]>
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> |
1744a21d | 13 | #include <linux/kmemcheck.h> |
7a8e76a3 SR |
14 | #include <linux/module.h> |
15 | #include <linux/percpu.h> | |
16 | #include <linux/mutex.h> | |
7a8e76a3 SR |
17 | #include <linux/init.h> |
18 | #include <linux/hash.h> | |
19 | #include <linux/list.h> | |
554f786e | 20 | #include <linux/cpu.h> |
7a8e76a3 SR |
21 | #include <linux/fs.h> |
22 | ||
182e9f5f SR |
23 | #include "trace.h" |
24 | ||
d1b182a8 SR |
25 | /* |
26 | * The ring buffer header is special. We must manually up keep it. | |
27 | */ | |
28 | int ring_buffer_print_entry_header(struct trace_seq *s) | |
29 | { | |
30 | int ret; | |
31 | ||
334d4169 LJ |
32 | ret = trace_seq_printf(s, "# compressed entry header\n"); |
33 | ret = trace_seq_printf(s, "\ttype_len : 5 bits\n"); | |
d1b182a8 SR |
34 | ret = trace_seq_printf(s, "\ttime_delta : 27 bits\n"); |
35 | ret = trace_seq_printf(s, "\tarray : 32 bits\n"); | |
36 | ret = trace_seq_printf(s, "\n"); | |
37 | ret = trace_seq_printf(s, "\tpadding : type == %d\n", | |
38 | RINGBUF_TYPE_PADDING); | |
39 | ret = trace_seq_printf(s, "\ttime_extend : type == %d\n", | |
40 | RINGBUF_TYPE_TIME_EXTEND); | |
334d4169 LJ |
41 | ret = trace_seq_printf(s, "\tdata max type_len == %d\n", |
42 | RINGBUF_TYPE_DATA_TYPE_LEN_MAX); | |
d1b182a8 SR |
43 | |
44 | return ret; | |
45 | } | |
46 | ||
5cc98548 SR |
47 | /* |
48 | * The ring buffer is made up of a list of pages. A separate list of pages is | |
49 | * allocated for each CPU. A writer may only write to a buffer that is | |
50 | * associated with the CPU it is currently executing on. A reader may read | |
51 | * from any per cpu buffer. | |
52 | * | |
53 | * The reader is special. For each per cpu buffer, the reader has its own | |
54 | * reader page. When a reader has read the entire reader page, this reader | |
55 | * page is swapped with another page in the ring buffer. | |
56 | * | |
57 | * Now, as long as the writer is off the reader page, the reader can do what | |
58 | * ever it wants with that page. The writer will never write to that page | |
59 | * again (as long as it is out of the ring buffer). | |
60 | * | |
61 | * Here's some silly ASCII art. | |
62 | * | |
63 | * +------+ | |
64 | * |reader| RING BUFFER | |
65 | * |page | | |
66 | * +------+ +---+ +---+ +---+ | |
67 | * | |-->| |-->| | | |
68 | * +---+ +---+ +---+ | |
69 | * ^ | | |
70 | * | | | |
71 | * +---------------+ | |
72 | * | |
73 | * | |
74 | * +------+ | |
75 | * |reader| RING BUFFER | |
76 | * |page |------------------v | |
77 | * +------+ +---+ +---+ +---+ | |
78 | * | |-->| |-->| | | |
79 | * +---+ +---+ +---+ | |
80 | * ^ | | |
81 | * | | | |
82 | * +---------------+ | |
83 | * | |
84 | * | |
85 | * +------+ | |
86 | * |reader| RING BUFFER | |
87 | * |page |------------------v | |
88 | * +------+ +---+ +---+ +---+ | |
89 | * ^ | |-->| |-->| | | |
90 | * | +---+ +---+ +---+ | |
91 | * | | | |
92 | * | | | |
93 | * +------------------------------+ | |
94 | * | |
95 | * | |
96 | * +------+ | |
97 | * |buffer| RING BUFFER | |
98 | * |page |------------------v | |
99 | * +------+ +---+ +---+ +---+ | |
100 | * ^ | | | |-->| | | |
101 | * | New +---+ +---+ +---+ | |
102 | * | Reader------^ | | |
103 | * | page | | |
104 | * +------------------------------+ | |
105 | * | |
106 | * | |
107 | * After we make this swap, the reader can hand this page off to the splice | |
108 | * code and be done with it. It can even allocate a new page if it needs to | |
109 | * and swap that into the ring buffer. | |
110 | * | |
111 | * We will be using cmpxchg soon to make all this lockless. | |
112 | * | |
113 | */ | |
114 | ||
033601a3 SR |
115 | /* |
116 | * A fast way to enable or disable all ring buffers is to | |
117 | * call tracing_on or tracing_off. Turning off the ring buffers | |
118 | * prevents all ring buffers from being recorded to. | |
119 | * Turning this switch on, makes it OK to write to the | |
120 | * ring buffer, if the ring buffer is enabled itself. | |
121 | * | |
122 | * There's three layers that must be on in order to write | |
123 | * to the ring buffer. | |
124 | * | |
125 | * 1) This global flag must be set. | |
126 | * 2) The ring buffer must be enabled for recording. | |
127 | * 3) The per cpu buffer must be enabled for recording. | |
128 | * | |
129 | * In case of an anomaly, this global flag has a bit set that | |
130 | * will permantly disable all ring buffers. | |
131 | */ | |
132 | ||
133 | /* | |
134 | * Global flag to disable all recording to ring buffers | |
135 | * This has two bits: ON, DISABLED | |
136 | * | |
137 | * ON DISABLED | |
138 | * ---- ---------- | |
139 | * 0 0 : ring buffers are off | |
140 | * 1 0 : ring buffers are on | |
141 | * X 1 : ring buffers are permanently disabled | |
142 | */ | |
143 | ||
144 | enum { | |
145 | RB_BUFFERS_ON_BIT = 0, | |
146 | RB_BUFFERS_DISABLED_BIT = 1, | |
147 | }; | |
148 | ||
149 | enum { | |
150 | RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT, | |
151 | RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT, | |
152 | }; | |
153 | ||
5e39841c | 154 | static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON; |
a3583244 | 155 | |
474d32b6 SR |
156 | #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data) |
157 | ||
a3583244 SR |
158 | /** |
159 | * tracing_on - enable all tracing buffers | |
160 | * | |
161 | * This function enables all tracing buffers that may have been | |
162 | * disabled with tracing_off. | |
163 | */ | |
164 | void tracing_on(void) | |
165 | { | |
033601a3 | 166 | set_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags); |
a3583244 | 167 | } |
c4f50183 | 168 | EXPORT_SYMBOL_GPL(tracing_on); |
a3583244 SR |
169 | |
170 | /** | |
171 | * tracing_off - turn off all tracing buffers | |
172 | * | |
173 | * This function stops all tracing buffers from recording data. | |
174 | * It does not disable any overhead the tracers themselves may | |
175 | * be causing. This function simply causes all recording to | |
176 | * the ring buffers to fail. | |
177 | */ | |
178 | void tracing_off(void) | |
179 | { | |
033601a3 SR |
180 | clear_bit(RB_BUFFERS_ON_BIT, &ring_buffer_flags); |
181 | } | |
c4f50183 | 182 | EXPORT_SYMBOL_GPL(tracing_off); |
033601a3 SR |
183 | |
184 | /** | |
185 | * tracing_off_permanent - permanently disable ring buffers | |
186 | * | |
187 | * This function, once called, will disable all ring buffers | |
c3706f00 | 188 | * permanently. |
033601a3 SR |
189 | */ |
190 | void tracing_off_permanent(void) | |
191 | { | |
192 | set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags); | |
a3583244 SR |
193 | } |
194 | ||
988ae9d6 SR |
195 | /** |
196 | * tracing_is_on - show state of ring buffers enabled | |
197 | */ | |
198 | int tracing_is_on(void) | |
199 | { | |
200 | return ring_buffer_flags == RB_BUFFERS_ON; | |
201 | } | |
202 | EXPORT_SYMBOL_GPL(tracing_is_on); | |
203 | ||
d06bbd66 IM |
204 | #include "trace.h" |
205 | ||
e3d6bf0a | 206 | #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array)) |
67d34724 | 207 | #define RB_ALIGNMENT 4U |
334d4169 | 208 | #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
c7b09308 | 209 | #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */ |
334d4169 LJ |
210 | |
211 | /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */ | |
212 | #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX | |
7a8e76a3 SR |
213 | |
214 | enum { | |
215 | RB_LEN_TIME_EXTEND = 8, | |
216 | RB_LEN_TIME_STAMP = 16, | |
217 | }; | |
218 | ||
2d622719 TZ |
219 | static inline int rb_null_event(struct ring_buffer_event *event) |
220 | { | |
334d4169 LJ |
221 | return event->type_len == RINGBUF_TYPE_PADDING |
222 | && event->time_delta == 0; | |
2d622719 TZ |
223 | } |
224 | ||
225 | static inline int rb_discarded_event(struct ring_buffer_event *event) | |
226 | { | |
334d4169 | 227 | return event->type_len == RINGBUF_TYPE_PADDING && event->time_delta; |
2d622719 TZ |
228 | } |
229 | ||
230 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
231 | { | |
334d4169 | 232 | event->type_len = RINGBUF_TYPE_PADDING; |
2d622719 TZ |
233 | event->time_delta = 0; |
234 | } | |
235 | ||
34a148bf | 236 | static unsigned |
2d622719 | 237 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
238 | { |
239 | unsigned length; | |
240 | ||
334d4169 LJ |
241 | if (event->type_len) |
242 | length = event->type_len * RB_ALIGNMENT; | |
2d622719 TZ |
243 | else |
244 | length = event->array[0]; | |
245 | return length + RB_EVNT_HDR_SIZE; | |
246 | } | |
247 | ||
248 | /* inline for ring buffer fast paths */ | |
249 | static unsigned | |
250 | rb_event_length(struct ring_buffer_event *event) | |
251 | { | |
334d4169 | 252 | switch (event->type_len) { |
7a8e76a3 | 253 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
254 | if (rb_null_event(event)) |
255 | /* undefined */ | |
256 | return -1; | |
334d4169 | 257 | return event->array[0] + RB_EVNT_HDR_SIZE; |
7a8e76a3 SR |
258 | |
259 | case RINGBUF_TYPE_TIME_EXTEND: | |
260 | return RB_LEN_TIME_EXTEND; | |
261 | ||
262 | case RINGBUF_TYPE_TIME_STAMP: | |
263 | return RB_LEN_TIME_STAMP; | |
264 | ||
265 | case RINGBUF_TYPE_DATA: | |
2d622719 | 266 | return rb_event_data_length(event); |
7a8e76a3 SR |
267 | default: |
268 | BUG(); | |
269 | } | |
270 | /* not hit */ | |
271 | return 0; | |
272 | } | |
273 | ||
274 | /** | |
275 | * ring_buffer_event_length - return the length of the event | |
276 | * @event: the event to get the length of | |
277 | */ | |
278 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
279 | { | |
465634ad | 280 | unsigned length = rb_event_length(event); |
334d4169 | 281 | if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
465634ad RR |
282 | return length; |
283 | length -= RB_EVNT_HDR_SIZE; | |
284 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
285 | length -= sizeof(event->array[0]); | |
286 | return length; | |
7a8e76a3 | 287 | } |
c4f50183 | 288 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
289 | |
290 | /* inline for ring buffer fast paths */ | |
34a148bf | 291 | static void * |
7a8e76a3 SR |
292 | rb_event_data(struct ring_buffer_event *event) |
293 | { | |
334d4169 | 294 | BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX); |
7a8e76a3 | 295 | /* If length is in len field, then array[0] has the data */ |
334d4169 | 296 | if (event->type_len) |
7a8e76a3 SR |
297 | return (void *)&event->array[0]; |
298 | /* Otherwise length is in array[0] and array[1] has the data */ | |
299 | return (void *)&event->array[1]; | |
300 | } | |
301 | ||
302 | /** | |
303 | * ring_buffer_event_data - return the data of the event | |
304 | * @event: the event to get the data from | |
305 | */ | |
306 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
307 | { | |
308 | return rb_event_data(event); | |
309 | } | |
c4f50183 | 310 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
311 | |
312 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 313 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
314 | |
315 | #define TS_SHIFT 27 | |
316 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
317 | #define TS_DELTA_TEST (~TS_MASK) | |
318 | ||
abc9b56d | 319 | struct buffer_data_page { |
e4c2ce82 | 320 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 321 | local_t commit; /* write committed index */ |
abc9b56d SR |
322 | unsigned char data[]; /* data of buffer page */ |
323 | }; | |
324 | ||
77ae365e SR |
325 | /* |
326 | * Note, the buffer_page list must be first. The buffer pages | |
327 | * are allocated in cache lines, which means that each buffer | |
328 | * page will be at the beginning of a cache line, and thus | |
329 | * the least significant bits will be zero. We use this to | |
330 | * add flags in the list struct pointers, to make the ring buffer | |
331 | * lockless. | |
332 | */ | |
abc9b56d | 333 | struct buffer_page { |
778c55d4 | 334 | struct list_head list; /* list of buffer pages */ |
abc9b56d | 335 | local_t write; /* index for next write */ |
6f807acd | 336 | unsigned read; /* index for next read */ |
778c55d4 | 337 | local_t entries; /* entries on this page */ |
abc9b56d | 338 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
339 | }; |
340 | ||
77ae365e SR |
341 | /* |
342 | * The buffer page counters, write and entries, must be reset | |
343 | * atomically when crossing page boundaries. To synchronize this | |
344 | * update, two counters are inserted into the number. One is | |
345 | * the actual counter for the write position or count on the page. | |
346 | * | |
347 | * The other is a counter of updaters. Before an update happens | |
348 | * the update partition of the counter is incremented. This will | |
349 | * allow the updater to update the counter atomically. | |
350 | * | |
351 | * The counter is 20 bits, and the state data is 12. | |
352 | */ | |
353 | #define RB_WRITE_MASK 0xfffff | |
354 | #define RB_WRITE_INTCNT (1 << 20) | |
355 | ||
044fa782 | 356 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 357 | { |
044fa782 | 358 | local_set(&bpage->commit, 0); |
abc9b56d SR |
359 | } |
360 | ||
474d32b6 SR |
361 | /** |
362 | * ring_buffer_page_len - the size of data on the page. | |
363 | * @page: The page to read | |
364 | * | |
365 | * Returns the amount of data on the page, including buffer page header. | |
366 | */ | |
ef7a4a16 SR |
367 | size_t ring_buffer_page_len(void *page) |
368 | { | |
474d32b6 SR |
369 | return local_read(&((struct buffer_data_page *)page)->commit) |
370 | + BUF_PAGE_HDR_SIZE; | |
ef7a4a16 SR |
371 | } |
372 | ||
ed56829c SR |
373 | /* |
374 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
375 | * this issue out. | |
376 | */ | |
34a148bf | 377 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 378 | { |
34a148bf | 379 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 380 | kfree(bpage); |
ed56829c SR |
381 | } |
382 | ||
7a8e76a3 SR |
383 | /* |
384 | * We need to fit the time_stamp delta into 27 bits. | |
385 | */ | |
386 | static inline int test_time_stamp(u64 delta) | |
387 | { | |
388 | if (delta & TS_DELTA_TEST) | |
389 | return 1; | |
390 | return 0; | |
391 | } | |
392 | ||
474d32b6 | 393 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 | 394 | |
be957c44 SR |
395 | /* Max payload is BUF_PAGE_SIZE - header (8bytes) */ |
396 | #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) | |
397 | ||
ea05b57c SR |
398 | /* Max number of timestamps that can fit on a page */ |
399 | #define RB_TIMESTAMPS_PER_PAGE (BUF_PAGE_SIZE / RB_LEN_TIME_STAMP) | |
400 | ||
d1b182a8 SR |
401 | int ring_buffer_print_page_header(struct trace_seq *s) |
402 | { | |
403 | struct buffer_data_page field; | |
404 | int ret; | |
405 | ||
406 | ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t" | |
407 | "offset:0;\tsize:%u;\n", | |
408 | (unsigned int)sizeof(field.time_stamp)); | |
409 | ||
410 | ret = trace_seq_printf(s, "\tfield: local_t commit;\t" | |
411 | "offset:%u;\tsize:%u;\n", | |
412 | (unsigned int)offsetof(typeof(field), commit), | |
413 | (unsigned int)sizeof(field.commit)); | |
414 | ||
415 | ret = trace_seq_printf(s, "\tfield: char data;\t" | |
416 | "offset:%u;\tsize:%u;\n", | |
417 | (unsigned int)offsetof(typeof(field), data), | |
418 | (unsigned int)BUF_PAGE_SIZE); | |
419 | ||
420 | return ret; | |
421 | } | |
422 | ||
7a8e76a3 SR |
423 | /* |
424 | * head_page == tail_page && head == tail then buffer is empty. | |
425 | */ | |
426 | struct ring_buffer_per_cpu { | |
427 | int cpu; | |
428 | struct ring_buffer *buffer; | |
77ae365e | 429 | spinlock_t reader_lock; /* serialize readers */ |
3e03fb7f | 430 | raw_spinlock_t lock; |
7a8e76a3 | 431 | struct lock_class_key lock_key; |
3adc54fa | 432 | struct list_head *pages; |
6f807acd SR |
433 | struct buffer_page *head_page; /* read from head */ |
434 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 435 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 436 | struct buffer_page *reader_page; |
77ae365e SR |
437 | local_t commit_overrun; |
438 | local_t overrun; | |
e4906eff | 439 | local_t entries; |
fa743953 SR |
440 | local_t committing; |
441 | local_t commits; | |
77ae365e | 442 | unsigned long read; |
7a8e76a3 SR |
443 | u64 write_stamp; |
444 | u64 read_stamp; | |
445 | atomic_t record_disabled; | |
446 | }; | |
447 | ||
448 | struct ring_buffer { | |
7a8e76a3 SR |
449 | unsigned pages; |
450 | unsigned flags; | |
451 | int cpus; | |
7a8e76a3 | 452 | atomic_t record_disabled; |
00f62f61 | 453 | cpumask_var_t cpumask; |
7a8e76a3 | 454 | |
1f8a6a10 PZ |
455 | struct lock_class_key *reader_lock_key; |
456 | ||
7a8e76a3 SR |
457 | struct mutex mutex; |
458 | ||
459 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 460 | |
59222efe | 461 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
462 | struct notifier_block cpu_notify; |
463 | #endif | |
37886f6a | 464 | u64 (*clock)(void); |
7a8e76a3 SR |
465 | }; |
466 | ||
467 | struct ring_buffer_iter { | |
468 | struct ring_buffer_per_cpu *cpu_buffer; | |
469 | unsigned long head; | |
470 | struct buffer_page *head_page; | |
471 | u64 read_stamp; | |
472 | }; | |
473 | ||
f536aafc | 474 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
bf41a158 | 475 | #define RB_WARN_ON(buffer, cond) \ |
3e89c7bb SR |
476 | ({ \ |
477 | int _____ret = unlikely(cond); \ | |
478 | if (_____ret) { \ | |
bf41a158 SR |
479 | atomic_inc(&buffer->record_disabled); \ |
480 | WARN_ON(1); \ | |
481 | } \ | |
3e89c7bb SR |
482 | _____ret; \ |
483 | }) | |
f536aafc | 484 | |
37886f6a SR |
485 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
486 | #define DEBUG_SHIFT 0 | |
487 | ||
88eb0125 SR |
488 | static inline u64 rb_time_stamp(struct ring_buffer *buffer, int cpu) |
489 | { | |
490 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
491 | return buffer->clock() << DEBUG_SHIFT; | |
492 | } | |
493 | ||
37886f6a SR |
494 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) |
495 | { | |
496 | u64 time; | |
497 | ||
498 | preempt_disable_notrace(); | |
88eb0125 | 499 | time = rb_time_stamp(buffer, cpu); |
37886f6a SR |
500 | preempt_enable_no_resched_notrace(); |
501 | ||
502 | return time; | |
503 | } | |
504 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
505 | ||
506 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
507 | int cpu, u64 *ts) | |
508 | { | |
509 | /* Just stupid testing the normalize function and deltas */ | |
510 | *ts >>= DEBUG_SHIFT; | |
511 | } | |
512 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
513 | ||
77ae365e SR |
514 | /* |
515 | * Making the ring buffer lockless makes things tricky. | |
516 | * Although writes only happen on the CPU that they are on, | |
517 | * and they only need to worry about interrupts. Reads can | |
518 | * happen on any CPU. | |
519 | * | |
520 | * The reader page is always off the ring buffer, but when the | |
521 | * reader finishes with a page, it needs to swap its page with | |
522 | * a new one from the buffer. The reader needs to take from | |
523 | * the head (writes go to the tail). But if a writer is in overwrite | |
524 | * mode and wraps, it must push the head page forward. | |
525 | * | |
526 | * Here lies the problem. | |
527 | * | |
528 | * The reader must be careful to replace only the head page, and | |
529 | * not another one. As described at the top of the file in the | |
530 | * ASCII art, the reader sets its old page to point to the next | |
531 | * page after head. It then sets the page after head to point to | |
532 | * the old reader page. But if the writer moves the head page | |
533 | * during this operation, the reader could end up with the tail. | |
534 | * | |
535 | * We use cmpxchg to help prevent this race. We also do something | |
536 | * special with the page before head. We set the LSB to 1. | |
537 | * | |
538 | * When the writer must push the page forward, it will clear the | |
539 | * bit that points to the head page, move the head, and then set | |
540 | * the bit that points to the new head page. | |
541 | * | |
542 | * We also don't want an interrupt coming in and moving the head | |
543 | * page on another writer. Thus we use the second LSB to catch | |
544 | * that too. Thus: | |
545 | * | |
546 | * head->list->prev->next bit 1 bit 0 | |
547 | * ------- ------- | |
548 | * Normal page 0 0 | |
549 | * Points to head page 0 1 | |
550 | * New head page 1 0 | |
551 | * | |
552 | * Note we can not trust the prev pointer of the head page, because: | |
553 | * | |
554 | * +----+ +-----+ +-----+ | |
555 | * | |------>| T |---X--->| N | | |
556 | * | |<------| | | | | |
557 | * +----+ +-----+ +-----+ | |
558 | * ^ ^ | | |
559 | * | +-----+ | | | |
560 | * +----------| R |----------+ | | |
561 | * | |<-----------+ | |
562 | * +-----+ | |
563 | * | |
564 | * Key: ---X--> HEAD flag set in pointer | |
565 | * T Tail page | |
566 | * R Reader page | |
567 | * N Next page | |
568 | * | |
569 | * (see __rb_reserve_next() to see where this happens) | |
570 | * | |
571 | * What the above shows is that the reader just swapped out | |
572 | * the reader page with a page in the buffer, but before it | |
573 | * could make the new header point back to the new page added | |
574 | * it was preempted by a writer. The writer moved forward onto | |
575 | * the new page added by the reader and is about to move forward | |
576 | * again. | |
577 | * | |
578 | * You can see, it is legitimate for the previous pointer of | |
579 | * the head (or any page) not to point back to itself. But only | |
580 | * temporarially. | |
581 | */ | |
582 | ||
583 | #define RB_PAGE_NORMAL 0UL | |
584 | #define RB_PAGE_HEAD 1UL | |
585 | #define RB_PAGE_UPDATE 2UL | |
586 | ||
587 | ||
588 | #define RB_FLAG_MASK 3UL | |
589 | ||
590 | /* PAGE_MOVED is not part of the mask */ | |
591 | #define RB_PAGE_MOVED 4UL | |
592 | ||
593 | /* | |
594 | * rb_list_head - remove any bit | |
595 | */ | |
596 | static struct list_head *rb_list_head(struct list_head *list) | |
597 | { | |
598 | unsigned long val = (unsigned long)list; | |
599 | ||
600 | return (struct list_head *)(val & ~RB_FLAG_MASK); | |
601 | } | |
602 | ||
603 | /* | |
604 | * rb_is_head_page - test if the give page is the head page | |
605 | * | |
606 | * Because the reader may move the head_page pointer, we can | |
607 | * not trust what the head page is (it may be pointing to | |
608 | * the reader page). But if the next page is a header page, | |
609 | * its flags will be non zero. | |
610 | */ | |
611 | static int inline | |
612 | rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
613 | struct buffer_page *page, struct list_head *list) | |
614 | { | |
615 | unsigned long val; | |
616 | ||
617 | val = (unsigned long)list->next; | |
618 | ||
619 | if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list) | |
620 | return RB_PAGE_MOVED; | |
621 | ||
622 | return val & RB_FLAG_MASK; | |
623 | } | |
624 | ||
625 | /* | |
626 | * rb_is_reader_page | |
627 | * | |
628 | * The unique thing about the reader page, is that, if the | |
629 | * writer is ever on it, the previous pointer never points | |
630 | * back to the reader page. | |
631 | */ | |
632 | static int rb_is_reader_page(struct buffer_page *page) | |
633 | { | |
634 | struct list_head *list = page->list.prev; | |
635 | ||
636 | return rb_list_head(list->next) != &page->list; | |
637 | } | |
638 | ||
639 | /* | |
640 | * rb_set_list_to_head - set a list_head to be pointing to head. | |
641 | */ | |
642 | static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer, | |
643 | struct list_head *list) | |
644 | { | |
645 | unsigned long *ptr; | |
646 | ||
647 | ptr = (unsigned long *)&list->next; | |
648 | *ptr |= RB_PAGE_HEAD; | |
649 | *ptr &= ~RB_PAGE_UPDATE; | |
650 | } | |
651 | ||
652 | /* | |
653 | * rb_head_page_activate - sets up head page | |
654 | */ | |
655 | static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer) | |
656 | { | |
657 | struct buffer_page *head; | |
658 | ||
659 | head = cpu_buffer->head_page; | |
660 | if (!head) | |
661 | return; | |
662 | ||
663 | /* | |
664 | * Set the previous list pointer to have the HEAD flag. | |
665 | */ | |
666 | rb_set_list_to_head(cpu_buffer, head->list.prev); | |
667 | } | |
668 | ||
669 | static void rb_list_head_clear(struct list_head *list) | |
670 | { | |
671 | unsigned long *ptr = (unsigned long *)&list->next; | |
672 | ||
673 | *ptr &= ~RB_FLAG_MASK; | |
674 | } | |
675 | ||
676 | /* | |
677 | * rb_head_page_dactivate - clears head page ptr (for free list) | |
678 | */ | |
679 | static void | |
680 | rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer) | |
681 | { | |
682 | struct list_head *hd; | |
683 | ||
684 | /* Go through the whole list and clear any pointers found. */ | |
685 | rb_list_head_clear(cpu_buffer->pages); | |
686 | ||
687 | list_for_each(hd, cpu_buffer->pages) | |
688 | rb_list_head_clear(hd); | |
689 | } | |
690 | ||
691 | static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer, | |
692 | struct buffer_page *head, | |
693 | struct buffer_page *prev, | |
694 | int old_flag, int new_flag) | |
695 | { | |
696 | struct list_head *list; | |
697 | unsigned long val = (unsigned long)&head->list; | |
698 | unsigned long ret; | |
699 | ||
700 | list = &prev->list; | |
701 | ||
702 | val &= ~RB_FLAG_MASK; | |
703 | ||
704 | ret = (unsigned long)cmpxchg(&list->next, | |
705 | val | old_flag, val | new_flag); | |
706 | ||
707 | /* check if the reader took the page */ | |
708 | if ((ret & ~RB_FLAG_MASK) != val) | |
709 | return RB_PAGE_MOVED; | |
710 | ||
711 | return ret & RB_FLAG_MASK; | |
712 | } | |
713 | ||
714 | static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer, | |
715 | struct buffer_page *head, | |
716 | struct buffer_page *prev, | |
717 | int old_flag) | |
718 | { | |
719 | return rb_head_page_set(cpu_buffer, head, prev, | |
720 | old_flag, RB_PAGE_UPDATE); | |
721 | } | |
722 | ||
723 | static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer, | |
724 | struct buffer_page *head, | |
725 | struct buffer_page *prev, | |
726 | int old_flag) | |
727 | { | |
728 | return rb_head_page_set(cpu_buffer, head, prev, | |
729 | old_flag, RB_PAGE_HEAD); | |
730 | } | |
731 | ||
732 | static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer, | |
733 | struct buffer_page *head, | |
734 | struct buffer_page *prev, | |
735 | int old_flag) | |
736 | { | |
737 | return rb_head_page_set(cpu_buffer, head, prev, | |
738 | old_flag, RB_PAGE_NORMAL); | |
739 | } | |
740 | ||
741 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
742 | struct buffer_page **bpage) | |
743 | { | |
744 | struct list_head *p = rb_list_head((*bpage)->list.next); | |
745 | ||
746 | *bpage = list_entry(p, struct buffer_page, list); | |
747 | } | |
748 | ||
749 | static struct buffer_page * | |
750 | rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer) | |
751 | { | |
752 | struct buffer_page *head; | |
753 | struct buffer_page *page; | |
754 | struct list_head *list; | |
755 | int i; | |
756 | ||
757 | if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page)) | |
758 | return NULL; | |
759 | ||
760 | /* sanity check */ | |
761 | list = cpu_buffer->pages; | |
762 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list)) | |
763 | return NULL; | |
764 | ||
765 | page = head = cpu_buffer->head_page; | |
766 | /* | |
767 | * It is possible that the writer moves the header behind | |
768 | * where we started, and we miss in one loop. | |
769 | * A second loop should grab the header, but we'll do | |
770 | * three loops just because I'm paranoid. | |
771 | */ | |
772 | for (i = 0; i < 3; i++) { | |
773 | do { | |
774 | if (rb_is_head_page(cpu_buffer, page, page->list.prev)) { | |
775 | cpu_buffer->head_page = page; | |
776 | return page; | |
777 | } | |
778 | rb_inc_page(cpu_buffer, &page); | |
779 | } while (page != head); | |
780 | } | |
781 | ||
782 | RB_WARN_ON(cpu_buffer, 1); | |
783 | ||
784 | return NULL; | |
785 | } | |
786 | ||
787 | static int rb_head_page_replace(struct buffer_page *old, | |
788 | struct buffer_page *new) | |
789 | { | |
790 | unsigned long *ptr = (unsigned long *)&old->list.prev->next; | |
791 | unsigned long val; | |
792 | unsigned long ret; | |
793 | ||
794 | val = *ptr & ~RB_FLAG_MASK; | |
795 | val |= RB_PAGE_HEAD; | |
796 | ||
797 | ret = cmpxchg(ptr, val, &new->list); | |
798 | ||
799 | return ret == val; | |
800 | } | |
801 | ||
802 | /* | |
803 | * rb_tail_page_update - move the tail page forward | |
804 | * | |
805 | * Returns 1 if moved tail page, 0 if someone else did. | |
806 | */ | |
807 | static int rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer, | |
808 | struct buffer_page *tail_page, | |
809 | struct buffer_page *next_page) | |
810 | { | |
811 | struct buffer_page *old_tail; | |
812 | unsigned long old_entries; | |
813 | unsigned long old_write; | |
814 | int ret = 0; | |
815 | ||
816 | /* | |
817 | * The tail page now needs to be moved forward. | |
818 | * | |
819 | * We need to reset the tail page, but without messing | |
820 | * with possible erasing of data brought in by interrupts | |
821 | * that have moved the tail page and are currently on it. | |
822 | * | |
823 | * We add a counter to the write field to denote this. | |
824 | */ | |
825 | old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write); | |
826 | old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries); | |
827 | ||
828 | /* | |
829 | * Just make sure we have seen our old_write and synchronize | |
830 | * with any interrupts that come in. | |
831 | */ | |
832 | barrier(); | |
833 | ||
834 | /* | |
835 | * If the tail page is still the same as what we think | |
836 | * it is, then it is up to us to update the tail | |
837 | * pointer. | |
838 | */ | |
839 | if (tail_page == cpu_buffer->tail_page) { | |
840 | /* Zero the write counter */ | |
841 | unsigned long val = old_write & ~RB_WRITE_MASK; | |
842 | unsigned long eval = old_entries & ~RB_WRITE_MASK; | |
843 | ||
844 | /* | |
845 | * This will only succeed if an interrupt did | |
846 | * not come in and change it. In which case, we | |
847 | * do not want to modify it. | |
848 | */ | |
849 | local_cmpxchg(&next_page->write, old_write, val); | |
850 | local_cmpxchg(&next_page->entries, old_entries, eval); | |
851 | ||
852 | /* | |
853 | * No need to worry about races with clearing out the commit. | |
854 | * it only can increment when a commit takes place. But that | |
855 | * only happens in the outer most nested commit. | |
856 | */ | |
857 | local_set(&next_page->page->commit, 0); | |
858 | ||
859 | old_tail = cmpxchg(&cpu_buffer->tail_page, | |
860 | tail_page, next_page); | |
861 | ||
862 | if (old_tail == tail_page) | |
863 | ret = 1; | |
864 | } | |
865 | ||
866 | return ret; | |
867 | } | |
868 | ||
869 | static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer, | |
870 | struct buffer_page *bpage) | |
871 | { | |
872 | unsigned long val = (unsigned long)bpage; | |
873 | ||
874 | if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK)) | |
875 | return 1; | |
876 | ||
877 | return 0; | |
878 | } | |
879 | ||
880 | /** | |
881 | * rb_check_list - make sure a pointer to a list has the last bits zero | |
882 | */ | |
883 | static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer, | |
884 | struct list_head *list) | |
885 | { | |
886 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev)) | |
887 | return 1; | |
888 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next)) | |
889 | return 1; | |
890 | return 0; | |
891 | } | |
892 | ||
7a8e76a3 SR |
893 | /** |
894 | * check_pages - integrity check of buffer pages | |
895 | * @cpu_buffer: CPU buffer with pages to test | |
896 | * | |
c3706f00 | 897 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
898 | * been corrupted. |
899 | */ | |
900 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
901 | { | |
3adc54fa | 902 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 903 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 904 | |
77ae365e SR |
905 | rb_head_page_deactivate(cpu_buffer); |
906 | ||
3e89c7bb SR |
907 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
908 | return -1; | |
909 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
910 | return -1; | |
7a8e76a3 | 911 | |
77ae365e SR |
912 | if (rb_check_list(cpu_buffer, head)) |
913 | return -1; | |
914 | ||
044fa782 | 915 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 916 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 917 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
918 | return -1; |
919 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 920 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 921 | return -1; |
77ae365e SR |
922 | if (rb_check_list(cpu_buffer, &bpage->list)) |
923 | return -1; | |
7a8e76a3 SR |
924 | } |
925 | ||
77ae365e SR |
926 | rb_head_page_activate(cpu_buffer); |
927 | ||
7a8e76a3 SR |
928 | return 0; |
929 | } | |
930 | ||
7a8e76a3 SR |
931 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
932 | unsigned nr_pages) | |
933 | { | |
044fa782 | 934 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
935 | unsigned long addr; |
936 | LIST_HEAD(pages); | |
937 | unsigned i; | |
938 | ||
3adc54fa SR |
939 | WARN_ON(!nr_pages); |
940 | ||
7a8e76a3 | 941 | for (i = 0; i < nr_pages; i++) { |
044fa782 | 942 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
aa1e0e3b | 943 | GFP_KERNEL, cpu_to_node(cpu_buffer->cpu)); |
044fa782 | 944 | if (!bpage) |
e4c2ce82 | 945 | goto free_pages; |
77ae365e SR |
946 | |
947 | rb_check_bpage(cpu_buffer, bpage); | |
948 | ||
044fa782 | 949 | list_add(&bpage->list, &pages); |
e4c2ce82 | 950 | |
7a8e76a3 SR |
951 | addr = __get_free_page(GFP_KERNEL); |
952 | if (!addr) | |
953 | goto free_pages; | |
044fa782 SR |
954 | bpage->page = (void *)addr; |
955 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
956 | } |
957 | ||
3adc54fa SR |
958 | /* |
959 | * The ring buffer page list is a circular list that does not | |
960 | * start and end with a list head. All page list items point to | |
961 | * other pages. | |
962 | */ | |
963 | cpu_buffer->pages = pages.next; | |
964 | list_del(&pages); | |
7a8e76a3 SR |
965 | |
966 | rb_check_pages(cpu_buffer); | |
967 | ||
968 | return 0; | |
969 | ||
970 | free_pages: | |
044fa782 SR |
971 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
972 | list_del_init(&bpage->list); | |
973 | free_buffer_page(bpage); | |
7a8e76a3 SR |
974 | } |
975 | return -ENOMEM; | |
976 | } | |
977 | ||
978 | static struct ring_buffer_per_cpu * | |
979 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) | |
980 | { | |
981 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 982 | struct buffer_page *bpage; |
d769041f | 983 | unsigned long addr; |
7a8e76a3 SR |
984 | int ret; |
985 | ||
986 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
987 | GFP_KERNEL, cpu_to_node(cpu)); | |
988 | if (!cpu_buffer) | |
989 | return NULL; | |
990 | ||
991 | cpu_buffer->cpu = cpu; | |
992 | cpu_buffer->buffer = buffer; | |
f83c9d0f | 993 | spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 994 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
3e03fb7f | 995 | cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; |
7a8e76a3 | 996 | |
044fa782 | 997 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 998 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 999 | if (!bpage) |
e4c2ce82 SR |
1000 | goto fail_free_buffer; |
1001 | ||
77ae365e SR |
1002 | rb_check_bpage(cpu_buffer, bpage); |
1003 | ||
044fa782 | 1004 | cpu_buffer->reader_page = bpage; |
d769041f SR |
1005 | addr = __get_free_page(GFP_KERNEL); |
1006 | if (!addr) | |
e4c2ce82 | 1007 | goto fail_free_reader; |
044fa782 SR |
1008 | bpage->page = (void *)addr; |
1009 | rb_init_page(bpage->page); | |
e4c2ce82 | 1010 | |
d769041f | 1011 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 1012 | |
7a8e76a3 SR |
1013 | ret = rb_allocate_pages(cpu_buffer, buffer->pages); |
1014 | if (ret < 0) | |
d769041f | 1015 | goto fail_free_reader; |
7a8e76a3 SR |
1016 | |
1017 | cpu_buffer->head_page | |
3adc54fa | 1018 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 1019 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 | 1020 | |
77ae365e SR |
1021 | rb_head_page_activate(cpu_buffer); |
1022 | ||
7a8e76a3 SR |
1023 | return cpu_buffer; |
1024 | ||
d769041f SR |
1025 | fail_free_reader: |
1026 | free_buffer_page(cpu_buffer->reader_page); | |
1027 | ||
7a8e76a3 SR |
1028 | fail_free_buffer: |
1029 | kfree(cpu_buffer); | |
1030 | return NULL; | |
1031 | } | |
1032 | ||
1033 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
1034 | { | |
3adc54fa | 1035 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1036 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1037 | |
d769041f SR |
1038 | free_buffer_page(cpu_buffer->reader_page); |
1039 | ||
77ae365e SR |
1040 | rb_head_page_deactivate(cpu_buffer); |
1041 | ||
3adc54fa SR |
1042 | if (head) { |
1043 | list_for_each_entry_safe(bpage, tmp, head, list) { | |
1044 | list_del_init(&bpage->list); | |
1045 | free_buffer_page(bpage); | |
1046 | } | |
1047 | bpage = list_entry(head, struct buffer_page, list); | |
044fa782 | 1048 | free_buffer_page(bpage); |
7a8e76a3 | 1049 | } |
3adc54fa | 1050 | |
7a8e76a3 SR |
1051 | kfree(cpu_buffer); |
1052 | } | |
1053 | ||
59222efe | 1054 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
1055 | static int rb_cpu_notify(struct notifier_block *self, |
1056 | unsigned long action, void *hcpu); | |
554f786e SR |
1057 | #endif |
1058 | ||
7a8e76a3 SR |
1059 | /** |
1060 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 1061 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
1062 | * @flags: attributes to set for the ring buffer. |
1063 | * | |
1064 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
1065 | * flag. This flag means that the buffer will overwrite old data | |
1066 | * when the buffer wraps. If this flag is not set, the buffer will | |
1067 | * drop data when the tail hits the head. | |
1068 | */ | |
1f8a6a10 PZ |
1069 | struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
1070 | struct lock_class_key *key) | |
7a8e76a3 SR |
1071 | { |
1072 | struct ring_buffer *buffer; | |
1073 | int bsize; | |
1074 | int cpu; | |
1075 | ||
1076 | /* keep it in its own cache line */ | |
1077 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
1078 | GFP_KERNEL); | |
1079 | if (!buffer) | |
1080 | return NULL; | |
1081 | ||
9e01c1b7 RR |
1082 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
1083 | goto fail_free_buffer; | |
1084 | ||
7a8e76a3 SR |
1085 | buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
1086 | buffer->flags = flags; | |
37886f6a | 1087 | buffer->clock = trace_clock_local; |
1f8a6a10 | 1088 | buffer->reader_lock_key = key; |
7a8e76a3 SR |
1089 | |
1090 | /* need at least two pages */ | |
5f78abee SR |
1091 | if (buffer->pages < 2) |
1092 | buffer->pages = 2; | |
7a8e76a3 | 1093 | |
3bf832ce FW |
1094 | /* |
1095 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
1096 | * in early initcall, it will not be notified of secondary cpus. | |
1097 | * In that off case, we need to allocate for all possible cpus. | |
1098 | */ | |
1099 | #ifdef CONFIG_HOTPLUG_CPU | |
554f786e SR |
1100 | get_online_cpus(); |
1101 | cpumask_copy(buffer->cpumask, cpu_online_mask); | |
3bf832ce FW |
1102 | #else |
1103 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
1104 | #endif | |
7a8e76a3 SR |
1105 | buffer->cpus = nr_cpu_ids; |
1106 | ||
1107 | bsize = sizeof(void *) * nr_cpu_ids; | |
1108 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
1109 | GFP_KERNEL); | |
1110 | if (!buffer->buffers) | |
9e01c1b7 | 1111 | goto fail_free_cpumask; |
7a8e76a3 SR |
1112 | |
1113 | for_each_buffer_cpu(buffer, cpu) { | |
1114 | buffer->buffers[cpu] = | |
1115 | rb_allocate_cpu_buffer(buffer, cpu); | |
1116 | if (!buffer->buffers[cpu]) | |
1117 | goto fail_free_buffers; | |
1118 | } | |
1119 | ||
59222efe | 1120 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1121 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
1122 | buffer->cpu_notify.priority = 0; | |
1123 | register_cpu_notifier(&buffer->cpu_notify); | |
1124 | #endif | |
1125 | ||
1126 | put_online_cpus(); | |
7a8e76a3 SR |
1127 | mutex_init(&buffer->mutex); |
1128 | ||
1129 | return buffer; | |
1130 | ||
1131 | fail_free_buffers: | |
1132 | for_each_buffer_cpu(buffer, cpu) { | |
1133 | if (buffer->buffers[cpu]) | |
1134 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1135 | } | |
1136 | kfree(buffer->buffers); | |
1137 | ||
9e01c1b7 RR |
1138 | fail_free_cpumask: |
1139 | free_cpumask_var(buffer->cpumask); | |
554f786e | 1140 | put_online_cpus(); |
9e01c1b7 | 1141 | |
7a8e76a3 SR |
1142 | fail_free_buffer: |
1143 | kfree(buffer); | |
1144 | return NULL; | |
1145 | } | |
1f8a6a10 | 1146 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
1147 | |
1148 | /** | |
1149 | * ring_buffer_free - free a ring buffer. | |
1150 | * @buffer: the buffer to free. | |
1151 | */ | |
1152 | void | |
1153 | ring_buffer_free(struct ring_buffer *buffer) | |
1154 | { | |
1155 | int cpu; | |
1156 | ||
554f786e SR |
1157 | get_online_cpus(); |
1158 | ||
59222efe | 1159 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1160 | unregister_cpu_notifier(&buffer->cpu_notify); |
1161 | #endif | |
1162 | ||
7a8e76a3 SR |
1163 | for_each_buffer_cpu(buffer, cpu) |
1164 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1165 | ||
554f786e SR |
1166 | put_online_cpus(); |
1167 | ||
9e01c1b7 RR |
1168 | free_cpumask_var(buffer->cpumask); |
1169 | ||
7a8e76a3 SR |
1170 | kfree(buffer); |
1171 | } | |
c4f50183 | 1172 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 1173 | |
37886f6a SR |
1174 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
1175 | u64 (*clock)(void)) | |
1176 | { | |
1177 | buffer->clock = clock; | |
1178 | } | |
1179 | ||
7a8e76a3 SR |
1180 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
1181 | ||
1182 | static void | |
1183 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages) | |
1184 | { | |
044fa782 | 1185 | struct buffer_page *bpage; |
7a8e76a3 SR |
1186 | struct list_head *p; |
1187 | unsigned i; | |
1188 | ||
1189 | atomic_inc(&cpu_buffer->record_disabled); | |
1190 | synchronize_sched(); | |
1191 | ||
77ae365e SR |
1192 | rb_head_page_deactivate(cpu_buffer); |
1193 | ||
7a8e76a3 | 1194 | for (i = 0; i < nr_pages; i++) { |
3adc54fa | 1195 | if (RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages))) |
3e89c7bb | 1196 | return; |
3adc54fa | 1197 | p = cpu_buffer->pages->next; |
044fa782 SR |
1198 | bpage = list_entry(p, struct buffer_page, list); |
1199 | list_del_init(&bpage->list); | |
1200 | free_buffer_page(bpage); | |
7a8e76a3 | 1201 | } |
3adc54fa | 1202 | if (RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages))) |
3e89c7bb | 1203 | return; |
7a8e76a3 SR |
1204 | |
1205 | rb_reset_cpu(cpu_buffer); | |
1206 | ||
1207 | rb_check_pages(cpu_buffer); | |
1208 | ||
1209 | atomic_dec(&cpu_buffer->record_disabled); | |
1210 | ||
1211 | } | |
1212 | ||
1213 | static void | |
1214 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
1215 | struct list_head *pages, unsigned nr_pages) | |
1216 | { | |
044fa782 | 1217 | struct buffer_page *bpage; |
7a8e76a3 SR |
1218 | struct list_head *p; |
1219 | unsigned i; | |
1220 | ||
1221 | atomic_inc(&cpu_buffer->record_disabled); | |
1222 | synchronize_sched(); | |
1223 | ||
77ae365e SR |
1224 | spin_lock_irq(&cpu_buffer->reader_lock); |
1225 | rb_head_page_deactivate(cpu_buffer); | |
1226 | ||
7a8e76a3 | 1227 | for (i = 0; i < nr_pages; i++) { |
3e89c7bb SR |
1228 | if (RB_WARN_ON(cpu_buffer, list_empty(pages))) |
1229 | return; | |
7a8e76a3 | 1230 | p = pages->next; |
044fa782 SR |
1231 | bpage = list_entry(p, struct buffer_page, list); |
1232 | list_del_init(&bpage->list); | |
3adc54fa | 1233 | list_add_tail(&bpage->list, cpu_buffer->pages); |
7a8e76a3 SR |
1234 | } |
1235 | rb_reset_cpu(cpu_buffer); | |
77ae365e | 1236 | spin_unlock_irq(&cpu_buffer->reader_lock); |
7a8e76a3 SR |
1237 | |
1238 | rb_check_pages(cpu_buffer); | |
1239 | ||
1240 | atomic_dec(&cpu_buffer->record_disabled); | |
1241 | } | |
1242 | ||
1243 | /** | |
1244 | * ring_buffer_resize - resize the ring buffer | |
1245 | * @buffer: the buffer to resize. | |
1246 | * @size: the new size. | |
1247 | * | |
1248 | * The tracer is responsible for making sure that the buffer is | |
1249 | * not being used while changing the size. | |
1250 | * Note: We may be able to change the above requirement by using | |
1251 | * RCU synchronizations. | |
1252 | * | |
1253 | * Minimum size is 2 * BUF_PAGE_SIZE. | |
1254 | * | |
1255 | * Returns -1 on failure. | |
1256 | */ | |
1257 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) | |
1258 | { | |
1259 | struct ring_buffer_per_cpu *cpu_buffer; | |
1260 | unsigned nr_pages, rm_pages, new_pages; | |
044fa782 | 1261 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
1262 | unsigned long buffer_size; |
1263 | unsigned long addr; | |
1264 | LIST_HEAD(pages); | |
1265 | int i, cpu; | |
1266 | ||
ee51a1de IM |
1267 | /* |
1268 | * Always succeed at resizing a non-existent buffer: | |
1269 | */ | |
1270 | if (!buffer) | |
1271 | return size; | |
1272 | ||
7a8e76a3 SR |
1273 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
1274 | size *= BUF_PAGE_SIZE; | |
1275 | buffer_size = buffer->pages * BUF_PAGE_SIZE; | |
1276 | ||
1277 | /* we need a minimum of two pages */ | |
1278 | if (size < BUF_PAGE_SIZE * 2) | |
1279 | size = BUF_PAGE_SIZE * 2; | |
1280 | ||
1281 | if (size == buffer_size) | |
1282 | return size; | |
1283 | ||
1284 | mutex_lock(&buffer->mutex); | |
554f786e | 1285 | get_online_cpus(); |
7a8e76a3 SR |
1286 | |
1287 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
1288 | ||
1289 | if (size < buffer_size) { | |
1290 | ||
1291 | /* easy case, just free pages */ | |
554f786e SR |
1292 | if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) |
1293 | goto out_fail; | |
7a8e76a3 SR |
1294 | |
1295 | rm_pages = buffer->pages - nr_pages; | |
1296 | ||
1297 | for_each_buffer_cpu(buffer, cpu) { | |
1298 | cpu_buffer = buffer->buffers[cpu]; | |
1299 | rb_remove_pages(cpu_buffer, rm_pages); | |
1300 | } | |
1301 | goto out; | |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * This is a bit more difficult. We only want to add pages | |
1306 | * when we can allocate enough for all CPUs. We do this | |
1307 | * by allocating all the pages and storing them on a local | |
1308 | * link list. If we succeed in our allocation, then we | |
1309 | * add these pages to the cpu_buffers. Otherwise we just free | |
1310 | * them all and return -ENOMEM; | |
1311 | */ | |
554f786e SR |
1312 | if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) |
1313 | goto out_fail; | |
f536aafc | 1314 | |
7a8e76a3 SR |
1315 | new_pages = nr_pages - buffer->pages; |
1316 | ||
1317 | for_each_buffer_cpu(buffer, cpu) { | |
1318 | for (i = 0; i < new_pages; i++) { | |
044fa782 | 1319 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), |
e4c2ce82 SR |
1320 | cache_line_size()), |
1321 | GFP_KERNEL, cpu_to_node(cpu)); | |
044fa782 | 1322 | if (!bpage) |
e4c2ce82 | 1323 | goto free_pages; |
044fa782 | 1324 | list_add(&bpage->list, &pages); |
7a8e76a3 SR |
1325 | addr = __get_free_page(GFP_KERNEL); |
1326 | if (!addr) | |
1327 | goto free_pages; | |
044fa782 SR |
1328 | bpage->page = (void *)addr; |
1329 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
1330 | } |
1331 | } | |
1332 | ||
1333 | for_each_buffer_cpu(buffer, cpu) { | |
1334 | cpu_buffer = buffer->buffers[cpu]; | |
1335 | rb_insert_pages(cpu_buffer, &pages, new_pages); | |
1336 | } | |
1337 | ||
554f786e SR |
1338 | if (RB_WARN_ON(buffer, !list_empty(&pages))) |
1339 | goto out_fail; | |
7a8e76a3 SR |
1340 | |
1341 | out: | |
1342 | buffer->pages = nr_pages; | |
554f786e | 1343 | put_online_cpus(); |
7a8e76a3 SR |
1344 | mutex_unlock(&buffer->mutex); |
1345 | ||
1346 | return size; | |
1347 | ||
1348 | free_pages: | |
044fa782 SR |
1349 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
1350 | list_del_init(&bpage->list); | |
1351 | free_buffer_page(bpage); | |
7a8e76a3 | 1352 | } |
554f786e | 1353 | put_online_cpus(); |
641d2f63 | 1354 | mutex_unlock(&buffer->mutex); |
7a8e76a3 | 1355 | return -ENOMEM; |
554f786e SR |
1356 | |
1357 | /* | |
1358 | * Something went totally wrong, and we are too paranoid | |
1359 | * to even clean up the mess. | |
1360 | */ | |
1361 | out_fail: | |
1362 | put_online_cpus(); | |
1363 | mutex_unlock(&buffer->mutex); | |
1364 | return -1; | |
7a8e76a3 | 1365 | } |
c4f50183 | 1366 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 1367 | |
8789a9e7 | 1368 | static inline void * |
044fa782 | 1369 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 1370 | { |
044fa782 | 1371 | return bpage->data + index; |
8789a9e7 SR |
1372 | } |
1373 | ||
044fa782 | 1374 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 1375 | { |
044fa782 | 1376 | return bpage->page->data + index; |
7a8e76a3 SR |
1377 | } |
1378 | ||
1379 | static inline struct ring_buffer_event * | |
d769041f | 1380 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1381 | { |
6f807acd SR |
1382 | return __rb_page_index(cpu_buffer->reader_page, |
1383 | cpu_buffer->reader_page->read); | |
1384 | } | |
1385 | ||
7a8e76a3 SR |
1386 | static inline struct ring_buffer_event * |
1387 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
1388 | { | |
6f807acd | 1389 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
1390 | } |
1391 | ||
77ae365e | 1392 | static inline unsigned long rb_page_write(struct buffer_page *bpage) |
bf41a158 | 1393 | { |
77ae365e | 1394 | return local_read(&bpage->write) & RB_WRITE_MASK; |
bf41a158 SR |
1395 | } |
1396 | ||
1397 | static inline unsigned rb_page_commit(struct buffer_page *bpage) | |
1398 | { | |
abc9b56d | 1399 | return local_read(&bpage->page->commit); |
bf41a158 SR |
1400 | } |
1401 | ||
77ae365e SR |
1402 | static inline unsigned long rb_page_entries(struct buffer_page *bpage) |
1403 | { | |
1404 | return local_read(&bpage->entries) & RB_WRITE_MASK; | |
1405 | } | |
1406 | ||
bf41a158 SR |
1407 | /* Size is determined by what has been commited */ |
1408 | static inline unsigned rb_page_size(struct buffer_page *bpage) | |
1409 | { | |
1410 | return rb_page_commit(bpage); | |
1411 | } | |
1412 | ||
1413 | static inline unsigned | |
1414 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
1415 | { | |
1416 | return rb_page_commit(cpu_buffer->commit_page); | |
1417 | } | |
1418 | ||
bf41a158 SR |
1419 | static inline unsigned |
1420 | rb_event_index(struct ring_buffer_event *event) | |
1421 | { | |
1422 | unsigned long addr = (unsigned long)event; | |
1423 | ||
22f470f8 | 1424 | return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE; |
bf41a158 SR |
1425 | } |
1426 | ||
0f0c85fc | 1427 | static inline int |
fa743953 SR |
1428 | rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1429 | struct ring_buffer_event *event) | |
bf41a158 SR |
1430 | { |
1431 | unsigned long addr = (unsigned long)event; | |
1432 | unsigned long index; | |
1433 | ||
1434 | index = rb_event_index(event); | |
1435 | addr &= PAGE_MASK; | |
1436 | ||
1437 | return cpu_buffer->commit_page->page == (void *)addr && | |
1438 | rb_commit_index(cpu_buffer) == index; | |
1439 | } | |
1440 | ||
34a148bf | 1441 | static void |
bf41a158 | 1442 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1443 | { |
77ae365e SR |
1444 | unsigned long max_count; |
1445 | ||
bf41a158 SR |
1446 | /* |
1447 | * We only race with interrupts and NMIs on this CPU. | |
1448 | * If we own the commit event, then we can commit | |
1449 | * all others that interrupted us, since the interruptions | |
1450 | * are in stack format (they finish before they come | |
1451 | * back to us). This allows us to do a simple loop to | |
1452 | * assign the commit to the tail. | |
1453 | */ | |
a8ccf1d6 | 1454 | again: |
77ae365e SR |
1455 | max_count = cpu_buffer->buffer->pages * 100; |
1456 | ||
bf41a158 | 1457 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
77ae365e SR |
1458 | if (RB_WARN_ON(cpu_buffer, !(--max_count))) |
1459 | return; | |
1460 | if (RB_WARN_ON(cpu_buffer, | |
1461 | rb_is_reader_page(cpu_buffer->tail_page))) | |
1462 | return; | |
1463 | local_set(&cpu_buffer->commit_page->page->commit, | |
1464 | rb_page_write(cpu_buffer->commit_page)); | |
bf41a158 | 1465 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); |
abc9b56d SR |
1466 | cpu_buffer->write_stamp = |
1467 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1468 | /* add barrier to keep gcc from optimizing too much */ |
1469 | barrier(); | |
1470 | } | |
1471 | while (rb_commit_index(cpu_buffer) != | |
1472 | rb_page_write(cpu_buffer->commit_page)) { | |
77ae365e SR |
1473 | |
1474 | local_set(&cpu_buffer->commit_page->page->commit, | |
1475 | rb_page_write(cpu_buffer->commit_page)); | |
1476 | RB_WARN_ON(cpu_buffer, | |
1477 | local_read(&cpu_buffer->commit_page->page->commit) & | |
1478 | ~RB_WRITE_MASK); | |
bf41a158 SR |
1479 | barrier(); |
1480 | } | |
a8ccf1d6 SR |
1481 | |
1482 | /* again, keep gcc from optimizing */ | |
1483 | barrier(); | |
1484 | ||
1485 | /* | |
1486 | * If an interrupt came in just after the first while loop | |
1487 | * and pushed the tail page forward, we will be left with | |
1488 | * a dangling commit that will never go forward. | |
1489 | */ | |
1490 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1491 | goto again; | |
7a8e76a3 SR |
1492 | } |
1493 | ||
d769041f | 1494 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1495 | { |
abc9b56d | 1496 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1497 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1498 | } |
1499 | ||
34a148bf | 1500 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1501 | { |
1502 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1503 | ||
1504 | /* | |
1505 | * The iterator could be on the reader page (it starts there). | |
1506 | * But the head could have moved, since the reader was | |
1507 | * found. Check for this case and assign the iterator | |
1508 | * to the head page instead of next. | |
1509 | */ | |
1510 | if (iter->head_page == cpu_buffer->reader_page) | |
77ae365e | 1511 | iter->head_page = rb_set_head_page(cpu_buffer); |
d769041f SR |
1512 | else |
1513 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1514 | ||
abc9b56d | 1515 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1516 | iter->head = 0; |
1517 | } | |
1518 | ||
1519 | /** | |
1520 | * ring_buffer_update_event - update event type and data | |
1521 | * @event: the even to update | |
1522 | * @type: the type of event | |
1523 | * @length: the size of the event field in the ring buffer | |
1524 | * | |
1525 | * Update the type and data fields of the event. The length | |
1526 | * is the actual size that is written to the ring buffer, | |
1527 | * and with this, we can determine what to place into the | |
1528 | * data field. | |
1529 | */ | |
34a148bf | 1530 | static void |
7a8e76a3 SR |
1531 | rb_update_event(struct ring_buffer_event *event, |
1532 | unsigned type, unsigned length) | |
1533 | { | |
334d4169 | 1534 | event->type_len = type; |
7a8e76a3 SR |
1535 | |
1536 | switch (type) { | |
1537 | ||
1538 | case RINGBUF_TYPE_PADDING: | |
7a8e76a3 | 1539 | case RINGBUF_TYPE_TIME_EXTEND: |
7a8e76a3 | 1540 | case RINGBUF_TYPE_TIME_STAMP: |
7a8e76a3 SR |
1541 | break; |
1542 | ||
334d4169 | 1543 | case 0: |
7a8e76a3 | 1544 | length -= RB_EVNT_HDR_SIZE; |
334d4169 | 1545 | if (length > RB_MAX_SMALL_DATA) |
7a8e76a3 | 1546 | event->array[0] = length; |
334d4169 LJ |
1547 | else |
1548 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
7a8e76a3 SR |
1549 | break; |
1550 | default: | |
1551 | BUG(); | |
1552 | } | |
1553 | } | |
1554 | ||
77ae365e SR |
1555 | /* |
1556 | * rb_handle_head_page - writer hit the head page | |
1557 | * | |
1558 | * Returns: +1 to retry page | |
1559 | * 0 to continue | |
1560 | * -1 on error | |
1561 | */ | |
1562 | static int | |
1563 | rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
1564 | struct buffer_page *tail_page, | |
1565 | struct buffer_page *next_page) | |
1566 | { | |
1567 | struct buffer_page *new_head; | |
1568 | int entries; | |
1569 | int type; | |
1570 | int ret; | |
1571 | ||
1572 | entries = rb_page_entries(next_page); | |
1573 | ||
1574 | /* | |
1575 | * The hard part is here. We need to move the head | |
1576 | * forward, and protect against both readers on | |
1577 | * other CPUs and writers coming in via interrupts. | |
1578 | */ | |
1579 | type = rb_head_page_set_update(cpu_buffer, next_page, tail_page, | |
1580 | RB_PAGE_HEAD); | |
1581 | ||
1582 | /* | |
1583 | * type can be one of four: | |
1584 | * NORMAL - an interrupt already moved it for us | |
1585 | * HEAD - we are the first to get here. | |
1586 | * UPDATE - we are the interrupt interrupting | |
1587 | * a current move. | |
1588 | * MOVED - a reader on another CPU moved the next | |
1589 | * pointer to its reader page. Give up | |
1590 | * and try again. | |
1591 | */ | |
1592 | ||
1593 | switch (type) { | |
1594 | case RB_PAGE_HEAD: | |
1595 | /* | |
1596 | * We changed the head to UPDATE, thus | |
1597 | * it is our responsibility to update | |
1598 | * the counters. | |
1599 | */ | |
1600 | local_add(entries, &cpu_buffer->overrun); | |
1601 | ||
1602 | /* | |
1603 | * The entries will be zeroed out when we move the | |
1604 | * tail page. | |
1605 | */ | |
1606 | ||
1607 | /* still more to do */ | |
1608 | break; | |
1609 | ||
1610 | case RB_PAGE_UPDATE: | |
1611 | /* | |
1612 | * This is an interrupt that interrupt the | |
1613 | * previous update. Still more to do. | |
1614 | */ | |
1615 | break; | |
1616 | case RB_PAGE_NORMAL: | |
1617 | /* | |
1618 | * An interrupt came in before the update | |
1619 | * and processed this for us. | |
1620 | * Nothing left to do. | |
1621 | */ | |
1622 | return 1; | |
1623 | case RB_PAGE_MOVED: | |
1624 | /* | |
1625 | * The reader is on another CPU and just did | |
1626 | * a swap with our next_page. | |
1627 | * Try again. | |
1628 | */ | |
1629 | return 1; | |
1630 | default: | |
1631 | RB_WARN_ON(cpu_buffer, 1); /* WTF??? */ | |
1632 | return -1; | |
1633 | } | |
1634 | ||
1635 | /* | |
1636 | * Now that we are here, the old head pointer is | |
1637 | * set to UPDATE. This will keep the reader from | |
1638 | * swapping the head page with the reader page. | |
1639 | * The reader (on another CPU) will spin till | |
1640 | * we are finished. | |
1641 | * | |
1642 | * We just need to protect against interrupts | |
1643 | * doing the job. We will set the next pointer | |
1644 | * to HEAD. After that, we set the old pointer | |
1645 | * to NORMAL, but only if it was HEAD before. | |
1646 | * otherwise we are an interrupt, and only | |
1647 | * want the outer most commit to reset it. | |
1648 | */ | |
1649 | new_head = next_page; | |
1650 | rb_inc_page(cpu_buffer, &new_head); | |
1651 | ||
1652 | ret = rb_head_page_set_head(cpu_buffer, new_head, next_page, | |
1653 | RB_PAGE_NORMAL); | |
1654 | ||
1655 | /* | |
1656 | * Valid returns are: | |
1657 | * HEAD - an interrupt came in and already set it. | |
1658 | * NORMAL - One of two things: | |
1659 | * 1) We really set it. | |
1660 | * 2) A bunch of interrupts came in and moved | |
1661 | * the page forward again. | |
1662 | */ | |
1663 | switch (ret) { | |
1664 | case RB_PAGE_HEAD: | |
1665 | case RB_PAGE_NORMAL: | |
1666 | /* OK */ | |
1667 | break; | |
1668 | default: | |
1669 | RB_WARN_ON(cpu_buffer, 1); | |
1670 | return -1; | |
1671 | } | |
1672 | ||
1673 | /* | |
1674 | * It is possible that an interrupt came in, | |
1675 | * set the head up, then more interrupts came in | |
1676 | * and moved it again. When we get back here, | |
1677 | * the page would have been set to NORMAL but we | |
1678 | * just set it back to HEAD. | |
1679 | * | |
1680 | * How do you detect this? Well, if that happened | |
1681 | * the tail page would have moved. | |
1682 | */ | |
1683 | if (ret == RB_PAGE_NORMAL) { | |
1684 | /* | |
1685 | * If the tail had moved passed next, then we need | |
1686 | * to reset the pointer. | |
1687 | */ | |
1688 | if (cpu_buffer->tail_page != tail_page && | |
1689 | cpu_buffer->tail_page != next_page) | |
1690 | rb_head_page_set_normal(cpu_buffer, new_head, | |
1691 | next_page, | |
1692 | RB_PAGE_HEAD); | |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * If this was the outer most commit (the one that | |
1697 | * changed the original pointer from HEAD to UPDATE), | |
1698 | * then it is up to us to reset it to NORMAL. | |
1699 | */ | |
1700 | if (type == RB_PAGE_HEAD) { | |
1701 | ret = rb_head_page_set_normal(cpu_buffer, next_page, | |
1702 | tail_page, | |
1703 | RB_PAGE_UPDATE); | |
1704 | if (RB_WARN_ON(cpu_buffer, | |
1705 | ret != RB_PAGE_UPDATE)) | |
1706 | return -1; | |
1707 | } | |
1708 | ||
1709 | return 0; | |
1710 | } | |
1711 | ||
34a148bf | 1712 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
1713 | { |
1714 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
1715 | ||
1716 | /* zero length can cause confusions */ | |
1717 | if (!length) | |
1718 | length = 1; | |
1719 | ||
1720 | if (length > RB_MAX_SMALL_DATA) | |
1721 | length += sizeof(event.array[0]); | |
1722 | ||
1723 | length += RB_EVNT_HDR_SIZE; | |
1724 | length = ALIGN(length, RB_ALIGNMENT); | |
1725 | ||
1726 | return length; | |
1727 | } | |
1728 | ||
c7b09308 SR |
1729 | static inline void |
1730 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
1731 | struct buffer_page *tail_page, | |
1732 | unsigned long tail, unsigned long length) | |
1733 | { | |
1734 | struct ring_buffer_event *event; | |
1735 | ||
1736 | /* | |
1737 | * Only the event that crossed the page boundary | |
1738 | * must fill the old tail_page with padding. | |
1739 | */ | |
1740 | if (tail >= BUF_PAGE_SIZE) { | |
1741 | local_sub(length, &tail_page->write); | |
1742 | return; | |
1743 | } | |
1744 | ||
1745 | event = __rb_page_index(tail_page, tail); | |
b0b7065b | 1746 | kmemcheck_annotate_bitfield(event, bitfield); |
c7b09308 SR |
1747 | |
1748 | /* | |
1749 | * If this event is bigger than the minimum size, then | |
1750 | * we need to be careful that we don't subtract the | |
1751 | * write counter enough to allow another writer to slip | |
1752 | * in on this page. | |
1753 | * We put in a discarded commit instead, to make sure | |
1754 | * that this space is not used again. | |
1755 | * | |
1756 | * If we are less than the minimum size, we don't need to | |
1757 | * worry about it. | |
1758 | */ | |
1759 | if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) { | |
1760 | /* No room for any events */ | |
1761 | ||
1762 | /* Mark the rest of the page with padding */ | |
1763 | rb_event_set_padding(event); | |
1764 | ||
1765 | /* Set the write back to the previous setting */ | |
1766 | local_sub(length, &tail_page->write); | |
1767 | return; | |
1768 | } | |
1769 | ||
1770 | /* Put in a discarded event */ | |
1771 | event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE; | |
1772 | event->type_len = RINGBUF_TYPE_PADDING; | |
1773 | /* time delta must be non zero */ | |
1774 | event->time_delta = 1; | |
1775 | /* Account for this as an entry */ | |
1776 | local_inc(&tail_page->entries); | |
1777 | local_inc(&cpu_buffer->entries); | |
1778 | ||
1779 | /* Set write to end of buffer */ | |
1780 | length = (tail + length) - BUF_PAGE_SIZE; | |
1781 | local_sub(length, &tail_page->write); | |
1782 | } | |
6634ff26 | 1783 | |
7a8e76a3 | 1784 | static struct ring_buffer_event * |
6634ff26 SR |
1785 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
1786 | unsigned long length, unsigned long tail, | |
1787 | struct buffer_page *commit_page, | |
1788 | struct buffer_page *tail_page, u64 *ts) | |
7a8e76a3 | 1789 | { |
7a8e76a3 | 1790 | struct ring_buffer *buffer = cpu_buffer->buffer; |
77ae365e SR |
1791 | struct buffer_page *next_page; |
1792 | int ret; | |
aa20ae84 SR |
1793 | |
1794 | next_page = tail_page; | |
1795 | ||
aa20ae84 SR |
1796 | rb_inc_page(cpu_buffer, &next_page); |
1797 | ||
aa20ae84 SR |
1798 | /* |
1799 | * If for some reason, we had an interrupt storm that made | |
1800 | * it all the way around the buffer, bail, and warn | |
1801 | * about it. | |
1802 | */ | |
1803 | if (unlikely(next_page == commit_page)) { | |
77ae365e | 1804 | local_inc(&cpu_buffer->commit_overrun); |
aa20ae84 SR |
1805 | goto out_reset; |
1806 | } | |
1807 | ||
77ae365e SR |
1808 | /* |
1809 | * This is where the fun begins! | |
1810 | * | |
1811 | * We are fighting against races between a reader that | |
1812 | * could be on another CPU trying to swap its reader | |
1813 | * page with the buffer head. | |
1814 | * | |
1815 | * We are also fighting against interrupts coming in and | |
1816 | * moving the head or tail on us as well. | |
1817 | * | |
1818 | * If the next page is the head page then we have filled | |
1819 | * the buffer, unless the commit page is still on the | |
1820 | * reader page. | |
1821 | */ | |
1822 | if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) { | |
aa20ae84 | 1823 | |
77ae365e SR |
1824 | /* |
1825 | * If the commit is not on the reader page, then | |
1826 | * move the header page. | |
1827 | */ | |
1828 | if (!rb_is_reader_page(cpu_buffer->commit_page)) { | |
1829 | /* | |
1830 | * If we are not in overwrite mode, | |
1831 | * this is easy, just stop here. | |
1832 | */ | |
1833 | if (!(buffer->flags & RB_FL_OVERWRITE)) | |
1834 | goto out_reset; | |
1835 | ||
1836 | ret = rb_handle_head_page(cpu_buffer, | |
1837 | tail_page, | |
1838 | next_page); | |
1839 | if (ret < 0) | |
1840 | goto out_reset; | |
1841 | if (ret) | |
1842 | goto out_again; | |
1843 | } else { | |
1844 | /* | |
1845 | * We need to be careful here too. The | |
1846 | * commit page could still be on the reader | |
1847 | * page. We could have a small buffer, and | |
1848 | * have filled up the buffer with events | |
1849 | * from interrupts and such, and wrapped. | |
1850 | * | |
1851 | * Note, if the tail page is also the on the | |
1852 | * reader_page, we let it move out. | |
1853 | */ | |
1854 | if (unlikely((cpu_buffer->commit_page != | |
1855 | cpu_buffer->tail_page) && | |
1856 | (cpu_buffer->commit_page == | |
1857 | cpu_buffer->reader_page))) { | |
1858 | local_inc(&cpu_buffer->commit_overrun); | |
1859 | goto out_reset; | |
1860 | } | |
aa20ae84 SR |
1861 | } |
1862 | } | |
1863 | ||
77ae365e SR |
1864 | ret = rb_tail_page_update(cpu_buffer, tail_page, next_page); |
1865 | if (ret) { | |
1866 | /* | |
1867 | * Nested commits always have zero deltas, so | |
1868 | * just reread the time stamp | |
1869 | */ | |
88eb0125 | 1870 | *ts = rb_time_stamp(buffer, cpu_buffer->cpu); |
77ae365e | 1871 | next_page->page->time_stamp = *ts; |
aa20ae84 SR |
1872 | } |
1873 | ||
77ae365e | 1874 | out_again: |
aa20ae84 | 1875 | |
77ae365e | 1876 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
aa20ae84 SR |
1877 | |
1878 | /* fail and let the caller try again */ | |
1879 | return ERR_PTR(-EAGAIN); | |
1880 | ||
45141d46 | 1881 | out_reset: |
6f3b3440 | 1882 | /* reset write */ |
c7b09308 | 1883 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
6f3b3440 | 1884 | |
bf41a158 | 1885 | return NULL; |
7a8e76a3 SR |
1886 | } |
1887 | ||
6634ff26 SR |
1888 | static struct ring_buffer_event * |
1889 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
1890 | unsigned type, unsigned long length, u64 *ts) | |
1891 | { | |
1892 | struct buffer_page *tail_page, *commit_page; | |
1893 | struct ring_buffer_event *event; | |
1894 | unsigned long tail, write; | |
1895 | ||
1896 | commit_page = cpu_buffer->commit_page; | |
1897 | /* we just need to protect against interrupts */ | |
1898 | barrier(); | |
1899 | tail_page = cpu_buffer->tail_page; | |
1900 | write = local_add_return(length, &tail_page->write); | |
77ae365e SR |
1901 | |
1902 | /* set write to only the index of the write */ | |
1903 | write &= RB_WRITE_MASK; | |
6634ff26 SR |
1904 | tail = write - length; |
1905 | ||
1906 | /* See if we shot pass the end of this buffer page */ | |
1907 | if (write > BUF_PAGE_SIZE) | |
1908 | return rb_move_tail(cpu_buffer, length, tail, | |
1909 | commit_page, tail_page, ts); | |
1910 | ||
1911 | /* We reserved something on the buffer */ | |
1912 | ||
6634ff26 | 1913 | event = __rb_page_index(tail_page, tail); |
1744a21d | 1914 | kmemcheck_annotate_bitfield(event, bitfield); |
6634ff26 SR |
1915 | rb_update_event(event, type, length); |
1916 | ||
1917 | /* The passed in type is zero for DATA */ | |
1918 | if (likely(!type)) | |
1919 | local_inc(&tail_page->entries); | |
1920 | ||
1921 | /* | |
fa743953 SR |
1922 | * If this is the first commit on the page, then update |
1923 | * its timestamp. | |
6634ff26 | 1924 | */ |
fa743953 SR |
1925 | if (!tail) |
1926 | tail_page->page->time_stamp = *ts; | |
6634ff26 SR |
1927 | |
1928 | return event; | |
1929 | } | |
1930 | ||
edd813bf SR |
1931 | static inline int |
1932 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
1933 | struct ring_buffer_event *event) | |
1934 | { | |
1935 | unsigned long new_index, old_index; | |
1936 | struct buffer_page *bpage; | |
1937 | unsigned long index; | |
1938 | unsigned long addr; | |
1939 | ||
1940 | new_index = rb_event_index(event); | |
1941 | old_index = new_index + rb_event_length(event); | |
1942 | addr = (unsigned long)event; | |
1943 | addr &= PAGE_MASK; | |
1944 | ||
1945 | bpage = cpu_buffer->tail_page; | |
1946 | ||
1947 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { | |
77ae365e SR |
1948 | unsigned long write_mask = |
1949 | local_read(&bpage->write) & ~RB_WRITE_MASK; | |
edd813bf SR |
1950 | /* |
1951 | * This is on the tail page. It is possible that | |
1952 | * a write could come in and move the tail page | |
1953 | * and write to the next page. That is fine | |
1954 | * because we just shorten what is on this page. | |
1955 | */ | |
77ae365e SR |
1956 | old_index += write_mask; |
1957 | new_index += write_mask; | |
edd813bf SR |
1958 | index = local_cmpxchg(&bpage->write, old_index, new_index); |
1959 | if (index == old_index) | |
1960 | return 1; | |
1961 | } | |
1962 | ||
1963 | /* could not discard */ | |
1964 | return 0; | |
1965 | } | |
1966 | ||
7a8e76a3 SR |
1967 | static int |
1968 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1969 | u64 *ts, u64 *delta) | |
1970 | { | |
1971 | struct ring_buffer_event *event; | |
1972 | static int once; | |
bf41a158 | 1973 | int ret; |
7a8e76a3 SR |
1974 | |
1975 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1976 | printk(KERN_WARNING "Delta way too big! %llu" | |
1977 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1978 | (unsigned long long)*delta, |
1979 | (unsigned long long)*ts, | |
1980 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1981 | WARN_ON(1); |
1982 | } | |
1983 | ||
1984 | /* | |
1985 | * The delta is too big, we to add a | |
1986 | * new timestamp. | |
1987 | */ | |
1988 | event = __rb_reserve_next(cpu_buffer, | |
1989 | RINGBUF_TYPE_TIME_EXTEND, | |
1990 | RB_LEN_TIME_EXTEND, | |
1991 | ts); | |
1992 | if (!event) | |
bf41a158 | 1993 | return -EBUSY; |
7a8e76a3 | 1994 | |
bf41a158 SR |
1995 | if (PTR_ERR(event) == -EAGAIN) |
1996 | return -EAGAIN; | |
1997 | ||
1998 | /* Only a commited time event can update the write stamp */ | |
fa743953 | 1999 | if (rb_event_is_commit(cpu_buffer, event)) { |
bf41a158 | 2000 | /* |
fa743953 SR |
2001 | * If this is the first on the page, then it was |
2002 | * updated with the page itself. Try to discard it | |
2003 | * and if we can't just make it zero. | |
bf41a158 SR |
2004 | */ |
2005 | if (rb_event_index(event)) { | |
2006 | event->time_delta = *delta & TS_MASK; | |
2007 | event->array[0] = *delta >> TS_SHIFT; | |
2008 | } else { | |
ea05b57c SR |
2009 | /* try to discard, since we do not need this */ |
2010 | if (!rb_try_to_discard(cpu_buffer, event)) { | |
2011 | /* nope, just zero it */ | |
2012 | event->time_delta = 0; | |
2013 | event->array[0] = 0; | |
2014 | } | |
bf41a158 | 2015 | } |
7a8e76a3 | 2016 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
2017 | /* let the caller know this was the commit */ |
2018 | ret = 1; | |
2019 | } else { | |
edd813bf SR |
2020 | /* Try to discard the event */ |
2021 | if (!rb_try_to_discard(cpu_buffer, event)) { | |
2022 | /* Darn, this is just wasted space */ | |
2023 | event->time_delta = 0; | |
2024 | event->array[0] = 0; | |
edd813bf | 2025 | } |
f57a8a19 | 2026 | ret = 0; |
7a8e76a3 SR |
2027 | } |
2028 | ||
bf41a158 SR |
2029 | *delta = 0; |
2030 | ||
2031 | return ret; | |
7a8e76a3 SR |
2032 | } |
2033 | ||
fa743953 SR |
2034 | static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer) |
2035 | { | |
2036 | local_inc(&cpu_buffer->committing); | |
2037 | local_inc(&cpu_buffer->commits); | |
2038 | } | |
2039 | ||
2040 | static void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer) | |
2041 | { | |
2042 | unsigned long commits; | |
2043 | ||
2044 | if (RB_WARN_ON(cpu_buffer, | |
2045 | !local_read(&cpu_buffer->committing))) | |
2046 | return; | |
2047 | ||
2048 | again: | |
2049 | commits = local_read(&cpu_buffer->commits); | |
2050 | /* synchronize with interrupts */ | |
2051 | barrier(); | |
2052 | if (local_read(&cpu_buffer->committing) == 1) | |
2053 | rb_set_commit_to_write(cpu_buffer); | |
2054 | ||
2055 | local_dec(&cpu_buffer->committing); | |
2056 | ||
2057 | /* synchronize with interrupts */ | |
2058 | barrier(); | |
2059 | ||
2060 | /* | |
2061 | * Need to account for interrupts coming in between the | |
2062 | * updating of the commit page and the clearing of the | |
2063 | * committing counter. | |
2064 | */ | |
2065 | if (unlikely(local_read(&cpu_buffer->commits) != commits) && | |
2066 | !local_read(&cpu_buffer->committing)) { | |
2067 | local_inc(&cpu_buffer->committing); | |
2068 | goto again; | |
2069 | } | |
2070 | } | |
2071 | ||
7a8e76a3 SR |
2072 | static struct ring_buffer_event * |
2073 | rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 2074 | unsigned long length) |
7a8e76a3 SR |
2075 | { |
2076 | struct ring_buffer_event *event; | |
168b6b1d | 2077 | u64 ts, delta = 0; |
bf41a158 | 2078 | int commit = 0; |
818e3dd3 | 2079 | int nr_loops = 0; |
7a8e76a3 | 2080 | |
fa743953 SR |
2081 | rb_start_commit(cpu_buffer); |
2082 | ||
be957c44 | 2083 | length = rb_calculate_event_length(length); |
bf41a158 | 2084 | again: |
818e3dd3 SR |
2085 | /* |
2086 | * We allow for interrupts to reenter here and do a trace. | |
2087 | * If one does, it will cause this original code to loop | |
2088 | * back here. Even with heavy interrupts happening, this | |
2089 | * should only happen a few times in a row. If this happens | |
2090 | * 1000 times in a row, there must be either an interrupt | |
2091 | * storm or we have something buggy. | |
2092 | * Bail! | |
2093 | */ | |
3e89c7bb | 2094 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
fa743953 | 2095 | goto out_fail; |
818e3dd3 | 2096 | |
88eb0125 | 2097 | ts = rb_time_stamp(cpu_buffer->buffer, cpu_buffer->cpu); |
7a8e76a3 | 2098 | |
bf41a158 SR |
2099 | /* |
2100 | * Only the first commit can update the timestamp. | |
2101 | * Yes there is a race here. If an interrupt comes in | |
2102 | * just after the conditional and it traces too, then it | |
2103 | * will also check the deltas. More than one timestamp may | |
2104 | * also be made. But only the entry that did the actual | |
2105 | * commit will be something other than zero. | |
2106 | */ | |
0f0c85fc SR |
2107 | if (likely(cpu_buffer->tail_page == cpu_buffer->commit_page && |
2108 | rb_page_write(cpu_buffer->tail_page) == | |
2109 | rb_commit_index(cpu_buffer))) { | |
168b6b1d | 2110 | u64 diff; |
bf41a158 | 2111 | |
168b6b1d | 2112 | diff = ts - cpu_buffer->write_stamp; |
7a8e76a3 | 2113 | |
168b6b1d | 2114 | /* make sure this diff is calculated here */ |
bf41a158 SR |
2115 | barrier(); |
2116 | ||
2117 | /* Did the write stamp get updated already? */ | |
2118 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
168b6b1d | 2119 | goto get_event; |
bf41a158 | 2120 | |
168b6b1d SR |
2121 | delta = diff; |
2122 | if (unlikely(test_time_stamp(delta))) { | |
7a8e76a3 | 2123 | |
bf41a158 | 2124 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
bf41a158 | 2125 | if (commit == -EBUSY) |
fa743953 | 2126 | goto out_fail; |
bf41a158 SR |
2127 | |
2128 | if (commit == -EAGAIN) | |
2129 | goto again; | |
2130 | ||
2131 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 2132 | } |
168b6b1d | 2133 | } |
7a8e76a3 | 2134 | |
168b6b1d | 2135 | get_event: |
1cd8d735 | 2136 | event = __rb_reserve_next(cpu_buffer, 0, length, &ts); |
168b6b1d | 2137 | if (unlikely(PTR_ERR(event) == -EAGAIN)) |
bf41a158 SR |
2138 | goto again; |
2139 | ||
fa743953 SR |
2140 | if (!event) |
2141 | goto out_fail; | |
7a8e76a3 | 2142 | |
fa743953 | 2143 | if (!rb_event_is_commit(cpu_buffer, event)) |
7a8e76a3 SR |
2144 | delta = 0; |
2145 | ||
2146 | event->time_delta = delta; | |
2147 | ||
2148 | return event; | |
fa743953 SR |
2149 | |
2150 | out_fail: | |
2151 | rb_end_commit(cpu_buffer); | |
2152 | return NULL; | |
7a8e76a3 SR |
2153 | } |
2154 | ||
1155de47 PM |
2155 | #ifdef CONFIG_TRACING |
2156 | ||
aa18efb2 | 2157 | #define TRACE_RECURSIVE_DEPTH 16 |
261842b7 SR |
2158 | |
2159 | static int trace_recursive_lock(void) | |
2160 | { | |
aa18efb2 | 2161 | current->trace_recursion++; |
261842b7 | 2162 | |
aa18efb2 SR |
2163 | if (likely(current->trace_recursion < TRACE_RECURSIVE_DEPTH)) |
2164 | return 0; | |
e057a5e5 | 2165 | |
aa18efb2 SR |
2166 | /* Disable all tracing before we do anything else */ |
2167 | tracing_off_permanent(); | |
261842b7 | 2168 | |
7d7d2b80 | 2169 | printk_once(KERN_WARNING "Tracing recursion: depth[%ld]:" |
aa18efb2 SR |
2170 | "HC[%lu]:SC[%lu]:NMI[%lu]\n", |
2171 | current->trace_recursion, | |
2172 | hardirq_count() >> HARDIRQ_SHIFT, | |
2173 | softirq_count() >> SOFTIRQ_SHIFT, | |
2174 | in_nmi()); | |
261842b7 | 2175 | |
aa18efb2 SR |
2176 | WARN_ON_ONCE(1); |
2177 | return -1; | |
261842b7 SR |
2178 | } |
2179 | ||
2180 | static void trace_recursive_unlock(void) | |
2181 | { | |
aa18efb2 | 2182 | WARN_ON_ONCE(!current->trace_recursion); |
261842b7 | 2183 | |
aa18efb2 | 2184 | current->trace_recursion--; |
261842b7 SR |
2185 | } |
2186 | ||
1155de47 PM |
2187 | #else |
2188 | ||
2189 | #define trace_recursive_lock() (0) | |
2190 | #define trace_recursive_unlock() do { } while (0) | |
2191 | ||
2192 | #endif | |
2193 | ||
bf41a158 SR |
2194 | static DEFINE_PER_CPU(int, rb_need_resched); |
2195 | ||
7a8e76a3 SR |
2196 | /** |
2197 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
2198 | * @buffer: the ring buffer to reserve from | |
2199 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
2200 | * |
2201 | * Returns a reseverd event on the ring buffer to copy directly to. | |
2202 | * The user of this interface will need to get the body to write into | |
2203 | * and can use the ring_buffer_event_data() interface. | |
2204 | * | |
2205 | * The length is the length of the data needed, not the event length | |
2206 | * which also includes the event header. | |
2207 | * | |
2208 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
2209 | * If NULL is returned, then nothing has been allocated or locked. | |
2210 | */ | |
2211 | struct ring_buffer_event * | |
0a987751 | 2212 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
2213 | { |
2214 | struct ring_buffer_per_cpu *cpu_buffer; | |
2215 | struct ring_buffer_event *event; | |
bf41a158 | 2216 | int cpu, resched; |
7a8e76a3 | 2217 | |
033601a3 | 2218 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2219 | return NULL; |
2220 | ||
7a8e76a3 SR |
2221 | if (atomic_read(&buffer->record_disabled)) |
2222 | return NULL; | |
2223 | ||
bf41a158 | 2224 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 2225 | resched = ftrace_preempt_disable(); |
bf41a158 | 2226 | |
261842b7 SR |
2227 | if (trace_recursive_lock()) |
2228 | goto out_nocheck; | |
2229 | ||
7a8e76a3 SR |
2230 | cpu = raw_smp_processor_id(); |
2231 | ||
9e01c1b7 | 2232 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2233 | goto out; |
7a8e76a3 SR |
2234 | |
2235 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2236 | |
2237 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 2238 | goto out; |
7a8e76a3 | 2239 | |
be957c44 | 2240 | if (length > BUF_MAX_DATA_SIZE) |
bf41a158 | 2241 | goto out; |
7a8e76a3 | 2242 | |
1cd8d735 | 2243 | event = rb_reserve_next_event(cpu_buffer, length); |
7a8e76a3 | 2244 | if (!event) |
d769041f | 2245 | goto out; |
7a8e76a3 | 2246 | |
bf41a158 SR |
2247 | /* |
2248 | * Need to store resched state on this cpu. | |
2249 | * Only the first needs to. | |
2250 | */ | |
2251 | ||
2252 | if (preempt_count() == 1) | |
2253 | per_cpu(rb_need_resched, cpu) = resched; | |
2254 | ||
7a8e76a3 SR |
2255 | return event; |
2256 | ||
d769041f | 2257 | out: |
261842b7 SR |
2258 | trace_recursive_unlock(); |
2259 | ||
2260 | out_nocheck: | |
182e9f5f | 2261 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
2262 | return NULL; |
2263 | } | |
c4f50183 | 2264 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 SR |
2265 | |
2266 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
2267 | struct ring_buffer_event *event) | |
2268 | { | |
e4906eff | 2269 | local_inc(&cpu_buffer->entries); |
bf41a158 | 2270 | |
fa743953 SR |
2271 | /* |
2272 | * The event first in the commit queue updates the | |
2273 | * time stamp. | |
2274 | */ | |
2275 | if (rb_event_is_commit(cpu_buffer, event)) | |
2276 | cpu_buffer->write_stamp += event->time_delta; | |
bf41a158 | 2277 | |
fa743953 | 2278 | rb_end_commit(cpu_buffer); |
7a8e76a3 SR |
2279 | } |
2280 | ||
2281 | /** | |
2282 | * ring_buffer_unlock_commit - commit a reserved | |
2283 | * @buffer: The buffer to commit to | |
2284 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
2285 | * |
2286 | * This commits the data to the ring buffer, and releases any locks held. | |
2287 | * | |
2288 | * Must be paired with ring_buffer_lock_reserve. | |
2289 | */ | |
2290 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 2291 | struct ring_buffer_event *event) |
7a8e76a3 SR |
2292 | { |
2293 | struct ring_buffer_per_cpu *cpu_buffer; | |
2294 | int cpu = raw_smp_processor_id(); | |
2295 | ||
2296 | cpu_buffer = buffer->buffers[cpu]; | |
2297 | ||
7a8e76a3 SR |
2298 | rb_commit(cpu_buffer, event); |
2299 | ||
261842b7 SR |
2300 | trace_recursive_unlock(); |
2301 | ||
bf41a158 SR |
2302 | /* |
2303 | * Only the last preempt count needs to restore preemption. | |
2304 | */ | |
182e9f5f SR |
2305 | if (preempt_count() == 1) |
2306 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
2307 | else | |
bf41a158 | 2308 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
2309 | |
2310 | return 0; | |
2311 | } | |
c4f50183 | 2312 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 2313 | |
f3b9aae1 FW |
2314 | static inline void rb_event_discard(struct ring_buffer_event *event) |
2315 | { | |
334d4169 LJ |
2316 | /* array[0] holds the actual length for the discarded event */ |
2317 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
2318 | event->type_len = RINGBUF_TYPE_PADDING; | |
f3b9aae1 FW |
2319 | /* time delta must be non zero */ |
2320 | if (!event->time_delta) | |
2321 | event->time_delta = 1; | |
2322 | } | |
2323 | ||
fa1b47dd SR |
2324 | /** |
2325 | * ring_buffer_event_discard - discard any event in the ring buffer | |
2326 | * @event: the event to discard | |
2327 | * | |
2328 | * Sometimes a event that is in the ring buffer needs to be ignored. | |
2329 | * This function lets the user discard an event in the ring buffer | |
2330 | * and then that event will not be read later. | |
2331 | * | |
2332 | * Note, it is up to the user to be careful with this, and protect | |
2333 | * against races. If the user discards an event that has been consumed | |
2334 | * it is possible that it could corrupt the ring buffer. | |
2335 | */ | |
2336 | void ring_buffer_event_discard(struct ring_buffer_event *event) | |
2337 | { | |
f3b9aae1 | 2338 | rb_event_discard(event); |
fa1b47dd SR |
2339 | } |
2340 | EXPORT_SYMBOL_GPL(ring_buffer_event_discard); | |
2341 | ||
2342 | /** | |
2343 | * ring_buffer_commit_discard - discard an event that has not been committed | |
2344 | * @buffer: the ring buffer | |
2345 | * @event: non committed event to discard | |
2346 | * | |
2347 | * This is similar to ring_buffer_event_discard but must only be | |
2348 | * performed on an event that has not been committed yet. The difference | |
2349 | * is that this will also try to free the event from the ring buffer | |
2350 | * if another event has not been added behind it. | |
2351 | * | |
2352 | * If another event has been added behind it, it will set the event | |
2353 | * up as discarded, and perform the commit. | |
2354 | * | |
2355 | * If this function is called, do not call ring_buffer_unlock_commit on | |
2356 | * the event. | |
2357 | */ | |
2358 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
2359 | struct ring_buffer_event *event) | |
2360 | { | |
2361 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
2362 | int cpu; |
2363 | ||
2364 | /* The event is discarded regardless */ | |
f3b9aae1 | 2365 | rb_event_discard(event); |
fa1b47dd | 2366 | |
fa743953 SR |
2367 | cpu = smp_processor_id(); |
2368 | cpu_buffer = buffer->buffers[cpu]; | |
2369 | ||
fa1b47dd SR |
2370 | /* |
2371 | * This must only be called if the event has not been | |
2372 | * committed yet. Thus we can assume that preemption | |
2373 | * is still disabled. | |
2374 | */ | |
fa743953 | 2375 | RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing)); |
fa1b47dd | 2376 | |
edd813bf SR |
2377 | if (!rb_try_to_discard(cpu_buffer, event)) |
2378 | goto out; | |
fa1b47dd SR |
2379 | |
2380 | /* | |
2381 | * The commit is still visible by the reader, so we | |
2382 | * must increment entries. | |
2383 | */ | |
e4906eff | 2384 | local_inc(&cpu_buffer->entries); |
fa1b47dd | 2385 | out: |
fa743953 | 2386 | rb_end_commit(cpu_buffer); |
fa1b47dd | 2387 | |
f3b9aae1 FW |
2388 | trace_recursive_unlock(); |
2389 | ||
fa1b47dd SR |
2390 | /* |
2391 | * Only the last preempt count needs to restore preemption. | |
2392 | */ | |
2393 | if (preempt_count() == 1) | |
2394 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
2395 | else | |
2396 | preempt_enable_no_resched_notrace(); | |
2397 | ||
2398 | } | |
2399 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
2400 | ||
7a8e76a3 SR |
2401 | /** |
2402 | * ring_buffer_write - write data to the buffer without reserving | |
2403 | * @buffer: The ring buffer to write to. | |
2404 | * @length: The length of the data being written (excluding the event header) | |
2405 | * @data: The data to write to the buffer. | |
2406 | * | |
2407 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
2408 | * one function. If you already have the data to write to the buffer, it | |
2409 | * may be easier to simply call this function. | |
2410 | * | |
2411 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
2412 | * and not the length of the event which would hold the header. | |
2413 | */ | |
2414 | int ring_buffer_write(struct ring_buffer *buffer, | |
2415 | unsigned long length, | |
2416 | void *data) | |
2417 | { | |
2418 | struct ring_buffer_per_cpu *cpu_buffer; | |
2419 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
2420 | void *body; |
2421 | int ret = -EBUSY; | |
bf41a158 | 2422 | int cpu, resched; |
7a8e76a3 | 2423 | |
033601a3 | 2424 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2425 | return -EBUSY; |
2426 | ||
7a8e76a3 SR |
2427 | if (atomic_read(&buffer->record_disabled)) |
2428 | return -EBUSY; | |
2429 | ||
182e9f5f | 2430 | resched = ftrace_preempt_disable(); |
bf41a158 | 2431 | |
7a8e76a3 SR |
2432 | cpu = raw_smp_processor_id(); |
2433 | ||
9e01c1b7 | 2434 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2435 | goto out; |
7a8e76a3 SR |
2436 | |
2437 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2438 | |
2439 | if (atomic_read(&cpu_buffer->record_disabled)) | |
2440 | goto out; | |
2441 | ||
be957c44 SR |
2442 | if (length > BUF_MAX_DATA_SIZE) |
2443 | goto out; | |
2444 | ||
2445 | event = rb_reserve_next_event(cpu_buffer, length); | |
7a8e76a3 SR |
2446 | if (!event) |
2447 | goto out; | |
2448 | ||
2449 | body = rb_event_data(event); | |
2450 | ||
2451 | memcpy(body, data, length); | |
2452 | ||
2453 | rb_commit(cpu_buffer, event); | |
2454 | ||
2455 | ret = 0; | |
2456 | out: | |
182e9f5f | 2457 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
2458 | |
2459 | return ret; | |
2460 | } | |
c4f50183 | 2461 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 2462 | |
34a148bf | 2463 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
2464 | { |
2465 | struct buffer_page *reader = cpu_buffer->reader_page; | |
77ae365e | 2466 | struct buffer_page *head = rb_set_head_page(cpu_buffer); |
bf41a158 SR |
2467 | struct buffer_page *commit = cpu_buffer->commit_page; |
2468 | ||
77ae365e SR |
2469 | /* In case of error, head will be NULL */ |
2470 | if (unlikely(!head)) | |
2471 | return 1; | |
2472 | ||
bf41a158 SR |
2473 | return reader->read == rb_page_commit(reader) && |
2474 | (commit == reader || | |
2475 | (commit == head && | |
2476 | head->read == rb_page_commit(commit))); | |
2477 | } | |
2478 | ||
7a8e76a3 SR |
2479 | /** |
2480 | * ring_buffer_record_disable - stop all writes into the buffer | |
2481 | * @buffer: The ring buffer to stop writes to. | |
2482 | * | |
2483 | * This prevents all writes to the buffer. Any attempt to write | |
2484 | * to the buffer after this will fail and return NULL. | |
2485 | * | |
2486 | * The caller should call synchronize_sched() after this. | |
2487 | */ | |
2488 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
2489 | { | |
2490 | atomic_inc(&buffer->record_disabled); | |
2491 | } | |
c4f50183 | 2492 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
2493 | |
2494 | /** | |
2495 | * ring_buffer_record_enable - enable writes to the buffer | |
2496 | * @buffer: The ring buffer to enable writes | |
2497 | * | |
2498 | * Note, multiple disables will need the same number of enables | |
2499 | * to truely enable the writing (much like preempt_disable). | |
2500 | */ | |
2501 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
2502 | { | |
2503 | atomic_dec(&buffer->record_disabled); | |
2504 | } | |
c4f50183 | 2505 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 SR |
2506 | |
2507 | /** | |
2508 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
2509 | * @buffer: The ring buffer to stop writes to. | |
2510 | * @cpu: The CPU buffer to stop | |
2511 | * | |
2512 | * This prevents all writes to the buffer. Any attempt to write | |
2513 | * to the buffer after this will fail and return NULL. | |
2514 | * | |
2515 | * The caller should call synchronize_sched() after this. | |
2516 | */ | |
2517 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
2518 | { | |
2519 | struct ring_buffer_per_cpu *cpu_buffer; | |
2520 | ||
9e01c1b7 | 2521 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2522 | return; |
7a8e76a3 SR |
2523 | |
2524 | cpu_buffer = buffer->buffers[cpu]; | |
2525 | atomic_inc(&cpu_buffer->record_disabled); | |
2526 | } | |
c4f50183 | 2527 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
2528 | |
2529 | /** | |
2530 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
2531 | * @buffer: The ring buffer to enable writes | |
2532 | * @cpu: The CPU to enable. | |
2533 | * | |
2534 | * Note, multiple disables will need the same number of enables | |
2535 | * to truely enable the writing (much like preempt_disable). | |
2536 | */ | |
2537 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
2538 | { | |
2539 | struct ring_buffer_per_cpu *cpu_buffer; | |
2540 | ||
9e01c1b7 | 2541 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2542 | return; |
7a8e76a3 SR |
2543 | |
2544 | cpu_buffer = buffer->buffers[cpu]; | |
2545 | atomic_dec(&cpu_buffer->record_disabled); | |
2546 | } | |
c4f50183 | 2547 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 SR |
2548 | |
2549 | /** | |
2550 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
2551 | * @buffer: The ring buffer | |
2552 | * @cpu: The per CPU buffer to get the entries from. | |
2553 | */ | |
2554 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
2555 | { | |
2556 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2557 | unsigned long ret; |
7a8e76a3 | 2558 | |
9e01c1b7 | 2559 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2560 | return 0; |
7a8e76a3 SR |
2561 | |
2562 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 2563 | ret = (local_read(&cpu_buffer->entries) - local_read(&cpu_buffer->overrun)) |
e4906eff | 2564 | - cpu_buffer->read; |
554f786e SR |
2565 | |
2566 | return ret; | |
7a8e76a3 | 2567 | } |
c4f50183 | 2568 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
2569 | |
2570 | /** | |
2571 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
2572 | * @buffer: The ring buffer | |
2573 | * @cpu: The per CPU buffer to get the number of overruns from | |
2574 | */ | |
2575 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
2576 | { | |
2577 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2578 | unsigned long ret; |
7a8e76a3 | 2579 | |
9e01c1b7 | 2580 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2581 | return 0; |
7a8e76a3 SR |
2582 | |
2583 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 2584 | ret = local_read(&cpu_buffer->overrun); |
554f786e SR |
2585 | |
2586 | return ret; | |
7a8e76a3 | 2587 | } |
c4f50183 | 2588 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 2589 | |
f0d2c681 SR |
2590 | /** |
2591 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits | |
2592 | * @buffer: The ring buffer | |
2593 | * @cpu: The per CPU buffer to get the number of overruns from | |
2594 | */ | |
2595 | unsigned long | |
2596 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
2597 | { | |
2598 | struct ring_buffer_per_cpu *cpu_buffer; | |
2599 | unsigned long ret; | |
2600 | ||
2601 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
2602 | return 0; | |
2603 | ||
2604 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 2605 | ret = local_read(&cpu_buffer->commit_overrun); |
f0d2c681 SR |
2606 | |
2607 | return ret; | |
2608 | } | |
2609 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
2610 | ||
7a8e76a3 SR |
2611 | /** |
2612 | * ring_buffer_entries - get the number of entries in a buffer | |
2613 | * @buffer: The ring buffer | |
2614 | * | |
2615 | * Returns the total number of entries in the ring buffer | |
2616 | * (all CPU entries) | |
2617 | */ | |
2618 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
2619 | { | |
2620 | struct ring_buffer_per_cpu *cpu_buffer; | |
2621 | unsigned long entries = 0; | |
2622 | int cpu; | |
2623 | ||
2624 | /* if you care about this being correct, lock the buffer */ | |
2625 | for_each_buffer_cpu(buffer, cpu) { | |
2626 | cpu_buffer = buffer->buffers[cpu]; | |
e4906eff | 2627 | entries += (local_read(&cpu_buffer->entries) - |
77ae365e | 2628 | local_read(&cpu_buffer->overrun)) - cpu_buffer->read; |
7a8e76a3 SR |
2629 | } |
2630 | ||
2631 | return entries; | |
2632 | } | |
c4f50183 | 2633 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
2634 | |
2635 | /** | |
2636 | * ring_buffer_overrun_cpu - get the number of overruns in buffer | |
2637 | * @buffer: The ring buffer | |
2638 | * | |
2639 | * Returns the total number of overruns in the ring buffer | |
2640 | * (all CPU entries) | |
2641 | */ | |
2642 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
2643 | { | |
2644 | struct ring_buffer_per_cpu *cpu_buffer; | |
2645 | unsigned long overruns = 0; | |
2646 | int cpu; | |
2647 | ||
2648 | /* if you care about this being correct, lock the buffer */ | |
2649 | for_each_buffer_cpu(buffer, cpu) { | |
2650 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 2651 | overruns += local_read(&cpu_buffer->overrun); |
7a8e76a3 SR |
2652 | } |
2653 | ||
2654 | return overruns; | |
2655 | } | |
c4f50183 | 2656 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 2657 | |
642edba5 | 2658 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
2659 | { |
2660 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2661 | ||
d769041f SR |
2662 | /* Iterator usage is expected to have record disabled */ |
2663 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
77ae365e SR |
2664 | iter->head_page = rb_set_head_page(cpu_buffer); |
2665 | if (unlikely(!iter->head_page)) | |
2666 | return; | |
2667 | iter->head = iter->head_page->read; | |
d769041f SR |
2668 | } else { |
2669 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 2670 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
2671 | } |
2672 | if (iter->head) | |
2673 | iter->read_stamp = cpu_buffer->read_stamp; | |
2674 | else | |
abc9b56d | 2675 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 2676 | } |
f83c9d0f | 2677 | |
642edba5 SR |
2678 | /** |
2679 | * ring_buffer_iter_reset - reset an iterator | |
2680 | * @iter: The iterator to reset | |
2681 | * | |
2682 | * Resets the iterator, so that it will start from the beginning | |
2683 | * again. | |
2684 | */ | |
2685 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
2686 | { | |
554f786e | 2687 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
2688 | unsigned long flags; |
2689 | ||
554f786e SR |
2690 | if (!iter) |
2691 | return; | |
2692 | ||
2693 | cpu_buffer = iter->cpu_buffer; | |
2694 | ||
642edba5 SR |
2695 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2696 | rb_iter_reset(iter); | |
f83c9d0f | 2697 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 2698 | } |
c4f50183 | 2699 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
2700 | |
2701 | /** | |
2702 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
2703 | * @iter: The iterator to check | |
2704 | */ | |
2705 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
2706 | { | |
2707 | struct ring_buffer_per_cpu *cpu_buffer; | |
2708 | ||
2709 | cpu_buffer = iter->cpu_buffer; | |
2710 | ||
bf41a158 SR |
2711 | return iter->head_page == cpu_buffer->commit_page && |
2712 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 2713 | } |
c4f50183 | 2714 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
2715 | |
2716 | static void | |
2717 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
2718 | struct ring_buffer_event *event) | |
2719 | { | |
2720 | u64 delta; | |
2721 | ||
334d4169 | 2722 | switch (event->type_len) { |
7a8e76a3 SR |
2723 | case RINGBUF_TYPE_PADDING: |
2724 | return; | |
2725 | ||
2726 | case RINGBUF_TYPE_TIME_EXTEND: | |
2727 | delta = event->array[0]; | |
2728 | delta <<= TS_SHIFT; | |
2729 | delta += event->time_delta; | |
2730 | cpu_buffer->read_stamp += delta; | |
2731 | return; | |
2732 | ||
2733 | case RINGBUF_TYPE_TIME_STAMP: | |
2734 | /* FIXME: not implemented */ | |
2735 | return; | |
2736 | ||
2737 | case RINGBUF_TYPE_DATA: | |
2738 | cpu_buffer->read_stamp += event->time_delta; | |
2739 | return; | |
2740 | ||
2741 | default: | |
2742 | BUG(); | |
2743 | } | |
2744 | return; | |
2745 | } | |
2746 | ||
2747 | static void | |
2748 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
2749 | struct ring_buffer_event *event) | |
2750 | { | |
2751 | u64 delta; | |
2752 | ||
334d4169 | 2753 | switch (event->type_len) { |
7a8e76a3 SR |
2754 | case RINGBUF_TYPE_PADDING: |
2755 | return; | |
2756 | ||
2757 | case RINGBUF_TYPE_TIME_EXTEND: | |
2758 | delta = event->array[0]; | |
2759 | delta <<= TS_SHIFT; | |
2760 | delta += event->time_delta; | |
2761 | iter->read_stamp += delta; | |
2762 | return; | |
2763 | ||
2764 | case RINGBUF_TYPE_TIME_STAMP: | |
2765 | /* FIXME: not implemented */ | |
2766 | return; | |
2767 | ||
2768 | case RINGBUF_TYPE_DATA: | |
2769 | iter->read_stamp += event->time_delta; | |
2770 | return; | |
2771 | ||
2772 | default: | |
2773 | BUG(); | |
2774 | } | |
2775 | return; | |
2776 | } | |
2777 | ||
d769041f SR |
2778 | static struct buffer_page * |
2779 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 2780 | { |
d769041f SR |
2781 | struct buffer_page *reader = NULL; |
2782 | unsigned long flags; | |
818e3dd3 | 2783 | int nr_loops = 0; |
77ae365e | 2784 | int ret; |
d769041f | 2785 | |
3e03fb7f SR |
2786 | local_irq_save(flags); |
2787 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
2788 | |
2789 | again: | |
818e3dd3 SR |
2790 | /* |
2791 | * This should normally only loop twice. But because the | |
2792 | * start of the reader inserts an empty page, it causes | |
2793 | * a case where we will loop three times. There should be no | |
2794 | * reason to loop four times (that I know of). | |
2795 | */ | |
3e89c7bb | 2796 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
2797 | reader = NULL; |
2798 | goto out; | |
2799 | } | |
2800 | ||
d769041f SR |
2801 | reader = cpu_buffer->reader_page; |
2802 | ||
2803 | /* If there's more to read, return this page */ | |
bf41a158 | 2804 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
2805 | goto out; |
2806 | ||
2807 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
2808 | if (RB_WARN_ON(cpu_buffer, |
2809 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
2810 | goto out; | |
d769041f SR |
2811 | |
2812 | /* check if we caught up to the tail */ | |
2813 | reader = NULL; | |
bf41a158 | 2814 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 2815 | goto out; |
7a8e76a3 SR |
2816 | |
2817 | /* | |
d769041f | 2818 | * Reset the reader page to size zero. |
7a8e76a3 | 2819 | */ |
77ae365e SR |
2820 | local_set(&cpu_buffer->reader_page->write, 0); |
2821 | local_set(&cpu_buffer->reader_page->entries, 0); | |
2822 | local_set(&cpu_buffer->reader_page->page->commit, 0); | |
7a8e76a3 | 2823 | |
77ae365e SR |
2824 | spin: |
2825 | /* | |
2826 | * Splice the empty reader page into the list around the head. | |
2827 | */ | |
2828 | reader = rb_set_head_page(cpu_buffer); | |
d769041f SR |
2829 | cpu_buffer->reader_page->list.next = reader->list.next; |
2830 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 | 2831 | |
3adc54fa SR |
2832 | /* |
2833 | * cpu_buffer->pages just needs to point to the buffer, it | |
2834 | * has no specific buffer page to point to. Lets move it out | |
2835 | * of our way so we don't accidently swap it. | |
2836 | */ | |
2837 | cpu_buffer->pages = reader->list.prev; | |
2838 | ||
77ae365e SR |
2839 | /* The reader page will be pointing to the new head */ |
2840 | rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list); | |
7a8e76a3 | 2841 | |
77ae365e SR |
2842 | /* |
2843 | * Here's the tricky part. | |
2844 | * | |
2845 | * We need to move the pointer past the header page. | |
2846 | * But we can only do that if a writer is not currently | |
2847 | * moving it. The page before the header page has the | |
2848 | * flag bit '1' set if it is pointing to the page we want. | |
2849 | * but if the writer is in the process of moving it | |
2850 | * than it will be '2' or already moved '0'. | |
2851 | */ | |
2852 | ||
2853 | ret = rb_head_page_replace(reader, cpu_buffer->reader_page); | |
7a8e76a3 SR |
2854 | |
2855 | /* | |
77ae365e | 2856 | * If we did not convert it, then we must try again. |
7a8e76a3 | 2857 | */ |
77ae365e SR |
2858 | if (!ret) |
2859 | goto spin; | |
7a8e76a3 | 2860 | |
77ae365e SR |
2861 | /* |
2862 | * Yeah! We succeeded in replacing the page. | |
2863 | * | |
2864 | * Now make the new head point back to the reader page. | |
2865 | */ | |
2866 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
2867 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); | |
d769041f SR |
2868 | |
2869 | /* Finally update the reader page to the new head */ | |
2870 | cpu_buffer->reader_page = reader; | |
2871 | rb_reset_reader_page(cpu_buffer); | |
2872 | ||
2873 | goto again; | |
2874 | ||
2875 | out: | |
3e03fb7f SR |
2876 | __raw_spin_unlock(&cpu_buffer->lock); |
2877 | local_irq_restore(flags); | |
d769041f SR |
2878 | |
2879 | return reader; | |
2880 | } | |
2881 | ||
2882 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
2883 | { | |
2884 | struct ring_buffer_event *event; | |
2885 | struct buffer_page *reader; | |
2886 | unsigned length; | |
2887 | ||
2888 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 2889 | |
d769041f | 2890 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
2891 | if (RB_WARN_ON(cpu_buffer, !reader)) |
2892 | return; | |
7a8e76a3 | 2893 | |
d769041f SR |
2894 | event = rb_reader_event(cpu_buffer); |
2895 | ||
334d4169 LJ |
2896 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX |
2897 | || rb_discarded_event(event)) | |
e4906eff | 2898 | cpu_buffer->read++; |
d769041f SR |
2899 | |
2900 | rb_update_read_stamp(cpu_buffer, event); | |
2901 | ||
2902 | length = rb_event_length(event); | |
6f807acd | 2903 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
2904 | } |
2905 | ||
2906 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
2907 | { | |
2908 | struct ring_buffer *buffer; | |
2909 | struct ring_buffer_per_cpu *cpu_buffer; | |
2910 | struct ring_buffer_event *event; | |
2911 | unsigned length; | |
2912 | ||
2913 | cpu_buffer = iter->cpu_buffer; | |
2914 | buffer = cpu_buffer->buffer; | |
2915 | ||
2916 | /* | |
2917 | * Check if we are at the end of the buffer. | |
2918 | */ | |
bf41a158 | 2919 | if (iter->head >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
2920 | /* discarded commits can make the page empty */ |
2921 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 2922 | return; |
d769041f | 2923 | rb_inc_iter(iter); |
7a8e76a3 SR |
2924 | return; |
2925 | } | |
2926 | ||
2927 | event = rb_iter_head_event(iter); | |
2928 | ||
2929 | length = rb_event_length(event); | |
2930 | ||
2931 | /* | |
2932 | * This should not be called to advance the header if we are | |
2933 | * at the tail of the buffer. | |
2934 | */ | |
3e89c7bb | 2935 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 2936 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
2937 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
2938 | return; | |
7a8e76a3 SR |
2939 | |
2940 | rb_update_iter_read_stamp(iter, event); | |
2941 | ||
2942 | iter->head += length; | |
2943 | ||
2944 | /* check for end of page padding */ | |
bf41a158 SR |
2945 | if ((iter->head >= rb_page_size(iter->head_page)) && |
2946 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
2947 | rb_advance_iter(iter); |
2948 | } | |
2949 | ||
f83c9d0f SR |
2950 | static struct ring_buffer_event * |
2951 | rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
7a8e76a3 SR |
2952 | { |
2953 | struct ring_buffer_per_cpu *cpu_buffer; | |
2954 | struct ring_buffer_event *event; | |
d769041f | 2955 | struct buffer_page *reader; |
818e3dd3 | 2956 | int nr_loops = 0; |
7a8e76a3 | 2957 | |
7a8e76a3 SR |
2958 | cpu_buffer = buffer->buffers[cpu]; |
2959 | ||
2960 | again: | |
818e3dd3 SR |
2961 | /* |
2962 | * We repeat when a timestamp is encountered. It is possible | |
2963 | * to get multiple timestamps from an interrupt entering just | |
ea05b57c SR |
2964 | * as one timestamp is about to be written, or from discarded |
2965 | * commits. The most that we can have is the number on a single page. | |
818e3dd3 | 2966 | */ |
ea05b57c | 2967 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE)) |
818e3dd3 | 2968 | return NULL; |
818e3dd3 | 2969 | |
d769041f SR |
2970 | reader = rb_get_reader_page(cpu_buffer); |
2971 | if (!reader) | |
7a8e76a3 SR |
2972 | return NULL; |
2973 | ||
d769041f | 2974 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 2975 | |
334d4169 | 2976 | switch (event->type_len) { |
7a8e76a3 | 2977 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
2978 | if (rb_null_event(event)) |
2979 | RB_WARN_ON(cpu_buffer, 1); | |
2980 | /* | |
2981 | * Because the writer could be discarding every | |
2982 | * event it creates (which would probably be bad) | |
2983 | * if we were to go back to "again" then we may never | |
2984 | * catch up, and will trigger the warn on, or lock | |
2985 | * the box. Return the padding, and we will release | |
2986 | * the current locks, and try again. | |
2987 | */ | |
d769041f | 2988 | rb_advance_reader(cpu_buffer); |
2d622719 | 2989 | return event; |
7a8e76a3 SR |
2990 | |
2991 | case RINGBUF_TYPE_TIME_EXTEND: | |
2992 | /* Internal data, OK to advance */ | |
d769041f | 2993 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2994 | goto again; |
2995 | ||
2996 | case RINGBUF_TYPE_TIME_STAMP: | |
2997 | /* FIXME: not implemented */ | |
d769041f | 2998 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
2999 | goto again; |
3000 | ||
3001 | case RINGBUF_TYPE_DATA: | |
3002 | if (ts) { | |
3003 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
37886f6a SR |
3004 | ring_buffer_normalize_time_stamp(buffer, |
3005 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
3006 | } |
3007 | return event; | |
3008 | ||
3009 | default: | |
3010 | BUG(); | |
3011 | } | |
3012 | ||
3013 | return NULL; | |
3014 | } | |
c4f50183 | 3015 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 3016 | |
f83c9d0f SR |
3017 | static struct ring_buffer_event * |
3018 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
3019 | { |
3020 | struct ring_buffer *buffer; | |
3021 | struct ring_buffer_per_cpu *cpu_buffer; | |
3022 | struct ring_buffer_event *event; | |
818e3dd3 | 3023 | int nr_loops = 0; |
7a8e76a3 SR |
3024 | |
3025 | if (ring_buffer_iter_empty(iter)) | |
3026 | return NULL; | |
3027 | ||
3028 | cpu_buffer = iter->cpu_buffer; | |
3029 | buffer = cpu_buffer->buffer; | |
3030 | ||
3031 | again: | |
818e3dd3 | 3032 | /* |
ea05b57c SR |
3033 | * We repeat when a timestamp is encountered. |
3034 | * We can get multiple timestamps by nested interrupts or also | |
3035 | * if filtering is on (discarding commits). Since discarding | |
3036 | * commits can be frequent we can get a lot of timestamps. | |
3037 | * But we limit them by not adding timestamps if they begin | |
3038 | * at the start of a page. | |
818e3dd3 | 3039 | */ |
ea05b57c | 3040 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE)) |
818e3dd3 | 3041 | return NULL; |
818e3dd3 | 3042 | |
7a8e76a3 SR |
3043 | if (rb_per_cpu_empty(cpu_buffer)) |
3044 | return NULL; | |
3045 | ||
3046 | event = rb_iter_head_event(iter); | |
3047 | ||
334d4169 | 3048 | switch (event->type_len) { |
7a8e76a3 | 3049 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3050 | if (rb_null_event(event)) { |
3051 | rb_inc_iter(iter); | |
3052 | goto again; | |
3053 | } | |
3054 | rb_advance_iter(iter); | |
3055 | return event; | |
7a8e76a3 SR |
3056 | |
3057 | case RINGBUF_TYPE_TIME_EXTEND: | |
3058 | /* Internal data, OK to advance */ | |
3059 | rb_advance_iter(iter); | |
3060 | goto again; | |
3061 | ||
3062 | case RINGBUF_TYPE_TIME_STAMP: | |
3063 | /* FIXME: not implemented */ | |
3064 | rb_advance_iter(iter); | |
3065 | goto again; | |
3066 | ||
3067 | case RINGBUF_TYPE_DATA: | |
3068 | if (ts) { | |
3069 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
3070 | ring_buffer_normalize_time_stamp(buffer, |
3071 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
3072 | } |
3073 | return event; | |
3074 | ||
3075 | default: | |
3076 | BUG(); | |
3077 | } | |
3078 | ||
3079 | return NULL; | |
3080 | } | |
c4f50183 | 3081 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 3082 | |
8d707e8e SR |
3083 | static inline int rb_ok_to_lock(void) |
3084 | { | |
3085 | /* | |
3086 | * If an NMI die dumps out the content of the ring buffer | |
3087 | * do not grab locks. We also permanently disable the ring | |
3088 | * buffer too. A one time deal is all you get from reading | |
3089 | * the ring buffer from an NMI. | |
3090 | */ | |
3091 | if (likely(!in_nmi() && !oops_in_progress)) | |
3092 | return 1; | |
3093 | ||
3094 | tracing_off_permanent(); | |
3095 | return 0; | |
3096 | } | |
3097 | ||
f83c9d0f SR |
3098 | /** |
3099 | * ring_buffer_peek - peek at the next event to be read | |
3100 | * @buffer: The ring buffer to read | |
3101 | * @cpu: The cpu to peak at | |
3102 | * @ts: The timestamp counter of this event. | |
3103 | * | |
3104 | * This will return the event that will be read next, but does | |
3105 | * not consume the data. | |
3106 | */ | |
3107 | struct ring_buffer_event * | |
3108 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
3109 | { | |
3110 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 3111 | struct ring_buffer_event *event; |
f83c9d0f | 3112 | unsigned long flags; |
8d707e8e | 3113 | int dolock; |
f83c9d0f | 3114 | |
554f786e | 3115 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3116 | return NULL; |
554f786e | 3117 | |
8d707e8e | 3118 | dolock = rb_ok_to_lock(); |
2d622719 | 3119 | again: |
8d707e8e SR |
3120 | local_irq_save(flags); |
3121 | if (dolock) | |
3122 | spin_lock(&cpu_buffer->reader_lock); | |
f83c9d0f | 3123 | event = rb_buffer_peek(buffer, cpu, ts); |
8d707e8e SR |
3124 | if (dolock) |
3125 | spin_unlock(&cpu_buffer->reader_lock); | |
3126 | local_irq_restore(flags); | |
f83c9d0f | 3127 | |
334d4169 | 3128 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
3129 | cpu_relax(); |
3130 | goto again; | |
3131 | } | |
3132 | ||
f83c9d0f SR |
3133 | return event; |
3134 | } | |
3135 | ||
3136 | /** | |
3137 | * ring_buffer_iter_peek - peek at the next event to be read | |
3138 | * @iter: The ring buffer iterator | |
3139 | * @ts: The timestamp counter of this event. | |
3140 | * | |
3141 | * This will return the event that will be read next, but does | |
3142 | * not increment the iterator. | |
3143 | */ | |
3144 | struct ring_buffer_event * | |
3145 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
3146 | { | |
3147 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3148 | struct ring_buffer_event *event; | |
3149 | unsigned long flags; | |
3150 | ||
2d622719 | 3151 | again: |
f83c9d0f SR |
3152 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3153 | event = rb_iter_peek(iter, ts); | |
3154 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
3155 | ||
334d4169 | 3156 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
3157 | cpu_relax(); |
3158 | goto again; | |
3159 | } | |
3160 | ||
f83c9d0f SR |
3161 | return event; |
3162 | } | |
3163 | ||
7a8e76a3 SR |
3164 | /** |
3165 | * ring_buffer_consume - return an event and consume it | |
3166 | * @buffer: The ring buffer to get the next event from | |
3167 | * | |
3168 | * Returns the next event in the ring buffer, and that event is consumed. | |
3169 | * Meaning, that sequential reads will keep returning a different event, | |
3170 | * and eventually empty the ring buffer if the producer is slower. | |
3171 | */ | |
3172 | struct ring_buffer_event * | |
3173 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
3174 | { | |
554f786e SR |
3175 | struct ring_buffer_per_cpu *cpu_buffer; |
3176 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 3177 | unsigned long flags; |
8d707e8e SR |
3178 | int dolock; |
3179 | ||
3180 | dolock = rb_ok_to_lock(); | |
7a8e76a3 | 3181 | |
2d622719 | 3182 | again: |
554f786e SR |
3183 | /* might be called in atomic */ |
3184 | preempt_disable(); | |
3185 | ||
9e01c1b7 | 3186 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 3187 | goto out; |
7a8e76a3 | 3188 | |
554f786e | 3189 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
3190 | local_irq_save(flags); |
3191 | if (dolock) | |
3192 | spin_lock(&cpu_buffer->reader_lock); | |
f83c9d0f SR |
3193 | |
3194 | event = rb_buffer_peek(buffer, cpu, ts); | |
7a8e76a3 | 3195 | if (!event) |
554f786e | 3196 | goto out_unlock; |
7a8e76a3 | 3197 | |
d769041f | 3198 | rb_advance_reader(cpu_buffer); |
7a8e76a3 | 3199 | |
554f786e | 3200 | out_unlock: |
8d707e8e SR |
3201 | if (dolock) |
3202 | spin_unlock(&cpu_buffer->reader_lock); | |
3203 | local_irq_restore(flags); | |
f83c9d0f | 3204 | |
554f786e SR |
3205 | out: |
3206 | preempt_enable(); | |
3207 | ||
334d4169 | 3208 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
3209 | cpu_relax(); |
3210 | goto again; | |
3211 | } | |
3212 | ||
7a8e76a3 SR |
3213 | return event; |
3214 | } | |
c4f50183 | 3215 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
3216 | |
3217 | /** | |
3218 | * ring_buffer_read_start - start a non consuming read of the buffer | |
3219 | * @buffer: The ring buffer to read from | |
3220 | * @cpu: The cpu buffer to iterate over | |
3221 | * | |
3222 | * This starts up an iteration through the buffer. It also disables | |
3223 | * the recording to the buffer until the reading is finished. | |
3224 | * This prevents the reading from being corrupted. This is not | |
3225 | * a consuming read, so a producer is not expected. | |
3226 | * | |
3227 | * Must be paired with ring_buffer_finish. | |
3228 | */ | |
3229 | struct ring_buffer_iter * | |
3230 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
3231 | { | |
3232 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3233 | struct ring_buffer_iter *iter; |
d769041f | 3234 | unsigned long flags; |
7a8e76a3 | 3235 | |
9e01c1b7 | 3236 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3237 | return NULL; |
7a8e76a3 SR |
3238 | |
3239 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
3240 | if (!iter) | |
8aabee57 | 3241 | return NULL; |
7a8e76a3 SR |
3242 | |
3243 | cpu_buffer = buffer->buffers[cpu]; | |
3244 | ||
3245 | iter->cpu_buffer = cpu_buffer; | |
3246 | ||
3247 | atomic_inc(&cpu_buffer->record_disabled); | |
3248 | synchronize_sched(); | |
3249 | ||
f83c9d0f | 3250 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 3251 | __raw_spin_lock(&cpu_buffer->lock); |
642edba5 | 3252 | rb_iter_reset(iter); |
3e03fb7f | 3253 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 3254 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
3255 | |
3256 | return iter; | |
3257 | } | |
c4f50183 | 3258 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
3259 | |
3260 | /** | |
3261 | * ring_buffer_finish - finish reading the iterator of the buffer | |
3262 | * @iter: The iterator retrieved by ring_buffer_start | |
3263 | * | |
3264 | * This re-enables the recording to the buffer, and frees the | |
3265 | * iterator. | |
3266 | */ | |
3267 | void | |
3268 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
3269 | { | |
3270 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3271 | ||
3272 | atomic_dec(&cpu_buffer->record_disabled); | |
3273 | kfree(iter); | |
3274 | } | |
c4f50183 | 3275 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
3276 | |
3277 | /** | |
3278 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
3279 | * @iter: The ring buffer iterator | |
3280 | * @ts: The time stamp of the event read. | |
3281 | * | |
3282 | * This reads the next event in the ring buffer and increments the iterator. | |
3283 | */ | |
3284 | struct ring_buffer_event * | |
3285 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
3286 | { | |
3287 | struct ring_buffer_event *event; | |
f83c9d0f SR |
3288 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
3289 | unsigned long flags; | |
7a8e76a3 | 3290 | |
2d622719 | 3291 | again: |
f83c9d0f SR |
3292 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3293 | event = rb_iter_peek(iter, ts); | |
7a8e76a3 | 3294 | if (!event) |
f83c9d0f | 3295 | goto out; |
7a8e76a3 SR |
3296 | |
3297 | rb_advance_iter(iter); | |
f83c9d0f SR |
3298 | out: |
3299 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 | 3300 | |
334d4169 | 3301 | if (event && event->type_len == RINGBUF_TYPE_PADDING) { |
2d622719 TZ |
3302 | cpu_relax(); |
3303 | goto again; | |
3304 | } | |
3305 | ||
7a8e76a3 SR |
3306 | return event; |
3307 | } | |
c4f50183 | 3308 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
3309 | |
3310 | /** | |
3311 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
3312 | * @buffer: The ring buffer. | |
3313 | */ | |
3314 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
3315 | { | |
3316 | return BUF_PAGE_SIZE * buffer->pages; | |
3317 | } | |
c4f50183 | 3318 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
3319 | |
3320 | static void | |
3321 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
3322 | { | |
77ae365e SR |
3323 | rb_head_page_deactivate(cpu_buffer); |
3324 | ||
7a8e76a3 | 3325 | cpu_buffer->head_page |
3adc54fa | 3326 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 3327 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 3328 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 3329 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 3330 | |
6f807acd | 3331 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
3332 | |
3333 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
3334 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
3335 | ||
3336 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
3337 | local_set(&cpu_buffer->reader_page->write, 0); | |
778c55d4 | 3338 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 3339 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 3340 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 3341 | |
77ae365e SR |
3342 | local_set(&cpu_buffer->commit_overrun, 0); |
3343 | local_set(&cpu_buffer->overrun, 0); | |
e4906eff | 3344 | local_set(&cpu_buffer->entries, 0); |
fa743953 SR |
3345 | local_set(&cpu_buffer->committing, 0); |
3346 | local_set(&cpu_buffer->commits, 0); | |
77ae365e | 3347 | cpu_buffer->read = 0; |
69507c06 SR |
3348 | |
3349 | cpu_buffer->write_stamp = 0; | |
3350 | cpu_buffer->read_stamp = 0; | |
77ae365e SR |
3351 | |
3352 | rb_head_page_activate(cpu_buffer); | |
7a8e76a3 SR |
3353 | } |
3354 | ||
3355 | /** | |
3356 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
3357 | * @buffer: The ring buffer to reset a per cpu buffer of | |
3358 | * @cpu: The CPU buffer to be reset | |
3359 | */ | |
3360 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
3361 | { | |
3362 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
3363 | unsigned long flags; | |
3364 | ||
9e01c1b7 | 3365 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3366 | return; |
7a8e76a3 | 3367 | |
41ede23e SR |
3368 | atomic_inc(&cpu_buffer->record_disabled); |
3369 | ||
f83c9d0f SR |
3370 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3371 | ||
3e03fb7f | 3372 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
3373 | |
3374 | rb_reset_cpu(cpu_buffer); | |
3375 | ||
3e03fb7f | 3376 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f SR |
3377 | |
3378 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
41ede23e SR |
3379 | |
3380 | atomic_dec(&cpu_buffer->record_disabled); | |
7a8e76a3 | 3381 | } |
c4f50183 | 3382 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
3383 | |
3384 | /** | |
3385 | * ring_buffer_reset - reset a ring buffer | |
3386 | * @buffer: The ring buffer to reset all cpu buffers | |
3387 | */ | |
3388 | void ring_buffer_reset(struct ring_buffer *buffer) | |
3389 | { | |
7a8e76a3 SR |
3390 | int cpu; |
3391 | ||
7a8e76a3 | 3392 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 3393 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 3394 | } |
c4f50183 | 3395 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
3396 | |
3397 | /** | |
3398 | * rind_buffer_empty - is the ring buffer empty? | |
3399 | * @buffer: The ring buffer to test | |
3400 | */ | |
3401 | int ring_buffer_empty(struct ring_buffer *buffer) | |
3402 | { | |
3403 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 3404 | unsigned long flags; |
8d707e8e | 3405 | int dolock; |
7a8e76a3 | 3406 | int cpu; |
d4788207 | 3407 | int ret; |
7a8e76a3 | 3408 | |
8d707e8e | 3409 | dolock = rb_ok_to_lock(); |
7a8e76a3 SR |
3410 | |
3411 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
3412 | for_each_buffer_cpu(buffer, cpu) { | |
3413 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e SR |
3414 | local_irq_save(flags); |
3415 | if (dolock) | |
3416 | spin_lock(&cpu_buffer->reader_lock); | |
d4788207 | 3417 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e SR |
3418 | if (dolock) |
3419 | spin_unlock(&cpu_buffer->reader_lock); | |
3420 | local_irq_restore(flags); | |
3421 | ||
d4788207 | 3422 | if (!ret) |
7a8e76a3 SR |
3423 | return 0; |
3424 | } | |
554f786e | 3425 | |
7a8e76a3 SR |
3426 | return 1; |
3427 | } | |
c4f50183 | 3428 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
3429 | |
3430 | /** | |
3431 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
3432 | * @buffer: The ring buffer | |
3433 | * @cpu: The CPU buffer to test | |
3434 | */ | |
3435 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
3436 | { | |
3437 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 3438 | unsigned long flags; |
8d707e8e | 3439 | int dolock; |
8aabee57 | 3440 | int ret; |
7a8e76a3 | 3441 | |
9e01c1b7 | 3442 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3443 | return 1; |
7a8e76a3 | 3444 | |
8d707e8e SR |
3445 | dolock = rb_ok_to_lock(); |
3446 | ||
7a8e76a3 | 3447 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
3448 | local_irq_save(flags); |
3449 | if (dolock) | |
3450 | spin_lock(&cpu_buffer->reader_lock); | |
554f786e | 3451 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e SR |
3452 | if (dolock) |
3453 | spin_unlock(&cpu_buffer->reader_lock); | |
3454 | local_irq_restore(flags); | |
554f786e SR |
3455 | |
3456 | return ret; | |
7a8e76a3 | 3457 | } |
c4f50183 | 3458 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 SR |
3459 | |
3460 | /** | |
3461 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
3462 | * @buffer_a: One buffer to swap with | |
3463 | * @buffer_b: The other buffer to swap with | |
3464 | * | |
3465 | * This function is useful for tracers that want to take a "snapshot" | |
3466 | * of a CPU buffer and has another back up buffer lying around. | |
3467 | * it is expected that the tracer handles the cpu buffer not being | |
3468 | * used at the moment. | |
3469 | */ | |
3470 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
3471 | struct ring_buffer *buffer_b, int cpu) | |
3472 | { | |
3473 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
3474 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
3475 | int ret = -EINVAL; |
3476 | ||
9e01c1b7 RR |
3477 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
3478 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 3479 | goto out; |
7a8e76a3 SR |
3480 | |
3481 | /* At least make sure the two buffers are somewhat the same */ | |
6d102bc6 | 3482 | if (buffer_a->pages != buffer_b->pages) |
554f786e SR |
3483 | goto out; |
3484 | ||
3485 | ret = -EAGAIN; | |
7a8e76a3 | 3486 | |
97b17efe | 3487 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 3488 | goto out; |
97b17efe SR |
3489 | |
3490 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 3491 | goto out; |
97b17efe SR |
3492 | |
3493 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 3494 | goto out; |
97b17efe | 3495 | |
7a8e76a3 SR |
3496 | cpu_buffer_a = buffer_a->buffers[cpu]; |
3497 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
3498 | ||
97b17efe | 3499 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 3500 | goto out; |
97b17efe SR |
3501 | |
3502 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 3503 | goto out; |
97b17efe | 3504 | |
7a8e76a3 SR |
3505 | /* |
3506 | * We can't do a synchronize_sched here because this | |
3507 | * function can be called in atomic context. | |
3508 | * Normally this will be called from the same CPU as cpu. | |
3509 | * If not it's up to the caller to protect this. | |
3510 | */ | |
3511 | atomic_inc(&cpu_buffer_a->record_disabled); | |
3512 | atomic_inc(&cpu_buffer_b->record_disabled); | |
3513 | ||
3514 | buffer_a->buffers[cpu] = cpu_buffer_b; | |
3515 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
3516 | ||
3517 | cpu_buffer_b->buffer = buffer_a; | |
3518 | cpu_buffer_a->buffer = buffer_b; | |
3519 | ||
3520 | atomic_dec(&cpu_buffer_a->record_disabled); | |
3521 | atomic_dec(&cpu_buffer_b->record_disabled); | |
3522 | ||
554f786e SR |
3523 | ret = 0; |
3524 | out: | |
554f786e | 3525 | return ret; |
7a8e76a3 | 3526 | } |
c4f50183 | 3527 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
7a8e76a3 | 3528 | |
8789a9e7 SR |
3529 | /** |
3530 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
3531 | * @buffer: the buffer to allocate for. | |
3532 | * | |
3533 | * This function is used in conjunction with ring_buffer_read_page. | |
3534 | * When reading a full page from the ring buffer, these functions | |
3535 | * can be used to speed up the process. The calling function should | |
3536 | * allocate a few pages first with this function. Then when it | |
3537 | * needs to get pages from the ring buffer, it passes the result | |
3538 | * of this function into ring_buffer_read_page, which will swap | |
3539 | * the page that was allocated, with the read page of the buffer. | |
3540 | * | |
3541 | * Returns: | |
3542 | * The page allocated, or NULL on error. | |
3543 | */ | |
3544 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer) | |
3545 | { | |
044fa782 | 3546 | struct buffer_data_page *bpage; |
ef7a4a16 | 3547 | unsigned long addr; |
8789a9e7 SR |
3548 | |
3549 | addr = __get_free_page(GFP_KERNEL); | |
3550 | if (!addr) | |
3551 | return NULL; | |
3552 | ||
044fa782 | 3553 | bpage = (void *)addr; |
8789a9e7 | 3554 | |
ef7a4a16 SR |
3555 | rb_init_page(bpage); |
3556 | ||
044fa782 | 3557 | return bpage; |
8789a9e7 | 3558 | } |
d6ce96da | 3559 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
3560 | |
3561 | /** | |
3562 | * ring_buffer_free_read_page - free an allocated read page | |
3563 | * @buffer: the buffer the page was allocate for | |
3564 | * @data: the page to free | |
3565 | * | |
3566 | * Free a page allocated from ring_buffer_alloc_read_page. | |
3567 | */ | |
3568 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
3569 | { | |
3570 | free_page((unsigned long)data); | |
3571 | } | |
d6ce96da | 3572 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
3573 | |
3574 | /** | |
3575 | * ring_buffer_read_page - extract a page from the ring buffer | |
3576 | * @buffer: buffer to extract from | |
3577 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 3578 | * @len: amount to extract |
8789a9e7 SR |
3579 | * @cpu: the cpu of the buffer to extract |
3580 | * @full: should the extraction only happen when the page is full. | |
3581 | * | |
3582 | * This function will pull out a page from the ring buffer and consume it. | |
3583 | * @data_page must be the address of the variable that was returned | |
3584 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
3585 | * to swap with a page in the ring buffer. | |
3586 | * | |
3587 | * for example: | |
b85fa01e | 3588 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
3589 | * if (!rpage) |
3590 | * return error; | |
ef7a4a16 | 3591 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
3592 | * if (ret >= 0) |
3593 | * process_page(rpage, ret); | |
8789a9e7 SR |
3594 | * |
3595 | * When @full is set, the function will not return true unless | |
3596 | * the writer is off the reader page. | |
3597 | * | |
3598 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
3599 | * The ring buffer can be used anywhere in the kernel and can not | |
3600 | * blindly call wake_up. The layer that uses the ring buffer must be | |
3601 | * responsible for that. | |
3602 | * | |
3603 | * Returns: | |
667d2412 LJ |
3604 | * >=0 if data has been transferred, returns the offset of consumed data. |
3605 | * <0 if no data has been transferred. | |
8789a9e7 SR |
3606 | */ |
3607 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 3608 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
3609 | { |
3610 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
3611 | struct ring_buffer_event *event; | |
044fa782 | 3612 | struct buffer_data_page *bpage; |
ef7a4a16 | 3613 | struct buffer_page *reader; |
8789a9e7 | 3614 | unsigned long flags; |
ef7a4a16 | 3615 | unsigned int commit; |
667d2412 | 3616 | unsigned int read; |
4f3640f8 | 3617 | u64 save_timestamp; |
667d2412 | 3618 | int ret = -1; |
8789a9e7 | 3619 | |
554f786e SR |
3620 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
3621 | goto out; | |
3622 | ||
474d32b6 SR |
3623 | /* |
3624 | * If len is not big enough to hold the page header, then | |
3625 | * we can not copy anything. | |
3626 | */ | |
3627 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 3628 | goto out; |
474d32b6 SR |
3629 | |
3630 | len -= BUF_PAGE_HDR_SIZE; | |
3631 | ||
8789a9e7 | 3632 | if (!data_page) |
554f786e | 3633 | goto out; |
8789a9e7 | 3634 | |
044fa782 SR |
3635 | bpage = *data_page; |
3636 | if (!bpage) | |
554f786e | 3637 | goto out; |
8789a9e7 SR |
3638 | |
3639 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
3640 | ||
ef7a4a16 SR |
3641 | reader = rb_get_reader_page(cpu_buffer); |
3642 | if (!reader) | |
554f786e | 3643 | goto out_unlock; |
8789a9e7 | 3644 | |
ef7a4a16 SR |
3645 | event = rb_reader_event(cpu_buffer); |
3646 | ||
3647 | read = reader->read; | |
3648 | commit = rb_page_commit(reader); | |
667d2412 | 3649 | |
8789a9e7 | 3650 | /* |
474d32b6 SR |
3651 | * If this page has been partially read or |
3652 | * if len is not big enough to read the rest of the page or | |
3653 | * a writer is still on the page, then | |
3654 | * we must copy the data from the page to the buffer. | |
3655 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 3656 | */ |
474d32b6 | 3657 | if (read || (len < (commit - read)) || |
ef7a4a16 | 3658 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 3659 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
3660 | unsigned int rpos = read; |
3661 | unsigned int pos = 0; | |
ef7a4a16 | 3662 | unsigned int size; |
8789a9e7 SR |
3663 | |
3664 | if (full) | |
554f786e | 3665 | goto out_unlock; |
8789a9e7 | 3666 | |
ef7a4a16 SR |
3667 | if (len > (commit - read)) |
3668 | len = (commit - read); | |
3669 | ||
3670 | size = rb_event_length(event); | |
3671 | ||
3672 | if (len < size) | |
554f786e | 3673 | goto out_unlock; |
ef7a4a16 | 3674 | |
4f3640f8 SR |
3675 | /* save the current timestamp, since the user will need it */ |
3676 | save_timestamp = cpu_buffer->read_stamp; | |
3677 | ||
ef7a4a16 SR |
3678 | /* Need to copy one event at a time */ |
3679 | do { | |
474d32b6 | 3680 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
3681 | |
3682 | len -= size; | |
3683 | ||
3684 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
3685 | rpos = reader->read; |
3686 | pos += size; | |
ef7a4a16 SR |
3687 | |
3688 | event = rb_reader_event(cpu_buffer); | |
3689 | size = rb_event_length(event); | |
3690 | } while (len > size); | |
667d2412 LJ |
3691 | |
3692 | /* update bpage */ | |
ef7a4a16 | 3693 | local_set(&bpage->commit, pos); |
4f3640f8 | 3694 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 3695 | |
474d32b6 SR |
3696 | /* we copied everything to the beginning */ |
3697 | read = 0; | |
8789a9e7 | 3698 | } else { |
afbab76a | 3699 | /* update the entry counter */ |
77ae365e | 3700 | cpu_buffer->read += rb_page_entries(reader); |
afbab76a | 3701 | |
8789a9e7 | 3702 | /* swap the pages */ |
044fa782 | 3703 | rb_init_page(bpage); |
ef7a4a16 SR |
3704 | bpage = reader->page; |
3705 | reader->page = *data_page; | |
3706 | local_set(&reader->write, 0); | |
778c55d4 | 3707 | local_set(&reader->entries, 0); |
ef7a4a16 | 3708 | reader->read = 0; |
044fa782 | 3709 | *data_page = bpage; |
8789a9e7 | 3710 | } |
667d2412 | 3711 | ret = read; |
8789a9e7 | 3712 | |
554f786e | 3713 | out_unlock: |
8789a9e7 SR |
3714 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
3715 | ||
554f786e | 3716 | out: |
8789a9e7 SR |
3717 | return ret; |
3718 | } | |
d6ce96da | 3719 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 3720 | |
1155de47 | 3721 | #ifdef CONFIG_TRACING |
a3583244 SR |
3722 | static ssize_t |
3723 | rb_simple_read(struct file *filp, char __user *ubuf, | |
3724 | size_t cnt, loff_t *ppos) | |
3725 | { | |
5e39841c | 3726 | unsigned long *p = filp->private_data; |
a3583244 SR |
3727 | char buf[64]; |
3728 | int r; | |
3729 | ||
033601a3 SR |
3730 | if (test_bit(RB_BUFFERS_DISABLED_BIT, p)) |
3731 | r = sprintf(buf, "permanently disabled\n"); | |
3732 | else | |
3733 | r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p)); | |
a3583244 SR |
3734 | |
3735 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3736 | } | |
3737 | ||
3738 | static ssize_t | |
3739 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
3740 | size_t cnt, loff_t *ppos) | |
3741 | { | |
5e39841c | 3742 | unsigned long *p = filp->private_data; |
a3583244 | 3743 | char buf[64]; |
5e39841c | 3744 | unsigned long val; |
a3583244 SR |
3745 | int ret; |
3746 | ||
3747 | if (cnt >= sizeof(buf)) | |
3748 | return -EINVAL; | |
3749 | ||
3750 | if (copy_from_user(&buf, ubuf, cnt)) | |
3751 | return -EFAULT; | |
3752 | ||
3753 | buf[cnt] = 0; | |
3754 | ||
3755 | ret = strict_strtoul(buf, 10, &val); | |
3756 | if (ret < 0) | |
3757 | return ret; | |
3758 | ||
033601a3 SR |
3759 | if (val) |
3760 | set_bit(RB_BUFFERS_ON_BIT, p); | |
3761 | else | |
3762 | clear_bit(RB_BUFFERS_ON_BIT, p); | |
a3583244 SR |
3763 | |
3764 | (*ppos)++; | |
3765 | ||
3766 | return cnt; | |
3767 | } | |
3768 | ||
5e2336a0 | 3769 | static const struct file_operations rb_simple_fops = { |
a3583244 SR |
3770 | .open = tracing_open_generic, |
3771 | .read = rb_simple_read, | |
3772 | .write = rb_simple_write, | |
3773 | }; | |
3774 | ||
3775 | ||
3776 | static __init int rb_init_debugfs(void) | |
3777 | { | |
3778 | struct dentry *d_tracer; | |
a3583244 SR |
3779 | |
3780 | d_tracer = tracing_init_dentry(); | |
3781 | ||
5452af66 FW |
3782 | trace_create_file("tracing_on", 0644, d_tracer, |
3783 | &ring_buffer_flags, &rb_simple_fops); | |
a3583244 SR |
3784 | |
3785 | return 0; | |
3786 | } | |
3787 | ||
3788 | fs_initcall(rb_init_debugfs); | |
1155de47 | 3789 | #endif |
554f786e | 3790 | |
59222efe | 3791 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
3792 | static int rb_cpu_notify(struct notifier_block *self, |
3793 | unsigned long action, void *hcpu) | |
554f786e SR |
3794 | { |
3795 | struct ring_buffer *buffer = | |
3796 | container_of(self, struct ring_buffer, cpu_notify); | |
3797 | long cpu = (long)hcpu; | |
3798 | ||
3799 | switch (action) { | |
3800 | case CPU_UP_PREPARE: | |
3801 | case CPU_UP_PREPARE_FROZEN: | |
3f237a79 | 3802 | if (cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e SR |
3803 | return NOTIFY_OK; |
3804 | ||
3805 | buffer->buffers[cpu] = | |
3806 | rb_allocate_cpu_buffer(buffer, cpu); | |
3807 | if (!buffer->buffers[cpu]) { | |
3808 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
3809 | cpu); | |
3810 | return NOTIFY_OK; | |
3811 | } | |
3812 | smp_wmb(); | |
3f237a79 | 3813 | cpumask_set_cpu(cpu, buffer->cpumask); |
554f786e SR |
3814 | break; |
3815 | case CPU_DOWN_PREPARE: | |
3816 | case CPU_DOWN_PREPARE_FROZEN: | |
3817 | /* | |
3818 | * Do nothing. | |
3819 | * If we were to free the buffer, then the user would | |
3820 | * lose any trace that was in the buffer. | |
3821 | */ | |
3822 | break; | |
3823 | default: | |
3824 | break; | |
3825 | } | |
3826 | return NOTIFY_OK; | |
3827 | } | |
3828 | #endif |