]>
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 | 207 | #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
c7b09308 | 208 | #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */ |
334d4169 LJ |
209 | |
210 | /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */ | |
211 | #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX | |
7a8e76a3 SR |
212 | |
213 | enum { | |
214 | RB_LEN_TIME_EXTEND = 8, | |
215 | RB_LEN_TIME_STAMP = 16, | |
216 | }; | |
217 | ||
2d622719 TZ |
218 | static inline int rb_null_event(struct ring_buffer_event *event) |
219 | { | |
334d4169 LJ |
220 | return event->type_len == RINGBUF_TYPE_PADDING |
221 | && event->time_delta == 0; | |
2d622719 TZ |
222 | } |
223 | ||
224 | static inline int rb_discarded_event(struct ring_buffer_event *event) | |
225 | { | |
334d4169 | 226 | return event->type_len == RINGBUF_TYPE_PADDING && event->time_delta; |
2d622719 TZ |
227 | } |
228 | ||
229 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
230 | { | |
334d4169 | 231 | event->type_len = RINGBUF_TYPE_PADDING; |
2d622719 TZ |
232 | event->time_delta = 0; |
233 | } | |
234 | ||
34a148bf | 235 | static unsigned |
2d622719 | 236 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
237 | { |
238 | unsigned length; | |
239 | ||
334d4169 LJ |
240 | if (event->type_len) |
241 | length = event->type_len * RB_ALIGNMENT; | |
2d622719 TZ |
242 | else |
243 | length = event->array[0]; | |
244 | return length + RB_EVNT_HDR_SIZE; | |
245 | } | |
246 | ||
247 | /* inline for ring buffer fast paths */ | |
248 | static unsigned | |
249 | rb_event_length(struct ring_buffer_event *event) | |
250 | { | |
334d4169 | 251 | switch (event->type_len) { |
7a8e76a3 | 252 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
253 | if (rb_null_event(event)) |
254 | /* undefined */ | |
255 | return -1; | |
334d4169 | 256 | return event->array[0] + RB_EVNT_HDR_SIZE; |
7a8e76a3 SR |
257 | |
258 | case RINGBUF_TYPE_TIME_EXTEND: | |
259 | return RB_LEN_TIME_EXTEND; | |
260 | ||
261 | case RINGBUF_TYPE_TIME_STAMP: | |
262 | return RB_LEN_TIME_STAMP; | |
263 | ||
264 | case RINGBUF_TYPE_DATA: | |
2d622719 | 265 | return rb_event_data_length(event); |
7a8e76a3 SR |
266 | default: |
267 | BUG(); | |
268 | } | |
269 | /* not hit */ | |
270 | return 0; | |
271 | } | |
272 | ||
273 | /** | |
274 | * ring_buffer_event_length - return the length of the event | |
275 | * @event: the event to get the length of | |
276 | */ | |
277 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
278 | { | |
465634ad | 279 | unsigned length = rb_event_length(event); |
334d4169 | 280 | if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
465634ad RR |
281 | return length; |
282 | length -= RB_EVNT_HDR_SIZE; | |
283 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
284 | length -= sizeof(event->array[0]); | |
285 | return length; | |
7a8e76a3 | 286 | } |
c4f50183 | 287 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
288 | |
289 | /* inline for ring buffer fast paths */ | |
34a148bf | 290 | static void * |
7a8e76a3 SR |
291 | rb_event_data(struct ring_buffer_event *event) |
292 | { | |
334d4169 | 293 | BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX); |
7a8e76a3 | 294 | /* If length is in len field, then array[0] has the data */ |
334d4169 | 295 | if (event->type_len) |
7a8e76a3 SR |
296 | return (void *)&event->array[0]; |
297 | /* Otherwise length is in array[0] and array[1] has the data */ | |
298 | return (void *)&event->array[1]; | |
299 | } | |
300 | ||
301 | /** | |
302 | * ring_buffer_event_data - return the data of the event | |
303 | * @event: the event to get the data from | |
304 | */ | |
305 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
306 | { | |
307 | return rb_event_data(event); | |
308 | } | |
c4f50183 | 309 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
310 | |
311 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 312 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
313 | |
314 | #define TS_SHIFT 27 | |
315 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
316 | #define TS_DELTA_TEST (~TS_MASK) | |
317 | ||
abc9b56d | 318 | struct buffer_data_page { |
e4c2ce82 | 319 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 320 | local_t commit; /* write committed index */ |
abc9b56d SR |
321 | unsigned char data[]; /* data of buffer page */ |
322 | }; | |
323 | ||
324 | struct buffer_page { | |
778c55d4 | 325 | struct list_head list; /* list of buffer pages */ |
abc9b56d | 326 | local_t write; /* index for next write */ |
6f807acd | 327 | unsigned read; /* index for next read */ |
778c55d4 | 328 | local_t entries; /* entries on this page */ |
abc9b56d | 329 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
330 | }; |
331 | ||
044fa782 | 332 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 333 | { |
044fa782 | 334 | local_set(&bpage->commit, 0); |
abc9b56d SR |
335 | } |
336 | ||
474d32b6 SR |
337 | /** |
338 | * ring_buffer_page_len - the size of data on the page. | |
339 | * @page: The page to read | |
340 | * | |
341 | * Returns the amount of data on the page, including buffer page header. | |
342 | */ | |
ef7a4a16 SR |
343 | size_t ring_buffer_page_len(void *page) |
344 | { | |
474d32b6 SR |
345 | return local_read(&((struct buffer_data_page *)page)->commit) |
346 | + BUF_PAGE_HDR_SIZE; | |
ef7a4a16 SR |
347 | } |
348 | ||
ed56829c SR |
349 | /* |
350 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
351 | * this issue out. | |
352 | */ | |
34a148bf | 353 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 354 | { |
34a148bf | 355 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 356 | kfree(bpage); |
ed56829c SR |
357 | } |
358 | ||
7a8e76a3 SR |
359 | /* |
360 | * We need to fit the time_stamp delta into 27 bits. | |
361 | */ | |
362 | static inline int test_time_stamp(u64 delta) | |
363 | { | |
364 | if (delta & TS_DELTA_TEST) | |
365 | return 1; | |
366 | return 0; | |
367 | } | |
368 | ||
474d32b6 | 369 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 | 370 | |
be957c44 SR |
371 | /* Max payload is BUF_PAGE_SIZE - header (8bytes) */ |
372 | #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) | |
373 | ||
ea05b57c SR |
374 | /* Max number of timestamps that can fit on a page */ |
375 | #define RB_TIMESTAMPS_PER_PAGE (BUF_PAGE_SIZE / RB_LEN_TIME_STAMP) | |
376 | ||
d1b182a8 SR |
377 | int ring_buffer_print_page_header(struct trace_seq *s) |
378 | { | |
379 | struct buffer_data_page field; | |
380 | int ret; | |
381 | ||
382 | ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t" | |
383 | "offset:0;\tsize:%u;\n", | |
384 | (unsigned int)sizeof(field.time_stamp)); | |
385 | ||
386 | ret = trace_seq_printf(s, "\tfield: local_t commit;\t" | |
387 | "offset:%u;\tsize:%u;\n", | |
388 | (unsigned int)offsetof(typeof(field), commit), | |
389 | (unsigned int)sizeof(field.commit)); | |
390 | ||
391 | ret = trace_seq_printf(s, "\tfield: char data;\t" | |
392 | "offset:%u;\tsize:%u;\n", | |
393 | (unsigned int)offsetof(typeof(field), data), | |
394 | (unsigned int)BUF_PAGE_SIZE); | |
395 | ||
396 | return ret; | |
397 | } | |
398 | ||
7a8e76a3 SR |
399 | /* |
400 | * head_page == tail_page && head == tail then buffer is empty. | |
401 | */ | |
402 | struct ring_buffer_per_cpu { | |
403 | int cpu; | |
404 | struct ring_buffer *buffer; | |
f83c9d0f | 405 | spinlock_t reader_lock; /* serialize readers */ |
3e03fb7f | 406 | raw_spinlock_t lock; |
7a8e76a3 SR |
407 | struct lock_class_key lock_key; |
408 | struct list_head pages; | |
6f807acd SR |
409 | struct buffer_page *head_page; /* read from head */ |
410 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 411 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 412 | struct buffer_page *reader_page; |
f0d2c681 SR |
413 | unsigned long nmi_dropped; |
414 | unsigned long commit_overrun; | |
7a8e76a3 | 415 | unsigned long overrun; |
e4906eff SR |
416 | unsigned long read; |
417 | local_t entries; | |
fa743953 SR |
418 | local_t committing; |
419 | local_t commits; | |
7a8e76a3 SR |
420 | u64 write_stamp; |
421 | u64 read_stamp; | |
422 | atomic_t record_disabled; | |
423 | }; | |
424 | ||
425 | struct ring_buffer { | |
7a8e76a3 SR |
426 | unsigned pages; |
427 | unsigned flags; | |
428 | int cpus; | |
7a8e76a3 | 429 | atomic_t record_disabled; |
00f62f61 | 430 | cpumask_var_t cpumask; |
7a8e76a3 | 431 | |
1f8a6a10 PZ |
432 | struct lock_class_key *reader_lock_key; |
433 | ||
7a8e76a3 SR |
434 | struct mutex mutex; |
435 | ||
436 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 437 | |
59222efe | 438 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
439 | struct notifier_block cpu_notify; |
440 | #endif | |
37886f6a | 441 | u64 (*clock)(void); |
7a8e76a3 SR |
442 | }; |
443 | ||
444 | struct ring_buffer_iter { | |
445 | struct ring_buffer_per_cpu *cpu_buffer; | |
446 | unsigned long head; | |
447 | struct buffer_page *head_page; | |
448 | u64 read_stamp; | |
449 | }; | |
450 | ||
f536aafc | 451 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
bf41a158 | 452 | #define RB_WARN_ON(buffer, cond) \ |
3e89c7bb SR |
453 | ({ \ |
454 | int _____ret = unlikely(cond); \ | |
455 | if (_____ret) { \ | |
bf41a158 SR |
456 | atomic_inc(&buffer->record_disabled); \ |
457 | WARN_ON(1); \ | |
458 | } \ | |
3e89c7bb SR |
459 | _____ret; \ |
460 | }) | |
f536aafc | 461 | |
37886f6a SR |
462 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
463 | #define DEBUG_SHIFT 0 | |
464 | ||
88eb0125 SR |
465 | static inline u64 rb_time_stamp(struct ring_buffer *buffer, int cpu) |
466 | { | |
467 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
468 | return buffer->clock() << DEBUG_SHIFT; | |
469 | } | |
470 | ||
37886f6a SR |
471 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) |
472 | { | |
473 | u64 time; | |
474 | ||
475 | preempt_disable_notrace(); | |
88eb0125 | 476 | time = rb_time_stamp(buffer, cpu); |
37886f6a SR |
477 | preempt_enable_no_resched_notrace(); |
478 | ||
479 | return time; | |
480 | } | |
481 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
482 | ||
483 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
484 | int cpu, u64 *ts) | |
485 | { | |
486 | /* Just stupid testing the normalize function and deltas */ | |
487 | *ts >>= DEBUG_SHIFT; | |
488 | } | |
489 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
490 | ||
7a8e76a3 SR |
491 | /** |
492 | * check_pages - integrity check of buffer pages | |
493 | * @cpu_buffer: CPU buffer with pages to test | |
494 | * | |
c3706f00 | 495 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
496 | * been corrupted. |
497 | */ | |
498 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
499 | { | |
500 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 501 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 502 | |
3e89c7bb SR |
503 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
504 | return -1; | |
505 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
506 | return -1; | |
7a8e76a3 | 507 | |
044fa782 | 508 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 509 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 510 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
511 | return -1; |
512 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 513 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 514 | return -1; |
7a8e76a3 SR |
515 | } |
516 | ||
517 | return 0; | |
518 | } | |
519 | ||
7a8e76a3 SR |
520 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
521 | unsigned nr_pages) | |
522 | { | |
523 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 524 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
525 | unsigned long addr; |
526 | LIST_HEAD(pages); | |
527 | unsigned i; | |
528 | ||
529 | for (i = 0; i < nr_pages; i++) { | |
044fa782 | 530 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
aa1e0e3b | 531 | GFP_KERNEL, cpu_to_node(cpu_buffer->cpu)); |
044fa782 | 532 | if (!bpage) |
e4c2ce82 | 533 | goto free_pages; |
044fa782 | 534 | list_add(&bpage->list, &pages); |
e4c2ce82 | 535 | |
7a8e76a3 SR |
536 | addr = __get_free_page(GFP_KERNEL); |
537 | if (!addr) | |
538 | goto free_pages; | |
044fa782 SR |
539 | bpage->page = (void *)addr; |
540 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
541 | } |
542 | ||
543 | list_splice(&pages, head); | |
544 | ||
545 | rb_check_pages(cpu_buffer); | |
546 | ||
547 | return 0; | |
548 | ||
549 | free_pages: | |
044fa782 SR |
550 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
551 | list_del_init(&bpage->list); | |
552 | free_buffer_page(bpage); | |
7a8e76a3 SR |
553 | } |
554 | return -ENOMEM; | |
555 | } | |
556 | ||
557 | static struct ring_buffer_per_cpu * | |
558 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) | |
559 | { | |
560 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 561 | struct buffer_page *bpage; |
d769041f | 562 | unsigned long addr; |
7a8e76a3 SR |
563 | int ret; |
564 | ||
565 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
566 | GFP_KERNEL, cpu_to_node(cpu)); | |
567 | if (!cpu_buffer) | |
568 | return NULL; | |
569 | ||
570 | cpu_buffer->cpu = cpu; | |
571 | cpu_buffer->buffer = buffer; | |
f83c9d0f | 572 | spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 573 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
3e03fb7f | 574 | cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; |
7a8e76a3 SR |
575 | INIT_LIST_HEAD(&cpu_buffer->pages); |
576 | ||
044fa782 | 577 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 578 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 579 | if (!bpage) |
e4c2ce82 SR |
580 | goto fail_free_buffer; |
581 | ||
044fa782 | 582 | cpu_buffer->reader_page = bpage; |
d769041f SR |
583 | addr = __get_free_page(GFP_KERNEL); |
584 | if (!addr) | |
e4c2ce82 | 585 | goto fail_free_reader; |
044fa782 SR |
586 | bpage->page = (void *)addr; |
587 | rb_init_page(bpage->page); | |
e4c2ce82 | 588 | |
d769041f | 589 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 590 | |
7a8e76a3 SR |
591 | ret = rb_allocate_pages(cpu_buffer, buffer->pages); |
592 | if (ret < 0) | |
d769041f | 593 | goto fail_free_reader; |
7a8e76a3 SR |
594 | |
595 | cpu_buffer->head_page | |
596 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 597 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 SR |
598 | |
599 | return cpu_buffer; | |
600 | ||
d769041f SR |
601 | fail_free_reader: |
602 | free_buffer_page(cpu_buffer->reader_page); | |
603 | ||
7a8e76a3 SR |
604 | fail_free_buffer: |
605 | kfree(cpu_buffer); | |
606 | return NULL; | |
607 | } | |
608 | ||
609 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
610 | { | |
611 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 612 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 613 | |
d769041f SR |
614 | free_buffer_page(cpu_buffer->reader_page); |
615 | ||
044fa782 SR |
616 | list_for_each_entry_safe(bpage, tmp, head, list) { |
617 | list_del_init(&bpage->list); | |
618 | free_buffer_page(bpage); | |
7a8e76a3 SR |
619 | } |
620 | kfree(cpu_buffer); | |
621 | } | |
622 | ||
59222efe | 623 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
624 | static int rb_cpu_notify(struct notifier_block *self, |
625 | unsigned long action, void *hcpu); | |
554f786e SR |
626 | #endif |
627 | ||
7a8e76a3 SR |
628 | /** |
629 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 630 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
631 | * @flags: attributes to set for the ring buffer. |
632 | * | |
633 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
634 | * flag. This flag means that the buffer will overwrite old data | |
635 | * when the buffer wraps. If this flag is not set, the buffer will | |
636 | * drop data when the tail hits the head. | |
637 | */ | |
1f8a6a10 PZ |
638 | struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
639 | struct lock_class_key *key) | |
7a8e76a3 SR |
640 | { |
641 | struct ring_buffer *buffer; | |
642 | int bsize; | |
643 | int cpu; | |
644 | ||
645 | /* keep it in its own cache line */ | |
646 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
647 | GFP_KERNEL); | |
648 | if (!buffer) | |
649 | return NULL; | |
650 | ||
9e01c1b7 RR |
651 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
652 | goto fail_free_buffer; | |
653 | ||
7a8e76a3 SR |
654 | buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
655 | buffer->flags = flags; | |
37886f6a | 656 | buffer->clock = trace_clock_local; |
1f8a6a10 | 657 | buffer->reader_lock_key = key; |
7a8e76a3 SR |
658 | |
659 | /* need at least two pages */ | |
5f78abee SR |
660 | if (buffer->pages < 2) |
661 | buffer->pages = 2; | |
7a8e76a3 | 662 | |
3bf832ce FW |
663 | /* |
664 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
665 | * in early initcall, it will not be notified of secondary cpus. | |
666 | * In that off case, we need to allocate for all possible cpus. | |
667 | */ | |
668 | #ifdef CONFIG_HOTPLUG_CPU | |
554f786e SR |
669 | get_online_cpus(); |
670 | cpumask_copy(buffer->cpumask, cpu_online_mask); | |
3bf832ce FW |
671 | #else |
672 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
673 | #endif | |
7a8e76a3 SR |
674 | buffer->cpus = nr_cpu_ids; |
675 | ||
676 | bsize = sizeof(void *) * nr_cpu_ids; | |
677 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
678 | GFP_KERNEL); | |
679 | if (!buffer->buffers) | |
9e01c1b7 | 680 | goto fail_free_cpumask; |
7a8e76a3 SR |
681 | |
682 | for_each_buffer_cpu(buffer, cpu) { | |
683 | buffer->buffers[cpu] = | |
684 | rb_allocate_cpu_buffer(buffer, cpu); | |
685 | if (!buffer->buffers[cpu]) | |
686 | goto fail_free_buffers; | |
687 | } | |
688 | ||
59222efe | 689 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
690 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
691 | buffer->cpu_notify.priority = 0; | |
692 | register_cpu_notifier(&buffer->cpu_notify); | |
693 | #endif | |
694 | ||
695 | put_online_cpus(); | |
7a8e76a3 SR |
696 | mutex_init(&buffer->mutex); |
697 | ||
698 | return buffer; | |
699 | ||
700 | fail_free_buffers: | |
701 | for_each_buffer_cpu(buffer, cpu) { | |
702 | if (buffer->buffers[cpu]) | |
703 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
704 | } | |
705 | kfree(buffer->buffers); | |
706 | ||
9e01c1b7 RR |
707 | fail_free_cpumask: |
708 | free_cpumask_var(buffer->cpumask); | |
554f786e | 709 | put_online_cpus(); |
9e01c1b7 | 710 | |
7a8e76a3 SR |
711 | fail_free_buffer: |
712 | kfree(buffer); | |
713 | return NULL; | |
714 | } | |
1f8a6a10 | 715 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
716 | |
717 | /** | |
718 | * ring_buffer_free - free a ring buffer. | |
719 | * @buffer: the buffer to free. | |
720 | */ | |
721 | void | |
722 | ring_buffer_free(struct ring_buffer *buffer) | |
723 | { | |
724 | int cpu; | |
725 | ||
554f786e SR |
726 | get_online_cpus(); |
727 | ||
59222efe | 728 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
729 | unregister_cpu_notifier(&buffer->cpu_notify); |
730 | #endif | |
731 | ||
7a8e76a3 SR |
732 | for_each_buffer_cpu(buffer, cpu) |
733 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
734 | ||
554f786e SR |
735 | put_online_cpus(); |
736 | ||
9e01c1b7 RR |
737 | free_cpumask_var(buffer->cpumask); |
738 | ||
7a8e76a3 SR |
739 | kfree(buffer); |
740 | } | |
c4f50183 | 741 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 742 | |
37886f6a SR |
743 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
744 | u64 (*clock)(void)) | |
745 | { | |
746 | buffer->clock = clock; | |
747 | } | |
748 | ||
7a8e76a3 SR |
749 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
750 | ||
751 | static void | |
752 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages) | |
753 | { | |
044fa782 | 754 | struct buffer_page *bpage; |
7a8e76a3 SR |
755 | struct list_head *p; |
756 | unsigned i; | |
757 | ||
758 | atomic_inc(&cpu_buffer->record_disabled); | |
759 | synchronize_sched(); | |
760 | ||
761 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
762 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
763 | return; | |
7a8e76a3 | 764 | p = cpu_buffer->pages.next; |
044fa782 SR |
765 | bpage = list_entry(p, struct buffer_page, list); |
766 | list_del_init(&bpage->list); | |
767 | free_buffer_page(bpage); | |
7a8e76a3 | 768 | } |
3e89c7bb SR |
769 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
770 | return; | |
7a8e76a3 SR |
771 | |
772 | rb_reset_cpu(cpu_buffer); | |
773 | ||
774 | rb_check_pages(cpu_buffer); | |
775 | ||
776 | atomic_dec(&cpu_buffer->record_disabled); | |
777 | ||
778 | } | |
779 | ||
780 | static void | |
781 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
782 | struct list_head *pages, unsigned nr_pages) | |
783 | { | |
044fa782 | 784 | struct buffer_page *bpage; |
7a8e76a3 SR |
785 | struct list_head *p; |
786 | unsigned i; | |
787 | ||
788 | atomic_inc(&cpu_buffer->record_disabled); | |
789 | synchronize_sched(); | |
790 | ||
791 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
792 | if (RB_WARN_ON(cpu_buffer, list_empty(pages))) |
793 | return; | |
7a8e76a3 | 794 | p = pages->next; |
044fa782 SR |
795 | bpage = list_entry(p, struct buffer_page, list); |
796 | list_del_init(&bpage->list); | |
797 | list_add_tail(&bpage->list, &cpu_buffer->pages); | |
7a8e76a3 SR |
798 | } |
799 | rb_reset_cpu(cpu_buffer); | |
800 | ||
801 | rb_check_pages(cpu_buffer); | |
802 | ||
803 | atomic_dec(&cpu_buffer->record_disabled); | |
804 | } | |
805 | ||
806 | /** | |
807 | * ring_buffer_resize - resize the ring buffer | |
808 | * @buffer: the buffer to resize. | |
809 | * @size: the new size. | |
810 | * | |
811 | * The tracer is responsible for making sure that the buffer is | |
812 | * not being used while changing the size. | |
813 | * Note: We may be able to change the above requirement by using | |
814 | * RCU synchronizations. | |
815 | * | |
816 | * Minimum size is 2 * BUF_PAGE_SIZE. | |
817 | * | |
818 | * Returns -1 on failure. | |
819 | */ | |
820 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) | |
821 | { | |
822 | struct ring_buffer_per_cpu *cpu_buffer; | |
823 | unsigned nr_pages, rm_pages, new_pages; | |
044fa782 | 824 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
825 | unsigned long buffer_size; |
826 | unsigned long addr; | |
827 | LIST_HEAD(pages); | |
828 | int i, cpu; | |
829 | ||
ee51a1de IM |
830 | /* |
831 | * Always succeed at resizing a non-existent buffer: | |
832 | */ | |
833 | if (!buffer) | |
834 | return size; | |
835 | ||
7a8e76a3 SR |
836 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
837 | size *= BUF_PAGE_SIZE; | |
838 | buffer_size = buffer->pages * BUF_PAGE_SIZE; | |
839 | ||
840 | /* we need a minimum of two pages */ | |
841 | if (size < BUF_PAGE_SIZE * 2) | |
842 | size = BUF_PAGE_SIZE * 2; | |
843 | ||
844 | if (size == buffer_size) | |
845 | return size; | |
846 | ||
847 | mutex_lock(&buffer->mutex); | |
554f786e | 848 | get_online_cpus(); |
7a8e76a3 SR |
849 | |
850 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
851 | ||
852 | if (size < buffer_size) { | |
853 | ||
854 | /* easy case, just free pages */ | |
554f786e SR |
855 | if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) |
856 | goto out_fail; | |
7a8e76a3 SR |
857 | |
858 | rm_pages = buffer->pages - nr_pages; | |
859 | ||
860 | for_each_buffer_cpu(buffer, cpu) { | |
861 | cpu_buffer = buffer->buffers[cpu]; | |
862 | rb_remove_pages(cpu_buffer, rm_pages); | |
863 | } | |
864 | goto out; | |
865 | } | |
866 | ||
867 | /* | |
868 | * This is a bit more difficult. We only want to add pages | |
869 | * when we can allocate enough for all CPUs. We do this | |
870 | * by allocating all the pages and storing them on a local | |
871 | * link list. If we succeed in our allocation, then we | |
872 | * add these pages to the cpu_buffers. Otherwise we just free | |
873 | * them all and return -ENOMEM; | |
874 | */ | |
554f786e SR |
875 | if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) |
876 | goto out_fail; | |
f536aafc | 877 | |
7a8e76a3 SR |
878 | new_pages = nr_pages - buffer->pages; |
879 | ||
880 | for_each_buffer_cpu(buffer, cpu) { | |
881 | for (i = 0; i < new_pages; i++) { | |
044fa782 | 882 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), |
e4c2ce82 SR |
883 | cache_line_size()), |
884 | GFP_KERNEL, cpu_to_node(cpu)); | |
044fa782 | 885 | if (!bpage) |
e4c2ce82 | 886 | goto free_pages; |
044fa782 | 887 | list_add(&bpage->list, &pages); |
7a8e76a3 SR |
888 | addr = __get_free_page(GFP_KERNEL); |
889 | if (!addr) | |
890 | goto free_pages; | |
044fa782 SR |
891 | bpage->page = (void *)addr; |
892 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
893 | } |
894 | } | |
895 | ||
896 | for_each_buffer_cpu(buffer, cpu) { | |
897 | cpu_buffer = buffer->buffers[cpu]; | |
898 | rb_insert_pages(cpu_buffer, &pages, new_pages); | |
899 | } | |
900 | ||
554f786e SR |
901 | if (RB_WARN_ON(buffer, !list_empty(&pages))) |
902 | goto out_fail; | |
7a8e76a3 SR |
903 | |
904 | out: | |
905 | buffer->pages = nr_pages; | |
554f786e | 906 | put_online_cpus(); |
7a8e76a3 SR |
907 | mutex_unlock(&buffer->mutex); |
908 | ||
909 | return size; | |
910 | ||
911 | free_pages: | |
044fa782 SR |
912 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
913 | list_del_init(&bpage->list); | |
914 | free_buffer_page(bpage); | |
7a8e76a3 | 915 | } |
554f786e | 916 | put_online_cpus(); |
641d2f63 | 917 | mutex_unlock(&buffer->mutex); |
7a8e76a3 | 918 | return -ENOMEM; |
554f786e SR |
919 | |
920 | /* | |
921 | * Something went totally wrong, and we are too paranoid | |
922 | * to even clean up the mess. | |
923 | */ | |
924 | out_fail: | |
925 | put_online_cpus(); | |
926 | mutex_unlock(&buffer->mutex); | |
927 | return -1; | |
7a8e76a3 | 928 | } |
c4f50183 | 929 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 930 | |
8789a9e7 | 931 | static inline void * |
044fa782 | 932 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 933 | { |
044fa782 | 934 | return bpage->data + index; |
8789a9e7 SR |
935 | } |
936 | ||
044fa782 | 937 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 938 | { |
044fa782 | 939 | return bpage->page->data + index; |
7a8e76a3 SR |
940 | } |
941 | ||
942 | static inline struct ring_buffer_event * | |
d769041f | 943 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 944 | { |
6f807acd SR |
945 | return __rb_page_index(cpu_buffer->reader_page, |
946 | cpu_buffer->reader_page->read); | |
947 | } | |
948 | ||
949 | static inline struct ring_buffer_event * | |
950 | rb_head_event(struct ring_buffer_per_cpu *cpu_buffer) | |
951 | { | |
952 | return __rb_page_index(cpu_buffer->head_page, | |
953 | cpu_buffer->head_page->read); | |
7a8e76a3 SR |
954 | } |
955 | ||
956 | static inline struct ring_buffer_event * | |
957 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
958 | { | |
6f807acd | 959 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
960 | } |
961 | ||
bf41a158 SR |
962 | static inline unsigned rb_page_write(struct buffer_page *bpage) |
963 | { | |
964 | return local_read(&bpage->write); | |
965 | } | |
966 | ||
967 | static inline unsigned rb_page_commit(struct buffer_page *bpage) | |
968 | { | |
abc9b56d | 969 | return local_read(&bpage->page->commit); |
bf41a158 SR |
970 | } |
971 | ||
972 | /* Size is determined by what has been commited */ | |
973 | static inline unsigned rb_page_size(struct buffer_page *bpage) | |
974 | { | |
975 | return rb_page_commit(bpage); | |
976 | } | |
977 | ||
978 | static inline unsigned | |
979 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
980 | { | |
981 | return rb_page_commit(cpu_buffer->commit_page); | |
982 | } | |
983 | ||
984 | static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer) | |
985 | { | |
986 | return rb_page_commit(cpu_buffer->head_page); | |
987 | } | |
988 | ||
7a8e76a3 | 989 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, |
044fa782 | 990 | struct buffer_page **bpage) |
7a8e76a3 | 991 | { |
044fa782 | 992 | struct list_head *p = (*bpage)->list.next; |
7a8e76a3 SR |
993 | |
994 | if (p == &cpu_buffer->pages) | |
995 | p = p->next; | |
996 | ||
044fa782 | 997 | *bpage = list_entry(p, struct buffer_page, list); |
7a8e76a3 SR |
998 | } |
999 | ||
bf41a158 SR |
1000 | static inline unsigned |
1001 | rb_event_index(struct ring_buffer_event *event) | |
1002 | { | |
1003 | unsigned long addr = (unsigned long)event; | |
1004 | ||
22f470f8 | 1005 | return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE; |
bf41a158 SR |
1006 | } |
1007 | ||
0f0c85fc | 1008 | static inline int |
fa743953 SR |
1009 | rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1010 | struct ring_buffer_event *event) | |
bf41a158 SR |
1011 | { |
1012 | unsigned long addr = (unsigned long)event; | |
1013 | unsigned long index; | |
1014 | ||
1015 | index = rb_event_index(event); | |
1016 | addr &= PAGE_MASK; | |
1017 | ||
1018 | return cpu_buffer->commit_page->page == (void *)addr && | |
1019 | rb_commit_index(cpu_buffer) == index; | |
1020 | } | |
1021 | ||
34a148bf | 1022 | static void |
bf41a158 | 1023 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1024 | { |
bf41a158 SR |
1025 | /* |
1026 | * We only race with interrupts and NMIs on this CPU. | |
1027 | * If we own the commit event, then we can commit | |
1028 | * all others that interrupted us, since the interruptions | |
1029 | * are in stack format (they finish before they come | |
1030 | * back to us). This allows us to do a simple loop to | |
1031 | * assign the commit to the tail. | |
1032 | */ | |
a8ccf1d6 | 1033 | again: |
bf41a158 | 1034 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
abc9b56d | 1035 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1036 | cpu_buffer->commit_page->write; |
1037 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
1038 | cpu_buffer->write_stamp = |
1039 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1040 | /* add barrier to keep gcc from optimizing too much */ |
1041 | barrier(); | |
1042 | } | |
1043 | while (rb_commit_index(cpu_buffer) != | |
1044 | rb_page_write(cpu_buffer->commit_page)) { | |
abc9b56d | 1045 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1046 | cpu_buffer->commit_page->write; |
1047 | barrier(); | |
1048 | } | |
a8ccf1d6 SR |
1049 | |
1050 | /* again, keep gcc from optimizing */ | |
1051 | barrier(); | |
1052 | ||
1053 | /* | |
1054 | * If an interrupt came in just after the first while loop | |
1055 | * and pushed the tail page forward, we will be left with | |
1056 | * a dangling commit that will never go forward. | |
1057 | */ | |
1058 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1059 | goto again; | |
7a8e76a3 SR |
1060 | } |
1061 | ||
d769041f | 1062 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1063 | { |
abc9b56d | 1064 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1065 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1066 | } |
1067 | ||
34a148bf | 1068 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1069 | { |
1070 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1071 | ||
1072 | /* | |
1073 | * The iterator could be on the reader page (it starts there). | |
1074 | * But the head could have moved, since the reader was | |
1075 | * found. Check for this case and assign the iterator | |
1076 | * to the head page instead of next. | |
1077 | */ | |
1078 | if (iter->head_page == cpu_buffer->reader_page) | |
1079 | iter->head_page = cpu_buffer->head_page; | |
1080 | else | |
1081 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1082 | ||
abc9b56d | 1083 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1084 | iter->head = 0; |
1085 | } | |
1086 | ||
1087 | /** | |
1088 | * ring_buffer_update_event - update event type and data | |
1089 | * @event: the even to update | |
1090 | * @type: the type of event | |
1091 | * @length: the size of the event field in the ring buffer | |
1092 | * | |
1093 | * Update the type and data fields of the event. The length | |
1094 | * is the actual size that is written to the ring buffer, | |
1095 | * and with this, we can determine what to place into the | |
1096 | * data field. | |
1097 | */ | |
34a148bf | 1098 | static void |
7a8e76a3 SR |
1099 | rb_update_event(struct ring_buffer_event *event, |
1100 | unsigned type, unsigned length) | |
1101 | { | |
334d4169 | 1102 | event->type_len = type; |
7a8e76a3 SR |
1103 | |
1104 | switch (type) { | |
1105 | ||
1106 | case RINGBUF_TYPE_PADDING: | |
7a8e76a3 | 1107 | case RINGBUF_TYPE_TIME_EXTEND: |
7a8e76a3 | 1108 | case RINGBUF_TYPE_TIME_STAMP: |
7a8e76a3 SR |
1109 | break; |
1110 | ||
334d4169 | 1111 | case 0: |
7a8e76a3 | 1112 | length -= RB_EVNT_HDR_SIZE; |
334d4169 | 1113 | if (length > RB_MAX_SMALL_DATA) |
7a8e76a3 | 1114 | event->array[0] = length; |
334d4169 LJ |
1115 | else |
1116 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
7a8e76a3 SR |
1117 | break; |
1118 | default: | |
1119 | BUG(); | |
1120 | } | |
1121 | } | |
1122 | ||
34a148bf | 1123 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
1124 | { |
1125 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
1126 | ||
1127 | /* zero length can cause confusions */ | |
1128 | if (!length) | |
1129 | length = 1; | |
1130 | ||
1131 | if (length > RB_MAX_SMALL_DATA) | |
1132 | length += sizeof(event.array[0]); | |
1133 | ||
1134 | length += RB_EVNT_HDR_SIZE; | |
1135 | length = ALIGN(length, RB_ALIGNMENT); | |
1136 | ||
1137 | return length; | |
1138 | } | |
1139 | ||
c7b09308 SR |
1140 | static inline void |
1141 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
1142 | struct buffer_page *tail_page, | |
1143 | unsigned long tail, unsigned long length) | |
1144 | { | |
1145 | struct ring_buffer_event *event; | |
1146 | ||
1147 | /* | |
1148 | * Only the event that crossed the page boundary | |
1149 | * must fill the old tail_page with padding. | |
1150 | */ | |
1151 | if (tail >= BUF_PAGE_SIZE) { | |
1152 | local_sub(length, &tail_page->write); | |
1153 | return; | |
1154 | } | |
1155 | ||
1156 | event = __rb_page_index(tail_page, tail); | |
1157 | ||
1158 | /* | |
1159 | * If this event is bigger than the minimum size, then | |
1160 | * we need to be careful that we don't subtract the | |
1161 | * write counter enough to allow another writer to slip | |
1162 | * in on this page. | |
1163 | * We put in a discarded commit instead, to make sure | |
1164 | * that this space is not used again. | |
1165 | * | |
1166 | * If we are less than the minimum size, we don't need to | |
1167 | * worry about it. | |
1168 | */ | |
1169 | if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) { | |
1170 | /* No room for any events */ | |
1171 | ||
1172 | /* Mark the rest of the page with padding */ | |
1173 | rb_event_set_padding(event); | |
1174 | ||
1175 | /* Set the write back to the previous setting */ | |
1176 | local_sub(length, &tail_page->write); | |
1177 | return; | |
1178 | } | |
1179 | ||
1180 | /* Put in a discarded event */ | |
1181 | event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE; | |
1182 | event->type_len = RINGBUF_TYPE_PADDING; | |
1183 | /* time delta must be non zero */ | |
1184 | event->time_delta = 1; | |
1185 | /* Account for this as an entry */ | |
1186 | local_inc(&tail_page->entries); | |
1187 | local_inc(&cpu_buffer->entries); | |
1188 | ||
1189 | /* Set write to end of buffer */ | |
1190 | length = (tail + length) - BUF_PAGE_SIZE; | |
1191 | local_sub(length, &tail_page->write); | |
1192 | } | |
6634ff26 | 1193 | |
7a8e76a3 | 1194 | static struct ring_buffer_event * |
6634ff26 SR |
1195 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
1196 | unsigned long length, unsigned long tail, | |
1197 | struct buffer_page *commit_page, | |
1198 | struct buffer_page *tail_page, u64 *ts) | |
7a8e76a3 | 1199 | { |
6634ff26 | 1200 | struct buffer_page *next_page, *head_page, *reader_page; |
7a8e76a3 | 1201 | struct ring_buffer *buffer = cpu_buffer->buffer; |
78d904b4 | 1202 | bool lock_taken = false; |
6634ff26 | 1203 | unsigned long flags; |
aa20ae84 SR |
1204 | |
1205 | next_page = tail_page; | |
1206 | ||
1207 | local_irq_save(flags); | |
1208 | /* | |
1209 | * Since the write to the buffer is still not | |
1210 | * fully lockless, we must be careful with NMIs. | |
1211 | * The locks in the writers are taken when a write | |
1212 | * crosses to a new page. The locks protect against | |
1213 | * races with the readers (this will soon be fixed | |
1214 | * with a lockless solution). | |
1215 | * | |
1216 | * Because we can not protect against NMIs, and we | |
1217 | * want to keep traces reentrant, we need to manage | |
1218 | * what happens when we are in an NMI. | |
1219 | * | |
1220 | * NMIs can happen after we take the lock. | |
1221 | * If we are in an NMI, only take the lock | |
1222 | * if it is not already taken. Otherwise | |
1223 | * simply fail. | |
1224 | */ | |
1225 | if (unlikely(in_nmi())) { | |
1226 | if (!__raw_spin_trylock(&cpu_buffer->lock)) { | |
1227 | cpu_buffer->nmi_dropped++; | |
1228 | goto out_reset; | |
1229 | } | |
1230 | } else | |
1231 | __raw_spin_lock(&cpu_buffer->lock); | |
1232 | ||
1233 | lock_taken = true; | |
1234 | ||
1235 | rb_inc_page(cpu_buffer, &next_page); | |
1236 | ||
1237 | head_page = cpu_buffer->head_page; | |
1238 | reader_page = cpu_buffer->reader_page; | |
1239 | ||
1240 | /* we grabbed the lock before incrementing */ | |
1241 | if (RB_WARN_ON(cpu_buffer, next_page == reader_page)) | |
1242 | goto out_reset; | |
1243 | ||
1244 | /* | |
1245 | * If for some reason, we had an interrupt storm that made | |
1246 | * it all the way around the buffer, bail, and warn | |
1247 | * about it. | |
1248 | */ | |
1249 | if (unlikely(next_page == commit_page)) { | |
1250 | cpu_buffer->commit_overrun++; | |
1251 | goto out_reset; | |
1252 | } | |
1253 | ||
1254 | if (next_page == head_page) { | |
1255 | if (!(buffer->flags & RB_FL_OVERWRITE)) | |
1256 | goto out_reset; | |
1257 | ||
1258 | /* tail_page has not moved yet? */ | |
1259 | if (tail_page == cpu_buffer->tail_page) { | |
1260 | /* count overflows */ | |
1261 | cpu_buffer->overrun += | |
1262 | local_read(&head_page->entries); | |
1263 | ||
1264 | rb_inc_page(cpu_buffer, &head_page); | |
1265 | cpu_buffer->head_page = head_page; | |
1266 | cpu_buffer->head_page->read = 0; | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * If the tail page is still the same as what we think | |
1272 | * it is, then it is up to us to update the tail | |
1273 | * pointer. | |
1274 | */ | |
1275 | if (tail_page == cpu_buffer->tail_page) { | |
1276 | local_set(&next_page->write, 0); | |
1277 | local_set(&next_page->entries, 0); | |
1278 | local_set(&next_page->page->commit, 0); | |
1279 | cpu_buffer->tail_page = next_page; | |
1280 | ||
1281 | /* reread the time stamp */ | |
88eb0125 | 1282 | *ts = rb_time_stamp(buffer, cpu_buffer->cpu); |
aa20ae84 SR |
1283 | cpu_buffer->tail_page->page->time_stamp = *ts; |
1284 | } | |
1285 | ||
c7b09308 | 1286 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
aa20ae84 | 1287 | |
aa20ae84 SR |
1288 | __raw_spin_unlock(&cpu_buffer->lock); |
1289 | local_irq_restore(flags); | |
1290 | ||
1291 | /* fail and let the caller try again */ | |
1292 | return ERR_PTR(-EAGAIN); | |
1293 | ||
45141d46 | 1294 | out_reset: |
6f3b3440 | 1295 | /* reset write */ |
c7b09308 | 1296 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
6f3b3440 | 1297 | |
78d904b4 SR |
1298 | if (likely(lock_taken)) |
1299 | __raw_spin_unlock(&cpu_buffer->lock); | |
3e03fb7f | 1300 | local_irq_restore(flags); |
bf41a158 | 1301 | return NULL; |
7a8e76a3 SR |
1302 | } |
1303 | ||
6634ff26 SR |
1304 | static struct ring_buffer_event * |
1305 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
1306 | unsigned type, unsigned long length, u64 *ts) | |
1307 | { | |
1308 | struct buffer_page *tail_page, *commit_page; | |
1309 | struct ring_buffer_event *event; | |
1310 | unsigned long tail, write; | |
1311 | ||
1312 | commit_page = cpu_buffer->commit_page; | |
1313 | /* we just need to protect against interrupts */ | |
1314 | barrier(); | |
1315 | tail_page = cpu_buffer->tail_page; | |
1316 | write = local_add_return(length, &tail_page->write); | |
1317 | tail = write - length; | |
1318 | ||
1319 | /* See if we shot pass the end of this buffer page */ | |
1320 | if (write > BUF_PAGE_SIZE) | |
1321 | return rb_move_tail(cpu_buffer, length, tail, | |
1322 | commit_page, tail_page, ts); | |
1323 | ||
1324 | /* We reserved something on the buffer */ | |
1325 | ||
6634ff26 SR |
1326 | event = __rb_page_index(tail_page, tail); |
1327 | rb_update_event(event, type, length); | |
1328 | ||
1329 | /* The passed in type is zero for DATA */ | |
1330 | if (likely(!type)) | |
1331 | local_inc(&tail_page->entries); | |
1332 | ||
1333 | /* | |
fa743953 SR |
1334 | * If this is the first commit on the page, then update |
1335 | * its timestamp. | |
6634ff26 | 1336 | */ |
fa743953 SR |
1337 | if (!tail) |
1338 | tail_page->page->time_stamp = *ts; | |
6634ff26 SR |
1339 | |
1340 | return event; | |
1341 | } | |
1342 | ||
edd813bf SR |
1343 | static inline int |
1344 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
1345 | struct ring_buffer_event *event) | |
1346 | { | |
1347 | unsigned long new_index, old_index; | |
1348 | struct buffer_page *bpage; | |
1349 | unsigned long index; | |
1350 | unsigned long addr; | |
1351 | ||
1352 | new_index = rb_event_index(event); | |
1353 | old_index = new_index + rb_event_length(event); | |
1354 | addr = (unsigned long)event; | |
1355 | addr &= PAGE_MASK; | |
1356 | ||
1357 | bpage = cpu_buffer->tail_page; | |
1358 | ||
1359 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { | |
1360 | /* | |
1361 | * This is on the tail page. It is possible that | |
1362 | * a write could come in and move the tail page | |
1363 | * and write to the next page. That is fine | |
1364 | * because we just shorten what is on this page. | |
1365 | */ | |
1366 | index = local_cmpxchg(&bpage->write, old_index, new_index); | |
1367 | if (index == old_index) | |
1368 | return 1; | |
1369 | } | |
1370 | ||
1371 | /* could not discard */ | |
1372 | return 0; | |
1373 | } | |
1374 | ||
7a8e76a3 SR |
1375 | static int |
1376 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1377 | u64 *ts, u64 *delta) | |
1378 | { | |
1379 | struct ring_buffer_event *event; | |
1380 | static int once; | |
bf41a158 | 1381 | int ret; |
7a8e76a3 SR |
1382 | |
1383 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1384 | printk(KERN_WARNING "Delta way too big! %llu" | |
1385 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1386 | (unsigned long long)*delta, |
1387 | (unsigned long long)*ts, | |
1388 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1389 | WARN_ON(1); |
1390 | } | |
1391 | ||
1392 | /* | |
1393 | * The delta is too big, we to add a | |
1394 | * new timestamp. | |
1395 | */ | |
1396 | event = __rb_reserve_next(cpu_buffer, | |
1397 | RINGBUF_TYPE_TIME_EXTEND, | |
1398 | RB_LEN_TIME_EXTEND, | |
1399 | ts); | |
1400 | if (!event) | |
bf41a158 | 1401 | return -EBUSY; |
7a8e76a3 | 1402 | |
bf41a158 SR |
1403 | if (PTR_ERR(event) == -EAGAIN) |
1404 | return -EAGAIN; | |
1405 | ||
1406 | /* Only a commited time event can update the write stamp */ | |
fa743953 | 1407 | if (rb_event_is_commit(cpu_buffer, event)) { |
bf41a158 | 1408 | /* |
fa743953 SR |
1409 | * If this is the first on the page, then it was |
1410 | * updated with the page itself. Try to discard it | |
1411 | * and if we can't just make it zero. | |
bf41a158 SR |
1412 | */ |
1413 | if (rb_event_index(event)) { | |
1414 | event->time_delta = *delta & TS_MASK; | |
1415 | event->array[0] = *delta >> TS_SHIFT; | |
1416 | } else { | |
ea05b57c SR |
1417 | /* try to discard, since we do not need this */ |
1418 | if (!rb_try_to_discard(cpu_buffer, event)) { | |
1419 | /* nope, just zero it */ | |
1420 | event->time_delta = 0; | |
1421 | event->array[0] = 0; | |
1422 | } | |
bf41a158 | 1423 | } |
7a8e76a3 | 1424 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
1425 | /* let the caller know this was the commit */ |
1426 | ret = 1; | |
1427 | } else { | |
edd813bf SR |
1428 | /* Try to discard the event */ |
1429 | if (!rb_try_to_discard(cpu_buffer, event)) { | |
1430 | /* Darn, this is just wasted space */ | |
1431 | event->time_delta = 0; | |
1432 | event->array[0] = 0; | |
edd813bf | 1433 | } |
f57a8a19 | 1434 | ret = 0; |
7a8e76a3 SR |
1435 | } |
1436 | ||
bf41a158 SR |
1437 | *delta = 0; |
1438 | ||
1439 | return ret; | |
7a8e76a3 SR |
1440 | } |
1441 | ||
fa743953 SR |
1442 | static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer) |
1443 | { | |
1444 | local_inc(&cpu_buffer->committing); | |
1445 | local_inc(&cpu_buffer->commits); | |
1446 | } | |
1447 | ||
1448 | static void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer) | |
1449 | { | |
1450 | unsigned long commits; | |
1451 | ||
1452 | if (RB_WARN_ON(cpu_buffer, | |
1453 | !local_read(&cpu_buffer->committing))) | |
1454 | return; | |
1455 | ||
1456 | again: | |
1457 | commits = local_read(&cpu_buffer->commits); | |
1458 | /* synchronize with interrupts */ | |
1459 | barrier(); | |
1460 | if (local_read(&cpu_buffer->committing) == 1) | |
1461 | rb_set_commit_to_write(cpu_buffer); | |
1462 | ||
1463 | local_dec(&cpu_buffer->committing); | |
1464 | ||
1465 | /* synchronize with interrupts */ | |
1466 | barrier(); | |
1467 | ||
1468 | /* | |
1469 | * Need to account for interrupts coming in between the | |
1470 | * updating of the commit page and the clearing of the | |
1471 | * committing counter. | |
1472 | */ | |
1473 | if (unlikely(local_read(&cpu_buffer->commits) != commits) && | |
1474 | !local_read(&cpu_buffer->committing)) { | |
1475 | local_inc(&cpu_buffer->committing); | |
1476 | goto again; | |
1477 | } | |
1478 | } | |
1479 | ||
7a8e76a3 SR |
1480 | static struct ring_buffer_event * |
1481 | rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 1482 | unsigned long length) |
7a8e76a3 SR |
1483 | { |
1484 | struct ring_buffer_event *event; | |
168b6b1d | 1485 | u64 ts, delta = 0; |
bf41a158 | 1486 | int commit = 0; |
818e3dd3 | 1487 | int nr_loops = 0; |
7a8e76a3 | 1488 | |
fa743953 SR |
1489 | rb_start_commit(cpu_buffer); |
1490 | ||
be957c44 | 1491 | length = rb_calculate_event_length(length); |
bf41a158 | 1492 | again: |
818e3dd3 SR |
1493 | /* |
1494 | * We allow for interrupts to reenter here and do a trace. | |
1495 | * If one does, it will cause this original code to loop | |
1496 | * back here. Even with heavy interrupts happening, this | |
1497 | * should only happen a few times in a row. If this happens | |
1498 | * 1000 times in a row, there must be either an interrupt | |
1499 | * storm or we have something buggy. | |
1500 | * Bail! | |
1501 | */ | |
3e89c7bb | 1502 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
fa743953 | 1503 | goto out_fail; |
818e3dd3 | 1504 | |
88eb0125 | 1505 | ts = rb_time_stamp(cpu_buffer->buffer, cpu_buffer->cpu); |
7a8e76a3 | 1506 | |
bf41a158 SR |
1507 | /* |
1508 | * Only the first commit can update the timestamp. | |
1509 | * Yes there is a race here. If an interrupt comes in | |
1510 | * just after the conditional and it traces too, then it | |
1511 | * will also check the deltas. More than one timestamp may | |
1512 | * also be made. But only the entry that did the actual | |
1513 | * commit will be something other than zero. | |
1514 | */ | |
0f0c85fc SR |
1515 | if (likely(cpu_buffer->tail_page == cpu_buffer->commit_page && |
1516 | rb_page_write(cpu_buffer->tail_page) == | |
1517 | rb_commit_index(cpu_buffer))) { | |
168b6b1d | 1518 | u64 diff; |
bf41a158 | 1519 | |
168b6b1d | 1520 | diff = ts - cpu_buffer->write_stamp; |
7a8e76a3 | 1521 | |
168b6b1d | 1522 | /* make sure this diff is calculated here */ |
bf41a158 SR |
1523 | barrier(); |
1524 | ||
1525 | /* Did the write stamp get updated already? */ | |
1526 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
168b6b1d | 1527 | goto get_event; |
bf41a158 | 1528 | |
168b6b1d SR |
1529 | delta = diff; |
1530 | if (unlikely(test_time_stamp(delta))) { | |
7a8e76a3 | 1531 | |
bf41a158 | 1532 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
bf41a158 | 1533 | if (commit == -EBUSY) |
fa743953 | 1534 | goto out_fail; |
bf41a158 SR |
1535 | |
1536 | if (commit == -EAGAIN) | |
1537 | goto again; | |
1538 | ||
1539 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 1540 | } |
168b6b1d | 1541 | } |
7a8e76a3 | 1542 | |
168b6b1d | 1543 | get_event: |
1cd8d735 | 1544 | event = __rb_reserve_next(cpu_buffer, 0, length, &ts); |
168b6b1d | 1545 | if (unlikely(PTR_ERR(event) == -EAGAIN)) |
bf41a158 SR |
1546 | goto again; |
1547 | ||
fa743953 SR |
1548 | if (!event) |
1549 | goto out_fail; | |
7a8e76a3 | 1550 | |
fa743953 | 1551 | if (!rb_event_is_commit(cpu_buffer, event)) |
7a8e76a3 SR |
1552 | delta = 0; |
1553 | ||
1554 | event->time_delta = delta; | |
1555 | ||
1556 | return event; | |
fa743953 SR |
1557 | |
1558 | out_fail: | |
1559 | rb_end_commit(cpu_buffer); | |
1560 | return NULL; | |
7a8e76a3 SR |
1561 | } |
1562 | ||
aa18efb2 | 1563 | #define TRACE_RECURSIVE_DEPTH 16 |
261842b7 SR |
1564 | |
1565 | static int trace_recursive_lock(void) | |
1566 | { | |
aa18efb2 | 1567 | current->trace_recursion++; |
261842b7 | 1568 | |
aa18efb2 SR |
1569 | if (likely(current->trace_recursion < TRACE_RECURSIVE_DEPTH)) |
1570 | return 0; | |
e057a5e5 | 1571 | |
aa18efb2 SR |
1572 | /* Disable all tracing before we do anything else */ |
1573 | tracing_off_permanent(); | |
261842b7 | 1574 | |
7d7d2b80 | 1575 | printk_once(KERN_WARNING "Tracing recursion: depth[%ld]:" |
aa18efb2 SR |
1576 | "HC[%lu]:SC[%lu]:NMI[%lu]\n", |
1577 | current->trace_recursion, | |
1578 | hardirq_count() >> HARDIRQ_SHIFT, | |
1579 | softirq_count() >> SOFTIRQ_SHIFT, | |
1580 | in_nmi()); | |
261842b7 | 1581 | |
aa18efb2 SR |
1582 | WARN_ON_ONCE(1); |
1583 | return -1; | |
261842b7 SR |
1584 | } |
1585 | ||
1586 | static void trace_recursive_unlock(void) | |
1587 | { | |
aa18efb2 | 1588 | WARN_ON_ONCE(!current->trace_recursion); |
261842b7 | 1589 | |
aa18efb2 | 1590 | current->trace_recursion--; |
261842b7 SR |
1591 | } |
1592 | ||
bf41a158 SR |
1593 | static DEFINE_PER_CPU(int, rb_need_resched); |
1594 | ||
7a8e76a3 SR |
1595 | /** |
1596 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
1597 | * @buffer: the ring buffer to reserve from | |
1598 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
1599 | * |
1600 | * Returns a reseverd event on the ring buffer to copy directly to. | |
1601 | * The user of this interface will need to get the body to write into | |
1602 | * and can use the ring_buffer_event_data() interface. | |
1603 | * | |
1604 | * The length is the length of the data needed, not the event length | |
1605 | * which also includes the event header. | |
1606 | * | |
1607 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
1608 | * If NULL is returned, then nothing has been allocated or locked. | |
1609 | */ | |
1610 | struct ring_buffer_event * | |
0a987751 | 1611 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
1612 | { |
1613 | struct ring_buffer_per_cpu *cpu_buffer; | |
1614 | struct ring_buffer_event *event; | |
bf41a158 | 1615 | int cpu, resched; |
7a8e76a3 | 1616 | |
033601a3 | 1617 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1618 | return NULL; |
1619 | ||
7a8e76a3 SR |
1620 | if (atomic_read(&buffer->record_disabled)) |
1621 | return NULL; | |
1622 | ||
bf41a158 | 1623 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 1624 | resched = ftrace_preempt_disable(); |
bf41a158 | 1625 | |
261842b7 SR |
1626 | if (trace_recursive_lock()) |
1627 | goto out_nocheck; | |
1628 | ||
7a8e76a3 SR |
1629 | cpu = raw_smp_processor_id(); |
1630 | ||
9e01c1b7 | 1631 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1632 | goto out; |
7a8e76a3 SR |
1633 | |
1634 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1635 | |
1636 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 1637 | goto out; |
7a8e76a3 | 1638 | |
be957c44 | 1639 | if (length > BUF_MAX_DATA_SIZE) |
bf41a158 | 1640 | goto out; |
7a8e76a3 | 1641 | |
1cd8d735 | 1642 | event = rb_reserve_next_event(cpu_buffer, length); |
7a8e76a3 | 1643 | if (!event) |
d769041f | 1644 | goto out; |
7a8e76a3 | 1645 | |
bf41a158 SR |
1646 | /* |
1647 | * Need to store resched state on this cpu. | |
1648 | * Only the first needs to. | |
1649 | */ | |
1650 | ||
1651 | if (preempt_count() == 1) | |
1652 | per_cpu(rb_need_resched, cpu) = resched; | |
1653 | ||
7a8e76a3 SR |
1654 | return event; |
1655 | ||
d769041f | 1656 | out: |
261842b7 SR |
1657 | trace_recursive_unlock(); |
1658 | ||
1659 | out_nocheck: | |
182e9f5f | 1660 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1661 | return NULL; |
1662 | } | |
c4f50183 | 1663 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 SR |
1664 | |
1665 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
1666 | struct ring_buffer_event *event) | |
1667 | { | |
e4906eff | 1668 | local_inc(&cpu_buffer->entries); |
bf41a158 | 1669 | |
fa743953 SR |
1670 | /* |
1671 | * The event first in the commit queue updates the | |
1672 | * time stamp. | |
1673 | */ | |
1674 | if (rb_event_is_commit(cpu_buffer, event)) | |
1675 | cpu_buffer->write_stamp += event->time_delta; | |
bf41a158 | 1676 | |
fa743953 | 1677 | rb_end_commit(cpu_buffer); |
7a8e76a3 SR |
1678 | } |
1679 | ||
1680 | /** | |
1681 | * ring_buffer_unlock_commit - commit a reserved | |
1682 | * @buffer: The buffer to commit to | |
1683 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
1684 | * |
1685 | * This commits the data to the ring buffer, and releases any locks held. | |
1686 | * | |
1687 | * Must be paired with ring_buffer_lock_reserve. | |
1688 | */ | |
1689 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 1690 | struct ring_buffer_event *event) |
7a8e76a3 SR |
1691 | { |
1692 | struct ring_buffer_per_cpu *cpu_buffer; | |
1693 | int cpu = raw_smp_processor_id(); | |
1694 | ||
1695 | cpu_buffer = buffer->buffers[cpu]; | |
1696 | ||
7a8e76a3 SR |
1697 | rb_commit(cpu_buffer, event); |
1698 | ||
261842b7 SR |
1699 | trace_recursive_unlock(); |
1700 | ||
bf41a158 SR |
1701 | /* |
1702 | * Only the last preempt count needs to restore preemption. | |
1703 | */ | |
182e9f5f SR |
1704 | if (preempt_count() == 1) |
1705 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1706 | else | |
bf41a158 | 1707 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
1708 | |
1709 | return 0; | |
1710 | } | |
c4f50183 | 1711 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 1712 | |
f3b9aae1 FW |
1713 | static inline void rb_event_discard(struct ring_buffer_event *event) |
1714 | { | |
334d4169 LJ |
1715 | /* array[0] holds the actual length for the discarded event */ |
1716 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
1717 | event->type_len = RINGBUF_TYPE_PADDING; | |
f3b9aae1 FW |
1718 | /* time delta must be non zero */ |
1719 | if (!event->time_delta) | |
1720 | event->time_delta = 1; | |
1721 | } | |
1722 | ||
fa1b47dd SR |
1723 | /** |
1724 | * ring_buffer_event_discard - discard any event in the ring buffer | |
1725 | * @event: the event to discard | |
1726 | * | |
1727 | * Sometimes a event that is in the ring buffer needs to be ignored. | |
1728 | * This function lets the user discard an event in the ring buffer | |
1729 | * and then that event will not be read later. | |
1730 | * | |
1731 | * Note, it is up to the user to be careful with this, and protect | |
1732 | * against races. If the user discards an event that has been consumed | |
1733 | * it is possible that it could corrupt the ring buffer. | |
1734 | */ | |
1735 | void ring_buffer_event_discard(struct ring_buffer_event *event) | |
1736 | { | |
f3b9aae1 | 1737 | rb_event_discard(event); |
fa1b47dd SR |
1738 | } |
1739 | EXPORT_SYMBOL_GPL(ring_buffer_event_discard); | |
1740 | ||
1741 | /** | |
1742 | * ring_buffer_commit_discard - discard an event that has not been committed | |
1743 | * @buffer: the ring buffer | |
1744 | * @event: non committed event to discard | |
1745 | * | |
1746 | * This is similar to ring_buffer_event_discard but must only be | |
1747 | * performed on an event that has not been committed yet. The difference | |
1748 | * is that this will also try to free the event from the ring buffer | |
1749 | * if another event has not been added behind it. | |
1750 | * | |
1751 | * If another event has been added behind it, it will set the event | |
1752 | * up as discarded, and perform the commit. | |
1753 | * | |
1754 | * If this function is called, do not call ring_buffer_unlock_commit on | |
1755 | * the event. | |
1756 | */ | |
1757 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
1758 | struct ring_buffer_event *event) | |
1759 | { | |
1760 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
1761 | int cpu; |
1762 | ||
1763 | /* The event is discarded regardless */ | |
f3b9aae1 | 1764 | rb_event_discard(event); |
fa1b47dd | 1765 | |
fa743953 SR |
1766 | cpu = smp_processor_id(); |
1767 | cpu_buffer = buffer->buffers[cpu]; | |
1768 | ||
fa1b47dd SR |
1769 | /* |
1770 | * This must only be called if the event has not been | |
1771 | * committed yet. Thus we can assume that preemption | |
1772 | * is still disabled. | |
1773 | */ | |
fa743953 | 1774 | RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing)); |
fa1b47dd | 1775 | |
edd813bf SR |
1776 | if (!rb_try_to_discard(cpu_buffer, event)) |
1777 | goto out; | |
fa1b47dd SR |
1778 | |
1779 | /* | |
1780 | * The commit is still visible by the reader, so we | |
1781 | * must increment entries. | |
1782 | */ | |
e4906eff | 1783 | local_inc(&cpu_buffer->entries); |
fa1b47dd | 1784 | out: |
fa743953 | 1785 | rb_end_commit(cpu_buffer); |
fa1b47dd | 1786 | |
f3b9aae1 FW |
1787 | trace_recursive_unlock(); |
1788 | ||
fa1b47dd SR |
1789 | /* |
1790 | * Only the last preempt count needs to restore preemption. | |
1791 | */ | |
1792 | if (preempt_count() == 1) | |
1793 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1794 | else | |
1795 | preempt_enable_no_resched_notrace(); | |
1796 | ||
1797 | } | |
1798 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
1799 | ||
7a8e76a3 SR |
1800 | /** |
1801 | * ring_buffer_write - write data to the buffer without reserving | |
1802 | * @buffer: The ring buffer to write to. | |
1803 | * @length: The length of the data being written (excluding the event header) | |
1804 | * @data: The data to write to the buffer. | |
1805 | * | |
1806 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
1807 | * one function. If you already have the data to write to the buffer, it | |
1808 | * may be easier to simply call this function. | |
1809 | * | |
1810 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
1811 | * and not the length of the event which would hold the header. | |
1812 | */ | |
1813 | int ring_buffer_write(struct ring_buffer *buffer, | |
1814 | unsigned long length, | |
1815 | void *data) | |
1816 | { | |
1817 | struct ring_buffer_per_cpu *cpu_buffer; | |
1818 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
1819 | void *body; |
1820 | int ret = -EBUSY; | |
bf41a158 | 1821 | int cpu, resched; |
7a8e76a3 | 1822 | |
033601a3 | 1823 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1824 | return -EBUSY; |
1825 | ||
7a8e76a3 SR |
1826 | if (atomic_read(&buffer->record_disabled)) |
1827 | return -EBUSY; | |
1828 | ||
182e9f5f | 1829 | resched = ftrace_preempt_disable(); |
bf41a158 | 1830 | |
7a8e76a3 SR |
1831 | cpu = raw_smp_processor_id(); |
1832 | ||
9e01c1b7 | 1833 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1834 | goto out; |
7a8e76a3 SR |
1835 | |
1836 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1837 | |
1838 | if (atomic_read(&cpu_buffer->record_disabled)) | |
1839 | goto out; | |
1840 | ||
be957c44 SR |
1841 | if (length > BUF_MAX_DATA_SIZE) |
1842 | goto out; | |
1843 | ||
1844 | event = rb_reserve_next_event(cpu_buffer, length); | |
7a8e76a3 SR |
1845 | if (!event) |
1846 | goto out; | |
1847 | ||
1848 | body = rb_event_data(event); | |
1849 | ||
1850 | memcpy(body, data, length); | |
1851 | ||
1852 | rb_commit(cpu_buffer, event); | |
1853 | ||
1854 | ret = 0; | |
1855 | out: | |
182e9f5f | 1856 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1857 | |
1858 | return ret; | |
1859 | } | |
c4f50183 | 1860 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 1861 | |
34a148bf | 1862 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
1863 | { |
1864 | struct buffer_page *reader = cpu_buffer->reader_page; | |
1865 | struct buffer_page *head = cpu_buffer->head_page; | |
1866 | struct buffer_page *commit = cpu_buffer->commit_page; | |
1867 | ||
1868 | return reader->read == rb_page_commit(reader) && | |
1869 | (commit == reader || | |
1870 | (commit == head && | |
1871 | head->read == rb_page_commit(commit))); | |
1872 | } | |
1873 | ||
7a8e76a3 SR |
1874 | /** |
1875 | * ring_buffer_record_disable - stop all writes into the buffer | |
1876 | * @buffer: The ring buffer to stop writes to. | |
1877 | * | |
1878 | * This prevents all writes to the buffer. Any attempt to write | |
1879 | * to the buffer after this will fail and return NULL. | |
1880 | * | |
1881 | * The caller should call synchronize_sched() after this. | |
1882 | */ | |
1883 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
1884 | { | |
1885 | atomic_inc(&buffer->record_disabled); | |
1886 | } | |
c4f50183 | 1887 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
1888 | |
1889 | /** | |
1890 | * ring_buffer_record_enable - enable writes to the buffer | |
1891 | * @buffer: The ring buffer to enable writes | |
1892 | * | |
1893 | * Note, multiple disables will need the same number of enables | |
1894 | * to truely enable the writing (much like preempt_disable). | |
1895 | */ | |
1896 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
1897 | { | |
1898 | atomic_dec(&buffer->record_disabled); | |
1899 | } | |
c4f50183 | 1900 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 SR |
1901 | |
1902 | /** | |
1903 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
1904 | * @buffer: The ring buffer to stop writes to. | |
1905 | * @cpu: The CPU buffer to stop | |
1906 | * | |
1907 | * This prevents all writes to the buffer. Any attempt to write | |
1908 | * to the buffer after this will fail and return NULL. | |
1909 | * | |
1910 | * The caller should call synchronize_sched() after this. | |
1911 | */ | |
1912 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
1913 | { | |
1914 | struct ring_buffer_per_cpu *cpu_buffer; | |
1915 | ||
9e01c1b7 | 1916 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1917 | return; |
7a8e76a3 SR |
1918 | |
1919 | cpu_buffer = buffer->buffers[cpu]; | |
1920 | atomic_inc(&cpu_buffer->record_disabled); | |
1921 | } | |
c4f50183 | 1922 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
1923 | |
1924 | /** | |
1925 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
1926 | * @buffer: The ring buffer to enable writes | |
1927 | * @cpu: The CPU to enable. | |
1928 | * | |
1929 | * Note, multiple disables will need the same number of enables | |
1930 | * to truely enable the writing (much like preempt_disable). | |
1931 | */ | |
1932 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
1933 | { | |
1934 | struct ring_buffer_per_cpu *cpu_buffer; | |
1935 | ||
9e01c1b7 | 1936 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1937 | return; |
7a8e76a3 SR |
1938 | |
1939 | cpu_buffer = buffer->buffers[cpu]; | |
1940 | atomic_dec(&cpu_buffer->record_disabled); | |
1941 | } | |
c4f50183 | 1942 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 SR |
1943 | |
1944 | /** | |
1945 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
1946 | * @buffer: The ring buffer | |
1947 | * @cpu: The per CPU buffer to get the entries from. | |
1948 | */ | |
1949 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
1950 | { | |
1951 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 1952 | unsigned long ret; |
7a8e76a3 | 1953 | |
9e01c1b7 | 1954 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1955 | return 0; |
7a8e76a3 SR |
1956 | |
1957 | cpu_buffer = buffer->buffers[cpu]; | |
e4906eff SR |
1958 | ret = (local_read(&cpu_buffer->entries) - cpu_buffer->overrun) |
1959 | - cpu_buffer->read; | |
554f786e SR |
1960 | |
1961 | return ret; | |
7a8e76a3 | 1962 | } |
c4f50183 | 1963 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
1964 | |
1965 | /** | |
1966 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
1967 | * @buffer: The ring buffer | |
1968 | * @cpu: The per CPU buffer to get the number of overruns from | |
1969 | */ | |
1970 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
1971 | { | |
1972 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 1973 | unsigned long ret; |
7a8e76a3 | 1974 | |
9e01c1b7 | 1975 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1976 | return 0; |
7a8e76a3 SR |
1977 | |
1978 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 1979 | ret = cpu_buffer->overrun; |
554f786e SR |
1980 | |
1981 | return ret; | |
7a8e76a3 | 1982 | } |
c4f50183 | 1983 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 1984 | |
f0d2c681 SR |
1985 | /** |
1986 | * ring_buffer_nmi_dropped_cpu - get the number of nmis that were dropped | |
1987 | * @buffer: The ring buffer | |
1988 | * @cpu: The per CPU buffer to get the number of overruns from | |
1989 | */ | |
1990 | unsigned long ring_buffer_nmi_dropped_cpu(struct ring_buffer *buffer, int cpu) | |
1991 | { | |
1992 | struct ring_buffer_per_cpu *cpu_buffer; | |
1993 | unsigned long ret; | |
1994 | ||
1995 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
1996 | return 0; | |
1997 | ||
1998 | cpu_buffer = buffer->buffers[cpu]; | |
1999 | ret = cpu_buffer->nmi_dropped; | |
2000 | ||
2001 | return ret; | |
2002 | } | |
2003 | EXPORT_SYMBOL_GPL(ring_buffer_nmi_dropped_cpu); | |
2004 | ||
2005 | /** | |
2006 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits | |
2007 | * @buffer: The ring buffer | |
2008 | * @cpu: The per CPU buffer to get the number of overruns from | |
2009 | */ | |
2010 | unsigned long | |
2011 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
2012 | { | |
2013 | struct ring_buffer_per_cpu *cpu_buffer; | |
2014 | unsigned long ret; | |
2015 | ||
2016 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
2017 | return 0; | |
2018 | ||
2019 | cpu_buffer = buffer->buffers[cpu]; | |
2020 | ret = cpu_buffer->commit_overrun; | |
2021 | ||
2022 | return ret; | |
2023 | } | |
2024 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
2025 | ||
7a8e76a3 SR |
2026 | /** |
2027 | * ring_buffer_entries - get the number of entries in a buffer | |
2028 | * @buffer: The ring buffer | |
2029 | * | |
2030 | * Returns the total number of entries in the ring buffer | |
2031 | * (all CPU entries) | |
2032 | */ | |
2033 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
2034 | { | |
2035 | struct ring_buffer_per_cpu *cpu_buffer; | |
2036 | unsigned long entries = 0; | |
2037 | int cpu; | |
2038 | ||
2039 | /* if you care about this being correct, lock the buffer */ | |
2040 | for_each_buffer_cpu(buffer, cpu) { | |
2041 | cpu_buffer = buffer->buffers[cpu]; | |
e4906eff SR |
2042 | entries += (local_read(&cpu_buffer->entries) - |
2043 | cpu_buffer->overrun) - cpu_buffer->read; | |
7a8e76a3 SR |
2044 | } |
2045 | ||
2046 | return entries; | |
2047 | } | |
c4f50183 | 2048 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
2049 | |
2050 | /** | |
2051 | * ring_buffer_overrun_cpu - get the number of overruns in buffer | |
2052 | * @buffer: The ring buffer | |
2053 | * | |
2054 | * Returns the total number of overruns in the ring buffer | |
2055 | * (all CPU entries) | |
2056 | */ | |
2057 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
2058 | { | |
2059 | struct ring_buffer_per_cpu *cpu_buffer; | |
2060 | unsigned long overruns = 0; | |
2061 | int cpu; | |
2062 | ||
2063 | /* if you care about this being correct, lock the buffer */ | |
2064 | for_each_buffer_cpu(buffer, cpu) { | |
2065 | cpu_buffer = buffer->buffers[cpu]; | |
2066 | overruns += cpu_buffer->overrun; | |
2067 | } | |
2068 | ||
2069 | return overruns; | |
2070 | } | |
c4f50183 | 2071 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 2072 | |
642edba5 | 2073 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
2074 | { |
2075 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2076 | ||
d769041f SR |
2077 | /* Iterator usage is expected to have record disabled */ |
2078 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
2079 | iter->head_page = cpu_buffer->head_page; | |
6f807acd | 2080 | iter->head = cpu_buffer->head_page->read; |
d769041f SR |
2081 | } else { |
2082 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 2083 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
2084 | } |
2085 | if (iter->head) | |
2086 | iter->read_stamp = cpu_buffer->read_stamp; | |
2087 | else | |
abc9b56d | 2088 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 2089 | } |
f83c9d0f | 2090 | |
642edba5 SR |
2091 | /** |
2092 | * ring_buffer_iter_reset - reset an iterator | |
2093 | * @iter: The iterator to reset | |
2094 | * | |
2095 | * Resets the iterator, so that it will start from the beginning | |
2096 | * again. | |
2097 | */ | |
2098 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
2099 | { | |
554f786e | 2100 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
2101 | unsigned long flags; |
2102 | ||
554f786e SR |
2103 | if (!iter) |
2104 | return; | |
2105 | ||
2106 | cpu_buffer = iter->cpu_buffer; | |
2107 | ||
642edba5 SR |
2108 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2109 | rb_iter_reset(iter); | |
f83c9d0f | 2110 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 2111 | } |
c4f50183 | 2112 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
2113 | |
2114 | /** | |
2115 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
2116 | * @iter: The iterator to check | |
2117 | */ | |
2118 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
2119 | { | |
2120 | struct ring_buffer_per_cpu *cpu_buffer; | |
2121 | ||
2122 | cpu_buffer = iter->cpu_buffer; | |
2123 | ||
bf41a158 SR |
2124 | return iter->head_page == cpu_buffer->commit_page && |
2125 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 2126 | } |
c4f50183 | 2127 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
2128 | |
2129 | static void | |
2130 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
2131 | struct ring_buffer_event *event) | |
2132 | { | |
2133 | u64 delta; | |
2134 | ||
334d4169 | 2135 | switch (event->type_len) { |
7a8e76a3 SR |
2136 | case RINGBUF_TYPE_PADDING: |
2137 | return; | |
2138 | ||
2139 | case RINGBUF_TYPE_TIME_EXTEND: | |
2140 | delta = event->array[0]; | |
2141 | delta <<= TS_SHIFT; | |
2142 | delta += event->time_delta; | |
2143 | cpu_buffer->read_stamp += delta; | |
2144 | return; | |
2145 | ||
2146 | case RINGBUF_TYPE_TIME_STAMP: | |
2147 | /* FIXME: not implemented */ | |
2148 | return; | |
2149 | ||
2150 | case RINGBUF_TYPE_DATA: | |
2151 | cpu_buffer->read_stamp += event->time_delta; | |
2152 | return; | |
2153 | ||
2154 | default: | |
2155 | BUG(); | |
2156 | } | |
2157 | return; | |
2158 | } | |
2159 | ||
2160 | static void | |
2161 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
2162 | struct ring_buffer_event *event) | |
2163 | { | |
2164 | u64 delta; | |
2165 | ||
334d4169 | 2166 | switch (event->type_len) { |
7a8e76a3 SR |
2167 | case RINGBUF_TYPE_PADDING: |
2168 | return; | |
2169 | ||
2170 | case RINGBUF_TYPE_TIME_EXTEND: | |
2171 | delta = event->array[0]; | |
2172 | delta <<= TS_SHIFT; | |
2173 | delta += event->time_delta; | |
2174 | iter->read_stamp += delta; | |
2175 | return; | |
2176 | ||
2177 | case RINGBUF_TYPE_TIME_STAMP: | |
2178 | /* FIXME: not implemented */ | |
2179 | return; | |
2180 | ||
2181 | case RINGBUF_TYPE_DATA: | |
2182 | iter->read_stamp += event->time_delta; | |
2183 | return; | |
2184 | ||
2185 | default: | |
2186 | BUG(); | |
2187 | } | |
2188 | return; | |
2189 | } | |
2190 | ||
d769041f SR |
2191 | static struct buffer_page * |
2192 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 2193 | { |
d769041f SR |
2194 | struct buffer_page *reader = NULL; |
2195 | unsigned long flags; | |
818e3dd3 | 2196 | int nr_loops = 0; |
d769041f | 2197 | |
3e03fb7f SR |
2198 | local_irq_save(flags); |
2199 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
2200 | |
2201 | again: | |
818e3dd3 SR |
2202 | /* |
2203 | * This should normally only loop twice. But because the | |
2204 | * start of the reader inserts an empty page, it causes | |
2205 | * a case where we will loop three times. There should be no | |
2206 | * reason to loop four times (that I know of). | |
2207 | */ | |
3e89c7bb | 2208 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
2209 | reader = NULL; |
2210 | goto out; | |
2211 | } | |
2212 | ||
d769041f SR |
2213 | reader = cpu_buffer->reader_page; |
2214 | ||
2215 | /* If there's more to read, return this page */ | |
bf41a158 | 2216 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
2217 | goto out; |
2218 | ||
2219 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
2220 | if (RB_WARN_ON(cpu_buffer, |
2221 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
2222 | goto out; | |
d769041f SR |
2223 | |
2224 | /* check if we caught up to the tail */ | |
2225 | reader = NULL; | |
bf41a158 | 2226 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 2227 | goto out; |
7a8e76a3 SR |
2228 | |
2229 | /* | |
d769041f SR |
2230 | * Splice the empty reader page into the list around the head. |
2231 | * Reset the reader page to size zero. | |
7a8e76a3 | 2232 | */ |
7a8e76a3 | 2233 | |
d769041f SR |
2234 | reader = cpu_buffer->head_page; |
2235 | cpu_buffer->reader_page->list.next = reader->list.next; | |
2236 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 SR |
2237 | |
2238 | local_set(&cpu_buffer->reader_page->write, 0); | |
778c55d4 | 2239 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 2240 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
7a8e76a3 | 2241 | |
d769041f SR |
2242 | /* Make the reader page now replace the head */ |
2243 | reader->list.prev->next = &cpu_buffer->reader_page->list; | |
2244 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
7a8e76a3 SR |
2245 | |
2246 | /* | |
d769041f SR |
2247 | * If the tail is on the reader, then we must set the head |
2248 | * to the inserted page, otherwise we set it one before. | |
7a8e76a3 | 2249 | */ |
d769041f | 2250 | cpu_buffer->head_page = cpu_buffer->reader_page; |
7a8e76a3 | 2251 | |
bf41a158 | 2252 | if (cpu_buffer->commit_page != reader) |
d769041f SR |
2253 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
2254 | ||
2255 | /* Finally update the reader page to the new head */ | |
2256 | cpu_buffer->reader_page = reader; | |
2257 | rb_reset_reader_page(cpu_buffer); | |
2258 | ||
2259 | goto again; | |
2260 | ||
2261 | out: | |
3e03fb7f SR |
2262 | __raw_spin_unlock(&cpu_buffer->lock); |
2263 | local_irq_restore(flags); | |
d769041f SR |
2264 | |
2265 | return reader; | |
2266 | } | |
2267 | ||
2268 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
2269 | { | |
2270 | struct ring_buffer_event *event; | |
2271 | struct buffer_page *reader; | |
2272 | unsigned length; | |
2273 | ||
2274 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 2275 | |
d769041f | 2276 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
2277 | if (RB_WARN_ON(cpu_buffer, !reader)) |
2278 | return; | |
7a8e76a3 | 2279 | |
d769041f SR |
2280 | event = rb_reader_event(cpu_buffer); |
2281 | ||
334d4169 LJ |
2282 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX |
2283 | || rb_discarded_event(event)) | |
e4906eff | 2284 | cpu_buffer->read++; |
d769041f SR |
2285 | |
2286 | rb_update_read_stamp(cpu_buffer, event); | |
2287 | ||
2288 | length = rb_event_length(event); | |
6f807acd | 2289 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
2290 | } |
2291 | ||
2292 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
2293 | { | |
2294 | struct ring_buffer *buffer; | |
2295 | struct ring_buffer_per_cpu *cpu_buffer; | |
2296 | struct ring_buffer_event *event; | |
2297 | unsigned length; | |
2298 | ||
2299 | cpu_buffer = iter->cpu_buffer; | |
2300 | buffer = cpu_buffer->buffer; | |
2301 | ||
2302 | /* | |
2303 | * Check if we are at the end of the buffer. | |
2304 | */ | |
bf41a158 | 2305 | if (iter->head >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
2306 | /* discarded commits can make the page empty */ |
2307 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 2308 | return; |
d769041f | 2309 | rb_inc_iter(iter); |
7a8e76a3 SR |
2310 | return; |
2311 | } | |
2312 | ||
2313 | event = rb_iter_head_event(iter); | |
2314 | ||
2315 | length = rb_event_length(event); | |
2316 | ||
2317 | /* | |
2318 | * This should not be called to advance the header if we are | |
2319 | * at the tail of the buffer. | |
2320 | */ | |
3e89c7bb | 2321 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 2322 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
2323 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
2324 | return; | |
7a8e76a3 SR |
2325 | |
2326 | rb_update_iter_read_stamp(iter, event); | |
2327 | ||
2328 | iter->head += length; | |
2329 | ||
2330 | /* check for end of page padding */ | |
bf41a158 SR |
2331 | if ((iter->head >= rb_page_size(iter->head_page)) && |
2332 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
2333 | rb_advance_iter(iter); |
2334 | } | |
2335 | ||
f83c9d0f SR |
2336 | static struct ring_buffer_event * |
2337 | rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
7a8e76a3 SR |
2338 | { |
2339 | struct ring_buffer_per_cpu *cpu_buffer; | |
2340 | struct ring_buffer_event *event; | |
d769041f | 2341 | struct buffer_page *reader; |
818e3dd3 | 2342 | int nr_loops = 0; |
7a8e76a3 | 2343 | |
7a8e76a3 SR |
2344 | cpu_buffer = buffer->buffers[cpu]; |
2345 | ||
2346 | again: | |
818e3dd3 SR |
2347 | /* |
2348 | * We repeat when a timestamp is encountered. It is possible | |
2349 | * to get multiple timestamps from an interrupt entering just | |
ea05b57c SR |
2350 | * as one timestamp is about to be written, or from discarded |
2351 | * commits. The most that we can have is the number on a single page. | |
818e3dd3 | 2352 | */ |
ea05b57c | 2353 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE)) |
818e3dd3 | 2354 | return NULL; |
818e3dd3 | 2355 | |
d769041f SR |
2356 | reader = rb_get_reader_page(cpu_buffer); |
2357 | if (!reader) | |
7a8e76a3 SR |
2358 | return NULL; |
2359 | ||
d769041f | 2360 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 2361 | |
334d4169 | 2362 | switch (event->type_len) { |
7a8e76a3 | 2363 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
2364 | if (rb_null_event(event)) |
2365 | RB_WARN_ON(cpu_buffer, 1); | |
2366 | /* | |
2367 | * Because the writer could be discarding every | |
2368 | * event it creates (which would probably be bad) | |
2369 | * if we were to go back to "again" then we may never | |
2370 | * catch up, and will trigger the warn on, or lock | |
2371 | * the box. Return the padding, and we will release | |
2372 | * the current locks, and try again. | |
2373 | */ | |
d769041f | 2374 | rb_advance_reader(cpu_buffer); |
2d622719 | 2375 | return event; |
7a8e76a3 SR |
2376 | |
2377 | case RINGBUF_TYPE_TIME_EXTEND: | |
2378 | /* Internal data, OK to advance */ | |
d769041f | 2379 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2380 | goto again; |
2381 | ||
2382 | case RINGBUF_TYPE_TIME_STAMP: | |
2383 | /* FIXME: not implemented */ | |
d769041f | 2384 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2385 | goto again; |
2386 | ||
2387 | case RINGBUF_TYPE_DATA: | |
2388 | if (ts) { | |
2389 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
37886f6a SR |
2390 | ring_buffer_normalize_time_stamp(buffer, |
2391 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2392 | } |
2393 | return event; | |
2394 | ||
2395 | default: | |
2396 | BUG(); | |
2397 | } | |
2398 | ||
2399 | return NULL; | |
2400 | } | |
c4f50183 | 2401 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 2402 | |
f83c9d0f SR |
2403 | static struct ring_buffer_event * |
2404 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
2405 | { |
2406 | struct ring_buffer *buffer; | |
2407 | struct ring_buffer_per_cpu *cpu_buffer; | |
2408 | struct ring_buffer_event *event; | |
818e3dd3 | 2409 | int nr_loops = 0; |
7a8e76a3 SR |
2410 | |
2411 | if (ring_buffer_iter_empty(iter)) | |
2412 | return NULL; | |
2413 | ||
2414 | cpu_buffer = iter->cpu_buffer; | |
2415 | buffer = cpu_buffer->buffer; | |
2416 | ||
2417 | again: | |
818e3dd3 | 2418 | /* |
ea05b57c SR |
2419 | * We repeat when a timestamp is encountered. |
2420 | * We can get multiple timestamps by nested interrupts or also | |
2421 | * if filtering is on (discarding commits). Since discarding | |
2422 | * commits can be frequent we can get a lot of timestamps. | |
2423 | * But we limit them by not adding timestamps if they begin | |
2424 | * at the start of a page. | |
818e3dd3 | 2425 | */ |
ea05b57c | 2426 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE)) |
818e3dd3 | 2427 | return NULL; |
818e3dd3 | 2428 | |
7a8e76a3 SR |
2429 | if (rb_per_cpu_empty(cpu_buffer)) |
2430 | return NULL; | |
2431 | ||
2432 | event = rb_iter_head_event(iter); | |
2433 | ||
334d4169 | 2434 | switch (event->type_len) { |
7a8e76a3 | 2435 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
2436 | if (rb_null_event(event)) { |
2437 | rb_inc_iter(iter); | |
2438 | goto again; | |
2439 | } | |
2440 | rb_advance_iter(iter); | |
2441 | return event; | |
7a8e76a3 SR |
2442 | |
2443 | case RINGBUF_TYPE_TIME_EXTEND: | |
2444 | /* Internal data, OK to advance */ | |
2445 | rb_advance_iter(iter); | |
2446 | goto again; | |
2447 | ||
2448 | case RINGBUF_TYPE_TIME_STAMP: | |
2449 | /* FIXME: not implemented */ | |
2450 | rb_advance_iter(iter); | |
2451 | goto again; | |
2452 | ||
2453 | case RINGBUF_TYPE_DATA: | |
2454 | if (ts) { | |
2455 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
2456 | ring_buffer_normalize_time_stamp(buffer, |
2457 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2458 | } |
2459 | return event; | |
2460 | ||
2461 | default: | |
2462 | BUG(); | |
2463 | } | |
2464 | ||
2465 | return NULL; | |
2466 | } | |
c4f50183 | 2467 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 2468 | |
f83c9d0f SR |
2469 | /** |
2470 | * ring_buffer_peek - peek at the next event to be read | |
2471 | * @buffer: The ring buffer to read | |
2472 | * @cpu: The cpu to peak at | |
2473 | * @ts: The timestamp counter of this event. | |
2474 | * | |
2475 | * This will return the event that will be read next, but does | |
2476 | * not consume the data. | |
2477 | */ | |
2478 | struct ring_buffer_event * | |
2479 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2480 | { | |
2481 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 2482 | struct ring_buffer_event *event; |
f83c9d0f SR |
2483 | unsigned long flags; |
2484 | ||
554f786e | 2485 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2486 | return NULL; |
554f786e | 2487 | |
2d622719 | 2488 | again: |
f83c9d0f SR |
2489 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2490 | event = rb_buffer_peek(buffer, cpu, ts); | |
2491 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2492 | ||
334d4169 | 2493 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2494 | cpu_relax(); |
2495 | goto again; | |
2496 | } | |
2497 | ||
f83c9d0f SR |
2498 | return event; |
2499 | } | |
2500 | ||
2501 | /** | |
2502 | * ring_buffer_iter_peek - peek at the next event to be read | |
2503 | * @iter: The ring buffer iterator | |
2504 | * @ts: The timestamp counter of this event. | |
2505 | * | |
2506 | * This will return the event that will be read next, but does | |
2507 | * not increment the iterator. | |
2508 | */ | |
2509 | struct ring_buffer_event * | |
2510 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
2511 | { | |
2512 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2513 | struct ring_buffer_event *event; | |
2514 | unsigned long flags; | |
2515 | ||
2d622719 | 2516 | again: |
f83c9d0f SR |
2517 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2518 | event = rb_iter_peek(iter, ts); | |
2519 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2520 | ||
334d4169 | 2521 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2522 | cpu_relax(); |
2523 | goto again; | |
2524 | } | |
2525 | ||
f83c9d0f SR |
2526 | return event; |
2527 | } | |
2528 | ||
7a8e76a3 SR |
2529 | /** |
2530 | * ring_buffer_consume - return an event and consume it | |
2531 | * @buffer: The ring buffer to get the next event from | |
2532 | * | |
2533 | * Returns the next event in the ring buffer, and that event is consumed. | |
2534 | * Meaning, that sequential reads will keep returning a different event, | |
2535 | * and eventually empty the ring buffer if the producer is slower. | |
2536 | */ | |
2537 | struct ring_buffer_event * | |
2538 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2539 | { | |
554f786e SR |
2540 | struct ring_buffer_per_cpu *cpu_buffer; |
2541 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 2542 | unsigned long flags; |
7a8e76a3 | 2543 | |
2d622719 | 2544 | again: |
554f786e SR |
2545 | /* might be called in atomic */ |
2546 | preempt_disable(); | |
2547 | ||
9e01c1b7 | 2548 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 2549 | goto out; |
7a8e76a3 | 2550 | |
554f786e | 2551 | cpu_buffer = buffer->buffers[cpu]; |
f83c9d0f SR |
2552 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2553 | ||
2554 | event = rb_buffer_peek(buffer, cpu, ts); | |
7a8e76a3 | 2555 | if (!event) |
554f786e | 2556 | goto out_unlock; |
7a8e76a3 | 2557 | |
d769041f | 2558 | rb_advance_reader(cpu_buffer); |
7a8e76a3 | 2559 | |
554f786e | 2560 | out_unlock: |
f83c9d0f SR |
2561 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
2562 | ||
554f786e SR |
2563 | out: |
2564 | preempt_enable(); | |
2565 | ||
334d4169 | 2566 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2567 | cpu_relax(); |
2568 | goto again; | |
2569 | } | |
2570 | ||
7a8e76a3 SR |
2571 | return event; |
2572 | } | |
c4f50183 | 2573 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
2574 | |
2575 | /** | |
2576 | * ring_buffer_read_start - start a non consuming read of the buffer | |
2577 | * @buffer: The ring buffer to read from | |
2578 | * @cpu: The cpu buffer to iterate over | |
2579 | * | |
2580 | * This starts up an iteration through the buffer. It also disables | |
2581 | * the recording to the buffer until the reading is finished. | |
2582 | * This prevents the reading from being corrupted. This is not | |
2583 | * a consuming read, so a producer is not expected. | |
2584 | * | |
2585 | * Must be paired with ring_buffer_finish. | |
2586 | */ | |
2587 | struct ring_buffer_iter * | |
2588 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
2589 | { | |
2590 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2591 | struct ring_buffer_iter *iter; |
d769041f | 2592 | unsigned long flags; |
7a8e76a3 | 2593 | |
9e01c1b7 | 2594 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2595 | return NULL; |
7a8e76a3 SR |
2596 | |
2597 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
2598 | if (!iter) | |
8aabee57 | 2599 | return NULL; |
7a8e76a3 SR |
2600 | |
2601 | cpu_buffer = buffer->buffers[cpu]; | |
2602 | ||
2603 | iter->cpu_buffer = cpu_buffer; | |
2604 | ||
2605 | atomic_inc(&cpu_buffer->record_disabled); | |
2606 | synchronize_sched(); | |
2607 | ||
f83c9d0f | 2608 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 2609 | __raw_spin_lock(&cpu_buffer->lock); |
642edba5 | 2610 | rb_iter_reset(iter); |
3e03fb7f | 2611 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 2612 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
2613 | |
2614 | return iter; | |
2615 | } | |
c4f50183 | 2616 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
2617 | |
2618 | /** | |
2619 | * ring_buffer_finish - finish reading the iterator of the buffer | |
2620 | * @iter: The iterator retrieved by ring_buffer_start | |
2621 | * | |
2622 | * This re-enables the recording to the buffer, and frees the | |
2623 | * iterator. | |
2624 | */ | |
2625 | void | |
2626 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
2627 | { | |
2628 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2629 | ||
2630 | atomic_dec(&cpu_buffer->record_disabled); | |
2631 | kfree(iter); | |
2632 | } | |
c4f50183 | 2633 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
2634 | |
2635 | /** | |
2636 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
2637 | * @iter: The ring buffer iterator | |
2638 | * @ts: The time stamp of the event read. | |
2639 | * | |
2640 | * This reads the next event in the ring buffer and increments the iterator. | |
2641 | */ | |
2642 | struct ring_buffer_event * | |
2643 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
2644 | { | |
2645 | struct ring_buffer_event *event; | |
f83c9d0f SR |
2646 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
2647 | unsigned long flags; | |
7a8e76a3 | 2648 | |
2d622719 | 2649 | again: |
f83c9d0f SR |
2650 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2651 | event = rb_iter_peek(iter, ts); | |
7a8e76a3 | 2652 | if (!event) |
f83c9d0f | 2653 | goto out; |
7a8e76a3 SR |
2654 | |
2655 | rb_advance_iter(iter); | |
f83c9d0f SR |
2656 | out: |
2657 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 2658 | |
334d4169 | 2659 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2660 | cpu_relax(); |
2661 | goto again; | |
2662 | } | |
2663 | ||
7a8e76a3 SR |
2664 | return event; |
2665 | } | |
c4f50183 | 2666 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
2667 | |
2668 | /** | |
2669 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
2670 | * @buffer: The ring buffer. | |
2671 | */ | |
2672 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
2673 | { | |
2674 | return BUF_PAGE_SIZE * buffer->pages; | |
2675 | } | |
c4f50183 | 2676 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
2677 | |
2678 | static void | |
2679 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
2680 | { | |
2681 | cpu_buffer->head_page | |
2682 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 2683 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 2684 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 2685 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 2686 | |
6f807acd | 2687 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
2688 | |
2689 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
2690 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
2691 | ||
2692 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
2693 | local_set(&cpu_buffer->reader_page->write, 0); | |
778c55d4 | 2694 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 2695 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 2696 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 2697 | |
f0d2c681 SR |
2698 | cpu_buffer->nmi_dropped = 0; |
2699 | cpu_buffer->commit_overrun = 0; | |
7a8e76a3 | 2700 | cpu_buffer->overrun = 0; |
e4906eff SR |
2701 | cpu_buffer->read = 0; |
2702 | local_set(&cpu_buffer->entries, 0); | |
fa743953 SR |
2703 | local_set(&cpu_buffer->committing, 0); |
2704 | local_set(&cpu_buffer->commits, 0); | |
69507c06 SR |
2705 | |
2706 | cpu_buffer->write_stamp = 0; | |
2707 | cpu_buffer->read_stamp = 0; | |
7a8e76a3 SR |
2708 | } |
2709 | ||
2710 | /** | |
2711 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
2712 | * @buffer: The ring buffer to reset a per cpu buffer of | |
2713 | * @cpu: The CPU buffer to be reset | |
2714 | */ | |
2715 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
2716 | { | |
2717 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2718 | unsigned long flags; | |
2719 | ||
9e01c1b7 | 2720 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2721 | return; |
7a8e76a3 | 2722 | |
41ede23e SR |
2723 | atomic_inc(&cpu_buffer->record_disabled); |
2724 | ||
f83c9d0f SR |
2725 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2726 | ||
3e03fb7f | 2727 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
2728 | |
2729 | rb_reset_cpu(cpu_buffer); | |
2730 | ||
3e03fb7f | 2731 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f SR |
2732 | |
2733 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
41ede23e SR |
2734 | |
2735 | atomic_dec(&cpu_buffer->record_disabled); | |
7a8e76a3 | 2736 | } |
c4f50183 | 2737 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
2738 | |
2739 | /** | |
2740 | * ring_buffer_reset - reset a ring buffer | |
2741 | * @buffer: The ring buffer to reset all cpu buffers | |
2742 | */ | |
2743 | void ring_buffer_reset(struct ring_buffer *buffer) | |
2744 | { | |
7a8e76a3 SR |
2745 | int cpu; |
2746 | ||
7a8e76a3 | 2747 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 2748 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 2749 | } |
c4f50183 | 2750 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
2751 | |
2752 | /** | |
2753 | * rind_buffer_empty - is the ring buffer empty? | |
2754 | * @buffer: The ring buffer to test | |
2755 | */ | |
2756 | int ring_buffer_empty(struct ring_buffer *buffer) | |
2757 | { | |
2758 | struct ring_buffer_per_cpu *cpu_buffer; | |
2759 | int cpu; | |
2760 | ||
2761 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
2762 | for_each_buffer_cpu(buffer, cpu) { | |
2763 | cpu_buffer = buffer->buffers[cpu]; | |
2764 | if (!rb_per_cpu_empty(cpu_buffer)) | |
2765 | return 0; | |
2766 | } | |
554f786e | 2767 | |
7a8e76a3 SR |
2768 | return 1; |
2769 | } | |
c4f50183 | 2770 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
2771 | |
2772 | /** | |
2773 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
2774 | * @buffer: The ring buffer | |
2775 | * @cpu: The CPU buffer to test | |
2776 | */ | |
2777 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
2778 | { | |
2779 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2780 | int ret; |
7a8e76a3 | 2781 | |
9e01c1b7 | 2782 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2783 | return 1; |
7a8e76a3 SR |
2784 | |
2785 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e SR |
2786 | ret = rb_per_cpu_empty(cpu_buffer); |
2787 | ||
554f786e SR |
2788 | |
2789 | return ret; | |
7a8e76a3 | 2790 | } |
c4f50183 | 2791 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 SR |
2792 | |
2793 | /** | |
2794 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
2795 | * @buffer_a: One buffer to swap with | |
2796 | * @buffer_b: The other buffer to swap with | |
2797 | * | |
2798 | * This function is useful for tracers that want to take a "snapshot" | |
2799 | * of a CPU buffer and has another back up buffer lying around. | |
2800 | * it is expected that the tracer handles the cpu buffer not being | |
2801 | * used at the moment. | |
2802 | */ | |
2803 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
2804 | struct ring_buffer *buffer_b, int cpu) | |
2805 | { | |
2806 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
2807 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
2808 | int ret = -EINVAL; |
2809 | ||
9e01c1b7 RR |
2810 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
2811 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 2812 | goto out; |
7a8e76a3 SR |
2813 | |
2814 | /* At least make sure the two buffers are somewhat the same */ | |
6d102bc6 | 2815 | if (buffer_a->pages != buffer_b->pages) |
554f786e SR |
2816 | goto out; |
2817 | ||
2818 | ret = -EAGAIN; | |
7a8e76a3 | 2819 | |
97b17efe | 2820 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 2821 | goto out; |
97b17efe SR |
2822 | |
2823 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 2824 | goto out; |
97b17efe SR |
2825 | |
2826 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 2827 | goto out; |
97b17efe | 2828 | |
7a8e76a3 SR |
2829 | cpu_buffer_a = buffer_a->buffers[cpu]; |
2830 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
2831 | ||
97b17efe | 2832 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 2833 | goto out; |
97b17efe SR |
2834 | |
2835 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 2836 | goto out; |
97b17efe | 2837 | |
7a8e76a3 SR |
2838 | /* |
2839 | * We can't do a synchronize_sched here because this | |
2840 | * function can be called in atomic context. | |
2841 | * Normally this will be called from the same CPU as cpu. | |
2842 | * If not it's up to the caller to protect this. | |
2843 | */ | |
2844 | atomic_inc(&cpu_buffer_a->record_disabled); | |
2845 | atomic_inc(&cpu_buffer_b->record_disabled); | |
2846 | ||
2847 | buffer_a->buffers[cpu] = cpu_buffer_b; | |
2848 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
2849 | ||
2850 | cpu_buffer_b->buffer = buffer_a; | |
2851 | cpu_buffer_a->buffer = buffer_b; | |
2852 | ||
2853 | atomic_dec(&cpu_buffer_a->record_disabled); | |
2854 | atomic_dec(&cpu_buffer_b->record_disabled); | |
2855 | ||
554f786e SR |
2856 | ret = 0; |
2857 | out: | |
554f786e | 2858 | return ret; |
7a8e76a3 | 2859 | } |
c4f50183 | 2860 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
7a8e76a3 | 2861 | |
8789a9e7 SR |
2862 | /** |
2863 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
2864 | * @buffer: the buffer to allocate for. | |
2865 | * | |
2866 | * This function is used in conjunction with ring_buffer_read_page. | |
2867 | * When reading a full page from the ring buffer, these functions | |
2868 | * can be used to speed up the process. The calling function should | |
2869 | * allocate a few pages first with this function. Then when it | |
2870 | * needs to get pages from the ring buffer, it passes the result | |
2871 | * of this function into ring_buffer_read_page, which will swap | |
2872 | * the page that was allocated, with the read page of the buffer. | |
2873 | * | |
2874 | * Returns: | |
2875 | * The page allocated, or NULL on error. | |
2876 | */ | |
2877 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer) | |
2878 | { | |
044fa782 | 2879 | struct buffer_data_page *bpage; |
ef7a4a16 | 2880 | unsigned long addr; |
8789a9e7 SR |
2881 | |
2882 | addr = __get_free_page(GFP_KERNEL); | |
2883 | if (!addr) | |
2884 | return NULL; | |
2885 | ||
044fa782 | 2886 | bpage = (void *)addr; |
8789a9e7 | 2887 | |
ef7a4a16 SR |
2888 | rb_init_page(bpage); |
2889 | ||
044fa782 | 2890 | return bpage; |
8789a9e7 | 2891 | } |
d6ce96da | 2892 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
2893 | |
2894 | /** | |
2895 | * ring_buffer_free_read_page - free an allocated read page | |
2896 | * @buffer: the buffer the page was allocate for | |
2897 | * @data: the page to free | |
2898 | * | |
2899 | * Free a page allocated from ring_buffer_alloc_read_page. | |
2900 | */ | |
2901 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
2902 | { | |
2903 | free_page((unsigned long)data); | |
2904 | } | |
d6ce96da | 2905 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
2906 | |
2907 | /** | |
2908 | * ring_buffer_read_page - extract a page from the ring buffer | |
2909 | * @buffer: buffer to extract from | |
2910 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 2911 | * @len: amount to extract |
8789a9e7 SR |
2912 | * @cpu: the cpu of the buffer to extract |
2913 | * @full: should the extraction only happen when the page is full. | |
2914 | * | |
2915 | * This function will pull out a page from the ring buffer and consume it. | |
2916 | * @data_page must be the address of the variable that was returned | |
2917 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
2918 | * to swap with a page in the ring buffer. | |
2919 | * | |
2920 | * for example: | |
b85fa01e | 2921 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
2922 | * if (!rpage) |
2923 | * return error; | |
ef7a4a16 | 2924 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
2925 | * if (ret >= 0) |
2926 | * process_page(rpage, ret); | |
8789a9e7 SR |
2927 | * |
2928 | * When @full is set, the function will not return true unless | |
2929 | * the writer is off the reader page. | |
2930 | * | |
2931 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
2932 | * The ring buffer can be used anywhere in the kernel and can not | |
2933 | * blindly call wake_up. The layer that uses the ring buffer must be | |
2934 | * responsible for that. | |
2935 | * | |
2936 | * Returns: | |
667d2412 LJ |
2937 | * >=0 if data has been transferred, returns the offset of consumed data. |
2938 | * <0 if no data has been transferred. | |
8789a9e7 SR |
2939 | */ |
2940 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 2941 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
2942 | { |
2943 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2944 | struct ring_buffer_event *event; | |
044fa782 | 2945 | struct buffer_data_page *bpage; |
ef7a4a16 | 2946 | struct buffer_page *reader; |
8789a9e7 | 2947 | unsigned long flags; |
ef7a4a16 | 2948 | unsigned int commit; |
667d2412 | 2949 | unsigned int read; |
4f3640f8 | 2950 | u64 save_timestamp; |
667d2412 | 2951 | int ret = -1; |
8789a9e7 | 2952 | |
554f786e SR |
2953 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
2954 | goto out; | |
2955 | ||
474d32b6 SR |
2956 | /* |
2957 | * If len is not big enough to hold the page header, then | |
2958 | * we can not copy anything. | |
2959 | */ | |
2960 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 2961 | goto out; |
474d32b6 SR |
2962 | |
2963 | len -= BUF_PAGE_HDR_SIZE; | |
2964 | ||
8789a9e7 | 2965 | if (!data_page) |
554f786e | 2966 | goto out; |
8789a9e7 | 2967 | |
044fa782 SR |
2968 | bpage = *data_page; |
2969 | if (!bpage) | |
554f786e | 2970 | goto out; |
8789a9e7 SR |
2971 | |
2972 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2973 | ||
ef7a4a16 SR |
2974 | reader = rb_get_reader_page(cpu_buffer); |
2975 | if (!reader) | |
554f786e | 2976 | goto out_unlock; |
8789a9e7 | 2977 | |
ef7a4a16 SR |
2978 | event = rb_reader_event(cpu_buffer); |
2979 | ||
2980 | read = reader->read; | |
2981 | commit = rb_page_commit(reader); | |
667d2412 | 2982 | |
8789a9e7 | 2983 | /* |
474d32b6 SR |
2984 | * If this page has been partially read or |
2985 | * if len is not big enough to read the rest of the page or | |
2986 | * a writer is still on the page, then | |
2987 | * we must copy the data from the page to the buffer. | |
2988 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 2989 | */ |
474d32b6 | 2990 | if (read || (len < (commit - read)) || |
ef7a4a16 | 2991 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 2992 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
2993 | unsigned int rpos = read; |
2994 | unsigned int pos = 0; | |
ef7a4a16 | 2995 | unsigned int size; |
8789a9e7 SR |
2996 | |
2997 | if (full) | |
554f786e | 2998 | goto out_unlock; |
8789a9e7 | 2999 | |
ef7a4a16 SR |
3000 | if (len > (commit - read)) |
3001 | len = (commit - read); | |
3002 | ||
3003 | size = rb_event_length(event); | |
3004 | ||
3005 | if (len < size) | |
554f786e | 3006 | goto out_unlock; |
ef7a4a16 | 3007 | |
4f3640f8 SR |
3008 | /* save the current timestamp, since the user will need it */ |
3009 | save_timestamp = cpu_buffer->read_stamp; | |
3010 | ||
ef7a4a16 SR |
3011 | /* Need to copy one event at a time */ |
3012 | do { | |
474d32b6 | 3013 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
3014 | |
3015 | len -= size; | |
3016 | ||
3017 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
3018 | rpos = reader->read; |
3019 | pos += size; | |
ef7a4a16 SR |
3020 | |
3021 | event = rb_reader_event(cpu_buffer); | |
3022 | size = rb_event_length(event); | |
3023 | } while (len > size); | |
667d2412 LJ |
3024 | |
3025 | /* update bpage */ | |
ef7a4a16 | 3026 | local_set(&bpage->commit, pos); |
4f3640f8 | 3027 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 3028 | |
474d32b6 SR |
3029 | /* we copied everything to the beginning */ |
3030 | read = 0; | |
8789a9e7 | 3031 | } else { |
afbab76a SR |
3032 | /* update the entry counter */ |
3033 | cpu_buffer->read += local_read(&reader->entries); | |
3034 | ||
8789a9e7 | 3035 | /* swap the pages */ |
044fa782 | 3036 | rb_init_page(bpage); |
ef7a4a16 SR |
3037 | bpage = reader->page; |
3038 | reader->page = *data_page; | |
3039 | local_set(&reader->write, 0); | |
778c55d4 | 3040 | local_set(&reader->entries, 0); |
ef7a4a16 | 3041 | reader->read = 0; |
044fa782 | 3042 | *data_page = bpage; |
8789a9e7 | 3043 | } |
667d2412 | 3044 | ret = read; |
8789a9e7 | 3045 | |
554f786e | 3046 | out_unlock: |
8789a9e7 SR |
3047 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
3048 | ||
554f786e | 3049 | out: |
8789a9e7 SR |
3050 | return ret; |
3051 | } | |
d6ce96da | 3052 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 3053 | |
a3583244 SR |
3054 | static ssize_t |
3055 | rb_simple_read(struct file *filp, char __user *ubuf, | |
3056 | size_t cnt, loff_t *ppos) | |
3057 | { | |
5e39841c | 3058 | unsigned long *p = filp->private_data; |
a3583244 SR |
3059 | char buf[64]; |
3060 | int r; | |
3061 | ||
033601a3 SR |
3062 | if (test_bit(RB_BUFFERS_DISABLED_BIT, p)) |
3063 | r = sprintf(buf, "permanently disabled\n"); | |
3064 | else | |
3065 | r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p)); | |
a3583244 SR |
3066 | |
3067 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3068 | } | |
3069 | ||
3070 | static ssize_t | |
3071 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
3072 | size_t cnt, loff_t *ppos) | |
3073 | { | |
5e39841c | 3074 | unsigned long *p = filp->private_data; |
a3583244 | 3075 | char buf[64]; |
5e39841c | 3076 | unsigned long val; |
a3583244 SR |
3077 | int ret; |
3078 | ||
3079 | if (cnt >= sizeof(buf)) | |
3080 | return -EINVAL; | |
3081 | ||
3082 | if (copy_from_user(&buf, ubuf, cnt)) | |
3083 | return -EFAULT; | |
3084 | ||
3085 | buf[cnt] = 0; | |
3086 | ||
3087 | ret = strict_strtoul(buf, 10, &val); | |
3088 | if (ret < 0) | |
3089 | return ret; | |
3090 | ||
033601a3 SR |
3091 | if (val) |
3092 | set_bit(RB_BUFFERS_ON_BIT, p); | |
3093 | else | |
3094 | clear_bit(RB_BUFFERS_ON_BIT, p); | |
a3583244 SR |
3095 | |
3096 | (*ppos)++; | |
3097 | ||
3098 | return cnt; | |
3099 | } | |
3100 | ||
5e2336a0 | 3101 | static const struct file_operations rb_simple_fops = { |
a3583244 SR |
3102 | .open = tracing_open_generic, |
3103 | .read = rb_simple_read, | |
3104 | .write = rb_simple_write, | |
3105 | }; | |
3106 | ||
3107 | ||
3108 | static __init int rb_init_debugfs(void) | |
3109 | { | |
3110 | struct dentry *d_tracer; | |
a3583244 SR |
3111 | |
3112 | d_tracer = tracing_init_dentry(); | |
3113 | ||
5452af66 FW |
3114 | trace_create_file("tracing_on", 0644, d_tracer, |
3115 | &ring_buffer_flags, &rb_simple_fops); | |
a3583244 SR |
3116 | |
3117 | return 0; | |
3118 | } | |
3119 | ||
3120 | fs_initcall(rb_init_debugfs); | |
554f786e | 3121 | |
59222efe | 3122 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
3123 | static int rb_cpu_notify(struct notifier_block *self, |
3124 | unsigned long action, void *hcpu) | |
554f786e SR |
3125 | { |
3126 | struct ring_buffer *buffer = | |
3127 | container_of(self, struct ring_buffer, cpu_notify); | |
3128 | long cpu = (long)hcpu; | |
3129 | ||
3130 | switch (action) { | |
3131 | case CPU_UP_PREPARE: | |
3132 | case CPU_UP_PREPARE_FROZEN: | |
3f237a79 | 3133 | if (cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e SR |
3134 | return NOTIFY_OK; |
3135 | ||
3136 | buffer->buffers[cpu] = | |
3137 | rb_allocate_cpu_buffer(buffer, cpu); | |
3138 | if (!buffer->buffers[cpu]) { | |
3139 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
3140 | cpu); | |
3141 | return NOTIFY_OK; | |
3142 | } | |
3143 | smp_wmb(); | |
3f237a79 | 3144 | cpumask_set_cpu(cpu, buffer->cpumask); |
554f786e SR |
3145 | break; |
3146 | case CPU_DOWN_PREPARE: | |
3147 | case CPU_DOWN_PREPARE_FROZEN: | |
3148 | /* | |
3149 | * Do nothing. | |
3150 | * If we were to free the buffer, then the user would | |
3151 | * lose any trace that was in the buffer. | |
3152 | */ | |
3153 | break; | |
3154 | default: | |
3155 | break; | |
3156 | } | |
3157 | return NOTIFY_OK; | |
3158 | } | |
3159 | #endif |