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; | |
7a8e76a3 SR |
418 | u64 write_stamp; |
419 | u64 read_stamp; | |
420 | atomic_t record_disabled; | |
421 | }; | |
422 | ||
423 | struct ring_buffer { | |
7a8e76a3 SR |
424 | unsigned pages; |
425 | unsigned flags; | |
426 | int cpus; | |
7a8e76a3 | 427 | atomic_t record_disabled; |
00f62f61 | 428 | cpumask_var_t cpumask; |
7a8e76a3 | 429 | |
1f8a6a10 PZ |
430 | struct lock_class_key *reader_lock_key; |
431 | ||
7a8e76a3 SR |
432 | struct mutex mutex; |
433 | ||
434 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 435 | |
59222efe | 436 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
437 | struct notifier_block cpu_notify; |
438 | #endif | |
37886f6a | 439 | u64 (*clock)(void); |
7a8e76a3 SR |
440 | }; |
441 | ||
442 | struct ring_buffer_iter { | |
443 | struct ring_buffer_per_cpu *cpu_buffer; | |
444 | unsigned long head; | |
445 | struct buffer_page *head_page; | |
446 | u64 read_stamp; | |
447 | }; | |
448 | ||
f536aafc | 449 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
bf41a158 | 450 | #define RB_WARN_ON(buffer, cond) \ |
3e89c7bb SR |
451 | ({ \ |
452 | int _____ret = unlikely(cond); \ | |
453 | if (_____ret) { \ | |
bf41a158 SR |
454 | atomic_inc(&buffer->record_disabled); \ |
455 | WARN_ON(1); \ | |
456 | } \ | |
3e89c7bb SR |
457 | _____ret; \ |
458 | }) | |
f536aafc | 459 | |
37886f6a SR |
460 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
461 | #define DEBUG_SHIFT 0 | |
462 | ||
88eb0125 SR |
463 | static inline u64 rb_time_stamp(struct ring_buffer *buffer, int cpu) |
464 | { | |
465 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
466 | return buffer->clock() << DEBUG_SHIFT; | |
467 | } | |
468 | ||
37886f6a SR |
469 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) |
470 | { | |
471 | u64 time; | |
472 | ||
473 | preempt_disable_notrace(); | |
88eb0125 | 474 | time = rb_time_stamp(buffer, cpu); |
37886f6a SR |
475 | preempt_enable_no_resched_notrace(); |
476 | ||
477 | return time; | |
478 | } | |
479 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
480 | ||
481 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
482 | int cpu, u64 *ts) | |
483 | { | |
484 | /* Just stupid testing the normalize function and deltas */ | |
485 | *ts >>= DEBUG_SHIFT; | |
486 | } | |
487 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
488 | ||
7a8e76a3 SR |
489 | /** |
490 | * check_pages - integrity check of buffer pages | |
491 | * @cpu_buffer: CPU buffer with pages to test | |
492 | * | |
c3706f00 | 493 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
494 | * been corrupted. |
495 | */ | |
496 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
497 | { | |
498 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 499 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 500 | |
3e89c7bb SR |
501 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
502 | return -1; | |
503 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
504 | return -1; | |
7a8e76a3 | 505 | |
044fa782 | 506 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 507 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 508 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
509 | return -1; |
510 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 511 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 512 | return -1; |
7a8e76a3 SR |
513 | } |
514 | ||
515 | return 0; | |
516 | } | |
517 | ||
7a8e76a3 SR |
518 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
519 | unsigned nr_pages) | |
520 | { | |
521 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 522 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
523 | unsigned long addr; |
524 | LIST_HEAD(pages); | |
525 | unsigned i; | |
526 | ||
527 | for (i = 0; i < nr_pages; i++) { | |
044fa782 | 528 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
aa1e0e3b | 529 | GFP_KERNEL, cpu_to_node(cpu_buffer->cpu)); |
044fa782 | 530 | if (!bpage) |
e4c2ce82 | 531 | goto free_pages; |
044fa782 | 532 | list_add(&bpage->list, &pages); |
e4c2ce82 | 533 | |
7a8e76a3 SR |
534 | addr = __get_free_page(GFP_KERNEL); |
535 | if (!addr) | |
536 | goto free_pages; | |
044fa782 SR |
537 | bpage->page = (void *)addr; |
538 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
539 | } |
540 | ||
541 | list_splice(&pages, head); | |
542 | ||
543 | rb_check_pages(cpu_buffer); | |
544 | ||
545 | return 0; | |
546 | ||
547 | free_pages: | |
044fa782 SR |
548 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
549 | list_del_init(&bpage->list); | |
550 | free_buffer_page(bpage); | |
7a8e76a3 SR |
551 | } |
552 | return -ENOMEM; | |
553 | } | |
554 | ||
555 | static struct ring_buffer_per_cpu * | |
556 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) | |
557 | { | |
558 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 559 | struct buffer_page *bpage; |
d769041f | 560 | unsigned long addr; |
7a8e76a3 SR |
561 | int ret; |
562 | ||
563 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
564 | GFP_KERNEL, cpu_to_node(cpu)); | |
565 | if (!cpu_buffer) | |
566 | return NULL; | |
567 | ||
568 | cpu_buffer->cpu = cpu; | |
569 | cpu_buffer->buffer = buffer; | |
f83c9d0f | 570 | spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 571 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
3e03fb7f | 572 | cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; |
7a8e76a3 SR |
573 | INIT_LIST_HEAD(&cpu_buffer->pages); |
574 | ||
044fa782 | 575 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 576 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 577 | if (!bpage) |
e4c2ce82 SR |
578 | goto fail_free_buffer; |
579 | ||
044fa782 | 580 | cpu_buffer->reader_page = bpage; |
d769041f SR |
581 | addr = __get_free_page(GFP_KERNEL); |
582 | if (!addr) | |
e4c2ce82 | 583 | goto fail_free_reader; |
044fa782 SR |
584 | bpage->page = (void *)addr; |
585 | rb_init_page(bpage->page); | |
e4c2ce82 | 586 | |
d769041f | 587 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 588 | |
7a8e76a3 SR |
589 | ret = rb_allocate_pages(cpu_buffer, buffer->pages); |
590 | if (ret < 0) | |
d769041f | 591 | goto fail_free_reader; |
7a8e76a3 SR |
592 | |
593 | cpu_buffer->head_page | |
594 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 595 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 SR |
596 | |
597 | return cpu_buffer; | |
598 | ||
d769041f SR |
599 | fail_free_reader: |
600 | free_buffer_page(cpu_buffer->reader_page); | |
601 | ||
7a8e76a3 SR |
602 | fail_free_buffer: |
603 | kfree(cpu_buffer); | |
604 | return NULL; | |
605 | } | |
606 | ||
607 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
608 | { | |
609 | struct list_head *head = &cpu_buffer->pages; | |
044fa782 | 610 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 611 | |
d769041f SR |
612 | free_buffer_page(cpu_buffer->reader_page); |
613 | ||
044fa782 SR |
614 | list_for_each_entry_safe(bpage, tmp, head, list) { |
615 | list_del_init(&bpage->list); | |
616 | free_buffer_page(bpage); | |
7a8e76a3 SR |
617 | } |
618 | kfree(cpu_buffer); | |
619 | } | |
620 | ||
a7b13743 SR |
621 | /* |
622 | * Causes compile errors if the struct buffer_page gets bigger | |
623 | * than the struct page. | |
624 | */ | |
625 | extern int ring_buffer_page_too_big(void); | |
626 | ||
59222efe | 627 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
628 | static int rb_cpu_notify(struct notifier_block *self, |
629 | unsigned long action, void *hcpu); | |
554f786e SR |
630 | #endif |
631 | ||
7a8e76a3 SR |
632 | /** |
633 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 634 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
635 | * @flags: attributes to set for the ring buffer. |
636 | * | |
637 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
638 | * flag. This flag means that the buffer will overwrite old data | |
639 | * when the buffer wraps. If this flag is not set, the buffer will | |
640 | * drop data when the tail hits the head. | |
641 | */ | |
1f8a6a10 PZ |
642 | struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
643 | struct lock_class_key *key) | |
7a8e76a3 SR |
644 | { |
645 | struct ring_buffer *buffer; | |
646 | int bsize; | |
647 | int cpu; | |
648 | ||
a7b13743 SR |
649 | /* Paranoid! Optimizes out when all is well */ |
650 | if (sizeof(struct buffer_page) > sizeof(struct page)) | |
651 | ring_buffer_page_too_big(); | |
652 | ||
653 | ||
7a8e76a3 SR |
654 | /* keep it in its own cache line */ |
655 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
656 | GFP_KERNEL); | |
657 | if (!buffer) | |
658 | return NULL; | |
659 | ||
9e01c1b7 RR |
660 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
661 | goto fail_free_buffer; | |
662 | ||
7a8e76a3 SR |
663 | buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
664 | buffer->flags = flags; | |
37886f6a | 665 | buffer->clock = trace_clock_local; |
1f8a6a10 | 666 | buffer->reader_lock_key = key; |
7a8e76a3 SR |
667 | |
668 | /* need at least two pages */ | |
669 | if (buffer->pages == 1) | |
670 | buffer->pages++; | |
671 | ||
3bf832ce FW |
672 | /* |
673 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
674 | * in early initcall, it will not be notified of secondary cpus. | |
675 | * In that off case, we need to allocate for all possible cpus. | |
676 | */ | |
677 | #ifdef CONFIG_HOTPLUG_CPU | |
554f786e SR |
678 | get_online_cpus(); |
679 | cpumask_copy(buffer->cpumask, cpu_online_mask); | |
3bf832ce FW |
680 | #else |
681 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
682 | #endif | |
7a8e76a3 SR |
683 | buffer->cpus = nr_cpu_ids; |
684 | ||
685 | bsize = sizeof(void *) * nr_cpu_ids; | |
686 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
687 | GFP_KERNEL); | |
688 | if (!buffer->buffers) | |
9e01c1b7 | 689 | goto fail_free_cpumask; |
7a8e76a3 SR |
690 | |
691 | for_each_buffer_cpu(buffer, cpu) { | |
692 | buffer->buffers[cpu] = | |
693 | rb_allocate_cpu_buffer(buffer, cpu); | |
694 | if (!buffer->buffers[cpu]) | |
695 | goto fail_free_buffers; | |
696 | } | |
697 | ||
59222efe | 698 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
699 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
700 | buffer->cpu_notify.priority = 0; | |
701 | register_cpu_notifier(&buffer->cpu_notify); | |
702 | #endif | |
703 | ||
704 | put_online_cpus(); | |
7a8e76a3 SR |
705 | mutex_init(&buffer->mutex); |
706 | ||
707 | return buffer; | |
708 | ||
709 | fail_free_buffers: | |
710 | for_each_buffer_cpu(buffer, cpu) { | |
711 | if (buffer->buffers[cpu]) | |
712 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
713 | } | |
714 | kfree(buffer->buffers); | |
715 | ||
9e01c1b7 RR |
716 | fail_free_cpumask: |
717 | free_cpumask_var(buffer->cpumask); | |
554f786e | 718 | put_online_cpus(); |
9e01c1b7 | 719 | |
7a8e76a3 SR |
720 | fail_free_buffer: |
721 | kfree(buffer); | |
722 | return NULL; | |
723 | } | |
1f8a6a10 | 724 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
725 | |
726 | /** | |
727 | * ring_buffer_free - free a ring buffer. | |
728 | * @buffer: the buffer to free. | |
729 | */ | |
730 | void | |
731 | ring_buffer_free(struct ring_buffer *buffer) | |
732 | { | |
733 | int cpu; | |
734 | ||
554f786e SR |
735 | get_online_cpus(); |
736 | ||
59222efe | 737 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
738 | unregister_cpu_notifier(&buffer->cpu_notify); |
739 | #endif | |
740 | ||
7a8e76a3 SR |
741 | for_each_buffer_cpu(buffer, cpu) |
742 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
743 | ||
554f786e SR |
744 | put_online_cpus(); |
745 | ||
9e01c1b7 RR |
746 | free_cpumask_var(buffer->cpumask); |
747 | ||
7a8e76a3 SR |
748 | kfree(buffer); |
749 | } | |
c4f50183 | 750 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 751 | |
37886f6a SR |
752 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
753 | u64 (*clock)(void)) | |
754 | { | |
755 | buffer->clock = clock; | |
756 | } | |
757 | ||
7a8e76a3 SR |
758 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
759 | ||
760 | static void | |
761 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages) | |
762 | { | |
044fa782 | 763 | struct buffer_page *bpage; |
7a8e76a3 SR |
764 | struct list_head *p; |
765 | unsigned i; | |
766 | ||
767 | atomic_inc(&cpu_buffer->record_disabled); | |
768 | synchronize_sched(); | |
769 | ||
770 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
771 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
772 | return; | |
7a8e76a3 | 773 | p = cpu_buffer->pages.next; |
044fa782 SR |
774 | bpage = list_entry(p, struct buffer_page, list); |
775 | list_del_init(&bpage->list); | |
776 | free_buffer_page(bpage); | |
7a8e76a3 | 777 | } |
3e89c7bb SR |
778 | if (RB_WARN_ON(cpu_buffer, list_empty(&cpu_buffer->pages))) |
779 | return; | |
7a8e76a3 SR |
780 | |
781 | rb_reset_cpu(cpu_buffer); | |
782 | ||
783 | rb_check_pages(cpu_buffer); | |
784 | ||
785 | atomic_dec(&cpu_buffer->record_disabled); | |
786 | ||
787 | } | |
788 | ||
789 | static void | |
790 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
791 | struct list_head *pages, unsigned nr_pages) | |
792 | { | |
044fa782 | 793 | struct buffer_page *bpage; |
7a8e76a3 SR |
794 | struct list_head *p; |
795 | unsigned i; | |
796 | ||
797 | atomic_inc(&cpu_buffer->record_disabled); | |
798 | synchronize_sched(); | |
799 | ||
800 | for (i = 0; i < nr_pages; i++) { | |
3e89c7bb SR |
801 | if (RB_WARN_ON(cpu_buffer, list_empty(pages))) |
802 | return; | |
7a8e76a3 | 803 | p = pages->next; |
044fa782 SR |
804 | bpage = list_entry(p, struct buffer_page, list); |
805 | list_del_init(&bpage->list); | |
806 | list_add_tail(&bpage->list, &cpu_buffer->pages); | |
7a8e76a3 SR |
807 | } |
808 | rb_reset_cpu(cpu_buffer); | |
809 | ||
810 | rb_check_pages(cpu_buffer); | |
811 | ||
812 | atomic_dec(&cpu_buffer->record_disabled); | |
813 | } | |
814 | ||
815 | /** | |
816 | * ring_buffer_resize - resize the ring buffer | |
817 | * @buffer: the buffer to resize. | |
818 | * @size: the new size. | |
819 | * | |
820 | * The tracer is responsible for making sure that the buffer is | |
821 | * not being used while changing the size. | |
822 | * Note: We may be able to change the above requirement by using | |
823 | * RCU synchronizations. | |
824 | * | |
825 | * Minimum size is 2 * BUF_PAGE_SIZE. | |
826 | * | |
827 | * Returns -1 on failure. | |
828 | */ | |
829 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) | |
830 | { | |
831 | struct ring_buffer_per_cpu *cpu_buffer; | |
832 | unsigned nr_pages, rm_pages, new_pages; | |
044fa782 | 833 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
834 | unsigned long buffer_size; |
835 | unsigned long addr; | |
836 | LIST_HEAD(pages); | |
837 | int i, cpu; | |
838 | ||
ee51a1de IM |
839 | /* |
840 | * Always succeed at resizing a non-existent buffer: | |
841 | */ | |
842 | if (!buffer) | |
843 | return size; | |
844 | ||
7a8e76a3 SR |
845 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
846 | size *= BUF_PAGE_SIZE; | |
847 | buffer_size = buffer->pages * BUF_PAGE_SIZE; | |
848 | ||
849 | /* we need a minimum of two pages */ | |
850 | if (size < BUF_PAGE_SIZE * 2) | |
851 | size = BUF_PAGE_SIZE * 2; | |
852 | ||
853 | if (size == buffer_size) | |
854 | return size; | |
855 | ||
856 | mutex_lock(&buffer->mutex); | |
554f786e | 857 | get_online_cpus(); |
7a8e76a3 SR |
858 | |
859 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
860 | ||
861 | if (size < buffer_size) { | |
862 | ||
863 | /* easy case, just free pages */ | |
554f786e SR |
864 | if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) |
865 | goto out_fail; | |
7a8e76a3 SR |
866 | |
867 | rm_pages = buffer->pages - nr_pages; | |
868 | ||
869 | for_each_buffer_cpu(buffer, cpu) { | |
870 | cpu_buffer = buffer->buffers[cpu]; | |
871 | rb_remove_pages(cpu_buffer, rm_pages); | |
872 | } | |
873 | goto out; | |
874 | } | |
875 | ||
876 | /* | |
877 | * This is a bit more difficult. We only want to add pages | |
878 | * when we can allocate enough for all CPUs. We do this | |
879 | * by allocating all the pages and storing them on a local | |
880 | * link list. If we succeed in our allocation, then we | |
881 | * add these pages to the cpu_buffers. Otherwise we just free | |
882 | * them all and return -ENOMEM; | |
883 | */ | |
554f786e SR |
884 | if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) |
885 | goto out_fail; | |
f536aafc | 886 | |
7a8e76a3 SR |
887 | new_pages = nr_pages - buffer->pages; |
888 | ||
889 | for_each_buffer_cpu(buffer, cpu) { | |
890 | for (i = 0; i < new_pages; i++) { | |
044fa782 | 891 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), |
e4c2ce82 SR |
892 | cache_line_size()), |
893 | GFP_KERNEL, cpu_to_node(cpu)); | |
044fa782 | 894 | if (!bpage) |
e4c2ce82 | 895 | goto free_pages; |
044fa782 | 896 | list_add(&bpage->list, &pages); |
7a8e76a3 SR |
897 | addr = __get_free_page(GFP_KERNEL); |
898 | if (!addr) | |
899 | goto free_pages; | |
044fa782 SR |
900 | bpage->page = (void *)addr; |
901 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
902 | } |
903 | } | |
904 | ||
905 | for_each_buffer_cpu(buffer, cpu) { | |
906 | cpu_buffer = buffer->buffers[cpu]; | |
907 | rb_insert_pages(cpu_buffer, &pages, new_pages); | |
908 | } | |
909 | ||
554f786e SR |
910 | if (RB_WARN_ON(buffer, !list_empty(&pages))) |
911 | goto out_fail; | |
7a8e76a3 SR |
912 | |
913 | out: | |
914 | buffer->pages = nr_pages; | |
554f786e | 915 | put_online_cpus(); |
7a8e76a3 SR |
916 | mutex_unlock(&buffer->mutex); |
917 | ||
918 | return size; | |
919 | ||
920 | free_pages: | |
044fa782 SR |
921 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
922 | list_del_init(&bpage->list); | |
923 | free_buffer_page(bpage); | |
7a8e76a3 | 924 | } |
554f786e | 925 | put_online_cpus(); |
641d2f63 | 926 | mutex_unlock(&buffer->mutex); |
7a8e76a3 | 927 | return -ENOMEM; |
554f786e SR |
928 | |
929 | /* | |
930 | * Something went totally wrong, and we are too paranoid | |
931 | * to even clean up the mess. | |
932 | */ | |
933 | out_fail: | |
934 | put_online_cpus(); | |
935 | mutex_unlock(&buffer->mutex); | |
936 | return -1; | |
7a8e76a3 | 937 | } |
c4f50183 | 938 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 939 | |
8789a9e7 | 940 | static inline void * |
044fa782 | 941 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 942 | { |
044fa782 | 943 | return bpage->data + index; |
8789a9e7 SR |
944 | } |
945 | ||
044fa782 | 946 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 947 | { |
044fa782 | 948 | return bpage->page->data + index; |
7a8e76a3 SR |
949 | } |
950 | ||
951 | static inline struct ring_buffer_event * | |
d769041f | 952 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 953 | { |
6f807acd SR |
954 | return __rb_page_index(cpu_buffer->reader_page, |
955 | cpu_buffer->reader_page->read); | |
956 | } | |
957 | ||
958 | static inline struct ring_buffer_event * | |
959 | rb_head_event(struct ring_buffer_per_cpu *cpu_buffer) | |
960 | { | |
961 | return __rb_page_index(cpu_buffer->head_page, | |
962 | cpu_buffer->head_page->read); | |
7a8e76a3 SR |
963 | } |
964 | ||
965 | static inline struct ring_buffer_event * | |
966 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
967 | { | |
6f807acd | 968 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
969 | } |
970 | ||
bf41a158 SR |
971 | static inline unsigned rb_page_write(struct buffer_page *bpage) |
972 | { | |
973 | return local_read(&bpage->write); | |
974 | } | |
975 | ||
976 | static inline unsigned rb_page_commit(struct buffer_page *bpage) | |
977 | { | |
abc9b56d | 978 | return local_read(&bpage->page->commit); |
bf41a158 SR |
979 | } |
980 | ||
981 | /* Size is determined by what has been commited */ | |
982 | static inline unsigned rb_page_size(struct buffer_page *bpage) | |
983 | { | |
984 | return rb_page_commit(bpage); | |
985 | } | |
986 | ||
987 | static inline unsigned | |
988 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
989 | { | |
990 | return rb_page_commit(cpu_buffer->commit_page); | |
991 | } | |
992 | ||
993 | static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer) | |
994 | { | |
995 | return rb_page_commit(cpu_buffer->head_page); | |
996 | } | |
997 | ||
7a8e76a3 | 998 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, |
044fa782 | 999 | struct buffer_page **bpage) |
7a8e76a3 | 1000 | { |
044fa782 | 1001 | struct list_head *p = (*bpage)->list.next; |
7a8e76a3 SR |
1002 | |
1003 | if (p == &cpu_buffer->pages) | |
1004 | p = p->next; | |
1005 | ||
044fa782 | 1006 | *bpage = list_entry(p, struct buffer_page, list); |
7a8e76a3 SR |
1007 | } |
1008 | ||
bf41a158 SR |
1009 | static inline unsigned |
1010 | rb_event_index(struct ring_buffer_event *event) | |
1011 | { | |
1012 | unsigned long addr = (unsigned long)event; | |
1013 | ||
1014 | return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE); | |
1015 | } | |
1016 | ||
0f0c85fc | 1017 | static inline int |
bf41a158 SR |
1018 | rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1019 | struct ring_buffer_event *event) | |
1020 | { | |
1021 | unsigned long addr = (unsigned long)event; | |
1022 | unsigned long index; | |
1023 | ||
1024 | index = rb_event_index(event); | |
1025 | addr &= PAGE_MASK; | |
1026 | ||
1027 | return cpu_buffer->commit_page->page == (void *)addr && | |
1028 | rb_commit_index(cpu_buffer) == index; | |
1029 | } | |
1030 | ||
34a148bf | 1031 | static void |
bf41a158 SR |
1032 | rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer, |
1033 | struct ring_buffer_event *event) | |
7a8e76a3 | 1034 | { |
bf41a158 SR |
1035 | unsigned long addr = (unsigned long)event; |
1036 | unsigned long index; | |
1037 | ||
1038 | index = rb_event_index(event); | |
1039 | addr &= PAGE_MASK; | |
1040 | ||
1041 | while (cpu_buffer->commit_page->page != (void *)addr) { | |
3e89c7bb SR |
1042 | if (RB_WARN_ON(cpu_buffer, |
1043 | cpu_buffer->commit_page == cpu_buffer->tail_page)) | |
1044 | return; | |
abc9b56d | 1045 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1046 | cpu_buffer->commit_page->write; |
1047 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
1048 | cpu_buffer->write_stamp = |
1049 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1050 | } |
1051 | ||
1052 | /* Now set the commit to the event's index */ | |
abc9b56d | 1053 | local_set(&cpu_buffer->commit_page->page->commit, index); |
7a8e76a3 SR |
1054 | } |
1055 | ||
34a148bf | 1056 | static void |
bf41a158 | 1057 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1058 | { |
bf41a158 SR |
1059 | /* |
1060 | * We only race with interrupts and NMIs on this CPU. | |
1061 | * If we own the commit event, then we can commit | |
1062 | * all others that interrupted us, since the interruptions | |
1063 | * are in stack format (they finish before they come | |
1064 | * back to us). This allows us to do a simple loop to | |
1065 | * assign the commit to the tail. | |
1066 | */ | |
a8ccf1d6 | 1067 | again: |
bf41a158 | 1068 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
abc9b56d | 1069 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1070 | cpu_buffer->commit_page->write; |
1071 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
abc9b56d SR |
1072 | cpu_buffer->write_stamp = |
1073 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1074 | /* add barrier to keep gcc from optimizing too much */ |
1075 | barrier(); | |
1076 | } | |
1077 | while (rb_commit_index(cpu_buffer) != | |
1078 | rb_page_write(cpu_buffer->commit_page)) { | |
abc9b56d | 1079 | cpu_buffer->commit_page->page->commit = |
bf41a158 SR |
1080 | cpu_buffer->commit_page->write; |
1081 | barrier(); | |
1082 | } | |
a8ccf1d6 SR |
1083 | |
1084 | /* again, keep gcc from optimizing */ | |
1085 | barrier(); | |
1086 | ||
1087 | /* | |
1088 | * If an interrupt came in just after the first while loop | |
1089 | * and pushed the tail page forward, we will be left with | |
1090 | * a dangling commit that will never go forward. | |
1091 | */ | |
1092 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1093 | goto again; | |
7a8e76a3 SR |
1094 | } |
1095 | ||
d769041f | 1096 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1097 | { |
abc9b56d | 1098 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1099 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1100 | } |
1101 | ||
34a148bf | 1102 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1103 | { |
1104 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1105 | ||
1106 | /* | |
1107 | * The iterator could be on the reader page (it starts there). | |
1108 | * But the head could have moved, since the reader was | |
1109 | * found. Check for this case and assign the iterator | |
1110 | * to the head page instead of next. | |
1111 | */ | |
1112 | if (iter->head_page == cpu_buffer->reader_page) | |
1113 | iter->head_page = cpu_buffer->head_page; | |
1114 | else | |
1115 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1116 | ||
abc9b56d | 1117 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1118 | iter->head = 0; |
1119 | } | |
1120 | ||
1121 | /** | |
1122 | * ring_buffer_update_event - update event type and data | |
1123 | * @event: the even to update | |
1124 | * @type: the type of event | |
1125 | * @length: the size of the event field in the ring buffer | |
1126 | * | |
1127 | * Update the type and data fields of the event. The length | |
1128 | * is the actual size that is written to the ring buffer, | |
1129 | * and with this, we can determine what to place into the | |
1130 | * data field. | |
1131 | */ | |
34a148bf | 1132 | static void |
7a8e76a3 SR |
1133 | rb_update_event(struct ring_buffer_event *event, |
1134 | unsigned type, unsigned length) | |
1135 | { | |
334d4169 | 1136 | event->type_len = type; |
7a8e76a3 SR |
1137 | |
1138 | switch (type) { | |
1139 | ||
1140 | case RINGBUF_TYPE_PADDING: | |
7a8e76a3 | 1141 | case RINGBUF_TYPE_TIME_EXTEND: |
7a8e76a3 | 1142 | case RINGBUF_TYPE_TIME_STAMP: |
7a8e76a3 SR |
1143 | break; |
1144 | ||
334d4169 | 1145 | case 0: |
7a8e76a3 | 1146 | length -= RB_EVNT_HDR_SIZE; |
334d4169 | 1147 | if (length > RB_MAX_SMALL_DATA) |
7a8e76a3 | 1148 | event->array[0] = length; |
334d4169 LJ |
1149 | else |
1150 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
7a8e76a3 SR |
1151 | break; |
1152 | default: | |
1153 | BUG(); | |
1154 | } | |
1155 | } | |
1156 | ||
34a148bf | 1157 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
1158 | { |
1159 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
1160 | ||
1161 | /* zero length can cause confusions */ | |
1162 | if (!length) | |
1163 | length = 1; | |
1164 | ||
1165 | if (length > RB_MAX_SMALL_DATA) | |
1166 | length += sizeof(event.array[0]); | |
1167 | ||
1168 | length += RB_EVNT_HDR_SIZE; | |
1169 | length = ALIGN(length, RB_ALIGNMENT); | |
1170 | ||
1171 | return length; | |
1172 | } | |
1173 | ||
c7b09308 SR |
1174 | static inline void |
1175 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
1176 | struct buffer_page *tail_page, | |
1177 | unsigned long tail, unsigned long length) | |
1178 | { | |
1179 | struct ring_buffer_event *event; | |
1180 | ||
1181 | /* | |
1182 | * Only the event that crossed the page boundary | |
1183 | * must fill the old tail_page with padding. | |
1184 | */ | |
1185 | if (tail >= BUF_PAGE_SIZE) { | |
1186 | local_sub(length, &tail_page->write); | |
1187 | return; | |
1188 | } | |
1189 | ||
1190 | event = __rb_page_index(tail_page, tail); | |
1191 | ||
1192 | /* | |
1193 | * If this event is bigger than the minimum size, then | |
1194 | * we need to be careful that we don't subtract the | |
1195 | * write counter enough to allow another writer to slip | |
1196 | * in on this page. | |
1197 | * We put in a discarded commit instead, to make sure | |
1198 | * that this space is not used again. | |
1199 | * | |
1200 | * If we are less than the minimum size, we don't need to | |
1201 | * worry about it. | |
1202 | */ | |
1203 | if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) { | |
1204 | /* No room for any events */ | |
1205 | ||
1206 | /* Mark the rest of the page with padding */ | |
1207 | rb_event_set_padding(event); | |
1208 | ||
1209 | /* Set the write back to the previous setting */ | |
1210 | local_sub(length, &tail_page->write); | |
1211 | return; | |
1212 | } | |
1213 | ||
1214 | /* Put in a discarded event */ | |
1215 | event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE; | |
1216 | event->type_len = RINGBUF_TYPE_PADDING; | |
1217 | /* time delta must be non zero */ | |
1218 | event->time_delta = 1; | |
1219 | /* Account for this as an entry */ | |
1220 | local_inc(&tail_page->entries); | |
1221 | local_inc(&cpu_buffer->entries); | |
1222 | ||
1223 | /* Set write to end of buffer */ | |
1224 | length = (tail + length) - BUF_PAGE_SIZE; | |
1225 | local_sub(length, &tail_page->write); | |
1226 | } | |
6634ff26 | 1227 | |
7a8e76a3 | 1228 | static struct ring_buffer_event * |
6634ff26 SR |
1229 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
1230 | unsigned long length, unsigned long tail, | |
1231 | struct buffer_page *commit_page, | |
1232 | struct buffer_page *tail_page, u64 *ts) | |
7a8e76a3 | 1233 | { |
6634ff26 | 1234 | struct buffer_page *next_page, *head_page, *reader_page; |
7a8e76a3 | 1235 | struct ring_buffer *buffer = cpu_buffer->buffer; |
78d904b4 | 1236 | bool lock_taken = false; |
6634ff26 | 1237 | unsigned long flags; |
aa20ae84 SR |
1238 | |
1239 | next_page = tail_page; | |
1240 | ||
1241 | local_irq_save(flags); | |
1242 | /* | |
1243 | * Since the write to the buffer is still not | |
1244 | * fully lockless, we must be careful with NMIs. | |
1245 | * The locks in the writers are taken when a write | |
1246 | * crosses to a new page. The locks protect against | |
1247 | * races with the readers (this will soon be fixed | |
1248 | * with a lockless solution). | |
1249 | * | |
1250 | * Because we can not protect against NMIs, and we | |
1251 | * want to keep traces reentrant, we need to manage | |
1252 | * what happens when we are in an NMI. | |
1253 | * | |
1254 | * NMIs can happen after we take the lock. | |
1255 | * If we are in an NMI, only take the lock | |
1256 | * if it is not already taken. Otherwise | |
1257 | * simply fail. | |
1258 | */ | |
1259 | if (unlikely(in_nmi())) { | |
1260 | if (!__raw_spin_trylock(&cpu_buffer->lock)) { | |
1261 | cpu_buffer->nmi_dropped++; | |
1262 | goto out_reset; | |
1263 | } | |
1264 | } else | |
1265 | __raw_spin_lock(&cpu_buffer->lock); | |
1266 | ||
1267 | lock_taken = true; | |
1268 | ||
1269 | rb_inc_page(cpu_buffer, &next_page); | |
1270 | ||
1271 | head_page = cpu_buffer->head_page; | |
1272 | reader_page = cpu_buffer->reader_page; | |
1273 | ||
1274 | /* we grabbed the lock before incrementing */ | |
1275 | if (RB_WARN_ON(cpu_buffer, next_page == reader_page)) | |
1276 | goto out_reset; | |
1277 | ||
1278 | /* | |
1279 | * If for some reason, we had an interrupt storm that made | |
1280 | * it all the way around the buffer, bail, and warn | |
1281 | * about it. | |
1282 | */ | |
1283 | if (unlikely(next_page == commit_page)) { | |
1284 | cpu_buffer->commit_overrun++; | |
1285 | goto out_reset; | |
1286 | } | |
1287 | ||
1288 | if (next_page == head_page) { | |
1289 | if (!(buffer->flags & RB_FL_OVERWRITE)) | |
1290 | goto out_reset; | |
1291 | ||
1292 | /* tail_page has not moved yet? */ | |
1293 | if (tail_page == cpu_buffer->tail_page) { | |
1294 | /* count overflows */ | |
1295 | cpu_buffer->overrun += | |
1296 | local_read(&head_page->entries); | |
1297 | ||
1298 | rb_inc_page(cpu_buffer, &head_page); | |
1299 | cpu_buffer->head_page = head_page; | |
1300 | cpu_buffer->head_page->read = 0; | |
1301 | } | |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * If the tail page is still the same as what we think | |
1306 | * it is, then it is up to us to update the tail | |
1307 | * pointer. | |
1308 | */ | |
1309 | if (tail_page == cpu_buffer->tail_page) { | |
1310 | local_set(&next_page->write, 0); | |
1311 | local_set(&next_page->entries, 0); | |
1312 | local_set(&next_page->page->commit, 0); | |
1313 | cpu_buffer->tail_page = next_page; | |
1314 | ||
1315 | /* reread the time stamp */ | |
88eb0125 | 1316 | *ts = rb_time_stamp(buffer, cpu_buffer->cpu); |
aa20ae84 SR |
1317 | cpu_buffer->tail_page->page->time_stamp = *ts; |
1318 | } | |
1319 | ||
c7b09308 | 1320 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
aa20ae84 SR |
1321 | |
1322 | /* | |
1323 | * If this was a commit entry that failed, | |
1324 | * increment that too | |
1325 | */ | |
1326 | if (tail_page == cpu_buffer->commit_page && | |
1327 | tail == rb_commit_index(cpu_buffer)) { | |
1328 | rb_set_commit_to_write(cpu_buffer); | |
1329 | } | |
1330 | ||
1331 | __raw_spin_unlock(&cpu_buffer->lock); | |
1332 | local_irq_restore(flags); | |
1333 | ||
1334 | /* fail and let the caller try again */ | |
1335 | return ERR_PTR(-EAGAIN); | |
1336 | ||
45141d46 | 1337 | out_reset: |
6f3b3440 | 1338 | /* reset write */ |
c7b09308 | 1339 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
6f3b3440 | 1340 | |
78d904b4 SR |
1341 | if (likely(lock_taken)) |
1342 | __raw_spin_unlock(&cpu_buffer->lock); | |
3e03fb7f | 1343 | local_irq_restore(flags); |
bf41a158 | 1344 | return NULL; |
7a8e76a3 SR |
1345 | } |
1346 | ||
6634ff26 SR |
1347 | static struct ring_buffer_event * |
1348 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
1349 | unsigned type, unsigned long length, u64 *ts) | |
1350 | { | |
1351 | struct buffer_page *tail_page, *commit_page; | |
1352 | struct ring_buffer_event *event; | |
1353 | unsigned long tail, write; | |
1354 | ||
1355 | commit_page = cpu_buffer->commit_page; | |
1356 | /* we just need to protect against interrupts */ | |
1357 | barrier(); | |
1358 | tail_page = cpu_buffer->tail_page; | |
1359 | write = local_add_return(length, &tail_page->write); | |
1360 | tail = write - length; | |
1361 | ||
1362 | /* See if we shot pass the end of this buffer page */ | |
1363 | if (write > BUF_PAGE_SIZE) | |
1364 | return rb_move_tail(cpu_buffer, length, tail, | |
1365 | commit_page, tail_page, ts); | |
1366 | ||
1367 | /* We reserved something on the buffer */ | |
1368 | ||
1369 | if (RB_WARN_ON(cpu_buffer, write > BUF_PAGE_SIZE)) | |
1370 | return NULL; | |
1371 | ||
1372 | event = __rb_page_index(tail_page, tail); | |
1373 | rb_update_event(event, type, length); | |
1374 | ||
1375 | /* The passed in type is zero for DATA */ | |
1376 | if (likely(!type)) | |
1377 | local_inc(&tail_page->entries); | |
1378 | ||
1379 | /* | |
1380 | * If this is a commit and the tail is zero, then update | |
1381 | * this page's time stamp. | |
1382 | */ | |
1383 | if (!tail && rb_is_commit(cpu_buffer, event)) | |
1384 | cpu_buffer->commit_page->page->time_stamp = *ts; | |
1385 | ||
1386 | return event; | |
1387 | } | |
1388 | ||
edd813bf SR |
1389 | static inline int |
1390 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
1391 | struct ring_buffer_event *event) | |
1392 | { | |
1393 | unsigned long new_index, old_index; | |
1394 | struct buffer_page *bpage; | |
1395 | unsigned long index; | |
1396 | unsigned long addr; | |
1397 | ||
1398 | new_index = rb_event_index(event); | |
1399 | old_index = new_index + rb_event_length(event); | |
1400 | addr = (unsigned long)event; | |
1401 | addr &= PAGE_MASK; | |
1402 | ||
1403 | bpage = cpu_buffer->tail_page; | |
1404 | ||
1405 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { | |
1406 | /* | |
1407 | * This is on the tail page. It is possible that | |
1408 | * a write could come in and move the tail page | |
1409 | * and write to the next page. That is fine | |
1410 | * because we just shorten what is on this page. | |
1411 | */ | |
1412 | index = local_cmpxchg(&bpage->write, old_index, new_index); | |
1413 | if (index == old_index) | |
1414 | return 1; | |
1415 | } | |
1416 | ||
1417 | /* could not discard */ | |
1418 | return 0; | |
1419 | } | |
1420 | ||
7a8e76a3 SR |
1421 | static int |
1422 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1423 | u64 *ts, u64 *delta) | |
1424 | { | |
1425 | struct ring_buffer_event *event; | |
1426 | static int once; | |
bf41a158 | 1427 | int ret; |
7a8e76a3 SR |
1428 | |
1429 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1430 | printk(KERN_WARNING "Delta way too big! %llu" | |
1431 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1432 | (unsigned long long)*delta, |
1433 | (unsigned long long)*ts, | |
1434 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1435 | WARN_ON(1); |
1436 | } | |
1437 | ||
1438 | /* | |
1439 | * The delta is too big, we to add a | |
1440 | * new timestamp. | |
1441 | */ | |
1442 | event = __rb_reserve_next(cpu_buffer, | |
1443 | RINGBUF_TYPE_TIME_EXTEND, | |
1444 | RB_LEN_TIME_EXTEND, | |
1445 | ts); | |
1446 | if (!event) | |
bf41a158 | 1447 | return -EBUSY; |
7a8e76a3 | 1448 | |
bf41a158 SR |
1449 | if (PTR_ERR(event) == -EAGAIN) |
1450 | return -EAGAIN; | |
1451 | ||
1452 | /* Only a commited time event can update the write stamp */ | |
1453 | if (rb_is_commit(cpu_buffer, event)) { | |
1454 | /* | |
1455 | * If this is the first on the page, then we need to | |
1456 | * update the page itself, and just put in a zero. | |
1457 | */ | |
1458 | if (rb_event_index(event)) { | |
1459 | event->time_delta = *delta & TS_MASK; | |
1460 | event->array[0] = *delta >> TS_SHIFT; | |
1461 | } else { | |
abc9b56d | 1462 | cpu_buffer->commit_page->page->time_stamp = *ts; |
ea05b57c SR |
1463 | /* try to discard, since we do not need this */ |
1464 | if (!rb_try_to_discard(cpu_buffer, event)) { | |
1465 | /* nope, just zero it */ | |
1466 | event->time_delta = 0; | |
1467 | event->array[0] = 0; | |
1468 | } | |
bf41a158 | 1469 | } |
7a8e76a3 | 1470 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
1471 | /* let the caller know this was the commit */ |
1472 | ret = 1; | |
1473 | } else { | |
edd813bf SR |
1474 | /* Try to discard the event */ |
1475 | if (!rb_try_to_discard(cpu_buffer, event)) { | |
1476 | /* Darn, this is just wasted space */ | |
1477 | event->time_delta = 0; | |
1478 | event->array[0] = 0; | |
edd813bf | 1479 | } |
f57a8a19 | 1480 | ret = 0; |
7a8e76a3 SR |
1481 | } |
1482 | ||
bf41a158 SR |
1483 | *delta = 0; |
1484 | ||
1485 | return ret; | |
7a8e76a3 SR |
1486 | } |
1487 | ||
1488 | static struct ring_buffer_event * | |
1489 | rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 1490 | unsigned long length) |
7a8e76a3 SR |
1491 | { |
1492 | struct ring_buffer_event *event; | |
168b6b1d | 1493 | u64 ts, delta = 0; |
bf41a158 | 1494 | int commit = 0; |
818e3dd3 | 1495 | int nr_loops = 0; |
7a8e76a3 | 1496 | |
be957c44 | 1497 | length = rb_calculate_event_length(length); |
bf41a158 | 1498 | again: |
818e3dd3 SR |
1499 | /* |
1500 | * We allow for interrupts to reenter here and do a trace. | |
1501 | * If one does, it will cause this original code to loop | |
1502 | * back here. Even with heavy interrupts happening, this | |
1503 | * should only happen a few times in a row. If this happens | |
1504 | * 1000 times in a row, there must be either an interrupt | |
1505 | * storm or we have something buggy. | |
1506 | * Bail! | |
1507 | */ | |
3e89c7bb | 1508 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
818e3dd3 | 1509 | return NULL; |
818e3dd3 | 1510 | |
88eb0125 | 1511 | ts = rb_time_stamp(cpu_buffer->buffer, cpu_buffer->cpu); |
7a8e76a3 | 1512 | |
bf41a158 SR |
1513 | /* |
1514 | * Only the first commit can update the timestamp. | |
1515 | * Yes there is a race here. If an interrupt comes in | |
1516 | * just after the conditional and it traces too, then it | |
1517 | * will also check the deltas. More than one timestamp may | |
1518 | * also be made. But only the entry that did the actual | |
1519 | * commit will be something other than zero. | |
1520 | */ | |
0f0c85fc SR |
1521 | if (likely(cpu_buffer->tail_page == cpu_buffer->commit_page && |
1522 | rb_page_write(cpu_buffer->tail_page) == | |
1523 | rb_commit_index(cpu_buffer))) { | |
168b6b1d | 1524 | u64 diff; |
bf41a158 | 1525 | |
168b6b1d | 1526 | diff = ts - cpu_buffer->write_stamp; |
7a8e76a3 | 1527 | |
168b6b1d | 1528 | /* make sure this diff is calculated here */ |
bf41a158 SR |
1529 | barrier(); |
1530 | ||
1531 | /* Did the write stamp get updated already? */ | |
1532 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
168b6b1d | 1533 | goto get_event; |
bf41a158 | 1534 | |
168b6b1d SR |
1535 | delta = diff; |
1536 | if (unlikely(test_time_stamp(delta))) { | |
7a8e76a3 | 1537 | |
bf41a158 | 1538 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
bf41a158 | 1539 | if (commit == -EBUSY) |
7a8e76a3 | 1540 | return NULL; |
bf41a158 SR |
1541 | |
1542 | if (commit == -EAGAIN) | |
1543 | goto again; | |
1544 | ||
1545 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 1546 | } |
168b6b1d | 1547 | } |
7a8e76a3 | 1548 | |
168b6b1d | 1549 | get_event: |
1cd8d735 | 1550 | event = __rb_reserve_next(cpu_buffer, 0, length, &ts); |
168b6b1d | 1551 | if (unlikely(PTR_ERR(event) == -EAGAIN)) |
bf41a158 SR |
1552 | goto again; |
1553 | ||
1554 | if (!event) { | |
1555 | if (unlikely(commit)) | |
1556 | /* | |
1557 | * Ouch! We needed a timestamp and it was commited. But | |
1558 | * we didn't get our event reserved. | |
1559 | */ | |
1560 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 | 1561 | return NULL; |
bf41a158 | 1562 | } |
7a8e76a3 | 1563 | |
bf41a158 SR |
1564 | /* |
1565 | * If the timestamp was commited, make the commit our entry | |
1566 | * now so that we will update it when needed. | |
1567 | */ | |
0f0c85fc | 1568 | if (unlikely(commit)) |
bf41a158 SR |
1569 | rb_set_commit_event(cpu_buffer, event); |
1570 | else if (!rb_is_commit(cpu_buffer, event)) | |
7a8e76a3 SR |
1571 | delta = 0; |
1572 | ||
1573 | event->time_delta = delta; | |
1574 | ||
1575 | return event; | |
1576 | } | |
1577 | ||
aa18efb2 | 1578 | #define TRACE_RECURSIVE_DEPTH 16 |
261842b7 SR |
1579 | |
1580 | static int trace_recursive_lock(void) | |
1581 | { | |
aa18efb2 | 1582 | current->trace_recursion++; |
261842b7 | 1583 | |
aa18efb2 SR |
1584 | if (likely(current->trace_recursion < TRACE_RECURSIVE_DEPTH)) |
1585 | return 0; | |
e057a5e5 | 1586 | |
aa18efb2 SR |
1587 | /* Disable all tracing before we do anything else */ |
1588 | tracing_off_permanent(); | |
261842b7 | 1589 | |
7d7d2b80 | 1590 | printk_once(KERN_WARNING "Tracing recursion: depth[%ld]:" |
aa18efb2 SR |
1591 | "HC[%lu]:SC[%lu]:NMI[%lu]\n", |
1592 | current->trace_recursion, | |
1593 | hardirq_count() >> HARDIRQ_SHIFT, | |
1594 | softirq_count() >> SOFTIRQ_SHIFT, | |
1595 | in_nmi()); | |
261842b7 | 1596 | |
aa18efb2 SR |
1597 | WARN_ON_ONCE(1); |
1598 | return -1; | |
261842b7 SR |
1599 | } |
1600 | ||
1601 | static void trace_recursive_unlock(void) | |
1602 | { | |
aa18efb2 | 1603 | WARN_ON_ONCE(!current->trace_recursion); |
261842b7 | 1604 | |
aa18efb2 | 1605 | current->trace_recursion--; |
261842b7 SR |
1606 | } |
1607 | ||
bf41a158 SR |
1608 | static DEFINE_PER_CPU(int, rb_need_resched); |
1609 | ||
7a8e76a3 SR |
1610 | /** |
1611 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
1612 | * @buffer: the ring buffer to reserve from | |
1613 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
1614 | * |
1615 | * Returns a reseverd event on the ring buffer to copy directly to. | |
1616 | * The user of this interface will need to get the body to write into | |
1617 | * and can use the ring_buffer_event_data() interface. | |
1618 | * | |
1619 | * The length is the length of the data needed, not the event length | |
1620 | * which also includes the event header. | |
1621 | * | |
1622 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
1623 | * If NULL is returned, then nothing has been allocated or locked. | |
1624 | */ | |
1625 | struct ring_buffer_event * | |
0a987751 | 1626 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
1627 | { |
1628 | struct ring_buffer_per_cpu *cpu_buffer; | |
1629 | struct ring_buffer_event *event; | |
bf41a158 | 1630 | int cpu, resched; |
7a8e76a3 | 1631 | |
033601a3 | 1632 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1633 | return NULL; |
1634 | ||
7a8e76a3 SR |
1635 | if (atomic_read(&buffer->record_disabled)) |
1636 | return NULL; | |
1637 | ||
bf41a158 | 1638 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 1639 | resched = ftrace_preempt_disable(); |
bf41a158 | 1640 | |
261842b7 SR |
1641 | if (trace_recursive_lock()) |
1642 | goto out_nocheck; | |
1643 | ||
7a8e76a3 SR |
1644 | cpu = raw_smp_processor_id(); |
1645 | ||
9e01c1b7 | 1646 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1647 | goto out; |
7a8e76a3 SR |
1648 | |
1649 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1650 | |
1651 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 1652 | goto out; |
7a8e76a3 | 1653 | |
be957c44 | 1654 | if (length > BUF_MAX_DATA_SIZE) |
bf41a158 | 1655 | goto out; |
7a8e76a3 | 1656 | |
1cd8d735 | 1657 | event = rb_reserve_next_event(cpu_buffer, length); |
7a8e76a3 | 1658 | if (!event) |
d769041f | 1659 | goto out; |
7a8e76a3 | 1660 | |
bf41a158 SR |
1661 | /* |
1662 | * Need to store resched state on this cpu. | |
1663 | * Only the first needs to. | |
1664 | */ | |
1665 | ||
1666 | if (preempt_count() == 1) | |
1667 | per_cpu(rb_need_resched, cpu) = resched; | |
1668 | ||
7a8e76a3 SR |
1669 | return event; |
1670 | ||
d769041f | 1671 | out: |
261842b7 SR |
1672 | trace_recursive_unlock(); |
1673 | ||
1674 | out_nocheck: | |
182e9f5f | 1675 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1676 | return NULL; |
1677 | } | |
c4f50183 | 1678 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 SR |
1679 | |
1680 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
1681 | struct ring_buffer_event *event) | |
1682 | { | |
e4906eff | 1683 | local_inc(&cpu_buffer->entries); |
bf41a158 SR |
1684 | |
1685 | /* Only process further if we own the commit */ | |
1686 | if (!rb_is_commit(cpu_buffer, event)) | |
1687 | return; | |
1688 | ||
1689 | cpu_buffer->write_stamp += event->time_delta; | |
1690 | ||
1691 | rb_set_commit_to_write(cpu_buffer); | |
7a8e76a3 SR |
1692 | } |
1693 | ||
1694 | /** | |
1695 | * ring_buffer_unlock_commit - commit a reserved | |
1696 | * @buffer: The buffer to commit to | |
1697 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
1698 | * |
1699 | * This commits the data to the ring buffer, and releases any locks held. | |
1700 | * | |
1701 | * Must be paired with ring_buffer_lock_reserve. | |
1702 | */ | |
1703 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 1704 | struct ring_buffer_event *event) |
7a8e76a3 SR |
1705 | { |
1706 | struct ring_buffer_per_cpu *cpu_buffer; | |
1707 | int cpu = raw_smp_processor_id(); | |
1708 | ||
1709 | cpu_buffer = buffer->buffers[cpu]; | |
1710 | ||
7a8e76a3 SR |
1711 | rb_commit(cpu_buffer, event); |
1712 | ||
261842b7 SR |
1713 | trace_recursive_unlock(); |
1714 | ||
bf41a158 SR |
1715 | /* |
1716 | * Only the last preempt count needs to restore preemption. | |
1717 | */ | |
182e9f5f SR |
1718 | if (preempt_count() == 1) |
1719 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1720 | else | |
bf41a158 | 1721 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
1722 | |
1723 | return 0; | |
1724 | } | |
c4f50183 | 1725 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 1726 | |
f3b9aae1 FW |
1727 | static inline void rb_event_discard(struct ring_buffer_event *event) |
1728 | { | |
334d4169 LJ |
1729 | /* array[0] holds the actual length for the discarded event */ |
1730 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
1731 | event->type_len = RINGBUF_TYPE_PADDING; | |
f3b9aae1 FW |
1732 | /* time delta must be non zero */ |
1733 | if (!event->time_delta) | |
1734 | event->time_delta = 1; | |
1735 | } | |
1736 | ||
fa1b47dd SR |
1737 | /** |
1738 | * ring_buffer_event_discard - discard any event in the ring buffer | |
1739 | * @event: the event to discard | |
1740 | * | |
1741 | * Sometimes a event that is in the ring buffer needs to be ignored. | |
1742 | * This function lets the user discard an event in the ring buffer | |
1743 | * and then that event will not be read later. | |
1744 | * | |
1745 | * Note, it is up to the user to be careful with this, and protect | |
1746 | * against races. If the user discards an event that has been consumed | |
1747 | * it is possible that it could corrupt the ring buffer. | |
1748 | */ | |
1749 | void ring_buffer_event_discard(struct ring_buffer_event *event) | |
1750 | { | |
f3b9aae1 | 1751 | rb_event_discard(event); |
fa1b47dd SR |
1752 | } |
1753 | EXPORT_SYMBOL_GPL(ring_buffer_event_discard); | |
1754 | ||
1755 | /** | |
1756 | * ring_buffer_commit_discard - discard an event that has not been committed | |
1757 | * @buffer: the ring buffer | |
1758 | * @event: non committed event to discard | |
1759 | * | |
1760 | * This is similar to ring_buffer_event_discard but must only be | |
1761 | * performed on an event that has not been committed yet. The difference | |
1762 | * is that this will also try to free the event from the ring buffer | |
1763 | * if another event has not been added behind it. | |
1764 | * | |
1765 | * If another event has been added behind it, it will set the event | |
1766 | * up as discarded, and perform the commit. | |
1767 | * | |
1768 | * If this function is called, do not call ring_buffer_unlock_commit on | |
1769 | * the event. | |
1770 | */ | |
1771 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
1772 | struct ring_buffer_event *event) | |
1773 | { | |
1774 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
1775 | int cpu; |
1776 | ||
1777 | /* The event is discarded regardless */ | |
f3b9aae1 | 1778 | rb_event_discard(event); |
fa1b47dd SR |
1779 | |
1780 | /* | |
1781 | * This must only be called if the event has not been | |
1782 | * committed yet. Thus we can assume that preemption | |
1783 | * is still disabled. | |
1784 | */ | |
74f4fd21 | 1785 | RB_WARN_ON(buffer, preemptible()); |
fa1b47dd SR |
1786 | |
1787 | cpu = smp_processor_id(); | |
1788 | cpu_buffer = buffer->buffers[cpu]; | |
1789 | ||
edd813bf SR |
1790 | if (!rb_try_to_discard(cpu_buffer, event)) |
1791 | goto out; | |
fa1b47dd SR |
1792 | |
1793 | /* | |
1794 | * The commit is still visible by the reader, so we | |
1795 | * must increment entries. | |
1796 | */ | |
e4906eff | 1797 | local_inc(&cpu_buffer->entries); |
fa1b47dd SR |
1798 | out: |
1799 | /* | |
1800 | * If a write came in and pushed the tail page | |
1801 | * we still need to update the commit pointer | |
1802 | * if we were the commit. | |
1803 | */ | |
1804 | if (rb_is_commit(cpu_buffer, event)) | |
1805 | rb_set_commit_to_write(cpu_buffer); | |
1806 | ||
f3b9aae1 FW |
1807 | trace_recursive_unlock(); |
1808 | ||
fa1b47dd SR |
1809 | /* |
1810 | * Only the last preempt count needs to restore preemption. | |
1811 | */ | |
1812 | if (preempt_count() == 1) | |
1813 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
1814 | else | |
1815 | preempt_enable_no_resched_notrace(); | |
1816 | ||
1817 | } | |
1818 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
1819 | ||
7a8e76a3 SR |
1820 | /** |
1821 | * ring_buffer_write - write data to the buffer without reserving | |
1822 | * @buffer: The ring buffer to write to. | |
1823 | * @length: The length of the data being written (excluding the event header) | |
1824 | * @data: The data to write to the buffer. | |
1825 | * | |
1826 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
1827 | * one function. If you already have the data to write to the buffer, it | |
1828 | * may be easier to simply call this function. | |
1829 | * | |
1830 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
1831 | * and not the length of the event which would hold the header. | |
1832 | */ | |
1833 | int ring_buffer_write(struct ring_buffer *buffer, | |
1834 | unsigned long length, | |
1835 | void *data) | |
1836 | { | |
1837 | struct ring_buffer_per_cpu *cpu_buffer; | |
1838 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
1839 | void *body; |
1840 | int ret = -EBUSY; | |
bf41a158 | 1841 | int cpu, resched; |
7a8e76a3 | 1842 | |
033601a3 | 1843 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
1844 | return -EBUSY; |
1845 | ||
7a8e76a3 SR |
1846 | if (atomic_read(&buffer->record_disabled)) |
1847 | return -EBUSY; | |
1848 | ||
182e9f5f | 1849 | resched = ftrace_preempt_disable(); |
bf41a158 | 1850 | |
7a8e76a3 SR |
1851 | cpu = raw_smp_processor_id(); |
1852 | ||
9e01c1b7 | 1853 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 1854 | goto out; |
7a8e76a3 SR |
1855 | |
1856 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
1857 | |
1858 | if (atomic_read(&cpu_buffer->record_disabled)) | |
1859 | goto out; | |
1860 | ||
be957c44 SR |
1861 | if (length > BUF_MAX_DATA_SIZE) |
1862 | goto out; | |
1863 | ||
1864 | event = rb_reserve_next_event(cpu_buffer, length); | |
7a8e76a3 SR |
1865 | if (!event) |
1866 | goto out; | |
1867 | ||
1868 | body = rb_event_data(event); | |
1869 | ||
1870 | memcpy(body, data, length); | |
1871 | ||
1872 | rb_commit(cpu_buffer, event); | |
1873 | ||
1874 | ret = 0; | |
1875 | out: | |
182e9f5f | 1876 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
1877 | |
1878 | return ret; | |
1879 | } | |
c4f50183 | 1880 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 1881 | |
34a148bf | 1882 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
1883 | { |
1884 | struct buffer_page *reader = cpu_buffer->reader_page; | |
1885 | struct buffer_page *head = cpu_buffer->head_page; | |
1886 | struct buffer_page *commit = cpu_buffer->commit_page; | |
1887 | ||
1888 | return reader->read == rb_page_commit(reader) && | |
1889 | (commit == reader || | |
1890 | (commit == head && | |
1891 | head->read == rb_page_commit(commit))); | |
1892 | } | |
1893 | ||
7a8e76a3 SR |
1894 | /** |
1895 | * ring_buffer_record_disable - stop all writes into the buffer | |
1896 | * @buffer: The ring buffer to stop writes to. | |
1897 | * | |
1898 | * This prevents all writes to the buffer. Any attempt to write | |
1899 | * to the buffer after this will fail and return NULL. | |
1900 | * | |
1901 | * The caller should call synchronize_sched() after this. | |
1902 | */ | |
1903 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
1904 | { | |
1905 | atomic_inc(&buffer->record_disabled); | |
1906 | } | |
c4f50183 | 1907 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
1908 | |
1909 | /** | |
1910 | * ring_buffer_record_enable - enable writes to the buffer | |
1911 | * @buffer: The ring buffer to enable writes | |
1912 | * | |
1913 | * Note, multiple disables will need the same number of enables | |
1914 | * to truely enable the writing (much like preempt_disable). | |
1915 | */ | |
1916 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
1917 | { | |
1918 | atomic_dec(&buffer->record_disabled); | |
1919 | } | |
c4f50183 | 1920 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 SR |
1921 | |
1922 | /** | |
1923 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
1924 | * @buffer: The ring buffer to stop writes to. | |
1925 | * @cpu: The CPU buffer to stop | |
1926 | * | |
1927 | * This prevents all writes to the buffer. Any attempt to write | |
1928 | * to the buffer after this will fail and return NULL. | |
1929 | * | |
1930 | * The caller should call synchronize_sched() after this. | |
1931 | */ | |
1932 | void ring_buffer_record_disable_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_inc(&cpu_buffer->record_disabled); | |
1941 | } | |
c4f50183 | 1942 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
1943 | |
1944 | /** | |
1945 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
1946 | * @buffer: The ring buffer to enable writes | |
1947 | * @cpu: The CPU to enable. | |
1948 | * | |
1949 | * Note, multiple disables will need the same number of enables | |
1950 | * to truely enable the writing (much like preempt_disable). | |
1951 | */ | |
1952 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
1953 | { | |
1954 | struct ring_buffer_per_cpu *cpu_buffer; | |
1955 | ||
9e01c1b7 | 1956 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1957 | return; |
7a8e76a3 SR |
1958 | |
1959 | cpu_buffer = buffer->buffers[cpu]; | |
1960 | atomic_dec(&cpu_buffer->record_disabled); | |
1961 | } | |
c4f50183 | 1962 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 SR |
1963 | |
1964 | /** | |
1965 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
1966 | * @buffer: The ring buffer | |
1967 | * @cpu: The per CPU buffer to get the entries from. | |
1968 | */ | |
1969 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
1970 | { | |
1971 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 1972 | unsigned long ret; |
7a8e76a3 | 1973 | |
9e01c1b7 | 1974 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1975 | return 0; |
7a8e76a3 SR |
1976 | |
1977 | cpu_buffer = buffer->buffers[cpu]; | |
e4906eff SR |
1978 | ret = (local_read(&cpu_buffer->entries) - cpu_buffer->overrun) |
1979 | - cpu_buffer->read; | |
554f786e SR |
1980 | |
1981 | return ret; | |
7a8e76a3 | 1982 | } |
c4f50183 | 1983 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
1984 | |
1985 | /** | |
1986 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
1987 | * @buffer: The ring buffer | |
1988 | * @cpu: The per CPU buffer to get the number of overruns from | |
1989 | */ | |
1990 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
1991 | { | |
1992 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 1993 | unsigned long ret; |
7a8e76a3 | 1994 | |
9e01c1b7 | 1995 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 1996 | return 0; |
7a8e76a3 SR |
1997 | |
1998 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 1999 | ret = cpu_buffer->overrun; |
554f786e SR |
2000 | |
2001 | return ret; | |
7a8e76a3 | 2002 | } |
c4f50183 | 2003 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 2004 | |
f0d2c681 SR |
2005 | /** |
2006 | * ring_buffer_nmi_dropped_cpu - get the number of nmis that were dropped | |
2007 | * @buffer: The ring buffer | |
2008 | * @cpu: The per CPU buffer to get the number of overruns from | |
2009 | */ | |
2010 | unsigned long ring_buffer_nmi_dropped_cpu(struct ring_buffer *buffer, int cpu) | |
2011 | { | |
2012 | struct ring_buffer_per_cpu *cpu_buffer; | |
2013 | unsigned long ret; | |
2014 | ||
2015 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
2016 | return 0; | |
2017 | ||
2018 | cpu_buffer = buffer->buffers[cpu]; | |
2019 | ret = cpu_buffer->nmi_dropped; | |
2020 | ||
2021 | return ret; | |
2022 | } | |
2023 | EXPORT_SYMBOL_GPL(ring_buffer_nmi_dropped_cpu); | |
2024 | ||
2025 | /** | |
2026 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits | |
2027 | * @buffer: The ring buffer | |
2028 | * @cpu: The per CPU buffer to get the number of overruns from | |
2029 | */ | |
2030 | unsigned long | |
2031 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
2032 | { | |
2033 | struct ring_buffer_per_cpu *cpu_buffer; | |
2034 | unsigned long ret; | |
2035 | ||
2036 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
2037 | return 0; | |
2038 | ||
2039 | cpu_buffer = buffer->buffers[cpu]; | |
2040 | ret = cpu_buffer->commit_overrun; | |
2041 | ||
2042 | return ret; | |
2043 | } | |
2044 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
2045 | ||
7a8e76a3 SR |
2046 | /** |
2047 | * ring_buffer_entries - get the number of entries in a buffer | |
2048 | * @buffer: The ring buffer | |
2049 | * | |
2050 | * Returns the total number of entries in the ring buffer | |
2051 | * (all CPU entries) | |
2052 | */ | |
2053 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
2054 | { | |
2055 | struct ring_buffer_per_cpu *cpu_buffer; | |
2056 | unsigned long entries = 0; | |
2057 | int cpu; | |
2058 | ||
2059 | /* if you care about this being correct, lock the buffer */ | |
2060 | for_each_buffer_cpu(buffer, cpu) { | |
2061 | cpu_buffer = buffer->buffers[cpu]; | |
e4906eff SR |
2062 | entries += (local_read(&cpu_buffer->entries) - |
2063 | cpu_buffer->overrun) - cpu_buffer->read; | |
7a8e76a3 SR |
2064 | } |
2065 | ||
2066 | return entries; | |
2067 | } | |
c4f50183 | 2068 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
2069 | |
2070 | /** | |
2071 | * ring_buffer_overrun_cpu - get the number of overruns in buffer | |
2072 | * @buffer: The ring buffer | |
2073 | * | |
2074 | * Returns the total number of overruns in the ring buffer | |
2075 | * (all CPU entries) | |
2076 | */ | |
2077 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
2078 | { | |
2079 | struct ring_buffer_per_cpu *cpu_buffer; | |
2080 | unsigned long overruns = 0; | |
2081 | int cpu; | |
2082 | ||
2083 | /* if you care about this being correct, lock the buffer */ | |
2084 | for_each_buffer_cpu(buffer, cpu) { | |
2085 | cpu_buffer = buffer->buffers[cpu]; | |
2086 | overruns += cpu_buffer->overrun; | |
2087 | } | |
2088 | ||
2089 | return overruns; | |
2090 | } | |
c4f50183 | 2091 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 2092 | |
642edba5 | 2093 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
2094 | { |
2095 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2096 | ||
d769041f SR |
2097 | /* Iterator usage is expected to have record disabled */ |
2098 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
2099 | iter->head_page = cpu_buffer->head_page; | |
6f807acd | 2100 | iter->head = cpu_buffer->head_page->read; |
d769041f SR |
2101 | } else { |
2102 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 2103 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
2104 | } |
2105 | if (iter->head) | |
2106 | iter->read_stamp = cpu_buffer->read_stamp; | |
2107 | else | |
abc9b56d | 2108 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 2109 | } |
f83c9d0f | 2110 | |
642edba5 SR |
2111 | /** |
2112 | * ring_buffer_iter_reset - reset an iterator | |
2113 | * @iter: The iterator to reset | |
2114 | * | |
2115 | * Resets the iterator, so that it will start from the beginning | |
2116 | * again. | |
2117 | */ | |
2118 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
2119 | { | |
554f786e | 2120 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
2121 | unsigned long flags; |
2122 | ||
554f786e SR |
2123 | if (!iter) |
2124 | return; | |
2125 | ||
2126 | cpu_buffer = iter->cpu_buffer; | |
2127 | ||
642edba5 SR |
2128 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2129 | rb_iter_reset(iter); | |
f83c9d0f | 2130 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 2131 | } |
c4f50183 | 2132 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
2133 | |
2134 | /** | |
2135 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
2136 | * @iter: The iterator to check | |
2137 | */ | |
2138 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
2139 | { | |
2140 | struct ring_buffer_per_cpu *cpu_buffer; | |
2141 | ||
2142 | cpu_buffer = iter->cpu_buffer; | |
2143 | ||
bf41a158 SR |
2144 | return iter->head_page == cpu_buffer->commit_page && |
2145 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 2146 | } |
c4f50183 | 2147 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
2148 | |
2149 | static void | |
2150 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
2151 | struct ring_buffer_event *event) | |
2152 | { | |
2153 | u64 delta; | |
2154 | ||
334d4169 | 2155 | switch (event->type_len) { |
7a8e76a3 SR |
2156 | case RINGBUF_TYPE_PADDING: |
2157 | return; | |
2158 | ||
2159 | case RINGBUF_TYPE_TIME_EXTEND: | |
2160 | delta = event->array[0]; | |
2161 | delta <<= TS_SHIFT; | |
2162 | delta += event->time_delta; | |
2163 | cpu_buffer->read_stamp += delta; | |
2164 | return; | |
2165 | ||
2166 | case RINGBUF_TYPE_TIME_STAMP: | |
2167 | /* FIXME: not implemented */ | |
2168 | return; | |
2169 | ||
2170 | case RINGBUF_TYPE_DATA: | |
2171 | cpu_buffer->read_stamp += event->time_delta; | |
2172 | return; | |
2173 | ||
2174 | default: | |
2175 | BUG(); | |
2176 | } | |
2177 | return; | |
2178 | } | |
2179 | ||
2180 | static void | |
2181 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
2182 | struct ring_buffer_event *event) | |
2183 | { | |
2184 | u64 delta; | |
2185 | ||
334d4169 | 2186 | switch (event->type_len) { |
7a8e76a3 SR |
2187 | case RINGBUF_TYPE_PADDING: |
2188 | return; | |
2189 | ||
2190 | case RINGBUF_TYPE_TIME_EXTEND: | |
2191 | delta = event->array[0]; | |
2192 | delta <<= TS_SHIFT; | |
2193 | delta += event->time_delta; | |
2194 | iter->read_stamp += delta; | |
2195 | return; | |
2196 | ||
2197 | case RINGBUF_TYPE_TIME_STAMP: | |
2198 | /* FIXME: not implemented */ | |
2199 | return; | |
2200 | ||
2201 | case RINGBUF_TYPE_DATA: | |
2202 | iter->read_stamp += event->time_delta; | |
2203 | return; | |
2204 | ||
2205 | default: | |
2206 | BUG(); | |
2207 | } | |
2208 | return; | |
2209 | } | |
2210 | ||
d769041f SR |
2211 | static struct buffer_page * |
2212 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 2213 | { |
d769041f SR |
2214 | struct buffer_page *reader = NULL; |
2215 | unsigned long flags; | |
818e3dd3 | 2216 | int nr_loops = 0; |
d769041f | 2217 | |
3e03fb7f SR |
2218 | local_irq_save(flags); |
2219 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
2220 | |
2221 | again: | |
818e3dd3 SR |
2222 | /* |
2223 | * This should normally only loop twice. But because the | |
2224 | * start of the reader inserts an empty page, it causes | |
2225 | * a case where we will loop three times. There should be no | |
2226 | * reason to loop four times (that I know of). | |
2227 | */ | |
3e89c7bb | 2228 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
2229 | reader = NULL; |
2230 | goto out; | |
2231 | } | |
2232 | ||
d769041f SR |
2233 | reader = cpu_buffer->reader_page; |
2234 | ||
2235 | /* If there's more to read, return this page */ | |
bf41a158 | 2236 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
2237 | goto out; |
2238 | ||
2239 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
2240 | if (RB_WARN_ON(cpu_buffer, |
2241 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
2242 | goto out; | |
d769041f SR |
2243 | |
2244 | /* check if we caught up to the tail */ | |
2245 | reader = NULL; | |
bf41a158 | 2246 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 2247 | goto out; |
7a8e76a3 SR |
2248 | |
2249 | /* | |
d769041f SR |
2250 | * Splice the empty reader page into the list around the head. |
2251 | * Reset the reader page to size zero. | |
7a8e76a3 | 2252 | */ |
7a8e76a3 | 2253 | |
d769041f SR |
2254 | reader = cpu_buffer->head_page; |
2255 | cpu_buffer->reader_page->list.next = reader->list.next; | |
2256 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 SR |
2257 | |
2258 | local_set(&cpu_buffer->reader_page->write, 0); | |
778c55d4 | 2259 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 2260 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
7a8e76a3 | 2261 | |
d769041f SR |
2262 | /* Make the reader page now replace the head */ |
2263 | reader->list.prev->next = &cpu_buffer->reader_page->list; | |
2264 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
7a8e76a3 SR |
2265 | |
2266 | /* | |
d769041f SR |
2267 | * If the tail is on the reader, then we must set the head |
2268 | * to the inserted page, otherwise we set it one before. | |
7a8e76a3 | 2269 | */ |
d769041f | 2270 | cpu_buffer->head_page = cpu_buffer->reader_page; |
7a8e76a3 | 2271 | |
bf41a158 | 2272 | if (cpu_buffer->commit_page != reader) |
d769041f SR |
2273 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
2274 | ||
2275 | /* Finally update the reader page to the new head */ | |
2276 | cpu_buffer->reader_page = reader; | |
2277 | rb_reset_reader_page(cpu_buffer); | |
2278 | ||
2279 | goto again; | |
2280 | ||
2281 | out: | |
3e03fb7f SR |
2282 | __raw_spin_unlock(&cpu_buffer->lock); |
2283 | local_irq_restore(flags); | |
d769041f SR |
2284 | |
2285 | return reader; | |
2286 | } | |
2287 | ||
2288 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
2289 | { | |
2290 | struct ring_buffer_event *event; | |
2291 | struct buffer_page *reader; | |
2292 | unsigned length; | |
2293 | ||
2294 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 2295 | |
d769041f | 2296 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
2297 | if (RB_WARN_ON(cpu_buffer, !reader)) |
2298 | return; | |
7a8e76a3 | 2299 | |
d769041f SR |
2300 | event = rb_reader_event(cpu_buffer); |
2301 | ||
334d4169 LJ |
2302 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX |
2303 | || rb_discarded_event(event)) | |
e4906eff | 2304 | cpu_buffer->read++; |
d769041f SR |
2305 | |
2306 | rb_update_read_stamp(cpu_buffer, event); | |
2307 | ||
2308 | length = rb_event_length(event); | |
6f807acd | 2309 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
2310 | } |
2311 | ||
2312 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
2313 | { | |
2314 | struct ring_buffer *buffer; | |
2315 | struct ring_buffer_per_cpu *cpu_buffer; | |
2316 | struct ring_buffer_event *event; | |
2317 | unsigned length; | |
2318 | ||
2319 | cpu_buffer = iter->cpu_buffer; | |
2320 | buffer = cpu_buffer->buffer; | |
2321 | ||
2322 | /* | |
2323 | * Check if we are at the end of the buffer. | |
2324 | */ | |
bf41a158 | 2325 | if (iter->head >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
2326 | /* discarded commits can make the page empty */ |
2327 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 2328 | return; |
d769041f | 2329 | rb_inc_iter(iter); |
7a8e76a3 SR |
2330 | return; |
2331 | } | |
2332 | ||
2333 | event = rb_iter_head_event(iter); | |
2334 | ||
2335 | length = rb_event_length(event); | |
2336 | ||
2337 | /* | |
2338 | * This should not be called to advance the header if we are | |
2339 | * at the tail of the buffer. | |
2340 | */ | |
3e89c7bb | 2341 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 2342 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
2343 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
2344 | return; | |
7a8e76a3 SR |
2345 | |
2346 | rb_update_iter_read_stamp(iter, event); | |
2347 | ||
2348 | iter->head += length; | |
2349 | ||
2350 | /* check for end of page padding */ | |
bf41a158 SR |
2351 | if ((iter->head >= rb_page_size(iter->head_page)) && |
2352 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
2353 | rb_advance_iter(iter); |
2354 | } | |
2355 | ||
f83c9d0f SR |
2356 | static struct ring_buffer_event * |
2357 | rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
7a8e76a3 SR |
2358 | { |
2359 | struct ring_buffer_per_cpu *cpu_buffer; | |
2360 | struct ring_buffer_event *event; | |
d769041f | 2361 | struct buffer_page *reader; |
818e3dd3 | 2362 | int nr_loops = 0; |
7a8e76a3 | 2363 | |
7a8e76a3 SR |
2364 | cpu_buffer = buffer->buffers[cpu]; |
2365 | ||
2366 | again: | |
818e3dd3 SR |
2367 | /* |
2368 | * We repeat when a timestamp is encountered. It is possible | |
2369 | * to get multiple timestamps from an interrupt entering just | |
ea05b57c SR |
2370 | * as one timestamp is about to be written, or from discarded |
2371 | * commits. The most that we can have is the number on a single page. | |
818e3dd3 | 2372 | */ |
ea05b57c | 2373 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE)) |
818e3dd3 | 2374 | return NULL; |
818e3dd3 | 2375 | |
d769041f SR |
2376 | reader = rb_get_reader_page(cpu_buffer); |
2377 | if (!reader) | |
7a8e76a3 SR |
2378 | return NULL; |
2379 | ||
d769041f | 2380 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 2381 | |
334d4169 | 2382 | switch (event->type_len) { |
7a8e76a3 | 2383 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
2384 | if (rb_null_event(event)) |
2385 | RB_WARN_ON(cpu_buffer, 1); | |
2386 | /* | |
2387 | * Because the writer could be discarding every | |
2388 | * event it creates (which would probably be bad) | |
2389 | * if we were to go back to "again" then we may never | |
2390 | * catch up, and will trigger the warn on, or lock | |
2391 | * the box. Return the padding, and we will release | |
2392 | * the current locks, and try again. | |
2393 | */ | |
d769041f | 2394 | rb_advance_reader(cpu_buffer); |
2d622719 | 2395 | return event; |
7a8e76a3 SR |
2396 | |
2397 | case RINGBUF_TYPE_TIME_EXTEND: | |
2398 | /* Internal data, OK to advance */ | |
d769041f | 2399 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2400 | goto again; |
2401 | ||
2402 | case RINGBUF_TYPE_TIME_STAMP: | |
2403 | /* FIXME: not implemented */ | |
d769041f | 2404 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2405 | goto again; |
2406 | ||
2407 | case RINGBUF_TYPE_DATA: | |
2408 | if (ts) { | |
2409 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
37886f6a SR |
2410 | ring_buffer_normalize_time_stamp(buffer, |
2411 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2412 | } |
2413 | return event; | |
2414 | ||
2415 | default: | |
2416 | BUG(); | |
2417 | } | |
2418 | ||
2419 | return NULL; | |
2420 | } | |
c4f50183 | 2421 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 2422 | |
f83c9d0f SR |
2423 | static struct ring_buffer_event * |
2424 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
2425 | { |
2426 | struct ring_buffer *buffer; | |
2427 | struct ring_buffer_per_cpu *cpu_buffer; | |
2428 | struct ring_buffer_event *event; | |
818e3dd3 | 2429 | int nr_loops = 0; |
7a8e76a3 SR |
2430 | |
2431 | if (ring_buffer_iter_empty(iter)) | |
2432 | return NULL; | |
2433 | ||
2434 | cpu_buffer = iter->cpu_buffer; | |
2435 | buffer = cpu_buffer->buffer; | |
2436 | ||
2437 | again: | |
818e3dd3 | 2438 | /* |
ea05b57c SR |
2439 | * We repeat when a timestamp is encountered. |
2440 | * We can get multiple timestamps by nested interrupts or also | |
2441 | * if filtering is on (discarding commits). Since discarding | |
2442 | * commits can be frequent we can get a lot of timestamps. | |
2443 | * But we limit them by not adding timestamps if they begin | |
2444 | * at the start of a page. | |
818e3dd3 | 2445 | */ |
ea05b57c | 2446 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE)) |
818e3dd3 | 2447 | return NULL; |
818e3dd3 | 2448 | |
7a8e76a3 SR |
2449 | if (rb_per_cpu_empty(cpu_buffer)) |
2450 | return NULL; | |
2451 | ||
2452 | event = rb_iter_head_event(iter); | |
2453 | ||
334d4169 | 2454 | switch (event->type_len) { |
7a8e76a3 | 2455 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
2456 | if (rb_null_event(event)) { |
2457 | rb_inc_iter(iter); | |
2458 | goto again; | |
2459 | } | |
2460 | rb_advance_iter(iter); | |
2461 | return event; | |
7a8e76a3 SR |
2462 | |
2463 | case RINGBUF_TYPE_TIME_EXTEND: | |
2464 | /* Internal data, OK to advance */ | |
2465 | rb_advance_iter(iter); | |
2466 | goto again; | |
2467 | ||
2468 | case RINGBUF_TYPE_TIME_STAMP: | |
2469 | /* FIXME: not implemented */ | |
2470 | rb_advance_iter(iter); | |
2471 | goto again; | |
2472 | ||
2473 | case RINGBUF_TYPE_DATA: | |
2474 | if (ts) { | |
2475 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
2476 | ring_buffer_normalize_time_stamp(buffer, |
2477 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
2478 | } |
2479 | return event; | |
2480 | ||
2481 | default: | |
2482 | BUG(); | |
2483 | } | |
2484 | ||
2485 | return NULL; | |
2486 | } | |
c4f50183 | 2487 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 2488 | |
f83c9d0f SR |
2489 | /** |
2490 | * ring_buffer_peek - peek at the next event to be read | |
2491 | * @buffer: The ring buffer to read | |
2492 | * @cpu: The cpu to peak at | |
2493 | * @ts: The timestamp counter of this event. | |
2494 | * | |
2495 | * This will return the event that will be read next, but does | |
2496 | * not consume the data. | |
2497 | */ | |
2498 | struct ring_buffer_event * | |
2499 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2500 | { | |
2501 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 2502 | struct ring_buffer_event *event; |
f83c9d0f SR |
2503 | unsigned long flags; |
2504 | ||
554f786e | 2505 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2506 | return NULL; |
554f786e | 2507 | |
2d622719 | 2508 | again: |
f83c9d0f SR |
2509 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2510 | event = rb_buffer_peek(buffer, cpu, ts); | |
2511 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2512 | ||
334d4169 | 2513 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2514 | cpu_relax(); |
2515 | goto again; | |
2516 | } | |
2517 | ||
f83c9d0f SR |
2518 | return event; |
2519 | } | |
2520 | ||
2521 | /** | |
2522 | * ring_buffer_iter_peek - peek at the next event to be read | |
2523 | * @iter: The ring buffer iterator | |
2524 | * @ts: The timestamp counter of this event. | |
2525 | * | |
2526 | * This will return the event that will be read next, but does | |
2527 | * not increment the iterator. | |
2528 | */ | |
2529 | struct ring_buffer_event * | |
2530 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
2531 | { | |
2532 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2533 | struct ring_buffer_event *event; | |
2534 | unsigned long flags; | |
2535 | ||
2d622719 | 2536 | again: |
f83c9d0f SR |
2537 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2538 | event = rb_iter_peek(iter, ts); | |
2539 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
2540 | ||
334d4169 | 2541 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2542 | cpu_relax(); |
2543 | goto again; | |
2544 | } | |
2545 | ||
f83c9d0f SR |
2546 | return event; |
2547 | } | |
2548 | ||
7a8e76a3 SR |
2549 | /** |
2550 | * ring_buffer_consume - return an event and consume it | |
2551 | * @buffer: The ring buffer to get the next event from | |
2552 | * | |
2553 | * Returns the next event in the ring buffer, and that event is consumed. | |
2554 | * Meaning, that sequential reads will keep returning a different event, | |
2555 | * and eventually empty the ring buffer if the producer is slower. | |
2556 | */ | |
2557 | struct ring_buffer_event * | |
2558 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
2559 | { | |
554f786e SR |
2560 | struct ring_buffer_per_cpu *cpu_buffer; |
2561 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 2562 | unsigned long flags; |
7a8e76a3 | 2563 | |
2d622719 | 2564 | again: |
554f786e SR |
2565 | /* might be called in atomic */ |
2566 | preempt_disable(); | |
2567 | ||
9e01c1b7 | 2568 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 2569 | goto out; |
7a8e76a3 | 2570 | |
554f786e | 2571 | cpu_buffer = buffer->buffers[cpu]; |
f83c9d0f SR |
2572 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2573 | ||
2574 | event = rb_buffer_peek(buffer, cpu, ts); | |
7a8e76a3 | 2575 | if (!event) |
554f786e | 2576 | goto out_unlock; |
7a8e76a3 | 2577 | |
d769041f | 2578 | rb_advance_reader(cpu_buffer); |
7a8e76a3 | 2579 | |
554f786e | 2580 | out_unlock: |
f83c9d0f SR |
2581 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
2582 | ||
554f786e SR |
2583 | out: |
2584 | preempt_enable(); | |
2585 | ||
334d4169 | 2586 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2587 | cpu_relax(); |
2588 | goto again; | |
2589 | } | |
2590 | ||
7a8e76a3 SR |
2591 | return event; |
2592 | } | |
c4f50183 | 2593 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
2594 | |
2595 | /** | |
2596 | * ring_buffer_read_start - start a non consuming read of the buffer | |
2597 | * @buffer: The ring buffer to read from | |
2598 | * @cpu: The cpu buffer to iterate over | |
2599 | * | |
2600 | * This starts up an iteration through the buffer. It also disables | |
2601 | * the recording to the buffer until the reading is finished. | |
2602 | * This prevents the reading from being corrupted. This is not | |
2603 | * a consuming read, so a producer is not expected. | |
2604 | * | |
2605 | * Must be paired with ring_buffer_finish. | |
2606 | */ | |
2607 | struct ring_buffer_iter * | |
2608 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
2609 | { | |
2610 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2611 | struct ring_buffer_iter *iter; |
d769041f | 2612 | unsigned long flags; |
7a8e76a3 | 2613 | |
9e01c1b7 | 2614 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2615 | return NULL; |
7a8e76a3 SR |
2616 | |
2617 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
2618 | if (!iter) | |
8aabee57 | 2619 | return NULL; |
7a8e76a3 SR |
2620 | |
2621 | cpu_buffer = buffer->buffers[cpu]; | |
2622 | ||
2623 | iter->cpu_buffer = cpu_buffer; | |
2624 | ||
2625 | atomic_inc(&cpu_buffer->record_disabled); | |
2626 | synchronize_sched(); | |
2627 | ||
f83c9d0f | 2628 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 2629 | __raw_spin_lock(&cpu_buffer->lock); |
642edba5 | 2630 | rb_iter_reset(iter); |
3e03fb7f | 2631 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 2632 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
2633 | |
2634 | return iter; | |
2635 | } | |
c4f50183 | 2636 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
2637 | |
2638 | /** | |
2639 | * ring_buffer_finish - finish reading the iterator of the buffer | |
2640 | * @iter: The iterator retrieved by ring_buffer_start | |
2641 | * | |
2642 | * This re-enables the recording to the buffer, and frees the | |
2643 | * iterator. | |
2644 | */ | |
2645 | void | |
2646 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
2647 | { | |
2648 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2649 | ||
2650 | atomic_dec(&cpu_buffer->record_disabled); | |
2651 | kfree(iter); | |
2652 | } | |
c4f50183 | 2653 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
2654 | |
2655 | /** | |
2656 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
2657 | * @iter: The ring buffer iterator | |
2658 | * @ts: The time stamp of the event read. | |
2659 | * | |
2660 | * This reads the next event in the ring buffer and increments the iterator. | |
2661 | */ | |
2662 | struct ring_buffer_event * | |
2663 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
2664 | { | |
2665 | struct ring_buffer_event *event; | |
f83c9d0f SR |
2666 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
2667 | unsigned long flags; | |
7a8e76a3 | 2668 | |
2d622719 | 2669 | again: |
f83c9d0f SR |
2670 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2671 | event = rb_iter_peek(iter, ts); | |
7a8e76a3 | 2672 | if (!event) |
f83c9d0f | 2673 | goto out; |
7a8e76a3 SR |
2674 | |
2675 | rb_advance_iter(iter); | |
f83c9d0f SR |
2676 | out: |
2677 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 2678 | |
334d4169 | 2679 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
2680 | cpu_relax(); |
2681 | goto again; | |
2682 | } | |
2683 | ||
7a8e76a3 SR |
2684 | return event; |
2685 | } | |
c4f50183 | 2686 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
2687 | |
2688 | /** | |
2689 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
2690 | * @buffer: The ring buffer. | |
2691 | */ | |
2692 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
2693 | { | |
2694 | return BUF_PAGE_SIZE * buffer->pages; | |
2695 | } | |
c4f50183 | 2696 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
2697 | |
2698 | static void | |
2699 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
2700 | { | |
2701 | cpu_buffer->head_page | |
2702 | = list_entry(cpu_buffer->pages.next, struct buffer_page, list); | |
bf41a158 | 2703 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 2704 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 2705 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 2706 | |
6f807acd | 2707 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
2708 | |
2709 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
2710 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
2711 | ||
2712 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
2713 | local_set(&cpu_buffer->reader_page->write, 0); | |
778c55d4 | 2714 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 2715 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 2716 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 2717 | |
f0d2c681 SR |
2718 | cpu_buffer->nmi_dropped = 0; |
2719 | cpu_buffer->commit_overrun = 0; | |
7a8e76a3 | 2720 | cpu_buffer->overrun = 0; |
e4906eff SR |
2721 | cpu_buffer->read = 0; |
2722 | local_set(&cpu_buffer->entries, 0); | |
69507c06 SR |
2723 | |
2724 | cpu_buffer->write_stamp = 0; | |
2725 | cpu_buffer->read_stamp = 0; | |
7a8e76a3 SR |
2726 | } |
2727 | ||
2728 | /** | |
2729 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
2730 | * @buffer: The ring buffer to reset a per cpu buffer of | |
2731 | * @cpu: The CPU buffer to be reset | |
2732 | */ | |
2733 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
2734 | { | |
2735 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2736 | unsigned long flags; | |
2737 | ||
9e01c1b7 | 2738 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2739 | return; |
7a8e76a3 | 2740 | |
41ede23e SR |
2741 | atomic_inc(&cpu_buffer->record_disabled); |
2742 | ||
f83c9d0f SR |
2743 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2744 | ||
3e03fb7f | 2745 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
2746 | |
2747 | rb_reset_cpu(cpu_buffer); | |
2748 | ||
3e03fb7f | 2749 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f SR |
2750 | |
2751 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
41ede23e SR |
2752 | |
2753 | atomic_dec(&cpu_buffer->record_disabled); | |
7a8e76a3 | 2754 | } |
c4f50183 | 2755 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
2756 | |
2757 | /** | |
2758 | * ring_buffer_reset - reset a ring buffer | |
2759 | * @buffer: The ring buffer to reset all cpu buffers | |
2760 | */ | |
2761 | void ring_buffer_reset(struct ring_buffer *buffer) | |
2762 | { | |
7a8e76a3 SR |
2763 | int cpu; |
2764 | ||
7a8e76a3 | 2765 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 2766 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 2767 | } |
c4f50183 | 2768 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
2769 | |
2770 | /** | |
2771 | * rind_buffer_empty - is the ring buffer empty? | |
2772 | * @buffer: The ring buffer to test | |
2773 | */ | |
2774 | int ring_buffer_empty(struct ring_buffer *buffer) | |
2775 | { | |
2776 | struct ring_buffer_per_cpu *cpu_buffer; | |
2777 | int cpu; | |
2778 | ||
2779 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
2780 | for_each_buffer_cpu(buffer, cpu) { | |
2781 | cpu_buffer = buffer->buffers[cpu]; | |
2782 | if (!rb_per_cpu_empty(cpu_buffer)) | |
2783 | return 0; | |
2784 | } | |
554f786e | 2785 | |
7a8e76a3 SR |
2786 | return 1; |
2787 | } | |
c4f50183 | 2788 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
2789 | |
2790 | /** | |
2791 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
2792 | * @buffer: The ring buffer | |
2793 | * @cpu: The CPU buffer to test | |
2794 | */ | |
2795 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
2796 | { | |
2797 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2798 | int ret; |
7a8e76a3 | 2799 | |
9e01c1b7 | 2800 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2801 | return 1; |
7a8e76a3 SR |
2802 | |
2803 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e SR |
2804 | ret = rb_per_cpu_empty(cpu_buffer); |
2805 | ||
554f786e SR |
2806 | |
2807 | return ret; | |
7a8e76a3 | 2808 | } |
c4f50183 | 2809 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 SR |
2810 | |
2811 | /** | |
2812 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
2813 | * @buffer_a: One buffer to swap with | |
2814 | * @buffer_b: The other buffer to swap with | |
2815 | * | |
2816 | * This function is useful for tracers that want to take a "snapshot" | |
2817 | * of a CPU buffer and has another back up buffer lying around. | |
2818 | * it is expected that the tracer handles the cpu buffer not being | |
2819 | * used at the moment. | |
2820 | */ | |
2821 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
2822 | struct ring_buffer *buffer_b, int cpu) | |
2823 | { | |
2824 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
2825 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
2826 | int ret = -EINVAL; |
2827 | ||
9e01c1b7 RR |
2828 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
2829 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 2830 | goto out; |
7a8e76a3 SR |
2831 | |
2832 | /* At least make sure the two buffers are somewhat the same */ | |
6d102bc6 | 2833 | if (buffer_a->pages != buffer_b->pages) |
554f786e SR |
2834 | goto out; |
2835 | ||
2836 | ret = -EAGAIN; | |
7a8e76a3 | 2837 | |
97b17efe | 2838 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 2839 | goto out; |
97b17efe SR |
2840 | |
2841 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 2842 | goto out; |
97b17efe SR |
2843 | |
2844 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 2845 | goto out; |
97b17efe | 2846 | |
7a8e76a3 SR |
2847 | cpu_buffer_a = buffer_a->buffers[cpu]; |
2848 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
2849 | ||
97b17efe | 2850 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 2851 | goto out; |
97b17efe SR |
2852 | |
2853 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 2854 | goto out; |
97b17efe | 2855 | |
7a8e76a3 SR |
2856 | /* |
2857 | * We can't do a synchronize_sched here because this | |
2858 | * function can be called in atomic context. | |
2859 | * Normally this will be called from the same CPU as cpu. | |
2860 | * If not it's up to the caller to protect this. | |
2861 | */ | |
2862 | atomic_inc(&cpu_buffer_a->record_disabled); | |
2863 | atomic_inc(&cpu_buffer_b->record_disabled); | |
2864 | ||
2865 | buffer_a->buffers[cpu] = cpu_buffer_b; | |
2866 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
2867 | ||
2868 | cpu_buffer_b->buffer = buffer_a; | |
2869 | cpu_buffer_a->buffer = buffer_b; | |
2870 | ||
2871 | atomic_dec(&cpu_buffer_a->record_disabled); | |
2872 | atomic_dec(&cpu_buffer_b->record_disabled); | |
2873 | ||
554f786e SR |
2874 | ret = 0; |
2875 | out: | |
554f786e | 2876 | return ret; |
7a8e76a3 | 2877 | } |
c4f50183 | 2878 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
7a8e76a3 | 2879 | |
8789a9e7 SR |
2880 | /** |
2881 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
2882 | * @buffer: the buffer to allocate for. | |
2883 | * | |
2884 | * This function is used in conjunction with ring_buffer_read_page. | |
2885 | * When reading a full page from the ring buffer, these functions | |
2886 | * can be used to speed up the process. The calling function should | |
2887 | * allocate a few pages first with this function. Then when it | |
2888 | * needs to get pages from the ring buffer, it passes the result | |
2889 | * of this function into ring_buffer_read_page, which will swap | |
2890 | * the page that was allocated, with the read page of the buffer. | |
2891 | * | |
2892 | * Returns: | |
2893 | * The page allocated, or NULL on error. | |
2894 | */ | |
2895 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer) | |
2896 | { | |
044fa782 | 2897 | struct buffer_data_page *bpage; |
ef7a4a16 | 2898 | unsigned long addr; |
8789a9e7 SR |
2899 | |
2900 | addr = __get_free_page(GFP_KERNEL); | |
2901 | if (!addr) | |
2902 | return NULL; | |
2903 | ||
044fa782 | 2904 | bpage = (void *)addr; |
8789a9e7 | 2905 | |
ef7a4a16 SR |
2906 | rb_init_page(bpage); |
2907 | ||
044fa782 | 2908 | return bpage; |
8789a9e7 | 2909 | } |
d6ce96da | 2910 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
2911 | |
2912 | /** | |
2913 | * ring_buffer_free_read_page - free an allocated read page | |
2914 | * @buffer: the buffer the page was allocate for | |
2915 | * @data: the page to free | |
2916 | * | |
2917 | * Free a page allocated from ring_buffer_alloc_read_page. | |
2918 | */ | |
2919 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
2920 | { | |
2921 | free_page((unsigned long)data); | |
2922 | } | |
d6ce96da | 2923 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
2924 | |
2925 | /** | |
2926 | * ring_buffer_read_page - extract a page from the ring buffer | |
2927 | * @buffer: buffer to extract from | |
2928 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 2929 | * @len: amount to extract |
8789a9e7 SR |
2930 | * @cpu: the cpu of the buffer to extract |
2931 | * @full: should the extraction only happen when the page is full. | |
2932 | * | |
2933 | * This function will pull out a page from the ring buffer and consume it. | |
2934 | * @data_page must be the address of the variable that was returned | |
2935 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
2936 | * to swap with a page in the ring buffer. | |
2937 | * | |
2938 | * for example: | |
b85fa01e | 2939 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
2940 | * if (!rpage) |
2941 | * return error; | |
ef7a4a16 | 2942 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
2943 | * if (ret >= 0) |
2944 | * process_page(rpage, ret); | |
8789a9e7 SR |
2945 | * |
2946 | * When @full is set, the function will not return true unless | |
2947 | * the writer is off the reader page. | |
2948 | * | |
2949 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
2950 | * The ring buffer can be used anywhere in the kernel and can not | |
2951 | * blindly call wake_up. The layer that uses the ring buffer must be | |
2952 | * responsible for that. | |
2953 | * | |
2954 | * Returns: | |
667d2412 LJ |
2955 | * >=0 if data has been transferred, returns the offset of consumed data. |
2956 | * <0 if no data has been transferred. | |
8789a9e7 SR |
2957 | */ |
2958 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 2959 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
2960 | { |
2961 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
2962 | struct ring_buffer_event *event; | |
044fa782 | 2963 | struct buffer_data_page *bpage; |
ef7a4a16 | 2964 | struct buffer_page *reader; |
8789a9e7 | 2965 | unsigned long flags; |
ef7a4a16 | 2966 | unsigned int commit; |
667d2412 | 2967 | unsigned int read; |
4f3640f8 | 2968 | u64 save_timestamp; |
667d2412 | 2969 | int ret = -1; |
8789a9e7 | 2970 | |
554f786e SR |
2971 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
2972 | goto out; | |
2973 | ||
474d32b6 SR |
2974 | /* |
2975 | * If len is not big enough to hold the page header, then | |
2976 | * we can not copy anything. | |
2977 | */ | |
2978 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 2979 | goto out; |
474d32b6 SR |
2980 | |
2981 | len -= BUF_PAGE_HDR_SIZE; | |
2982 | ||
8789a9e7 | 2983 | if (!data_page) |
554f786e | 2984 | goto out; |
8789a9e7 | 2985 | |
044fa782 SR |
2986 | bpage = *data_page; |
2987 | if (!bpage) | |
554f786e | 2988 | goto out; |
8789a9e7 SR |
2989 | |
2990 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
2991 | ||
ef7a4a16 SR |
2992 | reader = rb_get_reader_page(cpu_buffer); |
2993 | if (!reader) | |
554f786e | 2994 | goto out_unlock; |
8789a9e7 | 2995 | |
ef7a4a16 SR |
2996 | event = rb_reader_event(cpu_buffer); |
2997 | ||
2998 | read = reader->read; | |
2999 | commit = rb_page_commit(reader); | |
667d2412 | 3000 | |
8789a9e7 | 3001 | /* |
474d32b6 SR |
3002 | * If this page has been partially read or |
3003 | * if len is not big enough to read the rest of the page or | |
3004 | * a writer is still on the page, then | |
3005 | * we must copy the data from the page to the buffer. | |
3006 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 3007 | */ |
474d32b6 | 3008 | if (read || (len < (commit - read)) || |
ef7a4a16 | 3009 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 3010 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
3011 | unsigned int rpos = read; |
3012 | unsigned int pos = 0; | |
ef7a4a16 | 3013 | unsigned int size; |
8789a9e7 SR |
3014 | |
3015 | if (full) | |
554f786e | 3016 | goto out_unlock; |
8789a9e7 | 3017 | |
ef7a4a16 SR |
3018 | if (len > (commit - read)) |
3019 | len = (commit - read); | |
3020 | ||
3021 | size = rb_event_length(event); | |
3022 | ||
3023 | if (len < size) | |
554f786e | 3024 | goto out_unlock; |
ef7a4a16 | 3025 | |
4f3640f8 SR |
3026 | /* save the current timestamp, since the user will need it */ |
3027 | save_timestamp = cpu_buffer->read_stamp; | |
3028 | ||
ef7a4a16 SR |
3029 | /* Need to copy one event at a time */ |
3030 | do { | |
474d32b6 | 3031 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
3032 | |
3033 | len -= size; | |
3034 | ||
3035 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
3036 | rpos = reader->read; |
3037 | pos += size; | |
ef7a4a16 SR |
3038 | |
3039 | event = rb_reader_event(cpu_buffer); | |
3040 | size = rb_event_length(event); | |
3041 | } while (len > size); | |
667d2412 LJ |
3042 | |
3043 | /* update bpage */ | |
ef7a4a16 | 3044 | local_set(&bpage->commit, pos); |
4f3640f8 | 3045 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 3046 | |
474d32b6 SR |
3047 | /* we copied everything to the beginning */ |
3048 | read = 0; | |
8789a9e7 | 3049 | } else { |
afbab76a SR |
3050 | /* update the entry counter */ |
3051 | cpu_buffer->read += local_read(&reader->entries); | |
3052 | ||
8789a9e7 | 3053 | /* swap the pages */ |
044fa782 | 3054 | rb_init_page(bpage); |
ef7a4a16 SR |
3055 | bpage = reader->page; |
3056 | reader->page = *data_page; | |
3057 | local_set(&reader->write, 0); | |
778c55d4 | 3058 | local_set(&reader->entries, 0); |
ef7a4a16 | 3059 | reader->read = 0; |
044fa782 | 3060 | *data_page = bpage; |
8789a9e7 | 3061 | } |
667d2412 | 3062 | ret = read; |
8789a9e7 | 3063 | |
554f786e | 3064 | out_unlock: |
8789a9e7 SR |
3065 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
3066 | ||
554f786e | 3067 | out: |
8789a9e7 SR |
3068 | return ret; |
3069 | } | |
d6ce96da | 3070 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 3071 | |
a3583244 SR |
3072 | static ssize_t |
3073 | rb_simple_read(struct file *filp, char __user *ubuf, | |
3074 | size_t cnt, loff_t *ppos) | |
3075 | { | |
5e39841c | 3076 | unsigned long *p = filp->private_data; |
a3583244 SR |
3077 | char buf[64]; |
3078 | int r; | |
3079 | ||
033601a3 SR |
3080 | if (test_bit(RB_BUFFERS_DISABLED_BIT, p)) |
3081 | r = sprintf(buf, "permanently disabled\n"); | |
3082 | else | |
3083 | r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p)); | |
a3583244 SR |
3084 | |
3085 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3086 | } | |
3087 | ||
3088 | static ssize_t | |
3089 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
3090 | size_t cnt, loff_t *ppos) | |
3091 | { | |
5e39841c | 3092 | unsigned long *p = filp->private_data; |
a3583244 | 3093 | char buf[64]; |
5e39841c | 3094 | unsigned long val; |
a3583244 SR |
3095 | int ret; |
3096 | ||
3097 | if (cnt >= sizeof(buf)) | |
3098 | return -EINVAL; | |
3099 | ||
3100 | if (copy_from_user(&buf, ubuf, cnt)) | |
3101 | return -EFAULT; | |
3102 | ||
3103 | buf[cnt] = 0; | |
3104 | ||
3105 | ret = strict_strtoul(buf, 10, &val); | |
3106 | if (ret < 0) | |
3107 | return ret; | |
3108 | ||
033601a3 SR |
3109 | if (val) |
3110 | set_bit(RB_BUFFERS_ON_BIT, p); | |
3111 | else | |
3112 | clear_bit(RB_BUFFERS_ON_BIT, p); | |
a3583244 SR |
3113 | |
3114 | (*ppos)++; | |
3115 | ||
3116 | return cnt; | |
3117 | } | |
3118 | ||
5e2336a0 | 3119 | static const struct file_operations rb_simple_fops = { |
a3583244 SR |
3120 | .open = tracing_open_generic, |
3121 | .read = rb_simple_read, | |
3122 | .write = rb_simple_write, | |
3123 | }; | |
3124 | ||
3125 | ||
3126 | static __init int rb_init_debugfs(void) | |
3127 | { | |
3128 | struct dentry *d_tracer; | |
a3583244 SR |
3129 | |
3130 | d_tracer = tracing_init_dentry(); | |
3131 | ||
5452af66 FW |
3132 | trace_create_file("tracing_on", 0644, d_tracer, |
3133 | &ring_buffer_flags, &rb_simple_fops); | |
a3583244 SR |
3134 | |
3135 | return 0; | |
3136 | } | |
3137 | ||
3138 | fs_initcall(rb_init_debugfs); | |
554f786e | 3139 | |
59222efe | 3140 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
3141 | static int rb_cpu_notify(struct notifier_block *self, |
3142 | unsigned long action, void *hcpu) | |
554f786e SR |
3143 | { |
3144 | struct ring_buffer *buffer = | |
3145 | container_of(self, struct ring_buffer, cpu_notify); | |
3146 | long cpu = (long)hcpu; | |
3147 | ||
3148 | switch (action) { | |
3149 | case CPU_UP_PREPARE: | |
3150 | case CPU_UP_PREPARE_FROZEN: | |
3f237a79 | 3151 | if (cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e SR |
3152 | return NOTIFY_OK; |
3153 | ||
3154 | buffer->buffers[cpu] = | |
3155 | rb_allocate_cpu_buffer(buffer, cpu); | |
3156 | if (!buffer->buffers[cpu]) { | |
3157 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
3158 | cpu); | |
3159 | return NOTIFY_OK; | |
3160 | } | |
3161 | smp_wmb(); | |
3f237a79 | 3162 | cpumask_set_cpu(cpu, buffer->cpumask); |
554f786e SR |
3163 | break; |
3164 | case CPU_DOWN_PREPARE: | |
3165 | case CPU_DOWN_PREPARE_FROZEN: | |
3166 | /* | |
3167 | * Do nothing. | |
3168 | * If we were to free the buffer, then the user would | |
3169 | * lose any trace that was in the buffer. | |
3170 | */ | |
3171 | break; | |
3172 | default: | |
3173 | break; | |
3174 | } | |
3175 | return NOTIFY_OK; | |
3176 | } | |
3177 | #endif |