call ring_buffer_unlock_commit on | |
1674 | * the event. | |
1675 | */ | |
1676 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
1677 | struct ring_buffer_event *event) | |
1678 | { | |
1679 | struct ring_buffer_per_cpu *cpu_buffer; | |
1680 | unsigned long new_index, old_index; | |
1681 | struct buffer_page *bpage; | |
1682 | unsigned long index; | |
1683 | unsigned long addr; | |
1684 | int cpu; | |
1685 | ||
1686 | /* The event is discarded regardless */ | |
f3b9aae1 | 1687 | rb_event_discard(event); |
fa1b47dd SR |
1688 | |
1689 | /* | |
1690 | * This must only be called if the event has not been | |
1691 | * committed yet. Thus we can assume that preemption | |
1692 | * is still disabled. | |
1693 | */ | |
74f4fd21 | 1694 | RB_WARN_ON(buffer, preemptible()); |
fa1b47dd SR |
1695 | |
1696 | cpu = smp_processor_id(); | |
1697 | cpu_buffer = buffer->buffers[cpu]; | |
1698 | ||
1699 | new_index = rb_event_index(event); | |
1700 | old_index = new_index + rb_event_length(event); | |
1701 | addr = (unsigned long)event; | |
1702 | addr &= PAGE_MASK; | |
1703 | ||
1704 | bpage = cpu_buffer->tail_page; | |
1705 | ||
1706 | if (bpage == (void *)addr && rb_page_write(bpage) == old_index) { | |
1707 | /* | |
1708 | * This is on the tail page. It is possible that | |
1709 | * a write could come in and move the tail page | |
1710 | * and write to the next page. That is fine | |
1711 | * because we just shorten what is on this page. | |
1712 | */ | |
1713 | index = local_cmpxchg(&bpage->write, old_index, new_index); | |
1714 | if (index == old_index) | |
1715 | goto out; | |
1716 | } | |
1717 | ||
1718 | /* | |
1719 | * The commit is still visible by the reader, so we | |
1720 | * must increment entries. | |
1721 | */ | |
e4906eff | 1722 | local_inc(&cpu_buffer->entries); |
fa1b47dd SR |
1723 | out: |
1724 | /* | |
1725 | * If a write came in and pushed the tail page | |
1726 | * we still need to update the commit pointer | |
1727 | * if we were the commit. | |
1728 | */ | |
1729 | if (rb_is_commit(cpu_buffer, event)) | |
1730 | rb_set_commit_to_write(cpu_buffer); | |
1731 | ||
f3b9aae1 FW |
1732 | trace_recursive_unlock(); |
1733 | ||
fa1b47dd SR |
1734 | /* |
1735 | * Only the last preempt count needs to restore preemption. | |
1736 | */ | |
1737 | if (preempt_count() == 1) | |
1738 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1739 | else | |
1740 | preempt_enable_no_resched_notrace(); | |
1741 | ||
1742 | } | |
1743 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
1744 | ||
7a8e76a3 SR |
1745 | /** |
1746 | * ring_buffer_write - write data to the buffer without reserving | |
1747 | * @buffer: The ring buffer to write to. | |
1748 | * @length: The length of the data being written (excluding the event header) | |
1749 | * @data: The data to write to the buffer. | |
1750 | * | |
1751 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
1752 | * one function. If you already have the data to write to the buffer, it | |
1753 | * may be easier to simply call this function. | |
1754 | * | |
1755 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
1756 | * and not the length of the event which would hold the header. | |
1757 | */ | |
1758 | int ring_buffer_write(struct ring_buffer *buffer, | |
1759 | unsigned long length, | |
1760 | void *data) | |
1761 | { | |
1762 | struct ring_buffer_per_cpu *cpu_buffer; | |
1763 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
1764 | void *body; |
1765 | int ret = -EBUSY; | |
bf41a158 | 1766 | int cpu, resched; |
7a8e76a3 | 1767 | |
033601a3 | 1768 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1769 | return -EBUSY; |
1770 | ||
7a8e76a3 SR |
1771 | if (atomic_read(&buffer->record_disabled)) |
1772 | return -EBUSY; | |
1773 | ||
182e9f5f | 1774 | resched = ftrace_preempt_disable(); |
bf41a158 | 1775 | |
7a8e76a3 SR |
1776 | cpu = raw_smp_processor_id(); |
1777 | ||
9e01c1b7 | 1778 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1779 | goto out; |
7a8e76a3 SR |
1780 | |
1781 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1782 | |
1783 | if (atomic_read(&cpu_buffer->record_disabled)) | |
1784 | goto out; | |
1785 | ||
be957c44 SR |
1786 | if (length > BUF_MAX_DATA_SIZE) |
1787 | goto out; | |
1788 | ||
1789 | event = rb_reserve_next_event(cpu_buffer, length); | |
7a8e76a3 SR |
1790 | if (!event) |
1791 | goto out; | |
1792 | ||
1793 | body = rb_event_data(event); | |
1794 | ||
1795 | memcpy(body, data, length); | |
1796 | ||
1797 | rb_commit(cpu_buffer, event); | |
1798 | ||
1799 | ret = 0; | |
1800 | out: | |
182e9f5f | 1801 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1802 | |
1803 | return ret; | |
1804 | } | |
c4f50183 | 1805 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 1806 | |
34a148bf | 1807 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
1808 | { |
1809 | struct buffer_page *reader = cpu_buffer->reader_page; | |
1810 | struct buffer_page *head = cpu_buffer->head_page; | |
1811 | struct buffer_page *commit = cpu_buffer->commit_page; | |
1812 | ||
1813 | return reader->read == rb_page_commit(reader) && | |
1814 | (commit == reader || | |
1815 | (commit == head && | |
1816 | head->read == rb_page_commit(commit))); | |
1817 | } | |
1818 | ||
7a8e76a3 SR |
1819 | /** |
1820 | * ring_buffer_record_disable - stop all writes into the buffer | |
1821 | * @buffer: The ring buffer to stop writes to. | |
1822 | * | |
1823 | * This prevents all writes to the buffer. Any attempt to write | |
1824 | * to the buffer after this will fail and return NULL. | |
1825 | * | |
1826 | * The caller should call synchronize_sched() after this. | |
1827 | */ | |
1828 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
1829 | { | |
1830 | atomic_inc(&buffer->record_disabled); | |
1831 | } | |
c4f50183 | 1832 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
1833 | |
1834 | /** | |
1835 | * ring_buffer_record_enable - enable writes to the buffer | |
1836 | * @buffer: The ring buffer to enable writes | |
1837 | * | |
1838 | * Note, multiple disables will need the same number of enables | |
1839 | * to truely enable the writing (much like preempt_disable). | |
1840 | */ | |
1841 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
1842 | { | |
1843 | atomic_dec(&buffer->record_disabled); | |
1844 | } | |
c4f50183 | 1845 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 SR |
1846 | |
1847 | /** | |
1848 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
1849 | * @buffer: The ring buffer to stop writes to. | |
1850 | * @cpu: The CPU buffer to stop | |
1851 | * | |
1852 | * This prevents all writes to the buffer. Any attempt to write | |
1853 | * to the buffer after this will fail and return NULL. | |
1854 | * | |
1855 | * The caller should call synchronize_sched() after this. | |
1856 | */ | |
1857 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
1858 | { | |
1859 | struct ring_buffer_per_cpu *cpu_buffer; | |
1860 | ||
9e01c1b7 | 1861 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1862 | return; |
7a8e76a3 SR |
1863 | |
1864 | cpu_buffer = buffer->buffers[cpu]; | |
1865 | atomic_inc(&cpu_buffer->record_disabled); | |
1866 | } | |
c4f50183 | 1867 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
1868 | |
1869 | /** | |
1870 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
1871 | * @buffer: The ring buffer to enable writes | |
1872 | * @cpu: The CPU to enable. | |
1873 | * | |
1874 | * Note, multiple disables will need the same number of enables | |
1875 | * to truely enable the writing (much like preempt_disable). | |
1876 | */ | |
1877 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
1878 | { | |
1879 | struct ring_buffer_per_cpu *cpu_buffer; | |
1880 | ||
9e01c1b7 | 1881 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1882 | return; |
7a8e76a3 SR |
1883 | |
1884 | cpu_buffer = buffer->buffers[cpu]; | |
1885 | atomic_dec(&cpu_buffer->record_disabled); | |
1886 | } | |
c4f50183 | 1887 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 SR |
1888 | |
1889 | /** | |
1890 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
1891 | * @buffer: The ring buffer | |
1892 | * @cpu: The per CPU buffer to get the entries from. | |
1893 | */ | |
1894 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
1895 | { | |
1896 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 1897 | unsigned long ret; |
7a8e76a3 | 1898 | |
9e01c1b7 | 1899 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1900 | return 0; |
7a8e76a3 SR |
1901 | |
1902 | cpu_buffer = buffer->buffers[cpu]; | |
e4906eff SR |
1903 | ret = (local_read(&cpu_buffer->entries) - cpu_buffer->overrun) |
1904 | - cpu_buffer->read; | |
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 | 1929 | |
f0d2c681 SR |
1930 | /** |
1931 | * ring_buffer_nmi_dropped_cpu - get the number of nmis that were dropped | |
1932 | * @buffer: The ring buffer | |
1933 | * @cpu: The per CPU buffer to get the number of overruns from | |
1934 | */ | |
1935 | unsigned long ring_buffer_nmi_dropped_cpu(struct ring_buffer *buffer, int cpu) | |
1936 | { | |
1937 | struct ring_buffer_per_cpu *cpu_buffer; | |
1938 | unsigned long ret; | |
1939 | ||
1940 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
1941 | return 0; | |
1942 | ||
1943 | cpu_buffer = buffer->buffers[cpu]; | |
1944 | ret = cpu_buffer->nmi_dropped; | |
1945 | ||
1946 | return ret; | |
1947 | } | |
1948 | EXPORT_SYMBOL_GPL(ring_buffer_nmi_dropped_cpu); | |
1949 | ||
1950 | /** | |
1951 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits | |
1952 | * @buffer: The ring buffer | |
1953 | * @cpu: The per CPU buffer to get the number of overruns from | |
1954 | */ | |
1955 | unsigned long | |
1956 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
1957 | { | |
1958 | struct ring_buffer_per_cpu *cpu_buffer; | |
1959 | unsigned long ret; | |
1960 | ||
1961 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
1962 | return 0; | |
1963 | ||
1964 | cpu_buffer = buffer->buffers[cpu]; | |
1965 | ret = cpu_buffer->commit_overrun; | |
1966 | ||
1967 | return ret; | |
1968 | } | |
1969 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
1970 | ||
7a8e76a3 SR |
1971 | /** |
1972 | * ring_buffer_entries - get the number of entries in a buffer | |
1973 | * @buffer: The ring buffer | |
1974 | * | |
1975 | * Returns the total number of entries in the ring buffer | |
1976 | * (all CPU entries) | |
1977 | */ | |
1978 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
1979 | { | |
1980 | struct ring_buffer_per_cpu *cpu_buffer; | |
1981 | unsigned long entries = 0; | |
1982 | int cpu; | |
1983 | ||
1984 | /* if you care about this being correct, lock the buffer */ | |
1985 | for_each_buffer_cpu(buffer, cpu) { | |
1986 | cpu_buffer = buffer->buffers[cpu]; | |
e4906eff SR |
1987 | entries += (local_read(&cpu_buffer->entries) - |
1988 | cpu_buffer->overrun) - cpu_buffer->read; | |
7a8e76a3 SR |
1989 | } |
1990 | ||
1991 | return entries; | |
1992 | } | |
c4f50183 | 1993 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
1994 | |
1995 | /** | |
1996 | * ring_buffer_overrun_cpu - get the number of overruns in buffer | |
1997 | * @buffer: The ring buffer | |
1998 | * | |
1999 | * Returns the total number of overruns in the ring buffer | |
2000 | * (all CPU entries) | |
2001 | */ | |
2002 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
2003 | { | |
2004 | struct ring_buffer_per_cpu *cpu_buffer; | |
2005 | unsigned long overruns = 0; | |
2006 | int cpu; | |
2007 | ||
2008 | /* if you care about this being correct, lock the buffer */ | |
2009 | for_each_buffer_cpu(buffer, cpu) { | |
2010 | cpu_buffer = buffer->buffers[cpu]; | |
2011 | overruns += cpu_buffer->overrun; | |
2012 | } | |
2013 | ||
2014 | return overruns; | |
2015 | } | |
c4f50183 | 2016 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 2017 | |
642edba5 | 2018 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
2019 | { |
2020 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2021 | ||
d769041f SR |
2022 | /* Iterator usage is expected to have record disabled */ |
2023 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
2024 | iter->head_page = cpu_buffer->head_page; | |
6f807acd | 2025 | iter->head = cpu_buffer->head_page->read; |
d769041f SR |
2026 | } else { |
2027 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 2028 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
2029 | } |
2030 | if (iter->head) | |
2031 | iter->read_stamp = cpu_buffer->read_stamp; | |
2032 | else | |
abc9b56d | 2033 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 2034 | } |
f83c9d0f | 2035 | |
642edba5 SR |
2036 | /** |
2037 | * ring_buffer_iter_reset - reset an iterator | |
2038 | * @iter: The iterator to reset | |
2039 | * | |
2040 | * Resets the iterator, so that it will start from the beginning | |
2041 | * again. | |
2042 | */ | |
2043 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
2044 | { | |
554f786e | 2045 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
2046 | unsigned long flags; |
2047 | ||
554f786e SR |
2048 | if (!iter) |
2049 | return; | |
2050 | ||
2051 | cpu_buffer = iter->cpu_buffer; | |
2052 | ||
642edba5 SR |
2053 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2054 | rb_iter_reset(iter); | |
f83c9d0f | 2055 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 2056 | } |
c4f50183 | 2057 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
2058 | |
2059 | /** | |
2060 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
2061 | * @iter: The iterator to check | |
2062 | */ | |
2063 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
2064 | { | |
2065 | struct ring_buffer_per_cpu *cpu_buffer; | |
2066 | ||
2067 | cpu_buffer = iter->cpu_buffer; | |
2068 | ||
bf41a158 SR |
2069 | return iter->head_page == cpu_buffer->commit_page && |
2070 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 2071 | } |
c4f50183 | 2072 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
2073 | |
2074 | static void | |
2075 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
2076 | struct ring_buffer_event *event) | |
2077 | { | |
2078 | u64 delta; | |
2079 | ||
334d4169 | 2080 | switch (event->type_len) { |
7a8e76a3 SR |
2081 | case RINGBUF_TYPE_PADDING: |
2082 | return; | |
2083 | ||
2084 | case RINGBUF_TYPE_TIME_EXTEND: | |
2085 | delta = event->array[0]; | |
2086 | delta <<= TS_SHIFT; | |
2087 | delta += event->time_delta; | |
2088 | cpu_buffer->read_stamp += delta; | |
2089 | return; | |
2090 | ||
2091 | case RINGBUF_TYPE_TIME_STAMP: | |
2092 | /* FIXME: not implemented */ | |
2093 | return; | |
2094 | ||
2095 | case RINGBUF_TYPE_DATA: | |
2096 | cpu_buffer->read_stamp += event->time_delta; | |
2097 | return; | |
2098 | ||
2099 | default: | |
2100 | BUG(); | |
2101 | } | |
2102 | return; | |
2103 | } | |
2104 | ||
2105 | static void | |
2106 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
2107 | struct ring_buffer_event *event) | |
2108 | { | |
2109 | u64 delta; | |
2110 | ||
334d4169 | 2111 | switch (event->type_len) { |
7a8e76a3 SR |
2112 | case RINGBUF_TYPE_PADDING: |
2113 | return; | |
2114 | ||
2115 | case RINGBUF_TYPE_TIME_EXTEND: | |
2116 | delta = event->array[0]; | |
2117 | delta <<= TS_SHIFT; | |
2118 | delta += event->time_delta; | |
2119 | iter->read_stamp += delta; | |
2120 | return; | |
2121 | ||
2122 | case RINGBUF_TYPE_TIME_STAMP: | |
2123 | /* FIXME: not implemented */ | |
2124 | return; | |
2125 | ||
2126 | case RINGBUF_TYPE_DATA: | |
2127 | iter->read_stamp += event->time_delta; | |
2128 | return; | |
2129 | ||
2130 | default: | |
2131 | BUG(); | |
2132 | } | |
2133 | return; | |
2134 | } | |
2135 | ||
d769041f SR |
2136 | static struct buffer_page * |
2137 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 2138 | { |
d769041f SR |
2139 | struct buffer_page *reader = NULL; |
2140 | unsigned long flags; | |
818e3dd3 | 2141 | int nr_loops = 0; |
d769041f | 2142 | |
3e03fb7f SR |
2143 | local_irq_save(flags); |
2144 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
2145 | |
2146 | again: | |
818e3dd3 SR |
2147 | /* |
2148 | * This should normally only loop twice. But because the | |
2149 | * start of the reader inserts an empty page, it causes | |
2150 | * a case where we will loop three times. There should be no | |
2151 | * reason to loop four times (that I know of). | |
2152 | */ | |
3e89c7bb | 2153 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
2154 | reader = NULL; |
2155 | goto out; | |
2156 | } | |
2157 | ||
d769041f SR |
2158 | reader = cpu_buffer->reader_page; |
2159 | ||
2160 | /* If there's more to read, return this page */ | |
bf41a158 | 2161 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
2162 | goto out; |
2163 | ||
2164 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
2165 | if (RB_WARN_ON(cpu_buffer, |
2166 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
2167 | goto out; | |
d769041f SR |
2168 | |
2169 | /* check if we caught up to the tail */ | |
2170 | reader = NULL; | |
bf41a158 | 2171 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 2172 | goto out; |
7a8e76a3 SR |
2173 | |
2174 | /* | |
d769041f SR |
2175 | * Splice the empty reader page into the list around the head. |
2176 | * Reset the reader page to size zero. | |
7a8e76a3 | 2177 | */ |
7a8e76a3 | 2178 | |
d769041f SR |
2179 | reader = cpu_buffer->head_page; |
2180 | cpu_buffer->reader_page->list.next = reader->list.next; | |
2181 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 SR |
2182 | |
2183 | local_set(&cpu_buffer->reader_page->write, 0); | |
778c55d4 | 2184 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 2185 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
7a8e76a3 | 2186 | |
d769041f SR |
2187 | /* Make the reader page now replace the head */ |
2188 | reader->list.prev->next = &cpu_buffer->reader_page->list; | |
2189 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
7a8e76a3 SR |
2190 | |
2191 | /* | |
d769041f SR |
2192 | * If the tail is on the reader, then we must set the head |
2193 | * to the inserted page, otherwise we set it one before. | |
7a8e76a3 | 2194 | */ |
d769041f | 2195 | cpu_buffer->head_page = cpu_buffer->reader_page; |
7a8e76a3 | 2196 | |
bf41a158 | 2197 | if (cpu_buffer->commit_page != reader) |
d769041f SR |
2198 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
2199 | ||
2200 | /* Finally update the reader page to the new head */ | |
2201 | cpu_buffer->reader_page = reader; | |
2202 | rb_reset_reader_page(cpu_buffer); | |
2203 | ||
2204 | goto again; | |
2205 | ||
2206 | out: | |
3e03fb7f SR |
2207 | __raw_spin_unlock(&cpu_buffer->lock); |
2208 | local_irq_restore(flags); | |
d769041f SR |
2209 | |
2210 | return reader; | |
2211 | } | |
2212 | ||
2213 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
2214 | { | |
2215 | struct ring_buffer_event *event; | |
2216 | struct buffer_page *reader; | |
2217 | unsigned length; | |
2218 | ||
2219 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 2220 | |
d769041f | 2221 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
2222 | if (RB_WARN_ON(cpu_buffer, !reader)) |
2223 | return; | |
7a8e76a3 | 2224 | |
d769041f SR |
2225 | event = rb_reader_event(cpu_buffer); |
2226 | ||
334d4169 LJ |
2227 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX |
2228 | || rb_discarded_event(event)) | |
e4906eff | 2229 | cpu_buffer->read++; |
d769041f SR |
2230 | |
2231 | rb_update_read_stamp(cpu_buffer, event); | |
2232 | ||
2233 | length = rb_event_length(event); | |
6f807acd | 2234 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
2235 | } |
2236 | ||
2237 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
2238 | { | |
2239 | struct ring_buffer *buffer; | |
2240 | struct ring_buffer_per_cpu *cpu_buffer; | |
2241 | struct ring_buffer_event *event; | |
2242 | unsigned length; | |
2243 | ||
2244 | cpu_buffer = iter->cpu_buffer; | |
2245 | buffer = cpu_buffer->buffer; | |
2246 | ||
2247 | /* | |
2248 | * Check if we are at the end of the buffer. | |
2249 | */ | |
bf41a158 | 2250 | if (iter->head >= rb_page_size(iter->head_page)) { |
3e89c7bb SR |
2251 | if (RB_WARN_ON(buffer, |
2252 | iter->head_page == cpu_buffer->commit_page)) | |
2253 | return; | |
d769041f | 2254 | rb_inc_iter(iter); |
7a8e76a3 SR |
2255 | return; |
2256 | } | |
2257 | ||
2258 | event = rb_iter_head_event(iter); | |
2259 | ||
2260 | length = rb_event_length(event); | |
2261 | ||
2262 | /* | |
2263 | * This should not be called to advance the header if we are | |
2264 | * at the tail of the buffer. | |
2265 | */ | |
3e89c7bb | 2266 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 2267 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
2268 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
2269 | return; | |
7a8e76a3 SR |
2270 | |
2271 | rb_update_iter_read_stamp(iter, event); | |
2272 | ||
2273 | iter->head += length; | |
2274 | ||
2275 | /* check for end of page padding */ | |
bf41a158 SR |
2276 | if ((iter->head >= rb_page_size(iter->head_page)) && |
2277 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
2278 | rb_advance_iter(iter); |
2279 | } | |
2280 | ||
f83c9d0f SR |
2281 | static struct ring_buffer_event * |
2282 | rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
7a8e76a3 SR |
2283 | { |
2284 | struct ring_buffer_per_cpu *cpu_buffer; | |
2285 | struct ring_buffer_event *event; | |
d769041f | 2286 | struct buffer_page *reader; |
818e3dd3 | 2287 | int nr_loops = 0; |
7a8e76a3 | 2288 | |
7a8e76a3 SR |
2289 | cpu_buffer = buffer->buffers[cpu]; |
2290 | ||
2291 | again: | |
818e3dd3 SR |
2292 | /* |
2293 | * We repeat when a timestamp is encountered. It is possible | |
2294 | * to get multiple timestamps from an interrupt entering just | |
2295 | * as one timestamp is about to be written. The max times | |
2296 | * that this can happen is the number of nested interrupts we | |
2297 | * can have. Nesting 10 deep of interrupts is clearly | |
2298 | * an anomaly. | |
2299 | */ | |
3e89c7bb | 2300 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 2301 | return NULL; |
818e3dd3 | 2302 | |
d769041f SR |
2303 | reader = rb_get_reader_page(cpu_buffer); |
2304 | if (!reader) | |
7a8e76a3 SR |
2305 | return NULL; |
2306 | ||
d769041f | 2307 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 2308 | |
334d4169 | 2309 | switch (event->type_len) { |
7a8e76a3 | 2310 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
2311 | if (rb_null_event(event)) |
2312 | RB_WARN_ON(cpu_buffer, 1); | |
2313 | /* | |
2314 | * Because the writer could be discarding every | |
2315 | * event it creates (which would probably be bad) | |
2316 | * if we were to go back to "again" then we may never | |
2317 | * catch up, and will trigger the warn on, or lock | |
2318 | * the box. Return the padding, and we will release | |
2319 | * the current locks, and try again. | |
2320 | */ | |
d769041f | 2321 | rb_advance_reader(cpu_buffer); |
2d622719 | 2322 | return event; |
7a8e76a3 SR |
2323 | |
2324 | case RINGBUF_TYPE_TIME_EXTEND: | |
2325 | /* Internal data, OK to advance */ | |
d769041f | 2326 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2327 | goto again; |
2328 | ||
2329 | case RINGBUF_TYPE_TIME_STAMP: | |
2330 | /* FIXME: not implemented */ | |
d769041f | 2331 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2332 | goto again; |
2333 | ||
2334 | case RINGBUF_TYPE_DATA: | |
2335 | if (ts) { | |
2336 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
37886f6a SR |
2337 | ring_buffer_normalize_time_stamp(buffer, |
2338 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2339 | } |
2340 | return event; | |
2341 | ||
2342 | default: | |
2343 | BUG(); | |
2344 | } | |
2345 | ||
2346 | return NULL; | |
2347 | } | |
c4f50183 | 2348 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 2349 | |
f83c9d0f SR |
2350 | static struct ring_buffer_event * |
2351 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
2352 | { |
2353 | struct ring_buffer *buffer; | |
2354 | struct ring_buffer_per_cpu *cpu_buffer; | |
2355 | struct ring_buffer_event *event; | |
818e3dd3 | 2356 | int nr_loops = 0; |
7a8e76a3 SR |
2357 | |
2358 | if (ring_buffer_iter_empty(iter)) | |
2359 | return NULL; | |
2360 | ||
2361 | cpu_buffer = iter->cpu_buffer; | |
2362 | buffer = cpu_buffer->buffer; | |
2363 | ||
2364 | again: | |
818e3dd3 SR |
2365 | /* |
2366 | * We repeat when a timestamp is encountered. It is possible | |
2367 | * to get multiple timestamps from an interrupt entering just | |
2368 | * as one timestamp is about to be written. The max times | |
2369 | * that this can happen is the number of nested interrupts we | |
2370 | * can have. Nesting 10 deep of interrupts is clearly | |
2371 | * an anomaly. | |
2372 | */ | |
3e89c7bb | 2373 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 10)) |
818e3dd3 | 2374 | return NULL; |
818e3dd3 | 2375 | |
7a8e76a3 SR |
2376 | if (rb_per_cpu_empty(cpu_buffer)) |
2377 | return NULL; | |
2378 | ||
2379 | event = rb_iter_head_event(iter); | |
2380 | ||
334d4169 | 2381 | switch (event->type_len) { |
7a8e76a3 | 2382 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
2383 | if (rb_null_event(event)) { |
2384 | rb_inc_iter(iter); | |
2385 | goto again; | |
2386 | } | |
2387 | rb_advance_iter(iter); | |
2388 | return event; | |
7a8e76a3 SR |
2389 | |
2390 | case RINGBUF_TYPE_TIME_EXTEND: | |
2391 | /* Internal data, OK to advance */ | |
2392 | rb_advance_iter(iter); | |
2393 | goto again; | |
2394 | ||
2395 | case RINGBUF_TYPE_TIME_STAMP: | |
2396 | /* FIXME: not implemented */ | |
2397 | rb_advance_iter(iter); | |
2398 | goto again; | |
2399 | ||
2400 | case RINGBUF_TYPE_DATA: | |
2401 | if (ts) { | |
2402 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
2403 | ring_buffer_normalize_time_stamp(buffer, |
2404 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2405 | } |
2406 | return event; | |
2407 | ||
2408 | default: | |
2409 | BUG(); | |
2410 | } | |
2411 | ||
2412 | return NULL; | |
2413 | } | |
c4f50183 | 2414 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 2415 | |
f83c9d0f SR |
2416 | /** |
2417 | * ring_buffer_peek - peek at the next event to be read | |
2418 | * @buffer: The ring buffer to read | |
2419 | * @cpu: The cpu to peak at | |
2420 | * @ts: The timestamp counter of this event. | |
2421 | * | |
2422 | * This will return the event that will be read next, but does | |
2423 | * not consume the data. | |
2424 | */ | |
2425 | struct ring_buffer_event * | |
2426 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2427 | { | |
2428 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 2429 | struct ring_buffer_event *event; |
f83c9d0f SR |
2430 | unsigned long flags; |
2431 | ||
554f786e | 2432 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2433 | return NULL; |
554f786e | 2434 | |
2d622719 | 2435 | again: |
f83c9d0f SR |
2436 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2437 | event = rb_buffer_peek(buffer, cpu, ts); | |
2438 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2439 | ||
334d4169 | 2440 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2441 | cpu_relax(); |
2442 | goto again; | |
2443 | } | |
2444 | ||
f83c9d0f SR |
2445 | return event; |
2446 | } | |
2447 | ||
2448 | /** | |
2449 | * ring_buffer_iter_peek - peek at the next event to be read | |
2450 | * @iter: The ring buffer iterator | |
2451 | * @ts: The timestamp counter of this event. | |
2452 | * | |
2453 | * This will return the event that will be read next, but does | |
2454 | * not increment the iterator. | |
2455 | */ | |
2456 | struct ring_buffer_event * | |
2457 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
2458 | { | |
2459 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2460 | struct ring_buffer_event *event; | |
2461 | unsigned long flags; | |
2462 | ||
2d622719 | 2463 | again: |
f83c9d0f SR |
2464 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2465 | event = rb_iter_peek(iter, ts); | |
2466 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2467 | ||
334d4169 | 2468 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2469 | cpu_relax(); |
2470 | goto again; | |
2471 | } | |
2472 | ||
f83c9d0f SR |
2473 | return event; |
2474 | } | |
2475 | ||
7a8e76a3 SR |
2476 | /** |
2477 | * ring_buffer_consume - return an event and consume it | |
2478 | * @buffer: The ring buffer to get the next event from | |
2479 | * | |
2480 | * Returns the next event in the ring buffer, and that event is consumed. | |
2481 | * Meaning, that sequential reads will keep returning a different event, | |
2482 | * and eventually empty the ring buffer if the producer is slower. | |
2483 | */ | |
2484 | struct ring_buffer_event * | |
2485 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2486 | { | |
554f786e SR |
2487 | struct ring_buffer_per_cpu *cpu_buffer; |
2488 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 2489 | unsigned long flags; |
7a8e76a3 | 2490 | |
2d622719 | 2491 | again: |
554f786e SR |
2492 | /* might be called in atomic */ |
2493 | preempt_disable(); | |
2494 | ||
9e01c1b7 | 2495 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 2496 | goto out; |
7a8e76a3 | 2497 | |
554f786e | 2498 | cpu_buffer = buffer->buffers[cpu]; |
f83c9d0f SR |
2499 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2500 | ||
2501 | event = rb_buffer_peek(buffer, cpu, ts); | |
7a8e76a3 | 2502 | if (!event) |
554f786e | 2503 | goto out_unlock; |
7a8e76a3 | 2504 | |
d769041f | 2505 | rb_advance_reader(cpu_buffer); |
7a8e76a3 | 2506 | |
554f786e | 2507 | out_unlock: |
f83c9d0f SR |
2508 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
2509 | ||
554f786e SR |
2510 | out: |
2511 | preempt_enable(); | |
2512 | ||
334d4169 | 2513 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2514 | cpu_relax(); |
2515 | goto again; | |
2516 | } | |
2517 | ||
7a8e76a3 SR |
2518 | return event; |
2519 | } | |
c4f50183 | 2520 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
2521 | |
2522 | /** | |
2523 | * ring_buffer_read_start - start a non consuming read of the buffer | |
2524 | * @buffer: The ring buffer to read from | |
2525 | * @cpu: The cpu buffer to iterate over | |
2526 | * | |
2527 | * This starts up an iteration through the buffer. It also disables | |
2528 | * the recording to the buffer until the reading is finished. | |
2529 | * This prevents the reading from being corrupted. This is not | |
2530 | * a consuming read, so a producer is not expected. | |
2531 | * | |
2532 | * Must be paired with ring_buffer_finish. | |
2533 | */ | |
2534 | struct ring_buffer_iter * | |
2535 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
2536 | { | |
2537 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2538 | struct ring_buffer_iter *iter; |
d769041f | 2539 | unsigned long flags; |
7a8e76a3 | 2540 | |
9e01c1b7 | 2541 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2542 | return NULL; |
7a8e76a3 SR |
2543 | |
2544 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
2545 | if (!iter) | |
8aabee57 | 2546 | return NULL; |
7a8e76a3 SR |
2547 | |
2548 | cpu_buffer = buffer->buffers[cpu]; | |
2549 | ||
2550 | iter->cpu_buffer = cpu_buffer; | |
2551 | ||
2552 | atomic_inc(&cpu_buffer->record_disabled); | |
2553 | synchronize_sched(); | |
2554 | ||
f83c9d0f | 2555 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 2556 | __raw_spin_lock(&cpu_buffer->lock); |
642edba5 | 2557 | rb_iter_reset(iter); |
3e03fb7f | 2558 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 2559 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
2560 | |
2561 | return iter; | |
2562 | } | |
c4f50183 | 2563 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
2564 | |
2565 | /** | |
2566 | * ring_buffer_finish - finish reading the iterator of the buffer | |
2567 | * @iter: The iterator retrieved by ring_buffer_start | |
2568 | * | |
2569 | * This re-enables the recording to the buffer, and frees the | |
2570 | * iterator. | |
2571 | */ | |
2572 | void | |
2573 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
2574 | { | |
2575 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2576 | ||
2577 | atomic_dec(&cpu_buffer->record_disabled); | |
2578 | kfree(iter); | |
2579 | } | |
c4f50183 | 2580 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
2581 | |
2582 | /** | |
2583 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
2584 | * @iter: The ring buffer iterator | |
2585 | * @ts: The time stamp of the event read. | |
2586 | * | |
2587 | * This reads the next event in the ring buffer and increments the iterator. | |
2588 | */ | |
2589 | struct ring_buffer_event * | |
2590 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
2591 | { | |
2592 | struct ring_buffer_event *event; | |
f83c9d0f SR |
2593 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
2594 | unsigned long flags; | |
7a8e76a3 | 2595 | |
2d622719 | 2596 | again: |
f83c9d0f SR |
2597 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2598 | event = rb_iter_peek(iter, ts); | |
7a8e76a3 | 2599 | if (!event) |
f83c9d0f | 2600 | goto out; |
7a8e76a3 SR |
2601 | |
2602 | rb_advance_iter(iter); | |
f83c9d0f SR |
2603 | out: |
2604 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 2605 | |
334d4169 | 2606 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2607 | cpu_relax(); |
2608 | goto again; | |
2609 | } | |
2610 | ||
7a8e76a3 SR |
2611 | return event; |
2612 | } | |
c4f50183 | 2613 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
2614 | |
2615 | /** | |
2616 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
2617 | * @buffer: The ring buffer. | |
2618 | */ | |
2619 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
2620 | { | |
2621 | return BUF_PAGE_SIZE * buffer->pages; | |
2622 | } | |
c4f50183 | 2623 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
2624 | |
2625 | static void | |
2626 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
2627 | { | |
2628 | cpu_buffer->head_page | |
2629 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 2630 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 2631 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 2632 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 2633 | |
6f807acd | 2634 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
2635 | |
2636 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
2637 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
2638 | ||
2639 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
2640 | local_set(&cpu_buffer->reader_page->write, 0); | |
778c55d4 | 2641 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 2642 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 2643 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 2644 | |
f0d2c681 SR |
2645 | cpu_buffer->nmi_dropped = 0; |
2646 | cpu_buffer->commit_overrun = 0; | |
7a8e76a3 | 2647 | cpu_buffer->overrun = 0; |
e4906eff SR |
2648 | cpu_buffer->read = 0; |
2649 | local_set(&cpu_buffer->entries, 0); | |
69507c06 SR |
2650 | |
2651 | cpu_buffer->write_stamp = 0; | |
2652 | cpu_buffer->read_stamp = 0; | |
7a8e76a3 SR |
2653 | } |
2654 | ||
2655 | /** | |
2656 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
2657 | * @buffer: The ring buffer to reset a per cpu buffer of | |
2658 | * @cpu: The CPU buffer to be reset | |
2659 | */ | |
2660 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
2661 | { | |
2662 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2663 | unsigned long flags; | |
2664 | ||
9e01c1b7 | 2665 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2666 | return; |
7a8e76a3 | 2667 | |
41ede23e SR |
2668 | atomic_inc(&cpu_buffer->record_disabled); |
2669 | ||
f83c9d0f SR |
2670 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2671 | ||
3e03fb7f | 2672 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
2673 | |
2674 | rb_reset_cpu(cpu_buffer); | |
2675 | ||
3e03fb7f | 2676 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f SR |
2677 | |
2678 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
41ede23e SR |
2679 | |
2680 | atomic_dec(&cpu_buffer->record_disabled); | |
7a8e76a3 | 2681 | } |
c4f50183 | 2682 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
2683 | |
2684 | /** | |
2685 | * ring_buffer_reset - reset a ring buffer | |
2686 | * @buffer: The ring buffer to reset all cpu buffers | |
2687 | */ | |
2688 | void ring_buffer_reset(struct ring_buffer *buffer) | |
2689 | { | |
7a8e76a3 SR |
2690 | int cpu; |
2691 | ||
7a8e76a3 | 2692 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 2693 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 2694 | } |
c4f50183 | 2695 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
2696 | |
2697 | /** | |
2698 | * rind_buffer_empty - is the ring buffer empty? | |
2699 | * @buffer: The ring buffer to test | |
2700 | */ | |
2701 | int ring_buffer_empty(struct ring_buffer *buffer) | |
2702 | { | |
2703 | struct ring_buffer_per_cpu *cpu_buffer; | |
2704 | int cpu; | |
2705 | ||
2706 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
2707 | for_each_buffer_cpu(buffer, cpu) { | |
2708 | cpu_buffer = buffer->buffers[cpu]; | |
2709 | if (!rb_per_cpu_empty(cpu_buffer)) | |
2710 | return 0; | |
2711 | } | |
554f786e | 2712 | |
7a8e76a3 SR |
2713 | return 1; |
2714 | } | |
c4f50183 | 2715 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
2716 | |
2717 | /** | |
2718 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
2719 | * @buffer: The ring buffer | |
2720 | * @cpu: The CPU buffer to test | |
2721 | */ | |
2722 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
2723 | { | |
2724 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2725 | int ret; |
7a8e76a3 | 2726 | |
9e01c1b7 | 2727 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2728 | return 1; |
7a8e76a3 SR |
2729 | |
2730 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e SR |
2731 | ret = rb_per_cpu_empty(cpu_buffer); |
2732 | ||
554f786e SR |
2733 | |
2734 | return ret; | |
7a8e76a3 | 2735 | } |
c4f50183 | 2736 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 SR |
2737 | |
2738 | /** | |
2739 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
2740 | * @buffer_a: One buffer to swap with | |
2741 | * @buffer_b: The other buffer to swap with | |
2742 | * | |
2743 | * This function is useful for tracers that want to take a "snapshot" | |
2744 | * of a CPU buffer and has another back up buffer lying around. | |
2745 | * it is expected that the tracer handles the cpu buffer not being | |
2746 | * used at the moment. | |
2747 | */ | |
2748 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
2749 | struct ring_buffer *buffer_b, int cpu) | |
2750 | { | |
2751 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
2752 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
2753 | int ret = -EINVAL; |
2754 | ||
9e01c1b7 RR |
2755 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
2756 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 2757 | goto out; |
7a8e76a3 SR |
2758 | |
2759 | /* At least make sure the two buffers are somewhat the same */ | |
6d102bc6 | 2760 | if (buffer_a->pages != buffer_b->pages) |
554f786e SR |
2761 | goto out; |
2762 | ||
2763 | ret = -EAGAIN; | |
7a8e76a3 | 2764 | |
97b17efe | 2765 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 2766 | goto out; |
97b17efe SR |
2767 | |
2768 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 2769 | goto out; |
97b17efe SR |
2770 | |
2771 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 2772 | goto out; |
97b17efe | 2773 | |
7a8e76a3 SR |
2774 | cpu_buffer_a = buffer_a->buffers[cpu]; |
2775 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
2776 | ||
97b17efe | 2777 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 2778 | goto out; |
97b17efe SR |
2779 | |
2780 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 2781 | goto out; |
97b17efe | 2782 | |
7a8e76a3 SR |
2783 | /* |
2784 | * We can't do a synchronize_sched here because this | |
2785 | * function can be called in atomic context. | |
2786 | * Normally this will be called from the same CPU as cpu. | |
2787 | * If not it's up to the caller to protect this. | |
2788 | */ | |
2789 | atomic_inc(&cpu_buffer_a->record_disabled); | |
2790 | atomic_inc(&cpu_buffer_b->record_disabled); | |
2791 | ||
2792 | buffer_a->buffers[cpu] = cpu_buffer_b; | |
2793 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
2794 | ||
2795 | cpu_buffer_b->buffer = buffer_a; | |
2796 | cpu_buffer_a->buffer = buffer_b; | |
2797 | ||
2798 | atomic_dec(&cpu_buffer_a->record_disabled); | |
2799 | atomic_dec(&cpu_buffer_b->record_disabled); | |
2800 | ||
554f786e SR |
2801 | ret = 0; |
2802 | out: | |
554f786e | 2803 | return ret; |
7a8e76a3 | 2804 | } |
c4f50183 | 2805 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
7a8e76a3 | 2806 | |
8789a9e7 SR |
2807 | /** |
2808 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
2809 | * @buffer: the buffer to allocate for. | |
2810 | * | |
2811 | * This function is used in conjunction with ring_buffer_read_page. | |
2812 | * When reading a full page from the ring buffer, these functions | |
2813 | * can be used to speed up the process. The calling function should | |
2814 | * allocate a few pages first with this function. Then when it | |
2815 | * needs to get pages from the ring buffer, it passes the result | |
2816 | * of this function into ring_buffer_read_page, which will swap | |
2817 | * the page that was allocated, with the read page of the buffer. | |
2818 | * | |
2819 | * Returns: | |
2820 | * The page allocated, or NULL on error. | |
2821 | */ | |
2822 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer) | |
2823 | { | |
044fa782 | 2824 | struct buffer_data_page *bpage; |
ef7a4a16 | 2825 | unsigned long addr; |
8789a9e7 SR |
2826 | |
2827 | addr = __get_free_page(GFP_KERNEL); | |
2828 | if (!addr) | |
2829 | return NULL; | |
2830 | ||
044fa782 | 2831 | bpage = (void *)addr; |
8789a9e7 | 2832 | |
ef7a4a16 SR |
2833 | rb_init_page(bpage); |
2834 | ||
044fa782 | 2835 | return bpage; |
8789a9e7 | 2836 | } |
d6ce96da | 2837 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
2838 | |
2839 | /** | |
2840 | * ring_buffer_free_read_page - free an allocated read page | |
2841 | * @buffer: the buffer the page was allocate for | |
2842 | * @data: the page to free | |
2843 | * | |
2844 | * Free a page allocated from ring_buffer_alloc_read_page. | |
2845 | */ | |
2846 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
2847 | { | |
2848 | free_page((unsigned long)data); | |
2849 | } | |
d6ce96da | 2850 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
2851 | |
2852 | /** | |
2853 | * ring_buffer_read_page - extract a page from the ring buffer | |
2854 | * @buffer: buffer to extract from | |
2855 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 2856 | * @len: amount to extract |
8789a9e7 SR |
2857 | * @cpu: the cpu of the buffer to extract |
2858 | * @full: should the extraction only happen when the page is full. | |
2859 | * | |
2860 | * This function will pull out a page from the ring buffer and consume it. | |
2861 | * @data_page must be the address of the variable that was returned | |
2862 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
2863 | * to swap with a page in the ring buffer. | |
2864 | * | |
2865 | * for example: | |
b85fa01e | 2866 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
2867 | * if (!rpage) |
2868 | * return error; | |
ef7a4a16 | 2869 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
2870 | * if (ret >= 0) |
2871 | * process_page(rpage, ret); | |
8789a9e7 SR |
2872 | * |
2873 | * When @full is set, the function will not return true unless | |
2874 | * the writer is off the reader page. | |
2875 | * | |
2876 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
2877 | * The ring buffer can be used anywhere in the kernel and can not | |
2878 | * blindly call wake_up. The layer that uses the ring buffer must be | |
2879 | * responsible for that. | |
2880 | * | |
2881 | * Returns: | |
667d2412 LJ |
2882 | * >=0 if data has been transferred, returns the offset of consumed data. |
2883 | * <0 if no data has been transferred. | |
8789a9e7 SR |
2884 | */ |
2885 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 2886 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
2887 | { |
2888 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2889 | struct ring_buffer_event *event; | |
044fa782 | 2890 | struct buffer_data_page *bpage; |
ef7a4a16 | 2891 | struct buffer_page *reader; |
8789a9e7 | 2892 | unsigned long flags; |
ef7a4a16 | 2893 | unsigned int commit; |
667d2412 | 2894 | unsigned int read; |
4f3640f8 | 2895 | u64 save_timestamp; |
667d2412 | 2896 | int ret = -1; |
8789a9e7 | 2897 | |
554f786e SR |
2898 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
2899 | goto out; | |
2900 | ||
474d32b6 SR |
2901 | /* |
2902 | * If len is not big enough to hold the page header, then | |
2903 | * we can not copy anything. | |
2904 | */ | |
2905 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 2906 | goto out; |
474d32b6 SR |
2907 | |
2908 | len -= BUF_PAGE_HDR_SIZE; | |
2909 | ||
8789a9e7 | 2910 | if (!data_page) |
554f786e | 2911 | goto out; |
8789a9e7 | 2912 | |
044fa782 SR |
2913 | bpage = *data_page; |
2914 | if (!bpage) | |
554f786e | 2915 | goto out; |
8789a9e7 SR |
2916 | |
2917 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2918 | ||
ef7a4a16 SR |
2919 | reader = rb_get_reader_page(cpu_buffer); |
2920 | if (!reader) | |
554f786e | 2921 | goto out_unlock; |
8789a9e7 | 2922 | |
ef7a4a16 SR |
2923 | event = rb_reader_event(cpu_buffer); |
2924 | ||
2925 | read = reader->read; | |
2926 | commit = rb_page_commit(reader); | |
667d2412 | 2927 | |
8789a9e7 | 2928 | /* |
474d32b6 SR |
2929 | * If this page has been partially read or |
2930 | * if len is not big enough to read the rest of the page or | |
2931 | * a writer is still on the page, then | |
2932 | * we must copy the data from the page to the buffer. | |
2933 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 2934 | */ |
474d32b6 | 2935 | if (read || (len < (commit - read)) || |
ef7a4a16 | 2936 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 2937 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
2938 | unsigned int rpos = read; |
2939 | unsigned int pos = 0; | |
ef7a4a16 | 2940 | unsigned int size; |
8789a9e7 SR |
2941 | |
2942 | if (full) | |
554f786e | 2943 | goto out_unlock; |
8789a9e7 | 2944 | |
ef7a4a16 SR |
2945 | if (len > (commit - read)) |
2946 | len = (commit - read); | |
2947 | ||
2948 | size = rb_event_length(event); | |
2949 | ||
2950 | if (len < size) | |
554f786e | 2951 | goto out_unlock; |
ef7a4a16 | 2952 | |
4f3640f8 SR |
2953 | /* save the current timestamp, since the user will need it */ |
2954 | save_timestamp = cpu_buffer->read_stamp; | |
2955 | ||
ef7a4a16 SR |
2956 | /* Need to copy one event at a time */ |
2957 | do { | |
474d32b6 | 2958 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
2959 | |
2960 | len -= size; | |
2961 | ||
2962 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
2963 | rpos = reader->read; |
2964 | pos += size; | |
ef7a4a16 SR |
2965 | |
2966 | event = rb_reader_event(cpu_buffer); | |
2967 | size = rb_event_length(event); | |
2968 | } while (len > size); | |
667d2412 LJ |
2969 | |
2970 | /* update bpage */ | |
ef7a4a16 | 2971 | local_set(&bpage->commit, pos); |
4f3640f8 | 2972 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 2973 | |
474d32b6 SR |
2974 | /* we copied everything to the beginning */ |
2975 | read = 0; | |
8789a9e7 | 2976 | } else { |
afbab76a SR |
2977 | /* update the entry counter */ |
2978 | cpu_buffer->read += local_read(&reader->entries); | |
2979 | ||
8789a9e7 | 2980 | /* swap the pages */ |
044fa782 | 2981 | rb_init_page(bpage); |
ef7a4a16 SR |
2982 | bpage = reader->page; |
2983 | reader->page = *data_page; | |
2984 | local_set(&reader->write, 0); | |
778c55d4 | 2985 | local_set(&reader->entries, 0); |
ef7a4a16 | 2986 | reader->read = 0; |
044fa782 | 2987 | *data_page = bpage; |
8789a9e7 | 2988 | } |
667d2412 | 2989 | ret = read; |
8789a9e7 | 2990 | |
554f786e | 2991 | out_unlock: |
8789a9e7 SR |
2992 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
2993 | ||
554f786e | 2994 | out: |
8789a9e7 SR |
2995 | return ret; |
2996 | } | |
d6ce96da | 2997 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 2998 | |
a3583244 SR |
2999 | static ssize_t |
3000 | rb_simple_read(struct file *filp, char __user *ubuf, | |
3001 | size_t cnt, loff_t *ppos) | |
3002 | { | |
5e39841c | 3003 | unsigned long *p = filp->private_data; |
a3583244 SR |
3004 | char buf[64]; |
3005 | int r; | |
3006 | ||
033601a3 SR |
3007 | if (test_bit(RB_BUFFERS_DISABLED_BIT, p)) |
3008 | r = sprintf(buf, "permanently disabled\n"); | |
3009 | else | |
3010 | r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p)); | |
a3583244 SR |
3011 | |
3012 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3013 | } | |
3014 | ||
3015 | static ssize_t | |
3016 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
3017 | size_t cnt, loff_t *ppos) | |
3018 | { | |
5e39841c | 3019 | unsigned long *p = filp->private_data; |
a3583244 | 3020 | char buf[64]; |
5e39841c | 3021 | unsigned long val; |
a3583244 SR |
3022 | int ret; |
3023 | ||
3024 | if (cnt >= sizeof(buf)) | |
3025 | return -EINVAL; | |
3026 | ||
3027 | if (copy_from_user(&buf, ubuf, cnt)) | |
3028 | return -EFAULT; | |
3029 | ||
3030 | buf[cnt] = 0; | |
3031 | ||
3032 | ret = strict_strtoul(buf, 10, &val); | |
3033 | if (ret < 0) | |
3034 | return ret; | |
3035 | ||
033601a3 SR |
3036 | if (val) |
3037 | set_bit(RB_BUFFERS_ON_BIT, p); | |
3038 | else | |
3039 | clear_bit(RB_BUFFERS_ON_BIT, p); | |
a3583244 SR |
3040 | |
3041 | (*ppos)++; | |
3042 | ||
3043 | return cnt; | |
3044 | } | |
3045 | ||
5e2336a0 | 3046 | static const struct file_operations rb_simple_fops = { |
a3583244 SR |
3047 | .open = tracing_open_generic, |
3048 | .read = rb_simple_read, | |
3049 | .write = rb_simple_write, | |
3050 | }; | |
3051 | ||
3052 | ||
3053 | static __init int rb_init_debugfs(void) | |
3054 | { | |
3055 | struct dentry *d_tracer; | |
a3583244 SR |
3056 | |
3057 | d_tracer = tracing_init_dentry(); | |
3058 | ||
5452af66 FW |
3059 | trace_create_file("tracing_on", 0644, d_tracer, |
3060 | &ring_buffer_flags, &rb_simple_fops); | |
a3583244 SR |
3061 | |
3062 | return 0; | |
3063 | } | |
3064 | ||
3065 | fs_initcall(rb_init_debugfs); | |
554f786e | 3066 | |
59222efe | 3067 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
3068 | static int rb_cpu_notify(struct notifier_block *self, |
3069 | unsigned long action, void *hcpu) | |
554f786e SR |
3070 | { |
3071 | struct ring_buffer *buffer = | |
3072 | container_of(self, struct ring_buffer, cpu_notify); | |
3073 | long cpu = (long)hcpu; | |
3074 | ||
3075 | switch (action) { | |
3076 | case CPU_UP_PREPARE: | |
3077 | case CPU_UP_PREPARE_FROZEN: | |
3078 | if (cpu_isset(cpu, *buffer->cpumask)) | |
3079 | return NOTIFY_OK; | |
3080 | ||
3081 | buffer->buffers[cpu] = | |
3082 | rb_allocate_cpu_buffer(buffer, cpu); | |
3083 | if (!buffer->buffers[cpu]) { | |
3084 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
3085 | cpu); | |
3086 | return NOTIFY_OK; | |
3087 | } | |
3088 | smp_wmb(); | |
3089 | cpu_set(cpu, *buffer->cpumask); | |
3090 | break; | |
3091 | case CPU_DOWN_PREPARE: | |
3092 | case CPU_DOWN_PREPARE_FROZEN: | |
3093 | /* | |
3094 | * Do nothing. | |
3095 | * If we were to free the buffer, then the user would | |
3096 | * lose any trace that was in the buffer. | |
3097 | */ | |
3098 | break; | |
3099 | default: | |
3100 | break; | |
3101 | } | |
3102 | return NOTIFY_OK; | |
3103 | } | |
3104 | #endif |