]>
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 | ||
e3d6bf0a | 204 | #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array)) |
67d34724 | 205 | #define RB_ALIGNMENT 4U |
334d4169 | 206 | #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
c7b09308 | 207 | #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */ |
334d4169 LJ |
208 | |
209 | /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */ | |
210 | #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX | |
7a8e76a3 SR |
211 | |
212 | enum { | |
213 | RB_LEN_TIME_EXTEND = 8, | |
214 | RB_LEN_TIME_STAMP = 16, | |
215 | }; | |
216 | ||
2d622719 TZ |
217 | static inline int rb_null_event(struct ring_buffer_event *event) |
218 | { | |
a1863c21 | 219 | return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta; |
2d622719 TZ |
220 | } |
221 | ||
222 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
223 | { | |
a1863c21 | 224 | /* padding has a NULL time_delta */ |
334d4169 | 225 | event->type_len = RINGBUF_TYPE_PADDING; |
2d622719 TZ |
226 | event->time_delta = 0; |
227 | } | |
228 | ||
34a148bf | 229 | static unsigned |
2d622719 | 230 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
231 | { |
232 | unsigned length; | |
233 | ||
334d4169 LJ |
234 | if (event->type_len) |
235 | length = event->type_len * RB_ALIGNMENT; | |
2d622719 TZ |
236 | else |
237 | length = event->array[0]; | |
238 | return length + RB_EVNT_HDR_SIZE; | |
239 | } | |
240 | ||
241 | /* inline for ring buffer fast paths */ | |
242 | static unsigned | |
243 | rb_event_length(struct ring_buffer_event *event) | |
244 | { | |
334d4169 | 245 | switch (event->type_len) { |
7a8e76a3 | 246 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
247 | if (rb_null_event(event)) |
248 | /* undefined */ | |
249 | return -1; | |
334d4169 | 250 | return event->array[0] + RB_EVNT_HDR_SIZE; |
7a8e76a3 SR |
251 | |
252 | case RINGBUF_TYPE_TIME_EXTEND: | |
253 | return RB_LEN_TIME_EXTEND; | |
254 | ||
255 | case RINGBUF_TYPE_TIME_STAMP: | |
256 | return RB_LEN_TIME_STAMP; | |
257 | ||
258 | case RINGBUF_TYPE_DATA: | |
2d622719 | 259 | return rb_event_data_length(event); |
7a8e76a3 SR |
260 | default: |
261 | BUG(); | |
262 | } | |
263 | /* not hit */ | |
264 | return 0; | |
265 | } | |
266 | ||
267 | /** | |
268 | * ring_buffer_event_length - return the length of the event | |
269 | * @event: the event to get the length of | |
270 | */ | |
271 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
272 | { | |
465634ad | 273 | unsigned length = rb_event_length(event); |
334d4169 | 274 | if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
465634ad RR |
275 | return length; |
276 | length -= RB_EVNT_HDR_SIZE; | |
277 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
278 | length -= sizeof(event->array[0]); | |
279 | return length; | |
7a8e76a3 | 280 | } |
c4f50183 | 281 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
282 | |
283 | /* inline for ring buffer fast paths */ | |
34a148bf | 284 | static void * |
7a8e76a3 SR |
285 | rb_event_data(struct ring_buffer_event *event) |
286 | { | |
334d4169 | 287 | BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX); |
7a8e76a3 | 288 | /* If length is in len field, then array[0] has the data */ |
334d4169 | 289 | if (event->type_len) |
7a8e76a3 SR |
290 | return (void *)&event->array[0]; |
291 | /* Otherwise length is in array[0] and array[1] has the data */ | |
292 | return (void *)&event->array[1]; | |
293 | } | |
294 | ||
295 | /** | |
296 | * ring_buffer_event_data - return the data of the event | |
297 | * @event: the event to get the data from | |
298 | */ | |
299 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
300 | { | |
301 | return rb_event_data(event); | |
302 | } | |
c4f50183 | 303 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
304 | |
305 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 306 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
307 | |
308 | #define TS_SHIFT 27 | |
309 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
310 | #define TS_DELTA_TEST (~TS_MASK) | |
311 | ||
abc9b56d | 312 | struct buffer_data_page { |
e4c2ce82 | 313 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 314 | local_t commit; /* write committed index */ |
abc9b56d SR |
315 | unsigned char data[]; /* data of buffer page */ |
316 | }; | |
317 | ||
77ae365e SR |
318 | /* |
319 | * Note, the buffer_page list must be first. The buffer pages | |
320 | * are allocated in cache lines, which means that each buffer | |
321 | * page will be at the beginning of a cache line, and thus | |
322 | * the least significant bits will be zero. We use this to | |
323 | * add flags in the list struct pointers, to make the ring buffer | |
324 | * lockless. | |
325 | */ | |
abc9b56d | 326 | struct buffer_page { |
778c55d4 | 327 | struct list_head list; /* list of buffer pages */ |
abc9b56d | 328 | local_t write; /* index for next write */ |
6f807acd | 329 | unsigned read; /* index for next read */ |
778c55d4 | 330 | local_t entries; /* entries on this page */ |
abc9b56d | 331 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
332 | }; |
333 | ||
77ae365e SR |
334 | /* |
335 | * The buffer page counters, write and entries, must be reset | |
336 | * atomically when crossing page boundaries. To synchronize this | |
337 | * update, two counters are inserted into the number. One is | |
338 | * the actual counter for the write position or count on the page. | |
339 | * | |
340 | * The other is a counter of updaters. Before an update happens | |
341 | * the update partition of the counter is incremented. This will | |
342 | * allow the updater to update the counter atomically. | |
343 | * | |
344 | * The counter is 20 bits, and the state data is 12. | |
345 | */ | |
346 | #define RB_WRITE_MASK 0xfffff | |
347 | #define RB_WRITE_INTCNT (1 << 20) | |
348 | ||
044fa782 | 349 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 350 | { |
044fa782 | 351 | local_set(&bpage->commit, 0); |
abc9b56d SR |
352 | } |
353 | ||
474d32b6 SR |
354 | /** |
355 | * ring_buffer_page_len - the size of data on the page. | |
356 | * @page: The page to read | |
357 | * | |
358 | * Returns the amount of data on the page, including buffer page header. | |
359 | */ | |
ef7a4a16 SR |
360 | size_t ring_buffer_page_len(void *page) |
361 | { | |
474d32b6 SR |
362 | return local_read(&((struct buffer_data_page *)page)->commit) |
363 | + BUF_PAGE_HDR_SIZE; | |
ef7a4a16 SR |
364 | } |
365 | ||
ed56829c SR |
366 | /* |
367 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
368 | * this issue out. | |
369 | */ | |
34a148bf | 370 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 371 | { |
34a148bf | 372 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 373 | kfree(bpage); |
ed56829c SR |
374 | } |
375 | ||
7a8e76a3 SR |
376 | /* |
377 | * We need to fit the time_stamp delta into 27 bits. | |
378 | */ | |
379 | static inline int test_time_stamp(u64 delta) | |
380 | { | |
381 | if (delta & TS_DELTA_TEST) | |
382 | return 1; | |
383 | return 0; | |
384 | } | |
385 | ||
474d32b6 | 386 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 | 387 | |
be957c44 SR |
388 | /* Max payload is BUF_PAGE_SIZE - header (8bytes) */ |
389 | #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) | |
390 | ||
ea05b57c SR |
391 | /* Max number of timestamps that can fit on a page */ |
392 | #define RB_TIMESTAMPS_PER_PAGE (BUF_PAGE_SIZE / RB_LEN_TIME_STAMP) | |
393 | ||
d1b182a8 SR |
394 | int ring_buffer_print_page_header(struct trace_seq *s) |
395 | { | |
396 | struct buffer_data_page field; | |
397 | int ret; | |
398 | ||
399 | ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t" | |
26a50744 TZ |
400 | "offset:0;\tsize:%u;\tsigned:%u;\n", |
401 | (unsigned int)sizeof(field.time_stamp), | |
402 | (unsigned int)is_signed_type(u64)); | |
d1b182a8 SR |
403 | |
404 | ret = trace_seq_printf(s, "\tfield: local_t commit;\t" | |
26a50744 | 405 | "offset:%u;\tsize:%u;\tsigned:%u;\n", |
d1b182a8 | 406 | (unsigned int)offsetof(typeof(field), commit), |
26a50744 TZ |
407 | (unsigned int)sizeof(field.commit), |
408 | (unsigned int)is_signed_type(long)); | |
d1b182a8 SR |
409 | |
410 | ret = trace_seq_printf(s, "\tfield: char data;\t" | |
26a50744 | 411 | "offset:%u;\tsize:%u;\tsigned:%u;\n", |
d1b182a8 | 412 | (unsigned int)offsetof(typeof(field), data), |
26a50744 TZ |
413 | (unsigned int)BUF_PAGE_SIZE, |
414 | (unsigned int)is_signed_type(char)); | |
d1b182a8 SR |
415 | |
416 | return ret; | |
417 | } | |
418 | ||
7a8e76a3 SR |
419 | /* |
420 | * head_page == tail_page && head == tail then buffer is empty. | |
421 | */ | |
422 | struct ring_buffer_per_cpu { | |
423 | int cpu; | |
424 | struct ring_buffer *buffer; | |
77ae365e | 425 | spinlock_t reader_lock; /* serialize readers */ |
3e03fb7f | 426 | raw_spinlock_t lock; |
7a8e76a3 | 427 | struct lock_class_key lock_key; |
3adc54fa | 428 | struct list_head *pages; |
6f807acd SR |
429 | struct buffer_page *head_page; /* read from head */ |
430 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 431 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 432 | struct buffer_page *reader_page; |
77ae365e SR |
433 | local_t commit_overrun; |
434 | local_t overrun; | |
e4906eff | 435 | local_t entries; |
fa743953 SR |
436 | local_t committing; |
437 | local_t commits; | |
77ae365e | 438 | unsigned long read; |
7a8e76a3 SR |
439 | u64 write_stamp; |
440 | u64 read_stamp; | |
441 | atomic_t record_disabled; | |
442 | }; | |
443 | ||
444 | struct ring_buffer { | |
7a8e76a3 SR |
445 | unsigned pages; |
446 | unsigned flags; | |
447 | int cpus; | |
7a8e76a3 | 448 | atomic_t record_disabled; |
00f62f61 | 449 | cpumask_var_t cpumask; |
7a8e76a3 | 450 | |
1f8a6a10 PZ |
451 | struct lock_class_key *reader_lock_key; |
452 | ||
7a8e76a3 SR |
453 | struct mutex mutex; |
454 | ||
455 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 456 | |
59222efe | 457 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
458 | struct notifier_block cpu_notify; |
459 | #endif | |
37886f6a | 460 | u64 (*clock)(void); |
7a8e76a3 SR |
461 | }; |
462 | ||
463 | struct ring_buffer_iter { | |
464 | struct ring_buffer_per_cpu *cpu_buffer; | |
465 | unsigned long head; | |
466 | struct buffer_page *head_page; | |
467 | u64 read_stamp; | |
468 | }; | |
469 | ||
f536aafc | 470 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
077c5407 SR |
471 | #define RB_WARN_ON(b, cond) \ |
472 | ({ \ | |
473 | int _____ret = unlikely(cond); \ | |
474 | if (_____ret) { \ | |
475 | if (__same_type(*(b), struct ring_buffer_per_cpu)) { \ | |
476 | struct ring_buffer_per_cpu *__b = \ | |
477 | (void *)b; \ | |
478 | atomic_inc(&__b->buffer->record_disabled); \ | |
479 | } else \ | |
480 | atomic_inc(&b->record_disabled); \ | |
481 | WARN_ON(1); \ | |
482 | } \ | |
483 | _____ret; \ | |
3e89c7bb | 484 | }) |
f536aafc | 485 | |
37886f6a SR |
486 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
487 | #define DEBUG_SHIFT 0 | |
488 | ||
6d3f1e12 | 489 | static inline u64 rb_time_stamp(struct ring_buffer *buffer) |
88eb0125 SR |
490 | { |
491 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
492 | return buffer->clock() << DEBUG_SHIFT; | |
493 | } | |
494 | ||
37886f6a SR |
495 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) |
496 | { | |
497 | u64 time; | |
498 | ||
499 | preempt_disable_notrace(); | |
6d3f1e12 | 500 | time = rb_time_stamp(buffer); |
37886f6a SR |
501 | preempt_enable_no_resched_notrace(); |
502 | ||
503 | return time; | |
504 | } | |
505 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
506 | ||
507 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
508 | int cpu, u64 *ts) | |
509 | { | |
510 | /* Just stupid testing the normalize function and deltas */ | |
511 | *ts >>= DEBUG_SHIFT; | |
512 | } | |
513 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
514 | ||
77ae365e SR |
515 | /* |
516 | * Making the ring buffer lockless makes things tricky. | |
517 | * Although writes only happen on the CPU that they are on, | |
518 | * and they only need to worry about interrupts. Reads can | |
519 | * happen on any CPU. | |
520 | * | |
521 | * The reader page is always off the ring buffer, but when the | |
522 | * reader finishes with a page, it needs to swap its page with | |
523 | * a new one from the buffer. The reader needs to take from | |
524 | * the head (writes go to the tail). But if a writer is in overwrite | |
525 | * mode and wraps, it must push the head page forward. | |
526 | * | |
527 | * Here lies the problem. | |
528 | * | |
529 | * The reader must be careful to replace only the head page, and | |
530 | * not another one. As described at the top of the file in the | |
531 | * ASCII art, the reader sets its old page to point to the next | |
532 | * page after head. It then sets the page after head to point to | |
533 | * the old reader page. But if the writer moves the head page | |
534 | * during this operation, the reader could end up with the tail. | |
535 | * | |
536 | * We use cmpxchg to help prevent this race. We also do something | |
537 | * special with the page before head. We set the LSB to 1. | |
538 | * | |
539 | * When the writer must push the page forward, it will clear the | |
540 | * bit that points to the head page, move the head, and then set | |
541 | * the bit that points to the new head page. | |
542 | * | |
543 | * We also don't want an interrupt coming in and moving the head | |
544 | * page on another writer. Thus we use the second LSB to catch | |
545 | * that too. Thus: | |
546 | * | |
547 | * head->list->prev->next bit 1 bit 0 | |
548 | * ------- ------- | |
549 | * Normal page 0 0 | |
550 | * Points to head page 0 1 | |
551 | * New head page 1 0 | |
552 | * | |
553 | * Note we can not trust the prev pointer of the head page, because: | |
554 | * | |
555 | * +----+ +-----+ +-----+ | |
556 | * | |------>| T |---X--->| N | | |
557 | * | |<------| | | | | |
558 | * +----+ +-----+ +-----+ | |
559 | * ^ ^ | | |
560 | * | +-----+ | | | |
561 | * +----------| R |----------+ | | |
562 | * | |<-----------+ | |
563 | * +-----+ | |
564 | * | |
565 | * Key: ---X--> HEAD flag set in pointer | |
566 | * T Tail page | |
567 | * R Reader page | |
568 | * N Next page | |
569 | * | |
570 | * (see __rb_reserve_next() to see where this happens) | |
571 | * | |
572 | * What the above shows is that the reader just swapped out | |
573 | * the reader page with a page in the buffer, but before it | |
574 | * could make the new header point back to the new page added | |
575 | * it was preempted by a writer. The writer moved forward onto | |
576 | * the new page added by the reader and is about to move forward | |
577 | * again. | |
578 | * | |
579 | * You can see, it is legitimate for the previous pointer of | |
580 | * the head (or any page) not to point back to itself. But only | |
581 | * temporarially. | |
582 | */ | |
583 | ||
584 | #define RB_PAGE_NORMAL 0UL | |
585 | #define RB_PAGE_HEAD 1UL | |
586 | #define RB_PAGE_UPDATE 2UL | |
587 | ||
588 | ||
589 | #define RB_FLAG_MASK 3UL | |
590 | ||
591 | /* PAGE_MOVED is not part of the mask */ | |
592 | #define RB_PAGE_MOVED 4UL | |
593 | ||
594 | /* | |
595 | * rb_list_head - remove any bit | |
596 | */ | |
597 | static struct list_head *rb_list_head(struct list_head *list) | |
598 | { | |
599 | unsigned long val = (unsigned long)list; | |
600 | ||
601 | return (struct list_head *)(val & ~RB_FLAG_MASK); | |
602 | } | |
603 | ||
604 | /* | |
6d3f1e12 | 605 | * rb_is_head_page - test if the given page is the head page |
77ae365e SR |
606 | * |
607 | * Because the reader may move the head_page pointer, we can | |
608 | * not trust what the head page is (it may be pointing to | |
609 | * the reader page). But if the next page is a header page, | |
610 | * its flags will be non zero. | |
611 | */ | |
612 | static int inline | |
613 | rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
614 | struct buffer_page *page, struct list_head *list) | |
615 | { | |
616 | unsigned long val; | |
617 | ||
618 | val = (unsigned long)list->next; | |
619 | ||
620 | if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list) | |
621 | return RB_PAGE_MOVED; | |
622 | ||
623 | return val & RB_FLAG_MASK; | |
624 | } | |
625 | ||
626 | /* | |
627 | * rb_is_reader_page | |
628 | * | |
629 | * The unique thing about the reader page, is that, if the | |
630 | * writer is ever on it, the previous pointer never points | |
631 | * back to the reader page. | |
632 | */ | |
633 | static int rb_is_reader_page(struct buffer_page *page) | |
634 | { | |
635 | struct list_head *list = page->list.prev; | |
636 | ||
637 | return rb_list_head(list->next) != &page->list; | |
638 | } | |
639 | ||
640 | /* | |
641 | * rb_set_list_to_head - set a list_head to be pointing to head. | |
642 | */ | |
643 | static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer, | |
644 | struct list_head *list) | |
645 | { | |
646 | unsigned long *ptr; | |
647 | ||
648 | ptr = (unsigned long *)&list->next; | |
649 | *ptr |= RB_PAGE_HEAD; | |
650 | *ptr &= ~RB_PAGE_UPDATE; | |
651 | } | |
652 | ||
653 | /* | |
654 | * rb_head_page_activate - sets up head page | |
655 | */ | |
656 | static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer) | |
657 | { | |
658 | struct buffer_page *head; | |
659 | ||
660 | head = cpu_buffer->head_page; | |
661 | if (!head) | |
662 | return; | |
663 | ||
664 | /* | |
665 | * Set the previous list pointer to have the HEAD flag. | |
666 | */ | |
667 | rb_set_list_to_head(cpu_buffer, head->list.prev); | |
668 | } | |
669 | ||
670 | static void rb_list_head_clear(struct list_head *list) | |
671 | { | |
672 | unsigned long *ptr = (unsigned long *)&list->next; | |
673 | ||
674 | *ptr &= ~RB_FLAG_MASK; | |
675 | } | |
676 | ||
677 | /* | |
678 | * rb_head_page_dactivate - clears head page ptr (for free list) | |
679 | */ | |
680 | static void | |
681 | rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer) | |
682 | { | |
683 | struct list_head *hd; | |
684 | ||
685 | /* Go through the whole list and clear any pointers found. */ | |
686 | rb_list_head_clear(cpu_buffer->pages); | |
687 | ||
688 | list_for_each(hd, cpu_buffer->pages) | |
689 | rb_list_head_clear(hd); | |
690 | } | |
691 | ||
692 | static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer, | |
693 | struct buffer_page *head, | |
694 | struct buffer_page *prev, | |
695 | int old_flag, int new_flag) | |
696 | { | |
697 | struct list_head *list; | |
698 | unsigned long val = (unsigned long)&head->list; | |
699 | unsigned long ret; | |
700 | ||
701 | list = &prev->list; | |
702 | ||
703 | val &= ~RB_FLAG_MASK; | |
704 | ||
08a40816 SR |
705 | ret = cmpxchg((unsigned long *)&list->next, |
706 | val | old_flag, val | new_flag); | |
77ae365e SR |
707 | |
708 | /* check if the reader took the page */ | |
709 | if ((ret & ~RB_FLAG_MASK) != val) | |
710 | return RB_PAGE_MOVED; | |
711 | ||
712 | return ret & RB_FLAG_MASK; | |
713 | } | |
714 | ||
715 | static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer, | |
716 | struct buffer_page *head, | |
717 | struct buffer_page *prev, | |
718 | int old_flag) | |
719 | { | |
720 | return rb_head_page_set(cpu_buffer, head, prev, | |
721 | old_flag, RB_PAGE_UPDATE); | |
722 | } | |
723 | ||
724 | static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer, | |
725 | struct buffer_page *head, | |
726 | struct buffer_page *prev, | |
727 | int old_flag) | |
728 | { | |
729 | return rb_head_page_set(cpu_buffer, head, prev, | |
730 | old_flag, RB_PAGE_HEAD); | |
731 | } | |
732 | ||
733 | static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer, | |
734 | struct buffer_page *head, | |
735 | struct buffer_page *prev, | |
736 | int old_flag) | |
737 | { | |
738 | return rb_head_page_set(cpu_buffer, head, prev, | |
739 | old_flag, RB_PAGE_NORMAL); | |
740 | } | |
741 | ||
742 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
743 | struct buffer_page **bpage) | |
744 | { | |
745 | struct list_head *p = rb_list_head((*bpage)->list.next); | |
746 | ||
747 | *bpage = list_entry(p, struct buffer_page, list); | |
748 | } | |
749 | ||
750 | static struct buffer_page * | |
751 | rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer) | |
752 | { | |
753 | struct buffer_page *head; | |
754 | struct buffer_page *page; | |
755 | struct list_head *list; | |
756 | int i; | |
757 | ||
758 | if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page)) | |
759 | return NULL; | |
760 | ||
761 | /* sanity check */ | |
762 | list = cpu_buffer->pages; | |
763 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list)) | |
764 | return NULL; | |
765 | ||
766 | page = head = cpu_buffer->head_page; | |
767 | /* | |
768 | * It is possible that the writer moves the header behind | |
769 | * where we started, and we miss in one loop. | |
770 | * A second loop should grab the header, but we'll do | |
771 | * three loops just because I'm paranoid. | |
772 | */ | |
773 | for (i = 0; i < 3; i++) { | |
774 | do { | |
775 | if (rb_is_head_page(cpu_buffer, page, page->list.prev)) { | |
776 | cpu_buffer->head_page = page; | |
777 | return page; | |
778 | } | |
779 | rb_inc_page(cpu_buffer, &page); | |
780 | } while (page != head); | |
781 | } | |
782 | ||
783 | RB_WARN_ON(cpu_buffer, 1); | |
784 | ||
785 | return NULL; | |
786 | } | |
787 | ||
788 | static int rb_head_page_replace(struct buffer_page *old, | |
789 | struct buffer_page *new) | |
790 | { | |
791 | unsigned long *ptr = (unsigned long *)&old->list.prev->next; | |
792 | unsigned long val; | |
793 | unsigned long ret; | |
794 | ||
795 | val = *ptr & ~RB_FLAG_MASK; | |
796 | val |= RB_PAGE_HEAD; | |
797 | ||
08a40816 | 798 | ret = cmpxchg(ptr, val, (unsigned long)&new->list); |
77ae365e SR |
799 | |
800 | return ret == val; | |
801 | } | |
802 | ||
803 | /* | |
804 | * rb_tail_page_update - move the tail page forward | |
805 | * | |
806 | * Returns 1 if moved tail page, 0 if someone else did. | |
807 | */ | |
808 | static int rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer, | |
809 | struct buffer_page *tail_page, | |
810 | struct buffer_page *next_page) | |
811 | { | |
812 | struct buffer_page *old_tail; | |
813 | unsigned long old_entries; | |
814 | unsigned long old_write; | |
815 | int ret = 0; | |
816 | ||
817 | /* | |
818 | * The tail page now needs to be moved forward. | |
819 | * | |
820 | * We need to reset the tail page, but without messing | |
821 | * with possible erasing of data brought in by interrupts | |
822 | * that have moved the tail page and are currently on it. | |
823 | * | |
824 | * We add a counter to the write field to denote this. | |
825 | */ | |
826 | old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write); | |
827 | old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries); | |
828 | ||
829 | /* | |
830 | * Just make sure we have seen our old_write and synchronize | |
831 | * with any interrupts that come in. | |
832 | */ | |
833 | barrier(); | |
834 | ||
835 | /* | |
836 | * If the tail page is still the same as what we think | |
837 | * it is, then it is up to us to update the tail | |
838 | * pointer. | |
839 | */ | |
840 | if (tail_page == cpu_buffer->tail_page) { | |
841 | /* Zero the write counter */ | |
842 | unsigned long val = old_write & ~RB_WRITE_MASK; | |
843 | unsigned long eval = old_entries & ~RB_WRITE_MASK; | |
844 | ||
845 | /* | |
846 | * This will only succeed if an interrupt did | |
847 | * not come in and change it. In which case, we | |
848 | * do not want to modify it. | |
da706d8b LJ |
849 | * |
850 | * We add (void) to let the compiler know that we do not care | |
851 | * about the return value of these functions. We use the | |
852 | * cmpxchg to only update if an interrupt did not already | |
853 | * do it for us. If the cmpxchg fails, we don't care. | |
77ae365e | 854 | */ |
da706d8b LJ |
855 | (void)local_cmpxchg(&next_page->write, old_write, val); |
856 | (void)local_cmpxchg(&next_page->entries, old_entries, eval); | |
77ae365e SR |
857 | |
858 | /* | |
859 | * No need to worry about races with clearing out the commit. | |
860 | * it only can increment when a commit takes place. But that | |
861 | * only happens in the outer most nested commit. | |
862 | */ | |
863 | local_set(&next_page->page->commit, 0); | |
864 | ||
865 | old_tail = cmpxchg(&cpu_buffer->tail_page, | |
866 | tail_page, next_page); | |
867 | ||
868 | if (old_tail == tail_page) | |
869 | ret = 1; | |
870 | } | |
871 | ||
872 | return ret; | |
873 | } | |
874 | ||
875 | static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer, | |
876 | struct buffer_page *bpage) | |
877 | { | |
878 | unsigned long val = (unsigned long)bpage; | |
879 | ||
880 | if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK)) | |
881 | return 1; | |
882 | ||
883 | return 0; | |
884 | } | |
885 | ||
886 | /** | |
887 | * rb_check_list - make sure a pointer to a list has the last bits zero | |
888 | */ | |
889 | static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer, | |
890 | struct list_head *list) | |
891 | { | |
892 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev)) | |
893 | return 1; | |
894 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next)) | |
895 | return 1; | |
896 | return 0; | |
897 | } | |
898 | ||
7a8e76a3 SR |
899 | /** |
900 | * check_pages - integrity check of buffer pages | |
901 | * @cpu_buffer: CPU buffer with pages to test | |
902 | * | |
c3706f00 | 903 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
904 | * been corrupted. |
905 | */ | |
906 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
907 | { | |
3adc54fa | 908 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 909 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 910 | |
77ae365e SR |
911 | rb_head_page_deactivate(cpu_buffer); |
912 | ||
3e89c7bb SR |
913 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
914 | return -1; | |
915 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
916 | return -1; | |
7a8e76a3 | 917 | |
77ae365e SR |
918 | if (rb_check_list(cpu_buffer, head)) |
919 | return -1; | |
920 | ||
044fa782 | 921 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 922 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 923 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
924 | return -1; |
925 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 926 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 927 | return -1; |
77ae365e SR |
928 | if (rb_check_list(cpu_buffer, &bpage->list)) |
929 | return -1; | |
7a8e76a3 SR |
930 | } |
931 | ||
77ae365e SR |
932 | rb_head_page_activate(cpu_buffer); |
933 | ||
7a8e76a3 SR |
934 | return 0; |
935 | } | |
936 | ||
7a8e76a3 SR |
937 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, |
938 | unsigned nr_pages) | |
939 | { | |
044fa782 | 940 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
941 | unsigned long addr; |
942 | LIST_HEAD(pages); | |
943 | unsigned i; | |
944 | ||
3adc54fa SR |
945 | WARN_ON(!nr_pages); |
946 | ||
7a8e76a3 | 947 | for (i = 0; i < nr_pages; i++) { |
044fa782 | 948 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
aa1e0e3b | 949 | GFP_KERNEL, cpu_to_node(cpu_buffer->cpu)); |
044fa782 | 950 | if (!bpage) |
e4c2ce82 | 951 | goto free_pages; |
77ae365e SR |
952 | |
953 | rb_check_bpage(cpu_buffer, bpage); | |
954 | ||
044fa782 | 955 | list_add(&bpage->list, &pages); |
e4c2ce82 | 956 | |
7a8e76a3 SR |
957 | addr = __get_free_page(GFP_KERNEL); |
958 | if (!addr) | |
959 | goto free_pages; | |
044fa782 SR |
960 | bpage->page = (void *)addr; |
961 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
962 | } |
963 | ||
3adc54fa SR |
964 | /* |
965 | * The ring buffer page list is a circular list that does not | |
966 | * start and end with a list head. All page list items point to | |
967 | * other pages. | |
968 | */ | |
969 | cpu_buffer->pages = pages.next; | |
970 | list_del(&pages); | |
7a8e76a3 SR |
971 | |
972 | rb_check_pages(cpu_buffer); | |
973 | ||
974 | return 0; | |
975 | ||
976 | free_pages: | |
044fa782 SR |
977 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
978 | list_del_init(&bpage->list); | |
979 | free_buffer_page(bpage); | |
7a8e76a3 SR |
980 | } |
981 | return -ENOMEM; | |
982 | } | |
983 | ||
984 | static struct ring_buffer_per_cpu * | |
985 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu) | |
986 | { | |
987 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 988 | struct buffer_page *bpage; |
d769041f | 989 | unsigned long addr; |
7a8e76a3 SR |
990 | int ret; |
991 | ||
992 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
993 | GFP_KERNEL, cpu_to_node(cpu)); | |
994 | if (!cpu_buffer) | |
995 | return NULL; | |
996 | ||
997 | cpu_buffer->cpu = cpu; | |
998 | cpu_buffer->buffer = buffer; | |
f83c9d0f | 999 | spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 1000 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
3e03fb7f | 1001 | cpu_buffer->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; |
7a8e76a3 | 1002 | |
044fa782 | 1003 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 1004 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 1005 | if (!bpage) |
e4c2ce82 SR |
1006 | goto fail_free_buffer; |
1007 | ||
77ae365e SR |
1008 | rb_check_bpage(cpu_buffer, bpage); |
1009 | ||
044fa782 | 1010 | cpu_buffer->reader_page = bpage; |
d769041f SR |
1011 | addr = __get_free_page(GFP_KERNEL); |
1012 | if (!addr) | |
e4c2ce82 | 1013 | goto fail_free_reader; |
044fa782 SR |
1014 | bpage->page = (void *)addr; |
1015 | rb_init_page(bpage->page); | |
e4c2ce82 | 1016 | |
d769041f | 1017 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 1018 | |
7a8e76a3 SR |
1019 | ret = rb_allocate_pages(cpu_buffer, buffer->pages); |
1020 | if (ret < 0) | |
d769041f | 1021 | goto fail_free_reader; |
7a8e76a3 SR |
1022 | |
1023 | cpu_buffer->head_page | |
3adc54fa | 1024 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 1025 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 | 1026 | |
77ae365e SR |
1027 | rb_head_page_activate(cpu_buffer); |
1028 | ||
7a8e76a3 SR |
1029 | return cpu_buffer; |
1030 | ||
d769041f SR |
1031 | fail_free_reader: |
1032 | free_buffer_page(cpu_buffer->reader_page); | |
1033 | ||
7a8e76a3 SR |
1034 | fail_free_buffer: |
1035 | kfree(cpu_buffer); | |
1036 | return NULL; | |
1037 | } | |
1038 | ||
1039 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
1040 | { | |
3adc54fa | 1041 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1042 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1043 | |
d769041f SR |
1044 | free_buffer_page(cpu_buffer->reader_page); |
1045 | ||
77ae365e SR |
1046 | rb_head_page_deactivate(cpu_buffer); |
1047 | ||
3adc54fa SR |
1048 | if (head) { |
1049 | list_for_each_entry_safe(bpage, tmp, head, list) { | |
1050 | list_del_init(&bpage->list); | |
1051 | free_buffer_page(bpage); | |
1052 | } | |
1053 | bpage = list_entry(head, struct buffer_page, list); | |
044fa782 | 1054 | free_buffer_page(bpage); |
7a8e76a3 | 1055 | } |
3adc54fa | 1056 | |
7a8e76a3 SR |
1057 | kfree(cpu_buffer); |
1058 | } | |
1059 | ||
59222efe | 1060 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
1061 | static int rb_cpu_notify(struct notifier_block *self, |
1062 | unsigned long action, void *hcpu); | |
554f786e SR |
1063 | #endif |
1064 | ||
7a8e76a3 SR |
1065 | /** |
1066 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 1067 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
1068 | * @flags: attributes to set for the ring buffer. |
1069 | * | |
1070 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
1071 | * flag. This flag means that the buffer will overwrite old data | |
1072 | * when the buffer wraps. If this flag is not set, the buffer will | |
1073 | * drop data when the tail hits the head. | |
1074 | */ | |
1f8a6a10 PZ |
1075 | struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
1076 | struct lock_class_key *key) | |
7a8e76a3 SR |
1077 | { |
1078 | struct ring_buffer *buffer; | |
1079 | int bsize; | |
1080 | int cpu; | |
1081 | ||
1082 | /* keep it in its own cache line */ | |
1083 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
1084 | GFP_KERNEL); | |
1085 | if (!buffer) | |
1086 | return NULL; | |
1087 | ||
9e01c1b7 RR |
1088 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
1089 | goto fail_free_buffer; | |
1090 | ||
7a8e76a3 SR |
1091 | buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
1092 | buffer->flags = flags; | |
37886f6a | 1093 | buffer->clock = trace_clock_local; |
1f8a6a10 | 1094 | buffer->reader_lock_key = key; |
7a8e76a3 SR |
1095 | |
1096 | /* need at least two pages */ | |
5f78abee SR |
1097 | if (buffer->pages < 2) |
1098 | buffer->pages = 2; | |
7a8e76a3 | 1099 | |
3bf832ce FW |
1100 | /* |
1101 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
1102 | * in early initcall, it will not be notified of secondary cpus. | |
1103 | * In that off case, we need to allocate for all possible cpus. | |
1104 | */ | |
1105 | #ifdef CONFIG_HOTPLUG_CPU | |
554f786e SR |
1106 | get_online_cpus(); |
1107 | cpumask_copy(buffer->cpumask, cpu_online_mask); | |
3bf832ce FW |
1108 | #else |
1109 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
1110 | #endif | |
7a8e76a3 SR |
1111 | buffer->cpus = nr_cpu_ids; |
1112 | ||
1113 | bsize = sizeof(void *) * nr_cpu_ids; | |
1114 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
1115 | GFP_KERNEL); | |
1116 | if (!buffer->buffers) | |
9e01c1b7 | 1117 | goto fail_free_cpumask; |
7a8e76a3 SR |
1118 | |
1119 | for_each_buffer_cpu(buffer, cpu) { | |
1120 | buffer->buffers[cpu] = | |
1121 | rb_allocate_cpu_buffer(buffer, cpu); | |
1122 | if (!buffer->buffers[cpu]) | |
1123 | goto fail_free_buffers; | |
1124 | } | |
1125 | ||
59222efe | 1126 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1127 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
1128 | buffer->cpu_notify.priority = 0; | |
1129 | register_cpu_notifier(&buffer->cpu_notify); | |
1130 | #endif | |
1131 | ||
1132 | put_online_cpus(); | |
7a8e76a3 SR |
1133 | mutex_init(&buffer->mutex); |
1134 | ||
1135 | return buffer; | |
1136 | ||
1137 | fail_free_buffers: | |
1138 | for_each_buffer_cpu(buffer, cpu) { | |
1139 | if (buffer->buffers[cpu]) | |
1140 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1141 | } | |
1142 | kfree(buffer->buffers); | |
1143 | ||
9e01c1b7 RR |
1144 | fail_free_cpumask: |
1145 | free_cpumask_var(buffer->cpumask); | |
554f786e | 1146 | put_online_cpus(); |
9e01c1b7 | 1147 | |
7a8e76a3 SR |
1148 | fail_free_buffer: |
1149 | kfree(buffer); | |
1150 | return NULL; | |
1151 | } | |
1f8a6a10 | 1152 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
1153 | |
1154 | /** | |
1155 | * ring_buffer_free - free a ring buffer. | |
1156 | * @buffer: the buffer to free. | |
1157 | */ | |
1158 | void | |
1159 | ring_buffer_free(struct ring_buffer *buffer) | |
1160 | { | |
1161 | int cpu; | |
1162 | ||
554f786e SR |
1163 | get_online_cpus(); |
1164 | ||
59222efe | 1165 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1166 | unregister_cpu_notifier(&buffer->cpu_notify); |
1167 | #endif | |
1168 | ||
7a8e76a3 SR |
1169 | for_each_buffer_cpu(buffer, cpu) |
1170 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1171 | ||
554f786e SR |
1172 | put_online_cpus(); |
1173 | ||
bd3f0221 | 1174 | kfree(buffer->buffers); |
9e01c1b7 RR |
1175 | free_cpumask_var(buffer->cpumask); |
1176 | ||
7a8e76a3 SR |
1177 | kfree(buffer); |
1178 | } | |
c4f50183 | 1179 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 1180 | |
37886f6a SR |
1181 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
1182 | u64 (*clock)(void)) | |
1183 | { | |
1184 | buffer->clock = clock; | |
1185 | } | |
1186 | ||
7a8e76a3 SR |
1187 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
1188 | ||
1189 | static void | |
1190 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned nr_pages) | |
1191 | { | |
044fa782 | 1192 | struct buffer_page *bpage; |
7a8e76a3 SR |
1193 | struct list_head *p; |
1194 | unsigned i; | |
1195 | ||
1196 | atomic_inc(&cpu_buffer->record_disabled); | |
1197 | synchronize_sched(); | |
1198 | ||
f7112949 | 1199 | spin_lock_irq(&cpu_buffer->reader_lock); |
77ae365e SR |
1200 | rb_head_page_deactivate(cpu_buffer); |
1201 | ||
7a8e76a3 | 1202 | for (i = 0; i < nr_pages; i++) { |
3adc54fa | 1203 | if (RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages))) |
3e89c7bb | 1204 | return; |
3adc54fa | 1205 | p = cpu_buffer->pages->next; |
044fa782 SR |
1206 | bpage = list_entry(p, struct buffer_page, list); |
1207 | list_del_init(&bpage->list); | |
1208 | free_buffer_page(bpage); | |
7a8e76a3 | 1209 | } |
3adc54fa | 1210 | if (RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages))) |
3e89c7bb | 1211 | return; |
7a8e76a3 SR |
1212 | |
1213 | rb_reset_cpu(cpu_buffer); | |
f7112949 | 1214 | spin_unlock_irq(&cpu_buffer->reader_lock); |
7a8e76a3 SR |
1215 | |
1216 | rb_check_pages(cpu_buffer); | |
1217 | ||
1218 | atomic_dec(&cpu_buffer->record_disabled); | |
1219 | ||
1220 | } | |
1221 | ||
1222 | static void | |
1223 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
1224 | struct list_head *pages, unsigned nr_pages) | |
1225 | { | |
044fa782 | 1226 | struct buffer_page *bpage; |
7a8e76a3 SR |
1227 | struct list_head *p; |
1228 | unsigned i; | |
1229 | ||
1230 | atomic_inc(&cpu_buffer->record_disabled); | |
1231 | synchronize_sched(); | |
1232 | ||
77ae365e SR |
1233 | spin_lock_irq(&cpu_buffer->reader_lock); |
1234 | rb_head_page_deactivate(cpu_buffer); | |
1235 | ||
7a8e76a3 | 1236 | for (i = 0; i < nr_pages; i++) { |
3e89c7bb SR |
1237 | if (RB_WARN_ON(cpu_buffer, list_empty(pages))) |
1238 | return; | |
7a8e76a3 | 1239 | p = pages->next; |
044fa782 SR |
1240 | bpage = list_entry(p, struct buffer_page, list); |
1241 | list_del_init(&bpage->list); | |
3adc54fa | 1242 | list_add_tail(&bpage->list, cpu_buffer->pages); |
7a8e76a3 SR |
1243 | } |
1244 | rb_reset_cpu(cpu_buffer); | |
77ae365e | 1245 | spin_unlock_irq(&cpu_buffer->reader_lock); |
7a8e76a3 SR |
1246 | |
1247 | rb_check_pages(cpu_buffer); | |
1248 | ||
1249 | atomic_dec(&cpu_buffer->record_disabled); | |
1250 | } | |
1251 | ||
1252 | /** | |
1253 | * ring_buffer_resize - resize the ring buffer | |
1254 | * @buffer: the buffer to resize. | |
1255 | * @size: the new size. | |
1256 | * | |
1257 | * The tracer is responsible for making sure that the buffer is | |
1258 | * not being used while changing the size. | |
1259 | * Note: We may be able to change the above requirement by using | |
1260 | * RCU synchronizations. | |
1261 | * | |
1262 | * Minimum size is 2 * BUF_PAGE_SIZE. | |
1263 | * | |
1264 | * Returns -1 on failure. | |
1265 | */ | |
1266 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size) | |
1267 | { | |
1268 | struct ring_buffer_per_cpu *cpu_buffer; | |
1269 | unsigned nr_pages, rm_pages, new_pages; | |
044fa782 | 1270 | struct buffer_page *bpage, *tmp; |
7a8e76a3 SR |
1271 | unsigned long buffer_size; |
1272 | unsigned long addr; | |
1273 | LIST_HEAD(pages); | |
1274 | int i, cpu; | |
1275 | ||
ee51a1de IM |
1276 | /* |
1277 | * Always succeed at resizing a non-existent buffer: | |
1278 | */ | |
1279 | if (!buffer) | |
1280 | return size; | |
1281 | ||
7a8e76a3 SR |
1282 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
1283 | size *= BUF_PAGE_SIZE; | |
1284 | buffer_size = buffer->pages * BUF_PAGE_SIZE; | |
1285 | ||
1286 | /* we need a minimum of two pages */ | |
1287 | if (size < BUF_PAGE_SIZE * 2) | |
1288 | size = BUF_PAGE_SIZE * 2; | |
1289 | ||
1290 | if (size == buffer_size) | |
1291 | return size; | |
1292 | ||
1293 | mutex_lock(&buffer->mutex); | |
554f786e | 1294 | get_online_cpus(); |
7a8e76a3 SR |
1295 | |
1296 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); | |
1297 | ||
1298 | if (size < buffer_size) { | |
1299 | ||
1300 | /* easy case, just free pages */ | |
554f786e SR |
1301 | if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) |
1302 | goto out_fail; | |
7a8e76a3 SR |
1303 | |
1304 | rm_pages = buffer->pages - nr_pages; | |
1305 | ||
1306 | for_each_buffer_cpu(buffer, cpu) { | |
1307 | cpu_buffer = buffer->buffers[cpu]; | |
1308 | rb_remove_pages(cpu_buffer, rm_pages); | |
1309 | } | |
1310 | goto out; | |
1311 | } | |
1312 | ||
1313 | /* | |
1314 | * This is a bit more difficult. We only want to add pages | |
1315 | * when we can allocate enough for all CPUs. We do this | |
1316 | * by allocating all the pages and storing them on a local | |
1317 | * link list. If we succeed in our allocation, then we | |
1318 | * add these pages to the cpu_buffers. Otherwise we just free | |
1319 | * them all and return -ENOMEM; | |
1320 | */ | |
554f786e SR |
1321 | if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) |
1322 | goto out_fail; | |
f536aafc | 1323 | |
7a8e76a3 SR |
1324 | new_pages = nr_pages - buffer->pages; |
1325 | ||
1326 | for_each_buffer_cpu(buffer, cpu) { | |
1327 | for (i = 0; i < new_pages; i++) { | |
044fa782 | 1328 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), |
e4c2ce82 SR |
1329 | cache_line_size()), |
1330 | GFP_KERNEL, cpu_to_node(cpu)); | |
044fa782 | 1331 | if (!bpage) |
e4c2ce82 | 1332 | goto free_pages; |
044fa782 | 1333 | list_add(&bpage->list, &pages); |
7a8e76a3 SR |
1334 | addr = __get_free_page(GFP_KERNEL); |
1335 | if (!addr) | |
1336 | goto free_pages; | |
044fa782 SR |
1337 | bpage->page = (void *)addr; |
1338 | rb_init_page(bpage->page); | |
7a8e76a3 SR |
1339 | } |
1340 | } | |
1341 | ||
1342 | for_each_buffer_cpu(buffer, cpu) { | |
1343 | cpu_buffer = buffer->buffers[cpu]; | |
1344 | rb_insert_pages(cpu_buffer, &pages, new_pages); | |
1345 | } | |
1346 | ||
554f786e SR |
1347 | if (RB_WARN_ON(buffer, !list_empty(&pages))) |
1348 | goto out_fail; | |
7a8e76a3 SR |
1349 | |
1350 | out: | |
1351 | buffer->pages = nr_pages; | |
554f786e | 1352 | put_online_cpus(); |
7a8e76a3 SR |
1353 | mutex_unlock(&buffer->mutex); |
1354 | ||
1355 | return size; | |
1356 | ||
1357 | free_pages: | |
044fa782 SR |
1358 | list_for_each_entry_safe(bpage, tmp, &pages, list) { |
1359 | list_del_init(&bpage->list); | |
1360 | free_buffer_page(bpage); | |
7a8e76a3 | 1361 | } |
554f786e | 1362 | put_online_cpus(); |
641d2f63 | 1363 | mutex_unlock(&buffer->mutex); |
7a8e76a3 | 1364 | return -ENOMEM; |
554f786e SR |
1365 | |
1366 | /* | |
1367 | * Something went totally wrong, and we are too paranoid | |
1368 | * to even clean up the mess. | |
1369 | */ | |
1370 | out_fail: | |
1371 | put_online_cpus(); | |
1372 | mutex_unlock(&buffer->mutex); | |
1373 | return -1; | |
7a8e76a3 | 1374 | } |
c4f50183 | 1375 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 1376 | |
8789a9e7 | 1377 | static inline void * |
044fa782 | 1378 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 1379 | { |
044fa782 | 1380 | return bpage->data + index; |
8789a9e7 SR |
1381 | } |
1382 | ||
044fa782 | 1383 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 1384 | { |
044fa782 | 1385 | return bpage->page->data + index; |
7a8e76a3 SR |
1386 | } |
1387 | ||
1388 | static inline struct ring_buffer_event * | |
d769041f | 1389 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1390 | { |
6f807acd SR |
1391 | return __rb_page_index(cpu_buffer->reader_page, |
1392 | cpu_buffer->reader_page->read); | |
1393 | } | |
1394 | ||
7a8e76a3 SR |
1395 | static inline struct ring_buffer_event * |
1396 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
1397 | { | |
6f807acd | 1398 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
1399 | } |
1400 | ||
77ae365e | 1401 | static inline unsigned long rb_page_write(struct buffer_page *bpage) |
bf41a158 | 1402 | { |
77ae365e | 1403 | return local_read(&bpage->write) & RB_WRITE_MASK; |
bf41a158 SR |
1404 | } |
1405 | ||
1406 | static inline unsigned rb_page_commit(struct buffer_page *bpage) | |
1407 | { | |
abc9b56d | 1408 | return local_read(&bpage->page->commit); |
bf41a158 SR |
1409 | } |
1410 | ||
77ae365e SR |
1411 | static inline unsigned long rb_page_entries(struct buffer_page *bpage) |
1412 | { | |
1413 | return local_read(&bpage->entries) & RB_WRITE_MASK; | |
1414 | } | |
1415 | ||
bf41a158 SR |
1416 | /* Size is determined by what has been commited */ |
1417 | static inline unsigned rb_page_size(struct buffer_page *bpage) | |
1418 | { | |
1419 | return rb_page_commit(bpage); | |
1420 | } | |
1421 | ||
1422 | static inline unsigned | |
1423 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
1424 | { | |
1425 | return rb_page_commit(cpu_buffer->commit_page); | |
1426 | } | |
1427 | ||
bf41a158 SR |
1428 | static inline unsigned |
1429 | rb_event_index(struct ring_buffer_event *event) | |
1430 | { | |
1431 | unsigned long addr = (unsigned long)event; | |
1432 | ||
22f470f8 | 1433 | return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE; |
bf41a158 SR |
1434 | } |
1435 | ||
0f0c85fc | 1436 | static inline int |
fa743953 SR |
1437 | rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1438 | struct ring_buffer_event *event) | |
bf41a158 SR |
1439 | { |
1440 | unsigned long addr = (unsigned long)event; | |
1441 | unsigned long index; | |
1442 | ||
1443 | index = rb_event_index(event); | |
1444 | addr &= PAGE_MASK; | |
1445 | ||
1446 | return cpu_buffer->commit_page->page == (void *)addr && | |
1447 | rb_commit_index(cpu_buffer) == index; | |
1448 | } | |
1449 | ||
34a148bf | 1450 | static void |
bf41a158 | 1451 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1452 | { |
77ae365e SR |
1453 | unsigned long max_count; |
1454 | ||
bf41a158 SR |
1455 | /* |
1456 | * We only race with interrupts and NMIs on this CPU. | |
1457 | * If we own the commit event, then we can commit | |
1458 | * all others that interrupted us, since the interruptions | |
1459 | * are in stack format (they finish before they come | |
1460 | * back to us). This allows us to do a simple loop to | |
1461 | * assign the commit to the tail. | |
1462 | */ | |
a8ccf1d6 | 1463 | again: |
77ae365e SR |
1464 | max_count = cpu_buffer->buffer->pages * 100; |
1465 | ||
bf41a158 | 1466 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
77ae365e SR |
1467 | if (RB_WARN_ON(cpu_buffer, !(--max_count))) |
1468 | return; | |
1469 | if (RB_WARN_ON(cpu_buffer, | |
1470 | rb_is_reader_page(cpu_buffer->tail_page))) | |
1471 | return; | |
1472 | local_set(&cpu_buffer->commit_page->page->commit, | |
1473 | rb_page_write(cpu_buffer->commit_page)); | |
bf41a158 | 1474 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); |
abc9b56d SR |
1475 | cpu_buffer->write_stamp = |
1476 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1477 | /* add barrier to keep gcc from optimizing too much */ |
1478 | barrier(); | |
1479 | } | |
1480 | while (rb_commit_index(cpu_buffer) != | |
1481 | rb_page_write(cpu_buffer->commit_page)) { | |
77ae365e SR |
1482 | |
1483 | local_set(&cpu_buffer->commit_page->page->commit, | |
1484 | rb_page_write(cpu_buffer->commit_page)); | |
1485 | RB_WARN_ON(cpu_buffer, | |
1486 | local_read(&cpu_buffer->commit_page->page->commit) & | |
1487 | ~RB_WRITE_MASK); | |
bf41a158 SR |
1488 | barrier(); |
1489 | } | |
a8ccf1d6 SR |
1490 | |
1491 | /* again, keep gcc from optimizing */ | |
1492 | barrier(); | |
1493 | ||
1494 | /* | |
1495 | * If an interrupt came in just after the first while loop | |
1496 | * and pushed the tail page forward, we will be left with | |
1497 | * a dangling commit that will never go forward. | |
1498 | */ | |
1499 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1500 | goto again; | |
7a8e76a3 SR |
1501 | } |
1502 | ||
d769041f | 1503 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1504 | { |
abc9b56d | 1505 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1506 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1507 | } |
1508 | ||
34a148bf | 1509 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1510 | { |
1511 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1512 | ||
1513 | /* | |
1514 | * The iterator could be on the reader page (it starts there). | |
1515 | * But the head could have moved, since the reader was | |
1516 | * found. Check for this case and assign the iterator | |
1517 | * to the head page instead of next. | |
1518 | */ | |
1519 | if (iter->head_page == cpu_buffer->reader_page) | |
77ae365e | 1520 | iter->head_page = rb_set_head_page(cpu_buffer); |
d769041f SR |
1521 | else |
1522 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1523 | ||
abc9b56d | 1524 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1525 | iter->head = 0; |
1526 | } | |
1527 | ||
1528 | /** | |
1529 | * ring_buffer_update_event - update event type and data | |
1530 | * @event: the even to update | |
1531 | * @type: the type of event | |
1532 | * @length: the size of the event field in the ring buffer | |
1533 | * | |
1534 | * Update the type and data fields of the event. The length | |
1535 | * is the actual size that is written to the ring buffer, | |
1536 | * and with this, we can determine what to place into the | |
1537 | * data field. | |
1538 | */ | |
34a148bf | 1539 | static void |
7a8e76a3 SR |
1540 | rb_update_event(struct ring_buffer_event *event, |
1541 | unsigned type, unsigned length) | |
1542 | { | |
334d4169 | 1543 | event->type_len = type; |
7a8e76a3 SR |
1544 | |
1545 | switch (type) { | |
1546 | ||
1547 | case RINGBUF_TYPE_PADDING: | |
7a8e76a3 | 1548 | case RINGBUF_TYPE_TIME_EXTEND: |
7a8e76a3 | 1549 | case RINGBUF_TYPE_TIME_STAMP: |
7a8e76a3 SR |
1550 | break; |
1551 | ||
334d4169 | 1552 | case 0: |
7a8e76a3 | 1553 | length -= RB_EVNT_HDR_SIZE; |
334d4169 | 1554 | if (length > RB_MAX_SMALL_DATA) |
7a8e76a3 | 1555 | event->array[0] = length; |
334d4169 LJ |
1556 | else |
1557 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
7a8e76a3 SR |
1558 | break; |
1559 | default: | |
1560 | BUG(); | |
1561 | } | |
1562 | } | |
1563 | ||
77ae365e SR |
1564 | /* |
1565 | * rb_handle_head_page - writer hit the head page | |
1566 | * | |
1567 | * Returns: +1 to retry page | |
1568 | * 0 to continue | |
1569 | * -1 on error | |
1570 | */ | |
1571 | static int | |
1572 | rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
1573 | struct buffer_page *tail_page, | |
1574 | struct buffer_page *next_page) | |
1575 | { | |
1576 | struct buffer_page *new_head; | |
1577 | int entries; | |
1578 | int type; | |
1579 | int ret; | |
1580 | ||
1581 | entries = rb_page_entries(next_page); | |
1582 | ||
1583 | /* | |
1584 | * The hard part is here. We need to move the head | |
1585 | * forward, and protect against both readers on | |
1586 | * other CPUs and writers coming in via interrupts. | |
1587 | */ | |
1588 | type = rb_head_page_set_update(cpu_buffer, next_page, tail_page, | |
1589 | RB_PAGE_HEAD); | |
1590 | ||
1591 | /* | |
1592 | * type can be one of four: | |
1593 | * NORMAL - an interrupt already moved it for us | |
1594 | * HEAD - we are the first to get here. | |
1595 | * UPDATE - we are the interrupt interrupting | |
1596 | * a current move. | |
1597 | * MOVED - a reader on another CPU moved the next | |
1598 | * pointer to its reader page. Give up | |
1599 | * and try again. | |
1600 | */ | |
1601 | ||
1602 | switch (type) { | |
1603 | case RB_PAGE_HEAD: | |
1604 | /* | |
1605 | * We changed the head to UPDATE, thus | |
1606 | * it is our responsibility to update | |
1607 | * the counters. | |
1608 | */ | |
1609 | local_add(entries, &cpu_buffer->overrun); | |
1610 | ||
1611 | /* | |
1612 | * The entries will be zeroed out when we move the | |
1613 | * tail page. | |
1614 | */ | |
1615 | ||
1616 | /* still more to do */ | |
1617 | break; | |
1618 | ||
1619 | case RB_PAGE_UPDATE: | |
1620 | /* | |
1621 | * This is an interrupt that interrupt the | |
1622 | * previous update. Still more to do. | |
1623 | */ | |
1624 | break; | |
1625 | case RB_PAGE_NORMAL: | |
1626 | /* | |
1627 | * An interrupt came in before the update | |
1628 | * and processed this for us. | |
1629 | * Nothing left to do. | |
1630 | */ | |
1631 | return 1; | |
1632 | case RB_PAGE_MOVED: | |
1633 | /* | |
1634 | * The reader is on another CPU and just did | |
1635 | * a swap with our next_page. | |
1636 | * Try again. | |
1637 | */ | |
1638 | return 1; | |
1639 | default: | |
1640 | RB_WARN_ON(cpu_buffer, 1); /* WTF??? */ | |
1641 | return -1; | |
1642 | } | |
1643 | ||
1644 | /* | |
1645 | * Now that we are here, the old head pointer is | |
1646 | * set to UPDATE. This will keep the reader from | |
1647 | * swapping the head page with the reader page. | |
1648 | * The reader (on another CPU) will spin till | |
1649 | * we are finished. | |
1650 | * | |
1651 | * We just need to protect against interrupts | |
1652 | * doing the job. We will set the next pointer | |
1653 | * to HEAD. After that, we set the old pointer | |
1654 | * to NORMAL, but only if it was HEAD before. | |
1655 | * otherwise we are an interrupt, and only | |
1656 | * want the outer most commit to reset it. | |
1657 | */ | |
1658 | new_head = next_page; | |
1659 | rb_inc_page(cpu_buffer, &new_head); | |
1660 | ||
1661 | ret = rb_head_page_set_head(cpu_buffer, new_head, next_page, | |
1662 | RB_PAGE_NORMAL); | |
1663 | ||
1664 | /* | |
1665 | * Valid returns are: | |
1666 | * HEAD - an interrupt came in and already set it. | |
1667 | * NORMAL - One of two things: | |
1668 | * 1) We really set it. | |
1669 | * 2) A bunch of interrupts came in and moved | |
1670 | * the page forward again. | |
1671 | */ | |
1672 | switch (ret) { | |
1673 | case RB_PAGE_HEAD: | |
1674 | case RB_PAGE_NORMAL: | |
1675 | /* OK */ | |
1676 | break; | |
1677 | default: | |
1678 | RB_WARN_ON(cpu_buffer, 1); | |
1679 | return -1; | |
1680 | } | |
1681 | ||
1682 | /* | |
1683 | * It is possible that an interrupt came in, | |
1684 | * set the head up, then more interrupts came in | |
1685 | * and moved it again. When we get back here, | |
1686 | * the page would have been set to NORMAL but we | |
1687 | * just set it back to HEAD. | |
1688 | * | |
1689 | * How do you detect this? Well, if that happened | |
1690 | * the tail page would have moved. | |
1691 | */ | |
1692 | if (ret == RB_PAGE_NORMAL) { | |
1693 | /* | |
1694 | * If the tail had moved passed next, then we need | |
1695 | * to reset the pointer. | |
1696 | */ | |
1697 | if (cpu_buffer->tail_page != tail_page && | |
1698 | cpu_buffer->tail_page != next_page) | |
1699 | rb_head_page_set_normal(cpu_buffer, new_head, | |
1700 | next_page, | |
1701 | RB_PAGE_HEAD); | |
1702 | } | |
1703 | ||
1704 | /* | |
1705 | * If this was the outer most commit (the one that | |
1706 | * changed the original pointer from HEAD to UPDATE), | |
1707 | * then it is up to us to reset it to NORMAL. | |
1708 | */ | |
1709 | if (type == RB_PAGE_HEAD) { | |
1710 | ret = rb_head_page_set_normal(cpu_buffer, next_page, | |
1711 | tail_page, | |
1712 | RB_PAGE_UPDATE); | |
1713 | if (RB_WARN_ON(cpu_buffer, | |
1714 | ret != RB_PAGE_UPDATE)) | |
1715 | return -1; | |
1716 | } | |
1717 | ||
1718 | return 0; | |
1719 | } | |
1720 | ||
34a148bf | 1721 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
1722 | { |
1723 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
1724 | ||
1725 | /* zero length can cause confusions */ | |
1726 | if (!length) | |
1727 | length = 1; | |
1728 | ||
1729 | if (length > RB_MAX_SMALL_DATA) | |
1730 | length += sizeof(event.array[0]); | |
1731 | ||
1732 | length += RB_EVNT_HDR_SIZE; | |
1733 | length = ALIGN(length, RB_ALIGNMENT); | |
1734 | ||
1735 | return length; | |
1736 | } | |
1737 | ||
c7b09308 SR |
1738 | static inline void |
1739 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
1740 | struct buffer_page *tail_page, | |
1741 | unsigned long tail, unsigned long length) | |
1742 | { | |
1743 | struct ring_buffer_event *event; | |
1744 | ||
1745 | /* | |
1746 | * Only the event that crossed the page boundary | |
1747 | * must fill the old tail_page with padding. | |
1748 | */ | |
1749 | if (tail >= BUF_PAGE_SIZE) { | |
1750 | local_sub(length, &tail_page->write); | |
1751 | return; | |
1752 | } | |
1753 | ||
1754 | event = __rb_page_index(tail_page, tail); | |
b0b7065b | 1755 | kmemcheck_annotate_bitfield(event, bitfield); |
c7b09308 SR |
1756 | |
1757 | /* | |
1758 | * If this event is bigger than the minimum size, then | |
1759 | * we need to be careful that we don't subtract the | |
1760 | * write counter enough to allow another writer to slip | |
1761 | * in on this page. | |
1762 | * We put in a discarded commit instead, to make sure | |
1763 | * that this space is not used again. | |
1764 | * | |
1765 | * If we are less than the minimum size, we don't need to | |
1766 | * worry about it. | |
1767 | */ | |
1768 | if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) { | |
1769 | /* No room for any events */ | |
1770 | ||
1771 | /* Mark the rest of the page with padding */ | |
1772 | rb_event_set_padding(event); | |
1773 | ||
1774 | /* Set the write back to the previous setting */ | |
1775 | local_sub(length, &tail_page->write); | |
1776 | return; | |
1777 | } | |
1778 | ||
1779 | /* Put in a discarded event */ | |
1780 | event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE; | |
1781 | event->type_len = RINGBUF_TYPE_PADDING; | |
1782 | /* time delta must be non zero */ | |
1783 | event->time_delta = 1; | |
c7b09308 SR |
1784 | |
1785 | /* Set write to end of buffer */ | |
1786 | length = (tail + length) - BUF_PAGE_SIZE; | |
1787 | local_sub(length, &tail_page->write); | |
1788 | } | |
6634ff26 | 1789 | |
7a8e76a3 | 1790 | static struct ring_buffer_event * |
6634ff26 SR |
1791 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
1792 | unsigned long length, unsigned long tail, | |
6634ff26 | 1793 | struct buffer_page *tail_page, u64 *ts) |
7a8e76a3 | 1794 | { |
5a50e33c | 1795 | struct buffer_page *commit_page = cpu_buffer->commit_page; |
7a8e76a3 | 1796 | struct ring_buffer *buffer = cpu_buffer->buffer; |
77ae365e SR |
1797 | struct buffer_page *next_page; |
1798 | int ret; | |
aa20ae84 SR |
1799 | |
1800 | next_page = tail_page; | |
1801 | ||
aa20ae84 SR |
1802 | rb_inc_page(cpu_buffer, &next_page); |
1803 | ||
aa20ae84 SR |
1804 | /* |
1805 | * If for some reason, we had an interrupt storm that made | |
1806 | * it all the way around the buffer, bail, and warn | |
1807 | * about it. | |
1808 | */ | |
1809 | if (unlikely(next_page == commit_page)) { | |
77ae365e | 1810 | local_inc(&cpu_buffer->commit_overrun); |
aa20ae84 SR |
1811 | goto out_reset; |
1812 | } | |
1813 | ||
77ae365e SR |
1814 | /* |
1815 | * This is where the fun begins! | |
1816 | * | |
1817 | * We are fighting against races between a reader that | |
1818 | * could be on another CPU trying to swap its reader | |
1819 | * page with the buffer head. | |
1820 | * | |
1821 | * We are also fighting against interrupts coming in and | |
1822 | * moving the head or tail on us as well. | |
1823 | * | |
1824 | * If the next page is the head page then we have filled | |
1825 | * the buffer, unless the commit page is still on the | |
1826 | * reader page. | |
1827 | */ | |
1828 | if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) { | |
aa20ae84 | 1829 | |
77ae365e SR |
1830 | /* |
1831 | * If the commit is not on the reader page, then | |
1832 | * move the header page. | |
1833 | */ | |
1834 | if (!rb_is_reader_page(cpu_buffer->commit_page)) { | |
1835 | /* | |
1836 | * If we are not in overwrite mode, | |
1837 | * this is easy, just stop here. | |
1838 | */ | |
1839 | if (!(buffer->flags & RB_FL_OVERWRITE)) | |
1840 | goto out_reset; | |
1841 | ||
1842 | ret = rb_handle_head_page(cpu_buffer, | |
1843 | tail_page, | |
1844 | next_page); | |
1845 | if (ret < 0) | |
1846 | goto out_reset; | |
1847 | if (ret) | |
1848 | goto out_again; | |
1849 | } else { | |
1850 | /* | |
1851 | * We need to be careful here too. The | |
1852 | * commit page could still be on the reader | |
1853 | * page. We could have a small buffer, and | |
1854 | * have filled up the buffer with events | |
1855 | * from interrupts and such, and wrapped. | |
1856 | * | |
1857 | * Note, if the tail page is also the on the | |
1858 | * reader_page, we let it move out. | |
1859 | */ | |
1860 | if (unlikely((cpu_buffer->commit_page != | |
1861 | cpu_buffer->tail_page) && | |
1862 | (cpu_buffer->commit_page == | |
1863 | cpu_buffer->reader_page))) { | |
1864 | local_inc(&cpu_buffer->commit_overrun); | |
1865 | goto out_reset; | |
1866 | } | |
aa20ae84 SR |
1867 | } |
1868 | } | |
1869 | ||
77ae365e SR |
1870 | ret = rb_tail_page_update(cpu_buffer, tail_page, next_page); |
1871 | if (ret) { | |
1872 | /* | |
1873 | * Nested commits always have zero deltas, so | |
1874 | * just reread the time stamp | |
1875 | */ | |
6d3f1e12 | 1876 | *ts = rb_time_stamp(buffer); |
77ae365e | 1877 | next_page->page->time_stamp = *ts; |
aa20ae84 SR |
1878 | } |
1879 | ||
77ae365e | 1880 | out_again: |
aa20ae84 | 1881 | |
77ae365e | 1882 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
aa20ae84 SR |
1883 | |
1884 | /* fail and let the caller try again */ | |
1885 | return ERR_PTR(-EAGAIN); | |
1886 | ||
45141d46 | 1887 | out_reset: |
6f3b3440 | 1888 | /* reset write */ |
c7b09308 | 1889 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
6f3b3440 | 1890 | |
bf41a158 | 1891 | return NULL; |
7a8e76a3 SR |
1892 | } |
1893 | ||
6634ff26 SR |
1894 | static struct ring_buffer_event * |
1895 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
1896 | unsigned type, unsigned long length, u64 *ts) | |
1897 | { | |
5a50e33c | 1898 | struct buffer_page *tail_page; |
6634ff26 SR |
1899 | struct ring_buffer_event *event; |
1900 | unsigned long tail, write; | |
1901 | ||
6634ff26 SR |
1902 | tail_page = cpu_buffer->tail_page; |
1903 | write = local_add_return(length, &tail_page->write); | |
77ae365e SR |
1904 | |
1905 | /* set write to only the index of the write */ | |
1906 | write &= RB_WRITE_MASK; | |
6634ff26 SR |
1907 | tail = write - length; |
1908 | ||
1909 | /* See if we shot pass the end of this buffer page */ | |
1910 | if (write > BUF_PAGE_SIZE) | |
1911 | return rb_move_tail(cpu_buffer, length, tail, | |
5a50e33c | 1912 | tail_page, ts); |
6634ff26 SR |
1913 | |
1914 | /* We reserved something on the buffer */ | |
1915 | ||
6634ff26 | 1916 | event = __rb_page_index(tail_page, tail); |
1744a21d | 1917 | kmemcheck_annotate_bitfield(event, bitfield); |
6634ff26 SR |
1918 | rb_update_event(event, type, length); |
1919 | ||
1920 | /* The passed in type is zero for DATA */ | |
1921 | if (likely(!type)) | |
1922 | local_inc(&tail_page->entries); | |
1923 | ||
1924 | /* | |
fa743953 SR |
1925 | * If this is the first commit on the page, then update |
1926 | * its timestamp. | |
6634ff26 | 1927 | */ |
fa743953 SR |
1928 | if (!tail) |
1929 | tail_page->page->time_stamp = *ts; | |
6634ff26 SR |
1930 | |
1931 | return event; | |
1932 | } | |
1933 | ||
edd813bf SR |
1934 | static inline int |
1935 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
1936 | struct ring_buffer_event *event) | |
1937 | { | |
1938 | unsigned long new_index, old_index; | |
1939 | struct buffer_page *bpage; | |
1940 | unsigned long index; | |
1941 | unsigned long addr; | |
1942 | ||
1943 | new_index = rb_event_index(event); | |
1944 | old_index = new_index + rb_event_length(event); | |
1945 | addr = (unsigned long)event; | |
1946 | addr &= PAGE_MASK; | |
1947 | ||
1948 | bpage = cpu_buffer->tail_page; | |
1949 | ||
1950 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { | |
77ae365e SR |
1951 | unsigned long write_mask = |
1952 | local_read(&bpage->write) & ~RB_WRITE_MASK; | |
edd813bf SR |
1953 | /* |
1954 | * This is on the tail page. It is possible that | |
1955 | * a write could come in and move the tail page | |
1956 | * and write to the next page. That is fine | |
1957 | * because we just shorten what is on this page. | |
1958 | */ | |
77ae365e SR |
1959 | old_index += write_mask; |
1960 | new_index += write_mask; | |
edd813bf SR |
1961 | index = local_cmpxchg(&bpage->write, old_index, new_index); |
1962 | if (index == old_index) | |
1963 | return 1; | |
1964 | } | |
1965 | ||
1966 | /* could not discard */ | |
1967 | return 0; | |
1968 | } | |
1969 | ||
7a8e76a3 SR |
1970 | static int |
1971 | rb_add_time_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
1972 | u64 *ts, u64 *delta) | |
1973 | { | |
1974 | struct ring_buffer_event *event; | |
1975 | static int once; | |
bf41a158 | 1976 | int ret; |
7a8e76a3 SR |
1977 | |
1978 | if (unlikely(*delta > (1ULL << 59) && !once++)) { | |
1979 | printk(KERN_WARNING "Delta way too big! %llu" | |
1980 | " ts=%llu write stamp = %llu\n", | |
e2862c94 SR |
1981 | (unsigned long long)*delta, |
1982 | (unsigned long long)*ts, | |
1983 | (unsigned long long)cpu_buffer->write_stamp); | |
7a8e76a3 SR |
1984 | WARN_ON(1); |
1985 | } | |
1986 | ||
1987 | /* | |
1988 | * The delta is too big, we to add a | |
1989 | * new timestamp. | |
1990 | */ | |
1991 | event = __rb_reserve_next(cpu_buffer, | |
1992 | RINGBUF_TYPE_TIME_EXTEND, | |
1993 | RB_LEN_TIME_EXTEND, | |
1994 | ts); | |
1995 | if (!event) | |
bf41a158 | 1996 | return -EBUSY; |
7a8e76a3 | 1997 | |
bf41a158 SR |
1998 | if (PTR_ERR(event) == -EAGAIN) |
1999 | return -EAGAIN; | |
2000 | ||
2001 | /* Only a commited time event can update the write stamp */ | |
fa743953 | 2002 | if (rb_event_is_commit(cpu_buffer, event)) { |
bf41a158 | 2003 | /* |
fa743953 SR |
2004 | * If this is the first on the page, then it was |
2005 | * updated with the page itself. Try to discard it | |
2006 | * and if we can't just make it zero. | |
bf41a158 SR |
2007 | */ |
2008 | if (rb_event_index(event)) { | |
2009 | event->time_delta = *delta & TS_MASK; | |
2010 | event->array[0] = *delta >> TS_SHIFT; | |
2011 | } else { | |
ea05b57c SR |
2012 | /* try to discard, since we do not need this */ |
2013 | if (!rb_try_to_discard(cpu_buffer, event)) { | |
2014 | /* nope, just zero it */ | |
2015 | event->time_delta = 0; | |
2016 | event->array[0] = 0; | |
2017 | } | |
bf41a158 | 2018 | } |
7a8e76a3 | 2019 | cpu_buffer->write_stamp = *ts; |
bf41a158 SR |
2020 | /* let the caller know this was the commit */ |
2021 | ret = 1; | |
2022 | } else { | |
edd813bf SR |
2023 | /* Try to discard the event */ |
2024 | if (!rb_try_to_discard(cpu_buffer, event)) { | |
2025 | /* Darn, this is just wasted space */ | |
2026 | event->time_delta = 0; | |
2027 | event->array[0] = 0; | |
edd813bf | 2028 | } |
f57a8a19 | 2029 | ret = 0; |
7a8e76a3 SR |
2030 | } |
2031 | ||
bf41a158 SR |
2032 | *delta = 0; |
2033 | ||
2034 | return ret; | |
7a8e76a3 SR |
2035 | } |
2036 | ||
fa743953 SR |
2037 | static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer) |
2038 | { | |
2039 | local_inc(&cpu_buffer->committing); | |
2040 | local_inc(&cpu_buffer->commits); | |
2041 | } | |
2042 | ||
2043 | static void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer) | |
2044 | { | |
2045 | unsigned long commits; | |
2046 | ||
2047 | if (RB_WARN_ON(cpu_buffer, | |
2048 | !local_read(&cpu_buffer->committing))) | |
2049 | return; | |
2050 | ||
2051 | again: | |
2052 | commits = local_read(&cpu_buffer->commits); | |
2053 | /* synchronize with interrupts */ | |
2054 | barrier(); | |
2055 | if (local_read(&cpu_buffer->committing) == 1) | |
2056 | rb_set_commit_to_write(cpu_buffer); | |
2057 | ||
2058 | local_dec(&cpu_buffer->committing); | |
2059 | ||
2060 | /* synchronize with interrupts */ | |
2061 | barrier(); | |
2062 | ||
2063 | /* | |
2064 | * Need to account for interrupts coming in between the | |
2065 | * updating of the commit page and the clearing of the | |
2066 | * committing counter. | |
2067 | */ | |
2068 | if (unlikely(local_read(&cpu_buffer->commits) != commits) && | |
2069 | !local_read(&cpu_buffer->committing)) { | |
2070 | local_inc(&cpu_buffer->committing); | |
2071 | goto again; | |
2072 | } | |
2073 | } | |
2074 | ||
7a8e76a3 | 2075 | static struct ring_buffer_event * |
62f0b3eb SR |
2076 | rb_reserve_next_event(struct ring_buffer *buffer, |
2077 | struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 2078 | unsigned long length) |
7a8e76a3 SR |
2079 | { |
2080 | struct ring_buffer_event *event; | |
168b6b1d | 2081 | u64 ts, delta = 0; |
bf41a158 | 2082 | int commit = 0; |
818e3dd3 | 2083 | int nr_loops = 0; |
7a8e76a3 | 2084 | |
fa743953 SR |
2085 | rb_start_commit(cpu_buffer); |
2086 | ||
85bac32c | 2087 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
62f0b3eb SR |
2088 | /* |
2089 | * Due to the ability to swap a cpu buffer from a buffer | |
2090 | * it is possible it was swapped before we committed. | |
2091 | * (committing stops a swap). We check for it here and | |
2092 | * if it happened, we have to fail the write. | |
2093 | */ | |
2094 | barrier(); | |
2095 | if (unlikely(ACCESS_ONCE(cpu_buffer->buffer) != buffer)) { | |
2096 | local_dec(&cpu_buffer->committing); | |
2097 | local_dec(&cpu_buffer->commits); | |
2098 | return NULL; | |
2099 | } | |
85bac32c | 2100 | #endif |
62f0b3eb | 2101 | |
be957c44 | 2102 | length = rb_calculate_event_length(length); |
bf41a158 | 2103 | again: |
818e3dd3 SR |
2104 | /* |
2105 | * We allow for interrupts to reenter here and do a trace. | |
2106 | * If one does, it will cause this original code to loop | |
2107 | * back here. Even with heavy interrupts happening, this | |
2108 | * should only happen a few times in a row. If this happens | |
2109 | * 1000 times in a row, there must be either an interrupt | |
2110 | * storm or we have something buggy. | |
2111 | * Bail! | |
2112 | */ | |
3e89c7bb | 2113 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
fa743953 | 2114 | goto out_fail; |
818e3dd3 | 2115 | |
6d3f1e12 | 2116 | ts = rb_time_stamp(cpu_buffer->buffer); |
7a8e76a3 | 2117 | |
bf41a158 SR |
2118 | /* |
2119 | * Only the first commit can update the timestamp. | |
2120 | * Yes there is a race here. If an interrupt comes in | |
2121 | * just after the conditional and it traces too, then it | |
2122 | * will also check the deltas. More than one timestamp may | |
2123 | * also be made. But only the entry that did the actual | |
2124 | * commit will be something other than zero. | |
2125 | */ | |
0f0c85fc SR |
2126 | if (likely(cpu_buffer->tail_page == cpu_buffer->commit_page && |
2127 | rb_page_write(cpu_buffer->tail_page) == | |
2128 | rb_commit_index(cpu_buffer))) { | |
168b6b1d | 2129 | u64 diff; |
bf41a158 | 2130 | |
168b6b1d | 2131 | diff = ts - cpu_buffer->write_stamp; |
7a8e76a3 | 2132 | |
168b6b1d | 2133 | /* make sure this diff is calculated here */ |
bf41a158 SR |
2134 | barrier(); |
2135 | ||
2136 | /* Did the write stamp get updated already? */ | |
2137 | if (unlikely(ts < cpu_buffer->write_stamp)) | |
168b6b1d | 2138 | goto get_event; |
bf41a158 | 2139 | |
168b6b1d SR |
2140 | delta = diff; |
2141 | if (unlikely(test_time_stamp(delta))) { | |
7a8e76a3 | 2142 | |
bf41a158 | 2143 | commit = rb_add_time_stamp(cpu_buffer, &ts, &delta); |
bf41a158 | 2144 | if (commit == -EBUSY) |
fa743953 | 2145 | goto out_fail; |
bf41a158 SR |
2146 | |
2147 | if (commit == -EAGAIN) | |
2148 | goto again; | |
2149 | ||
2150 | RB_WARN_ON(cpu_buffer, commit < 0); | |
7a8e76a3 | 2151 | } |
168b6b1d | 2152 | } |
7a8e76a3 | 2153 | |
168b6b1d | 2154 | get_event: |
1cd8d735 | 2155 | event = __rb_reserve_next(cpu_buffer, 0, length, &ts); |
168b6b1d | 2156 | if (unlikely(PTR_ERR(event) == -EAGAIN)) |
bf41a158 SR |
2157 | goto again; |
2158 | ||
fa743953 SR |
2159 | if (!event) |
2160 | goto out_fail; | |
7a8e76a3 | 2161 | |
fa743953 | 2162 | if (!rb_event_is_commit(cpu_buffer, event)) |
7a8e76a3 SR |
2163 | delta = 0; |
2164 | ||
2165 | event->time_delta = delta; | |
2166 | ||
2167 | return event; | |
fa743953 SR |
2168 | |
2169 | out_fail: | |
2170 | rb_end_commit(cpu_buffer); | |
2171 | return NULL; | |
7a8e76a3 SR |
2172 | } |
2173 | ||
1155de47 PM |
2174 | #ifdef CONFIG_TRACING |
2175 | ||
aa18efb2 | 2176 | #define TRACE_RECURSIVE_DEPTH 16 |
261842b7 SR |
2177 | |
2178 | static int trace_recursive_lock(void) | |
2179 | { | |
aa18efb2 | 2180 | current->trace_recursion++; |
261842b7 | 2181 | |
aa18efb2 SR |
2182 | if (likely(current->trace_recursion < TRACE_RECURSIVE_DEPTH)) |
2183 | return 0; | |
e057a5e5 | 2184 | |
aa18efb2 SR |
2185 | /* Disable all tracing before we do anything else */ |
2186 | tracing_off_permanent(); | |
261842b7 | 2187 | |
7d7d2b80 | 2188 | printk_once(KERN_WARNING "Tracing recursion: depth[%ld]:" |
aa18efb2 SR |
2189 | "HC[%lu]:SC[%lu]:NMI[%lu]\n", |
2190 | current->trace_recursion, | |
2191 | hardirq_count() >> HARDIRQ_SHIFT, | |
2192 | softirq_count() >> SOFTIRQ_SHIFT, | |
2193 | in_nmi()); | |
261842b7 | 2194 | |
aa18efb2 SR |
2195 | WARN_ON_ONCE(1); |
2196 | return -1; | |
261842b7 SR |
2197 | } |
2198 | ||
2199 | static void trace_recursive_unlock(void) | |
2200 | { | |
aa18efb2 | 2201 | WARN_ON_ONCE(!current->trace_recursion); |
261842b7 | 2202 | |
aa18efb2 | 2203 | current->trace_recursion--; |
261842b7 SR |
2204 | } |
2205 | ||
1155de47 PM |
2206 | #else |
2207 | ||
2208 | #define trace_recursive_lock() (0) | |
2209 | #define trace_recursive_unlock() do { } while (0) | |
2210 | ||
2211 | #endif | |
2212 | ||
bf41a158 SR |
2213 | static DEFINE_PER_CPU(int, rb_need_resched); |
2214 | ||
7a8e76a3 SR |
2215 | /** |
2216 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
2217 | * @buffer: the ring buffer to reserve from | |
2218 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
2219 | * |
2220 | * Returns a reseverd event on the ring buffer to copy directly to. | |
2221 | * The user of this interface will need to get the body to write into | |
2222 | * and can use the ring_buffer_event_data() interface. | |
2223 | * | |
2224 | * The length is the length of the data needed, not the event length | |
2225 | * which also includes the event header. | |
2226 | * | |
2227 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
2228 | * If NULL is returned, then nothing has been allocated or locked. | |
2229 | */ | |
2230 | struct ring_buffer_event * | |
0a987751 | 2231 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
2232 | { |
2233 | struct ring_buffer_per_cpu *cpu_buffer; | |
2234 | struct ring_buffer_event *event; | |
bf41a158 | 2235 | int cpu, resched; |
7a8e76a3 | 2236 | |
033601a3 | 2237 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2238 | return NULL; |
2239 | ||
7a8e76a3 SR |
2240 | if (atomic_read(&buffer->record_disabled)) |
2241 | return NULL; | |
2242 | ||
bf41a158 | 2243 | /* If we are tracing schedule, we don't want to recurse */ |
182e9f5f | 2244 | resched = ftrace_preempt_disable(); |
bf41a158 | 2245 | |
261842b7 SR |
2246 | if (trace_recursive_lock()) |
2247 | goto out_nocheck; | |
2248 | ||
7a8e76a3 SR |
2249 | cpu = raw_smp_processor_id(); |
2250 | ||
9e01c1b7 | 2251 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2252 | goto out; |
7a8e76a3 SR |
2253 | |
2254 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2255 | |
2256 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 2257 | goto out; |
7a8e76a3 | 2258 | |
be957c44 | 2259 | if (length > BUF_MAX_DATA_SIZE) |
bf41a158 | 2260 | goto out; |
7a8e76a3 | 2261 | |
62f0b3eb | 2262 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 2263 | if (!event) |
d769041f | 2264 | goto out; |
7a8e76a3 | 2265 | |
bf41a158 SR |
2266 | /* |
2267 | * Need to store resched state on this cpu. | |
2268 | * Only the first needs to. | |
2269 | */ | |
2270 | ||
2271 | if (preempt_count() == 1) | |
2272 | per_cpu(rb_need_resched, cpu) = resched; | |
2273 | ||
7a8e76a3 SR |
2274 | return event; |
2275 | ||
d769041f | 2276 | out: |
261842b7 SR |
2277 | trace_recursive_unlock(); |
2278 | ||
2279 | out_nocheck: | |
182e9f5f | 2280 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
2281 | return NULL; |
2282 | } | |
c4f50183 | 2283 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 | 2284 | |
a1863c21 SR |
2285 | static void |
2286 | rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
7a8e76a3 SR |
2287 | struct ring_buffer_event *event) |
2288 | { | |
fa743953 SR |
2289 | /* |
2290 | * The event first in the commit queue updates the | |
2291 | * time stamp. | |
2292 | */ | |
2293 | if (rb_event_is_commit(cpu_buffer, event)) | |
2294 | cpu_buffer->write_stamp += event->time_delta; | |
a1863c21 | 2295 | } |
bf41a158 | 2296 | |
a1863c21 SR |
2297 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, |
2298 | struct ring_buffer_event *event) | |
2299 | { | |
2300 | local_inc(&cpu_buffer->entries); | |
2301 | rb_update_write_stamp(cpu_buffer, event); | |
fa743953 | 2302 | rb_end_commit(cpu_buffer); |
7a8e76a3 SR |
2303 | } |
2304 | ||
2305 | /** | |
2306 | * ring_buffer_unlock_commit - commit a reserved | |
2307 | * @buffer: The buffer to commit to | |
2308 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
2309 | * |
2310 | * This commits the data to the ring buffer, and releases any locks held. | |
2311 | * | |
2312 | * Must be paired with ring_buffer_lock_reserve. | |
2313 | */ | |
2314 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 2315 | struct ring_buffer_event *event) |
7a8e76a3 SR |
2316 | { |
2317 | struct ring_buffer_per_cpu *cpu_buffer; | |
2318 | int cpu = raw_smp_processor_id(); | |
2319 | ||
2320 | cpu_buffer = buffer->buffers[cpu]; | |
2321 | ||
7a8e76a3 SR |
2322 | rb_commit(cpu_buffer, event); |
2323 | ||
261842b7 SR |
2324 | trace_recursive_unlock(); |
2325 | ||
bf41a158 SR |
2326 | /* |
2327 | * Only the last preempt count needs to restore preemption. | |
2328 | */ | |
182e9f5f SR |
2329 | if (preempt_count() == 1) |
2330 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
2331 | else | |
bf41a158 | 2332 | preempt_enable_no_resched_notrace(); |
7a8e76a3 SR |
2333 | |
2334 | return 0; | |
2335 | } | |
c4f50183 | 2336 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 2337 | |
f3b9aae1 FW |
2338 | static inline void rb_event_discard(struct ring_buffer_event *event) |
2339 | { | |
334d4169 LJ |
2340 | /* array[0] holds the actual length for the discarded event */ |
2341 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
2342 | event->type_len = RINGBUF_TYPE_PADDING; | |
f3b9aae1 FW |
2343 | /* time delta must be non zero */ |
2344 | if (!event->time_delta) | |
2345 | event->time_delta = 1; | |
2346 | } | |
2347 | ||
a1863c21 SR |
2348 | /* |
2349 | * Decrement the entries to the page that an event is on. | |
2350 | * The event does not even need to exist, only the pointer | |
2351 | * to the page it is on. This may only be called before the commit | |
2352 | * takes place. | |
2353 | */ | |
2354 | static inline void | |
2355 | rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer, | |
2356 | struct ring_buffer_event *event) | |
2357 | { | |
2358 | unsigned long addr = (unsigned long)event; | |
2359 | struct buffer_page *bpage = cpu_buffer->commit_page; | |
2360 | struct buffer_page *start; | |
2361 | ||
2362 | addr &= PAGE_MASK; | |
2363 | ||
2364 | /* Do the likely case first */ | |
2365 | if (likely(bpage->page == (void *)addr)) { | |
2366 | local_dec(&bpage->entries); | |
2367 | return; | |
2368 | } | |
2369 | ||
2370 | /* | |
2371 | * Because the commit page may be on the reader page we | |
2372 | * start with the next page and check the end loop there. | |
2373 | */ | |
2374 | rb_inc_page(cpu_buffer, &bpage); | |
2375 | start = bpage; | |
2376 | do { | |
2377 | if (bpage->page == (void *)addr) { | |
2378 | local_dec(&bpage->entries); | |
2379 | return; | |
2380 | } | |
2381 | rb_inc_page(cpu_buffer, &bpage); | |
2382 | } while (bpage != start); | |
2383 | ||
2384 | /* commit not part of this buffer?? */ | |
2385 | RB_WARN_ON(cpu_buffer, 1); | |
2386 | } | |
2387 | ||
fa1b47dd SR |
2388 | /** |
2389 | * ring_buffer_commit_discard - discard an event that has not been committed | |
2390 | * @buffer: the ring buffer | |
2391 | * @event: non committed event to discard | |
2392 | * | |
dc892f73 SR |
2393 | * Sometimes an event that is in the ring buffer needs to be ignored. |
2394 | * This function lets the user discard an event in the ring buffer | |
2395 | * and then that event will not be read later. | |
2396 | * | |
2397 | * This function only works if it is called before the the item has been | |
2398 | * committed. It will try to free the event from the ring buffer | |
fa1b47dd SR |
2399 | * if another event has not been added behind it. |
2400 | * | |
2401 | * If another event has been added behind it, it will set the event | |
2402 | * up as discarded, and perform the commit. | |
2403 | * | |
2404 | * If this function is called, do not call ring_buffer_unlock_commit on | |
2405 | * the event. | |
2406 | */ | |
2407 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
2408 | struct ring_buffer_event *event) | |
2409 | { | |
2410 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
2411 | int cpu; |
2412 | ||
2413 | /* The event is discarded regardless */ | |
f3b9aae1 | 2414 | rb_event_discard(event); |
fa1b47dd | 2415 | |
fa743953 SR |
2416 | cpu = smp_processor_id(); |
2417 | cpu_buffer = buffer->buffers[cpu]; | |
2418 | ||
fa1b47dd SR |
2419 | /* |
2420 | * This must only be called if the event has not been | |
2421 | * committed yet. Thus we can assume that preemption | |
2422 | * is still disabled. | |
2423 | */ | |
fa743953 | 2424 | RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing)); |
fa1b47dd | 2425 | |
a1863c21 | 2426 | rb_decrement_entry(cpu_buffer, event); |
0f2541d2 | 2427 | if (rb_try_to_discard(cpu_buffer, event)) |
edd813bf | 2428 | goto out; |
fa1b47dd SR |
2429 | |
2430 | /* | |
2431 | * The commit is still visible by the reader, so we | |
a1863c21 | 2432 | * must still update the timestamp. |
fa1b47dd | 2433 | */ |
a1863c21 | 2434 | rb_update_write_stamp(cpu_buffer, event); |
fa1b47dd | 2435 | out: |
fa743953 | 2436 | rb_end_commit(cpu_buffer); |
fa1b47dd | 2437 | |
f3b9aae1 FW |
2438 | trace_recursive_unlock(); |
2439 | ||
fa1b47dd SR |
2440 | /* |
2441 | * Only the last preempt count needs to restore preemption. | |
2442 | */ | |
2443 | if (preempt_count() == 1) | |
2444 | ftrace_preempt_enable(per_cpu(rb_need_resched, cpu)); | |
2445 | else | |
2446 | preempt_enable_no_resched_notrace(); | |
2447 | ||
2448 | } | |
2449 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
2450 | ||
7a8e76a3 SR |
2451 | /** |
2452 | * ring_buffer_write - write data to the buffer without reserving | |
2453 | * @buffer: The ring buffer to write to. | |
2454 | * @length: The length of the data being written (excluding the event header) | |
2455 | * @data: The data to write to the buffer. | |
2456 | * | |
2457 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
2458 | * one function. If you already have the data to write to the buffer, it | |
2459 | * may be easier to simply call this function. | |
2460 | * | |
2461 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
2462 | * and not the length of the event which would hold the header. | |
2463 | */ | |
2464 | int ring_buffer_write(struct ring_buffer *buffer, | |
2465 | unsigned long length, | |
2466 | void *data) | |
2467 | { | |
2468 | struct ring_buffer_per_cpu *cpu_buffer; | |
2469 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
2470 | void *body; |
2471 | int ret = -EBUSY; | |
bf41a158 | 2472 | int cpu, resched; |
7a8e76a3 | 2473 | |
033601a3 | 2474 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2475 | return -EBUSY; |
2476 | ||
7a8e76a3 SR |
2477 | if (atomic_read(&buffer->record_disabled)) |
2478 | return -EBUSY; | |
2479 | ||
182e9f5f | 2480 | resched = ftrace_preempt_disable(); |
bf41a158 | 2481 | |
7a8e76a3 SR |
2482 | cpu = raw_smp_processor_id(); |
2483 | ||
9e01c1b7 | 2484 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2485 | goto out; |
7a8e76a3 SR |
2486 | |
2487 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2488 | |
2489 | if (atomic_read(&cpu_buffer->record_disabled)) | |
2490 | goto out; | |
2491 | ||
be957c44 SR |
2492 | if (length > BUF_MAX_DATA_SIZE) |
2493 | goto out; | |
2494 | ||
62f0b3eb | 2495 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 SR |
2496 | if (!event) |
2497 | goto out; | |
2498 | ||
2499 | body = rb_event_data(event); | |
2500 | ||
2501 | memcpy(body, data, length); | |
2502 | ||
2503 | rb_commit(cpu_buffer, event); | |
2504 | ||
2505 | ret = 0; | |
2506 | out: | |
182e9f5f | 2507 | ftrace_preempt_enable(resched); |
7a8e76a3 SR |
2508 | |
2509 | return ret; | |
2510 | } | |
c4f50183 | 2511 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 2512 | |
34a148bf | 2513 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
2514 | { |
2515 | struct buffer_page *reader = cpu_buffer->reader_page; | |
77ae365e | 2516 | struct buffer_page *head = rb_set_head_page(cpu_buffer); |
bf41a158 SR |
2517 | struct buffer_page *commit = cpu_buffer->commit_page; |
2518 | ||
77ae365e SR |
2519 | /* In case of error, head will be NULL */ |
2520 | if (unlikely(!head)) | |
2521 | return 1; | |
2522 | ||
bf41a158 SR |
2523 | return reader->read == rb_page_commit(reader) && |
2524 | (commit == reader || | |
2525 | (commit == head && | |
2526 | head->read == rb_page_commit(commit))); | |
2527 | } | |
2528 | ||
7a8e76a3 SR |
2529 | /** |
2530 | * ring_buffer_record_disable - stop all writes into the buffer | |
2531 | * @buffer: The ring buffer to stop writes to. | |
2532 | * | |
2533 | * This prevents all writes to the buffer. Any attempt to write | |
2534 | * to the buffer after this will fail and return NULL. | |
2535 | * | |
2536 | * The caller should call synchronize_sched() after this. | |
2537 | */ | |
2538 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
2539 | { | |
2540 | atomic_inc(&buffer->record_disabled); | |
2541 | } | |
c4f50183 | 2542 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
2543 | |
2544 | /** | |
2545 | * ring_buffer_record_enable - enable writes to the buffer | |
2546 | * @buffer: The ring buffer to enable writes | |
2547 | * | |
2548 | * Note, multiple disables will need the same number of enables | |
2549 | * to truely enable the writing (much like preempt_disable). | |
2550 | */ | |
2551 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
2552 | { | |
2553 | atomic_dec(&buffer->record_disabled); | |
2554 | } | |
c4f50183 | 2555 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 SR |
2556 | |
2557 | /** | |
2558 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
2559 | * @buffer: The ring buffer to stop writes to. | |
2560 | * @cpu: The CPU buffer to stop | |
2561 | * | |
2562 | * This prevents all writes to the buffer. Any attempt to write | |
2563 | * to the buffer after this will fail and return NULL. | |
2564 | * | |
2565 | * The caller should call synchronize_sched() after this. | |
2566 | */ | |
2567 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
2568 | { | |
2569 | struct ring_buffer_per_cpu *cpu_buffer; | |
2570 | ||
9e01c1b7 | 2571 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2572 | return; |
7a8e76a3 SR |
2573 | |
2574 | cpu_buffer = buffer->buffers[cpu]; | |
2575 | atomic_inc(&cpu_buffer->record_disabled); | |
2576 | } | |
c4f50183 | 2577 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
2578 | |
2579 | /** | |
2580 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
2581 | * @buffer: The ring buffer to enable writes | |
2582 | * @cpu: The CPU to enable. | |
2583 | * | |
2584 | * Note, multiple disables will need the same number of enables | |
2585 | * to truely enable the writing (much like preempt_disable). | |
2586 | */ | |
2587 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
2588 | { | |
2589 | struct ring_buffer_per_cpu *cpu_buffer; | |
2590 | ||
9e01c1b7 | 2591 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2592 | return; |
7a8e76a3 SR |
2593 | |
2594 | cpu_buffer = buffer->buffers[cpu]; | |
2595 | atomic_dec(&cpu_buffer->record_disabled); | |
2596 | } | |
c4f50183 | 2597 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 SR |
2598 | |
2599 | /** | |
2600 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
2601 | * @buffer: The ring buffer | |
2602 | * @cpu: The per CPU buffer to get the entries from. | |
2603 | */ | |
2604 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
2605 | { | |
2606 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2607 | unsigned long ret; |
7a8e76a3 | 2608 | |
9e01c1b7 | 2609 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2610 | return 0; |
7a8e76a3 SR |
2611 | |
2612 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 2613 | ret = (local_read(&cpu_buffer->entries) - local_read(&cpu_buffer->overrun)) |
e4906eff | 2614 | - cpu_buffer->read; |
554f786e SR |
2615 | |
2616 | return ret; | |
7a8e76a3 | 2617 | } |
c4f50183 | 2618 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
2619 | |
2620 | /** | |
2621 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
2622 | * @buffer: The ring buffer | |
2623 | * @cpu: The per CPU buffer to get the number of overruns from | |
2624 | */ | |
2625 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
2626 | { | |
2627 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2628 | unsigned long ret; |
7a8e76a3 | 2629 | |
9e01c1b7 | 2630 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2631 | return 0; |
7a8e76a3 SR |
2632 | |
2633 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 2634 | ret = local_read(&cpu_buffer->overrun); |
554f786e SR |
2635 | |
2636 | return ret; | |
7a8e76a3 | 2637 | } |
c4f50183 | 2638 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 2639 | |
f0d2c681 SR |
2640 | /** |
2641 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits | |
2642 | * @buffer: The ring buffer | |
2643 | * @cpu: The per CPU buffer to get the number of overruns from | |
2644 | */ | |
2645 | unsigned long | |
2646 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
2647 | { | |
2648 | struct ring_buffer_per_cpu *cpu_buffer; | |
2649 | unsigned long ret; | |
2650 | ||
2651 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
2652 | return 0; | |
2653 | ||
2654 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 2655 | ret = local_read(&cpu_buffer->commit_overrun); |
f0d2c681 SR |
2656 | |
2657 | return ret; | |
2658 | } | |
2659 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
2660 | ||
7a8e76a3 SR |
2661 | /** |
2662 | * ring_buffer_entries - get the number of entries in a buffer | |
2663 | * @buffer: The ring buffer | |
2664 | * | |
2665 | * Returns the total number of entries in the ring buffer | |
2666 | * (all CPU entries) | |
2667 | */ | |
2668 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
2669 | { | |
2670 | struct ring_buffer_per_cpu *cpu_buffer; | |
2671 | unsigned long entries = 0; | |
2672 | int cpu; | |
2673 | ||
2674 | /* if you care about this being correct, lock the buffer */ | |
2675 | for_each_buffer_cpu(buffer, cpu) { | |
2676 | cpu_buffer = buffer->buffers[cpu]; | |
e4906eff | 2677 | entries += (local_read(&cpu_buffer->entries) - |
77ae365e | 2678 | local_read(&cpu_buffer->overrun)) - cpu_buffer->read; |
7a8e76a3 SR |
2679 | } |
2680 | ||
2681 | return entries; | |
2682 | } | |
c4f50183 | 2683 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
2684 | |
2685 | /** | |
67b394f7 | 2686 | * ring_buffer_overruns - get the number of overruns in buffer |
7a8e76a3 SR |
2687 | * @buffer: The ring buffer |
2688 | * | |
2689 | * Returns the total number of overruns in the ring buffer | |
2690 | * (all CPU entries) | |
2691 | */ | |
2692 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
2693 | { | |
2694 | struct ring_buffer_per_cpu *cpu_buffer; | |
2695 | unsigned long overruns = 0; | |
2696 | int cpu; | |
2697 | ||
2698 | /* if you care about this being correct, lock the buffer */ | |
2699 | for_each_buffer_cpu(buffer, cpu) { | |
2700 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 2701 | overruns += local_read(&cpu_buffer->overrun); |
7a8e76a3 SR |
2702 | } |
2703 | ||
2704 | return overruns; | |
2705 | } | |
c4f50183 | 2706 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 2707 | |
642edba5 | 2708 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
2709 | { |
2710 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
2711 | ||
d769041f SR |
2712 | /* Iterator usage is expected to have record disabled */ |
2713 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
77ae365e SR |
2714 | iter->head_page = rb_set_head_page(cpu_buffer); |
2715 | if (unlikely(!iter->head_page)) | |
2716 | return; | |
2717 | iter->head = iter->head_page->read; | |
d769041f SR |
2718 | } else { |
2719 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 2720 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
2721 | } |
2722 | if (iter->head) | |
2723 | iter->read_stamp = cpu_buffer->read_stamp; | |
2724 | else | |
abc9b56d | 2725 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 2726 | } |
f83c9d0f | 2727 | |
642edba5 SR |
2728 | /** |
2729 | * ring_buffer_iter_reset - reset an iterator | |
2730 | * @iter: The iterator to reset | |
2731 | * | |
2732 | * Resets the iterator, so that it will start from the beginning | |
2733 | * again. | |
2734 | */ | |
2735 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
2736 | { | |
554f786e | 2737 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
2738 | unsigned long flags; |
2739 | ||
554f786e SR |
2740 | if (!iter) |
2741 | return; | |
2742 | ||
2743 | cpu_buffer = iter->cpu_buffer; | |
2744 | ||
642edba5 SR |
2745 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
2746 | rb_iter_reset(iter); | |
f83c9d0f | 2747 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 2748 | } |
c4f50183 | 2749 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
2750 | |
2751 | /** | |
2752 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
2753 | * @iter: The iterator to check | |
2754 | */ | |
2755 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
2756 | { | |
2757 | struct ring_buffer_per_cpu *cpu_buffer; | |
2758 | ||
2759 | cpu_buffer = iter->cpu_buffer; | |
2760 | ||
bf41a158 SR |
2761 | return iter->head_page == cpu_buffer->commit_page && |
2762 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 2763 | } |
c4f50183 | 2764 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
2765 | |
2766 | static void | |
2767 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
2768 | struct ring_buffer_event *event) | |
2769 | { | |
2770 | u64 delta; | |
2771 | ||
334d4169 | 2772 | switch (event->type_len) { |
7a8e76a3 SR |
2773 | case RINGBUF_TYPE_PADDING: |
2774 | return; | |
2775 | ||
2776 | case RINGBUF_TYPE_TIME_EXTEND: | |
2777 | delta = event->array[0]; | |
2778 | delta <<= TS_SHIFT; | |
2779 | delta += event->time_delta; | |
2780 | cpu_buffer->read_stamp += delta; | |
2781 | return; | |
2782 | ||
2783 | case RINGBUF_TYPE_TIME_STAMP: | |
2784 | /* FIXME: not implemented */ | |
2785 | return; | |
2786 | ||
2787 | case RINGBUF_TYPE_DATA: | |
2788 | cpu_buffer->read_stamp += event->time_delta; | |
2789 | return; | |
2790 | ||
2791 | default: | |
2792 | BUG(); | |
2793 | } | |
2794 | return; | |
2795 | } | |
2796 | ||
2797 | static void | |
2798 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
2799 | struct ring_buffer_event *event) | |
2800 | { | |
2801 | u64 delta; | |
2802 | ||
334d4169 | 2803 | switch (event->type_len) { |
7a8e76a3 SR |
2804 | case RINGBUF_TYPE_PADDING: |
2805 | return; | |
2806 | ||
2807 | case RINGBUF_TYPE_TIME_EXTEND: | |
2808 | delta = event->array[0]; | |
2809 | delta <<= TS_SHIFT; | |
2810 | delta += event->time_delta; | |
2811 | iter->read_stamp += delta; | |
2812 | return; | |
2813 | ||
2814 | case RINGBUF_TYPE_TIME_STAMP: | |
2815 | /* FIXME: not implemented */ | |
2816 | return; | |
2817 | ||
2818 | case RINGBUF_TYPE_DATA: | |
2819 | iter->read_stamp += event->time_delta; | |
2820 | return; | |
2821 | ||
2822 | default: | |
2823 | BUG(); | |
2824 | } | |
2825 | return; | |
2826 | } | |
2827 | ||
d769041f SR |
2828 | static struct buffer_page * |
2829 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 2830 | { |
d769041f SR |
2831 | struct buffer_page *reader = NULL; |
2832 | unsigned long flags; | |
818e3dd3 | 2833 | int nr_loops = 0; |
77ae365e | 2834 | int ret; |
d769041f | 2835 | |
3e03fb7f SR |
2836 | local_irq_save(flags); |
2837 | __raw_spin_lock(&cpu_buffer->lock); | |
d769041f SR |
2838 | |
2839 | again: | |
818e3dd3 SR |
2840 | /* |
2841 | * This should normally only loop twice. But because the | |
2842 | * start of the reader inserts an empty page, it causes | |
2843 | * a case where we will loop three times. There should be no | |
2844 | * reason to loop four times (that I know of). | |
2845 | */ | |
3e89c7bb | 2846 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
2847 | reader = NULL; |
2848 | goto out; | |
2849 | } | |
2850 | ||
d769041f SR |
2851 | reader = cpu_buffer->reader_page; |
2852 | ||
2853 | /* If there's more to read, return this page */ | |
bf41a158 | 2854 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
2855 | goto out; |
2856 | ||
2857 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
2858 | if (RB_WARN_ON(cpu_buffer, |
2859 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
2860 | goto out; | |
d769041f SR |
2861 | |
2862 | /* check if we caught up to the tail */ | |
2863 | reader = NULL; | |
bf41a158 | 2864 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 2865 | goto out; |
7a8e76a3 SR |
2866 | |
2867 | /* | |
d769041f | 2868 | * Reset the reader page to size zero. |
7a8e76a3 | 2869 | */ |
77ae365e SR |
2870 | local_set(&cpu_buffer->reader_page->write, 0); |
2871 | local_set(&cpu_buffer->reader_page->entries, 0); | |
2872 | local_set(&cpu_buffer->reader_page->page->commit, 0); | |
7a8e76a3 | 2873 | |
77ae365e SR |
2874 | spin: |
2875 | /* | |
2876 | * Splice the empty reader page into the list around the head. | |
2877 | */ | |
2878 | reader = rb_set_head_page(cpu_buffer); | |
d769041f SR |
2879 | cpu_buffer->reader_page->list.next = reader->list.next; |
2880 | cpu_buffer->reader_page->list.prev = reader->list.prev; | |
bf41a158 | 2881 | |
3adc54fa SR |
2882 | /* |
2883 | * cpu_buffer->pages just needs to point to the buffer, it | |
2884 | * has no specific buffer page to point to. Lets move it out | |
2885 | * of our way so we don't accidently swap it. | |
2886 | */ | |
2887 | cpu_buffer->pages = reader->list.prev; | |
2888 | ||
77ae365e SR |
2889 | /* The reader page will be pointing to the new head */ |
2890 | rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list); | |
7a8e76a3 | 2891 | |
77ae365e SR |
2892 | /* |
2893 | * Here's the tricky part. | |
2894 | * | |
2895 | * We need to move the pointer past the header page. | |
2896 | * But we can only do that if a writer is not currently | |
2897 | * moving it. The page before the header page has the | |
2898 | * flag bit '1' set if it is pointing to the page we want. | |
2899 | * but if the writer is in the process of moving it | |
2900 | * than it will be '2' or already moved '0'. | |
2901 | */ | |
2902 | ||
2903 | ret = rb_head_page_replace(reader, cpu_buffer->reader_page); | |
7a8e76a3 SR |
2904 | |
2905 | /* | |
77ae365e | 2906 | * If we did not convert it, then we must try again. |
7a8e76a3 | 2907 | */ |
77ae365e SR |
2908 | if (!ret) |
2909 | goto spin; | |
7a8e76a3 | 2910 | |
77ae365e SR |
2911 | /* |
2912 | * Yeah! We succeeded in replacing the page. | |
2913 | * | |
2914 | * Now make the new head point back to the reader page. | |
2915 | */ | |
2916 | reader->list.next->prev = &cpu_buffer->reader_page->list; | |
2917 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); | |
d769041f SR |
2918 | |
2919 | /* Finally update the reader page to the new head */ | |
2920 | cpu_buffer->reader_page = reader; | |
2921 | rb_reset_reader_page(cpu_buffer); | |
2922 | ||
2923 | goto again; | |
2924 | ||
2925 | out: | |
3e03fb7f SR |
2926 | __raw_spin_unlock(&cpu_buffer->lock); |
2927 | local_irq_restore(flags); | |
d769041f SR |
2928 | |
2929 | return reader; | |
2930 | } | |
2931 | ||
2932 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
2933 | { | |
2934 | struct ring_buffer_event *event; | |
2935 | struct buffer_page *reader; | |
2936 | unsigned length; | |
2937 | ||
2938 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 2939 | |
d769041f | 2940 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
2941 | if (RB_WARN_ON(cpu_buffer, !reader)) |
2942 | return; | |
7a8e76a3 | 2943 | |
d769041f SR |
2944 | event = rb_reader_event(cpu_buffer); |
2945 | ||
a1863c21 | 2946 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
e4906eff | 2947 | cpu_buffer->read++; |
d769041f SR |
2948 | |
2949 | rb_update_read_stamp(cpu_buffer, event); | |
2950 | ||
2951 | length = rb_event_length(event); | |
6f807acd | 2952 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
2953 | } |
2954 | ||
2955 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
2956 | { | |
2957 | struct ring_buffer *buffer; | |
2958 | struct ring_buffer_per_cpu *cpu_buffer; | |
2959 | struct ring_buffer_event *event; | |
2960 | unsigned length; | |
2961 | ||
2962 | cpu_buffer = iter->cpu_buffer; | |
2963 | buffer = cpu_buffer->buffer; | |
2964 | ||
2965 | /* | |
2966 | * Check if we are at the end of the buffer. | |
2967 | */ | |
bf41a158 | 2968 | if (iter->head >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
2969 | /* discarded commits can make the page empty */ |
2970 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 2971 | return; |
d769041f | 2972 | rb_inc_iter(iter); |
7a8e76a3 SR |
2973 | return; |
2974 | } | |
2975 | ||
2976 | event = rb_iter_head_event(iter); | |
2977 | ||
2978 | length = rb_event_length(event); | |
2979 | ||
2980 | /* | |
2981 | * This should not be called to advance the header if we are | |
2982 | * at the tail of the buffer. | |
2983 | */ | |
3e89c7bb | 2984 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 2985 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
2986 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
2987 | return; | |
7a8e76a3 SR |
2988 | |
2989 | rb_update_iter_read_stamp(iter, event); | |
2990 | ||
2991 | iter->head += length; | |
2992 | ||
2993 | /* check for end of page padding */ | |
bf41a158 SR |
2994 | if ((iter->head >= rb_page_size(iter->head_page)) && |
2995 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
2996 | rb_advance_iter(iter); |
2997 | } | |
2998 | ||
f83c9d0f | 2999 | static struct ring_buffer_event * |
d8eeb2d3 | 3000 | rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts) |
7a8e76a3 | 3001 | { |
7a8e76a3 | 3002 | struct ring_buffer_event *event; |
d769041f | 3003 | struct buffer_page *reader; |
818e3dd3 | 3004 | int nr_loops = 0; |
7a8e76a3 | 3005 | |
7a8e76a3 | 3006 | again: |
818e3dd3 SR |
3007 | /* |
3008 | * We repeat when a timestamp is encountered. It is possible | |
3009 | * to get multiple timestamps from an interrupt entering just | |
ea05b57c SR |
3010 | * as one timestamp is about to be written, or from discarded |
3011 | * commits. The most that we can have is the number on a single page. | |
818e3dd3 | 3012 | */ |
ea05b57c | 3013 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE)) |
818e3dd3 | 3014 | return NULL; |
818e3dd3 | 3015 | |
d769041f SR |
3016 | reader = rb_get_reader_page(cpu_buffer); |
3017 | if (!reader) | |
7a8e76a3 SR |
3018 | return NULL; |
3019 | ||
d769041f | 3020 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 3021 | |
334d4169 | 3022 | switch (event->type_len) { |
7a8e76a3 | 3023 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3024 | if (rb_null_event(event)) |
3025 | RB_WARN_ON(cpu_buffer, 1); | |
3026 | /* | |
3027 | * Because the writer could be discarding every | |
3028 | * event it creates (which would probably be bad) | |
3029 | * if we were to go back to "again" then we may never | |
3030 | * catch up, and will trigger the warn on, or lock | |
3031 | * the box. Return the padding, and we will release | |
3032 | * the current locks, and try again. | |
3033 | */ | |
2d622719 | 3034 | return event; |
7a8e76a3 SR |
3035 | |
3036 | case RINGBUF_TYPE_TIME_EXTEND: | |
3037 | /* Internal data, OK to advance */ | |
d769041f | 3038 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3039 | goto again; |
3040 | ||
3041 | case RINGBUF_TYPE_TIME_STAMP: | |
3042 | /* FIXME: not implemented */ | |
d769041f | 3043 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3044 | goto again; |
3045 | ||
3046 | case RINGBUF_TYPE_DATA: | |
3047 | if (ts) { | |
3048 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
d8eeb2d3 | 3049 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, |
37886f6a | 3050 | cpu_buffer->cpu, ts); |
7a8e76a3 SR |
3051 | } |
3052 | return event; | |
3053 | ||
3054 | default: | |
3055 | BUG(); | |
3056 | } | |
3057 | ||
3058 | return NULL; | |
3059 | } | |
c4f50183 | 3060 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 3061 | |
f83c9d0f SR |
3062 | static struct ring_buffer_event * |
3063 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
3064 | { |
3065 | struct ring_buffer *buffer; | |
3066 | struct ring_buffer_per_cpu *cpu_buffer; | |
3067 | struct ring_buffer_event *event; | |
818e3dd3 | 3068 | int nr_loops = 0; |
7a8e76a3 SR |
3069 | |
3070 | if (ring_buffer_iter_empty(iter)) | |
3071 | return NULL; | |
3072 | ||
3073 | cpu_buffer = iter->cpu_buffer; | |
3074 | buffer = cpu_buffer->buffer; | |
3075 | ||
3076 | again: | |
818e3dd3 | 3077 | /* |
ea05b57c SR |
3078 | * We repeat when a timestamp is encountered. |
3079 | * We can get multiple timestamps by nested interrupts or also | |
3080 | * if filtering is on (discarding commits). Since discarding | |
3081 | * commits can be frequent we can get a lot of timestamps. | |
3082 | * But we limit them by not adding timestamps if they begin | |
3083 | * at the start of a page. | |
818e3dd3 | 3084 | */ |
ea05b57c | 3085 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > RB_TIMESTAMPS_PER_PAGE)) |
818e3dd3 | 3086 | return NULL; |
818e3dd3 | 3087 | |
7a8e76a3 SR |
3088 | if (rb_per_cpu_empty(cpu_buffer)) |
3089 | return NULL; | |
3090 | ||
3091 | event = rb_iter_head_event(iter); | |
3092 | ||
334d4169 | 3093 | switch (event->type_len) { |
7a8e76a3 | 3094 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3095 | if (rb_null_event(event)) { |
3096 | rb_inc_iter(iter); | |
3097 | goto again; | |
3098 | } | |
3099 | rb_advance_iter(iter); | |
3100 | return event; | |
7a8e76a3 SR |
3101 | |
3102 | case RINGBUF_TYPE_TIME_EXTEND: | |
3103 | /* Internal data, OK to advance */ | |
3104 | rb_advance_iter(iter); | |
3105 | goto again; | |
3106 | ||
3107 | case RINGBUF_TYPE_TIME_STAMP: | |
3108 | /* FIXME: not implemented */ | |
3109 | rb_advance_iter(iter); | |
3110 | goto again; | |
3111 | ||
3112 | case RINGBUF_TYPE_DATA: | |
3113 | if (ts) { | |
3114 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
3115 | ring_buffer_normalize_time_stamp(buffer, |
3116 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
3117 | } |
3118 | return event; | |
3119 | ||
3120 | default: | |
3121 | BUG(); | |
3122 | } | |
3123 | ||
3124 | return NULL; | |
3125 | } | |
c4f50183 | 3126 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 3127 | |
8d707e8e SR |
3128 | static inline int rb_ok_to_lock(void) |
3129 | { | |
3130 | /* | |
3131 | * If an NMI die dumps out the content of the ring buffer | |
3132 | * do not grab locks. We also permanently disable the ring | |
3133 | * buffer too. A one time deal is all you get from reading | |
3134 | * the ring buffer from an NMI. | |
3135 | */ | |
464e85eb | 3136 | if (likely(!in_nmi())) |
8d707e8e SR |
3137 | return 1; |
3138 | ||
3139 | tracing_off_permanent(); | |
3140 | return 0; | |
3141 | } | |
3142 | ||
f83c9d0f SR |
3143 | /** |
3144 | * ring_buffer_peek - peek at the next event to be read | |
3145 | * @buffer: The ring buffer to read | |
3146 | * @cpu: The cpu to peak at | |
3147 | * @ts: The timestamp counter of this event. | |
3148 | * | |
3149 | * This will return the event that will be read next, but does | |
3150 | * not consume the data. | |
3151 | */ | |
3152 | struct ring_buffer_event * | |
3153 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts) | |
3154 | { | |
3155 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 3156 | struct ring_buffer_event *event; |
f83c9d0f | 3157 | unsigned long flags; |
8d707e8e | 3158 | int dolock; |
f83c9d0f | 3159 | |
554f786e | 3160 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3161 | return NULL; |
554f786e | 3162 | |
8d707e8e | 3163 | dolock = rb_ok_to_lock(); |
2d622719 | 3164 | again: |
8d707e8e SR |
3165 | local_irq_save(flags); |
3166 | if (dolock) | |
3167 | spin_lock(&cpu_buffer->reader_lock); | |
d8eeb2d3 | 3168 | event = rb_buffer_peek(cpu_buffer, ts); |
469535a5 RR |
3169 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
3170 | rb_advance_reader(cpu_buffer); | |
8d707e8e SR |
3171 | if (dolock) |
3172 | spin_unlock(&cpu_buffer->reader_lock); | |
3173 | local_irq_restore(flags); | |
f83c9d0f | 3174 | |
1b959e18 | 3175 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3176 | goto again; |
2d622719 | 3177 | |
f83c9d0f SR |
3178 | return event; |
3179 | } | |
3180 | ||
3181 | /** | |
3182 | * ring_buffer_iter_peek - peek at the next event to be read | |
3183 | * @iter: The ring buffer iterator | |
3184 | * @ts: The timestamp counter of this event. | |
3185 | * | |
3186 | * This will return the event that will be read next, but does | |
3187 | * not increment the iterator. | |
3188 | */ | |
3189 | struct ring_buffer_event * | |
3190 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
3191 | { | |
3192 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3193 | struct ring_buffer_event *event; | |
3194 | unsigned long flags; | |
3195 | ||
2d622719 | 3196 | again: |
f83c9d0f SR |
3197 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3198 | event = rb_iter_peek(iter, ts); | |
3199 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
3200 | ||
1b959e18 | 3201 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3202 | goto again; |
2d622719 | 3203 | |
f83c9d0f SR |
3204 | return event; |
3205 | } | |
3206 | ||
7a8e76a3 SR |
3207 | /** |
3208 | * ring_buffer_consume - return an event and consume it | |
3209 | * @buffer: The ring buffer to get the next event from | |
3210 | * | |
3211 | * Returns the next event in the ring buffer, and that event is consumed. | |
3212 | * Meaning, that sequential reads will keep returning a different event, | |
3213 | * and eventually empty the ring buffer if the producer is slower. | |
3214 | */ | |
3215 | struct ring_buffer_event * | |
3216 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) | |
3217 | { | |
554f786e SR |
3218 | struct ring_buffer_per_cpu *cpu_buffer; |
3219 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 3220 | unsigned long flags; |
8d707e8e SR |
3221 | int dolock; |
3222 | ||
3223 | dolock = rb_ok_to_lock(); | |
7a8e76a3 | 3224 | |
2d622719 | 3225 | again: |
554f786e SR |
3226 | /* might be called in atomic */ |
3227 | preempt_disable(); | |
3228 | ||
9e01c1b7 | 3229 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 3230 | goto out; |
7a8e76a3 | 3231 | |
554f786e | 3232 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
3233 | local_irq_save(flags); |
3234 | if (dolock) | |
3235 | spin_lock(&cpu_buffer->reader_lock); | |
f83c9d0f | 3236 | |
d8eeb2d3 | 3237 | event = rb_buffer_peek(cpu_buffer, ts); |
469535a5 RR |
3238 | if (event) |
3239 | rb_advance_reader(cpu_buffer); | |
7a8e76a3 | 3240 | |
8d707e8e SR |
3241 | if (dolock) |
3242 | spin_unlock(&cpu_buffer->reader_lock); | |
3243 | local_irq_restore(flags); | |
f83c9d0f | 3244 | |
554f786e SR |
3245 | out: |
3246 | preempt_enable(); | |
3247 | ||
1b959e18 | 3248 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3249 | goto again; |
2d622719 | 3250 | |
7a8e76a3 SR |
3251 | return event; |
3252 | } | |
c4f50183 | 3253 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
3254 | |
3255 | /** | |
3256 | * ring_buffer_read_start - start a non consuming read of the buffer | |
3257 | * @buffer: The ring buffer to read from | |
3258 | * @cpu: The cpu buffer to iterate over | |
3259 | * | |
3260 | * This starts up an iteration through the buffer. It also disables | |
3261 | * the recording to the buffer until the reading is finished. | |
3262 | * This prevents the reading from being corrupted. This is not | |
3263 | * a consuming read, so a producer is not expected. | |
3264 | * | |
3265 | * Must be paired with ring_buffer_finish. | |
3266 | */ | |
3267 | struct ring_buffer_iter * | |
3268 | ring_buffer_read_start(struct ring_buffer *buffer, int cpu) | |
3269 | { | |
3270 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3271 | struct ring_buffer_iter *iter; |
d769041f | 3272 | unsigned long flags; |
7a8e76a3 | 3273 | |
9e01c1b7 | 3274 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3275 | return NULL; |
7a8e76a3 SR |
3276 | |
3277 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
3278 | if (!iter) | |
8aabee57 | 3279 | return NULL; |
7a8e76a3 SR |
3280 | |
3281 | cpu_buffer = buffer->buffers[cpu]; | |
3282 | ||
3283 | iter->cpu_buffer = cpu_buffer; | |
3284 | ||
3285 | atomic_inc(&cpu_buffer->record_disabled); | |
3286 | synchronize_sched(); | |
3287 | ||
f83c9d0f | 3288 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3e03fb7f | 3289 | __raw_spin_lock(&cpu_buffer->lock); |
642edba5 | 3290 | rb_iter_reset(iter); |
3e03fb7f | 3291 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 3292 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
3293 | |
3294 | return iter; | |
3295 | } | |
c4f50183 | 3296 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
3297 | |
3298 | /** | |
3299 | * ring_buffer_finish - finish reading the iterator of the buffer | |
3300 | * @iter: The iterator retrieved by ring_buffer_start | |
3301 | * | |
3302 | * This re-enables the recording to the buffer, and frees the | |
3303 | * iterator. | |
3304 | */ | |
3305 | void | |
3306 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
3307 | { | |
3308 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3309 | ||
3310 | atomic_dec(&cpu_buffer->record_disabled); | |
3311 | kfree(iter); | |
3312 | } | |
c4f50183 | 3313 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
3314 | |
3315 | /** | |
3316 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
3317 | * @iter: The ring buffer iterator | |
3318 | * @ts: The time stamp of the event read. | |
3319 | * | |
3320 | * This reads the next event in the ring buffer and increments the iterator. | |
3321 | */ | |
3322 | struct ring_buffer_event * | |
3323 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
3324 | { | |
3325 | struct ring_buffer_event *event; | |
f83c9d0f SR |
3326 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
3327 | unsigned long flags; | |
7a8e76a3 | 3328 | |
f83c9d0f | 3329 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
7e9391cf | 3330 | again: |
f83c9d0f | 3331 | event = rb_iter_peek(iter, ts); |
7a8e76a3 | 3332 | if (!event) |
f83c9d0f | 3333 | goto out; |
7a8e76a3 | 3334 | |
7e9391cf SR |
3335 | if (event->type_len == RINGBUF_TYPE_PADDING) |
3336 | goto again; | |
3337 | ||
7a8e76a3 | 3338 | rb_advance_iter(iter); |
f83c9d0f SR |
3339 | out: |
3340 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
7a8e76a3 SR |
3341 | |
3342 | return event; | |
3343 | } | |
c4f50183 | 3344 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
3345 | |
3346 | /** | |
3347 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
3348 | * @buffer: The ring buffer. | |
3349 | */ | |
3350 | unsigned long ring_buffer_size(struct ring_buffer *buffer) | |
3351 | { | |
3352 | return BUF_PAGE_SIZE * buffer->pages; | |
3353 | } | |
c4f50183 | 3354 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
3355 | |
3356 | static void | |
3357 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
3358 | { | |
77ae365e SR |
3359 | rb_head_page_deactivate(cpu_buffer); |
3360 | ||
7a8e76a3 | 3361 | cpu_buffer->head_page |
3adc54fa | 3362 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 3363 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 3364 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 3365 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 3366 | |
6f807acd | 3367 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
3368 | |
3369 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
3370 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
3371 | ||
3372 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
3373 | local_set(&cpu_buffer->reader_page->write, 0); | |
778c55d4 | 3374 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 3375 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 3376 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 3377 | |
77ae365e SR |
3378 | local_set(&cpu_buffer->commit_overrun, 0); |
3379 | local_set(&cpu_buffer->overrun, 0); | |
e4906eff | 3380 | local_set(&cpu_buffer->entries, 0); |
fa743953 SR |
3381 | local_set(&cpu_buffer->committing, 0); |
3382 | local_set(&cpu_buffer->commits, 0); | |
77ae365e | 3383 | cpu_buffer->read = 0; |
69507c06 SR |
3384 | |
3385 | cpu_buffer->write_stamp = 0; | |
3386 | cpu_buffer->read_stamp = 0; | |
77ae365e SR |
3387 | |
3388 | rb_head_page_activate(cpu_buffer); | |
7a8e76a3 SR |
3389 | } |
3390 | ||
3391 | /** | |
3392 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
3393 | * @buffer: The ring buffer to reset a per cpu buffer of | |
3394 | * @cpu: The CPU buffer to be reset | |
3395 | */ | |
3396 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
3397 | { | |
3398 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
3399 | unsigned long flags; | |
3400 | ||
9e01c1b7 | 3401 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3402 | return; |
7a8e76a3 | 3403 | |
41ede23e SR |
3404 | atomic_inc(&cpu_buffer->record_disabled); |
3405 | ||
f83c9d0f SR |
3406 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
3407 | ||
41b6a95d SR |
3408 | if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing))) |
3409 | goto out; | |
3410 | ||
3e03fb7f | 3411 | __raw_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
3412 | |
3413 | rb_reset_cpu(cpu_buffer); | |
3414 | ||
3e03fb7f | 3415 | __raw_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 3416 | |
41b6a95d | 3417 | out: |
f83c9d0f | 3418 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
41ede23e SR |
3419 | |
3420 | atomic_dec(&cpu_buffer->record_disabled); | |
7a8e76a3 | 3421 | } |
c4f50183 | 3422 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
3423 | |
3424 | /** | |
3425 | * ring_buffer_reset - reset a ring buffer | |
3426 | * @buffer: The ring buffer to reset all cpu buffers | |
3427 | */ | |
3428 | void ring_buffer_reset(struct ring_buffer *buffer) | |
3429 | { | |
7a8e76a3 SR |
3430 | int cpu; |
3431 | ||
7a8e76a3 | 3432 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 3433 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 3434 | } |
c4f50183 | 3435 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
3436 | |
3437 | /** | |
3438 | * rind_buffer_empty - is the ring buffer empty? | |
3439 | * @buffer: The ring buffer to test | |
3440 | */ | |
3441 | int ring_buffer_empty(struct ring_buffer *buffer) | |
3442 | { | |
3443 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 3444 | unsigned long flags; |
8d707e8e | 3445 | int dolock; |
7a8e76a3 | 3446 | int cpu; |
d4788207 | 3447 | int ret; |
7a8e76a3 | 3448 | |
8d707e8e | 3449 | dolock = rb_ok_to_lock(); |
7a8e76a3 SR |
3450 | |
3451 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
3452 | for_each_buffer_cpu(buffer, cpu) { | |
3453 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e SR |
3454 | local_irq_save(flags); |
3455 | if (dolock) | |
3456 | spin_lock(&cpu_buffer->reader_lock); | |
d4788207 | 3457 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e SR |
3458 | if (dolock) |
3459 | spin_unlock(&cpu_buffer->reader_lock); | |
3460 | local_irq_restore(flags); | |
3461 | ||
d4788207 | 3462 | if (!ret) |
7a8e76a3 SR |
3463 | return 0; |
3464 | } | |
554f786e | 3465 | |
7a8e76a3 SR |
3466 | return 1; |
3467 | } | |
c4f50183 | 3468 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
3469 | |
3470 | /** | |
3471 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
3472 | * @buffer: The ring buffer | |
3473 | * @cpu: The CPU buffer to test | |
3474 | */ | |
3475 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
3476 | { | |
3477 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 3478 | unsigned long flags; |
8d707e8e | 3479 | int dolock; |
8aabee57 | 3480 | int ret; |
7a8e76a3 | 3481 | |
9e01c1b7 | 3482 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3483 | return 1; |
7a8e76a3 | 3484 | |
8d707e8e SR |
3485 | dolock = rb_ok_to_lock(); |
3486 | ||
7a8e76a3 | 3487 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
3488 | local_irq_save(flags); |
3489 | if (dolock) | |
3490 | spin_lock(&cpu_buffer->reader_lock); | |
554f786e | 3491 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e SR |
3492 | if (dolock) |
3493 | spin_unlock(&cpu_buffer->reader_lock); | |
3494 | local_irq_restore(flags); | |
554f786e SR |
3495 | |
3496 | return ret; | |
7a8e76a3 | 3497 | } |
c4f50183 | 3498 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 | 3499 | |
85bac32c | 3500 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
7a8e76a3 SR |
3501 | /** |
3502 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
3503 | * @buffer_a: One buffer to swap with | |
3504 | * @buffer_b: The other buffer to swap with | |
3505 | * | |
3506 | * This function is useful for tracers that want to take a "snapshot" | |
3507 | * of a CPU buffer and has another back up buffer lying around. | |
3508 | * it is expected that the tracer handles the cpu buffer not being | |
3509 | * used at the moment. | |
3510 | */ | |
3511 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
3512 | struct ring_buffer *buffer_b, int cpu) | |
3513 | { | |
3514 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
3515 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
3516 | int ret = -EINVAL; |
3517 | ||
9e01c1b7 RR |
3518 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
3519 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 3520 | goto out; |
7a8e76a3 SR |
3521 | |
3522 | /* At least make sure the two buffers are somewhat the same */ | |
6d102bc6 | 3523 | if (buffer_a->pages != buffer_b->pages) |
554f786e SR |
3524 | goto out; |
3525 | ||
3526 | ret = -EAGAIN; | |
7a8e76a3 | 3527 | |
97b17efe | 3528 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 3529 | goto out; |
97b17efe SR |
3530 | |
3531 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 3532 | goto out; |
97b17efe SR |
3533 | |
3534 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 3535 | goto out; |
97b17efe | 3536 | |
7a8e76a3 SR |
3537 | cpu_buffer_a = buffer_a->buffers[cpu]; |
3538 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
3539 | ||
97b17efe | 3540 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 3541 | goto out; |
97b17efe SR |
3542 | |
3543 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 3544 | goto out; |
97b17efe | 3545 | |
7a8e76a3 SR |
3546 | /* |
3547 | * We can't do a synchronize_sched here because this | |
3548 | * function can be called in atomic context. | |
3549 | * Normally this will be called from the same CPU as cpu. | |
3550 | * If not it's up to the caller to protect this. | |
3551 | */ | |
3552 | atomic_inc(&cpu_buffer_a->record_disabled); | |
3553 | atomic_inc(&cpu_buffer_b->record_disabled); | |
3554 | ||
98277991 SR |
3555 | ret = -EBUSY; |
3556 | if (local_read(&cpu_buffer_a->committing)) | |
3557 | goto out_dec; | |
3558 | if (local_read(&cpu_buffer_b->committing)) | |
3559 | goto out_dec; | |
3560 | ||
7a8e76a3 SR |
3561 | buffer_a->buffers[cpu] = cpu_buffer_b; |
3562 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
3563 | ||
3564 | cpu_buffer_b->buffer = buffer_a; | |
3565 | cpu_buffer_a->buffer = buffer_b; | |
3566 | ||
98277991 SR |
3567 | ret = 0; |
3568 | ||
3569 | out_dec: | |
7a8e76a3 SR |
3570 | atomic_dec(&cpu_buffer_a->record_disabled); |
3571 | atomic_dec(&cpu_buffer_b->record_disabled); | |
554f786e | 3572 | out: |
554f786e | 3573 | return ret; |
7a8e76a3 | 3574 | } |
c4f50183 | 3575 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
85bac32c | 3576 | #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */ |
7a8e76a3 | 3577 | |
8789a9e7 SR |
3578 | /** |
3579 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
3580 | * @buffer: the buffer to allocate for. | |
3581 | * | |
3582 | * This function is used in conjunction with ring_buffer_read_page. | |
3583 | * When reading a full page from the ring buffer, these functions | |
3584 | * can be used to speed up the process. The calling function should | |
3585 | * allocate a few pages first with this function. Then when it | |
3586 | * needs to get pages from the ring buffer, it passes the result | |
3587 | * of this function into ring_buffer_read_page, which will swap | |
3588 | * the page that was allocated, with the read page of the buffer. | |
3589 | * | |
3590 | * Returns: | |
3591 | * The page allocated, or NULL on error. | |
3592 | */ | |
3593 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer) | |
3594 | { | |
044fa782 | 3595 | struct buffer_data_page *bpage; |
ef7a4a16 | 3596 | unsigned long addr; |
8789a9e7 SR |
3597 | |
3598 | addr = __get_free_page(GFP_KERNEL); | |
3599 | if (!addr) | |
3600 | return NULL; | |
3601 | ||
044fa782 | 3602 | bpage = (void *)addr; |
8789a9e7 | 3603 | |
ef7a4a16 SR |
3604 | rb_init_page(bpage); |
3605 | ||
044fa782 | 3606 | return bpage; |
8789a9e7 | 3607 | } |
d6ce96da | 3608 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
3609 | |
3610 | /** | |
3611 | * ring_buffer_free_read_page - free an allocated read page | |
3612 | * @buffer: the buffer the page was allocate for | |
3613 | * @data: the page to free | |
3614 | * | |
3615 | * Free a page allocated from ring_buffer_alloc_read_page. | |
3616 | */ | |
3617 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
3618 | { | |
3619 | free_page((unsigned long)data); | |
3620 | } | |
d6ce96da | 3621 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
3622 | |
3623 | /** | |
3624 | * ring_buffer_read_page - extract a page from the ring buffer | |
3625 | * @buffer: buffer to extract from | |
3626 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 3627 | * @len: amount to extract |
8789a9e7 SR |
3628 | * @cpu: the cpu of the buffer to extract |
3629 | * @full: should the extraction only happen when the page is full. | |
3630 | * | |
3631 | * This function will pull out a page from the ring buffer and consume it. | |
3632 | * @data_page must be the address of the variable that was returned | |
3633 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
3634 | * to swap with a page in the ring buffer. | |
3635 | * | |
3636 | * for example: | |
b85fa01e | 3637 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
3638 | * if (!rpage) |
3639 | * return error; | |
ef7a4a16 | 3640 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
3641 | * if (ret >= 0) |
3642 | * process_page(rpage, ret); | |
8789a9e7 SR |
3643 | * |
3644 | * When @full is set, the function will not return true unless | |
3645 | * the writer is off the reader page. | |
3646 | * | |
3647 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
3648 | * The ring buffer can be used anywhere in the kernel and can not | |
3649 | * blindly call wake_up. The layer that uses the ring buffer must be | |
3650 | * responsible for that. | |
3651 | * | |
3652 | * Returns: | |
667d2412 LJ |
3653 | * >=0 if data has been transferred, returns the offset of consumed data. |
3654 | * <0 if no data has been transferred. | |
8789a9e7 SR |
3655 | */ |
3656 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 3657 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
3658 | { |
3659 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
3660 | struct ring_buffer_event *event; | |
044fa782 | 3661 | struct buffer_data_page *bpage; |
ef7a4a16 | 3662 | struct buffer_page *reader; |
8789a9e7 | 3663 | unsigned long flags; |
ef7a4a16 | 3664 | unsigned int commit; |
667d2412 | 3665 | unsigned int read; |
4f3640f8 | 3666 | u64 save_timestamp; |
667d2412 | 3667 | int ret = -1; |
8789a9e7 | 3668 | |
554f786e SR |
3669 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
3670 | goto out; | |
3671 | ||
474d32b6 SR |
3672 | /* |
3673 | * If len is not big enough to hold the page header, then | |
3674 | * we can not copy anything. | |
3675 | */ | |
3676 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 3677 | goto out; |
474d32b6 SR |
3678 | |
3679 | len -= BUF_PAGE_HDR_SIZE; | |
3680 | ||
8789a9e7 | 3681 | if (!data_page) |
554f786e | 3682 | goto out; |
8789a9e7 | 3683 | |
044fa782 SR |
3684 | bpage = *data_page; |
3685 | if (!bpage) | |
554f786e | 3686 | goto out; |
8789a9e7 SR |
3687 | |
3688 | spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
3689 | ||
ef7a4a16 SR |
3690 | reader = rb_get_reader_page(cpu_buffer); |
3691 | if (!reader) | |
554f786e | 3692 | goto out_unlock; |
8789a9e7 | 3693 | |
ef7a4a16 SR |
3694 | event = rb_reader_event(cpu_buffer); |
3695 | ||
3696 | read = reader->read; | |
3697 | commit = rb_page_commit(reader); | |
667d2412 | 3698 | |
8789a9e7 | 3699 | /* |
474d32b6 SR |
3700 | * If this page has been partially read or |
3701 | * if len is not big enough to read the rest of the page or | |
3702 | * a writer is still on the page, then | |
3703 | * we must copy the data from the page to the buffer. | |
3704 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 3705 | */ |
474d32b6 | 3706 | if (read || (len < (commit - read)) || |
ef7a4a16 | 3707 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 3708 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
3709 | unsigned int rpos = read; |
3710 | unsigned int pos = 0; | |
ef7a4a16 | 3711 | unsigned int size; |
8789a9e7 SR |
3712 | |
3713 | if (full) | |
554f786e | 3714 | goto out_unlock; |
8789a9e7 | 3715 | |
ef7a4a16 SR |
3716 | if (len > (commit - read)) |
3717 | len = (commit - read); | |
3718 | ||
3719 | size = rb_event_length(event); | |
3720 | ||
3721 | if (len < size) | |
554f786e | 3722 | goto out_unlock; |
ef7a4a16 | 3723 | |
4f3640f8 SR |
3724 | /* save the current timestamp, since the user will need it */ |
3725 | save_timestamp = cpu_buffer->read_stamp; | |
3726 | ||
ef7a4a16 SR |
3727 | /* Need to copy one event at a time */ |
3728 | do { | |
474d32b6 | 3729 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
3730 | |
3731 | len -= size; | |
3732 | ||
3733 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
3734 | rpos = reader->read; |
3735 | pos += size; | |
ef7a4a16 SR |
3736 | |
3737 | event = rb_reader_event(cpu_buffer); | |
3738 | size = rb_event_length(event); | |
3739 | } while (len > size); | |
667d2412 LJ |
3740 | |
3741 | /* update bpage */ | |
ef7a4a16 | 3742 | local_set(&bpage->commit, pos); |
4f3640f8 | 3743 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 3744 | |
474d32b6 SR |
3745 | /* we copied everything to the beginning */ |
3746 | read = 0; | |
8789a9e7 | 3747 | } else { |
afbab76a | 3748 | /* update the entry counter */ |
77ae365e | 3749 | cpu_buffer->read += rb_page_entries(reader); |
afbab76a | 3750 | |
8789a9e7 | 3751 | /* swap the pages */ |
044fa782 | 3752 | rb_init_page(bpage); |
ef7a4a16 SR |
3753 | bpage = reader->page; |
3754 | reader->page = *data_page; | |
3755 | local_set(&reader->write, 0); | |
778c55d4 | 3756 | local_set(&reader->entries, 0); |
ef7a4a16 | 3757 | reader->read = 0; |
044fa782 | 3758 | *data_page = bpage; |
8789a9e7 | 3759 | } |
667d2412 | 3760 | ret = read; |
8789a9e7 | 3761 | |
554f786e | 3762 | out_unlock: |
8789a9e7 SR |
3763 | spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
3764 | ||
554f786e | 3765 | out: |
8789a9e7 SR |
3766 | return ret; |
3767 | } | |
d6ce96da | 3768 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 3769 | |
1155de47 | 3770 | #ifdef CONFIG_TRACING |
a3583244 SR |
3771 | static ssize_t |
3772 | rb_simple_read(struct file *filp, char __user *ubuf, | |
3773 | size_t cnt, loff_t *ppos) | |
3774 | { | |
5e39841c | 3775 | unsigned long *p = filp->private_data; |
a3583244 SR |
3776 | char buf[64]; |
3777 | int r; | |
3778 | ||
033601a3 SR |
3779 | if (test_bit(RB_BUFFERS_DISABLED_BIT, p)) |
3780 | r = sprintf(buf, "permanently disabled\n"); | |
3781 | else | |
3782 | r = sprintf(buf, "%d\n", test_bit(RB_BUFFERS_ON_BIT, p)); | |
a3583244 SR |
3783 | |
3784 | return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
3785 | } | |
3786 | ||
3787 | static ssize_t | |
3788 | rb_simple_write(struct file *filp, const char __user *ubuf, | |
3789 | size_t cnt, loff_t *ppos) | |
3790 | { | |
5e39841c | 3791 | unsigned long *p = filp->private_data; |
a3583244 | 3792 | char buf[64]; |
5e39841c | 3793 | unsigned long val; |
a3583244 SR |
3794 | int ret; |
3795 | ||
3796 | if (cnt >= sizeof(buf)) | |
3797 | return -EINVAL; | |
3798 | ||
3799 | if (copy_from_user(&buf, ubuf, cnt)) | |
3800 | return -EFAULT; | |
3801 | ||
3802 | buf[cnt] = 0; | |
3803 | ||
3804 | ret = strict_strtoul(buf, 10, &val); | |
3805 | if (ret < 0) | |
3806 | return ret; | |
3807 | ||
033601a3 SR |
3808 | if (val) |
3809 | set_bit(RB_BUFFERS_ON_BIT, p); | |
3810 | else | |
3811 | clear_bit(RB_BUFFERS_ON_BIT, p); | |
a3583244 SR |
3812 | |
3813 | (*ppos)++; | |
3814 | ||
3815 | return cnt; | |
3816 | } | |
3817 | ||
5e2336a0 | 3818 | static const struct file_operations rb_simple_fops = { |
a3583244 SR |
3819 | .open = tracing_open_generic, |
3820 | .read = rb_simple_read, | |
3821 | .write = rb_simple_write, | |
3822 | }; | |
3823 | ||
3824 | ||
3825 | static __init int rb_init_debugfs(void) | |
3826 | { | |
3827 | struct dentry *d_tracer; | |
a3583244 SR |
3828 | |
3829 | d_tracer = tracing_init_dentry(); | |
3830 | ||
5452af66 FW |
3831 | trace_create_file("tracing_on", 0644, d_tracer, |
3832 | &ring_buffer_flags, &rb_simple_fops); | |
a3583244 SR |
3833 | |
3834 | return 0; | |
3835 | } | |
3836 | ||
3837 | fs_initcall(rb_init_debugfs); | |
1155de47 | 3838 | #endif |
554f786e | 3839 | |
59222efe | 3840 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
3841 | static int rb_cpu_notify(struct notifier_block *self, |
3842 | unsigned long action, void *hcpu) | |
554f786e SR |
3843 | { |
3844 | struct ring_buffer *buffer = | |
3845 | container_of(self, struct ring_buffer, cpu_notify); | |
3846 | long cpu = (long)hcpu; | |
3847 | ||
3848 | switch (action) { | |
3849 | case CPU_UP_PREPARE: | |
3850 | case CPU_UP_PREPARE_FROZEN: | |
3f237a79 | 3851 | if (cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e SR |
3852 | return NOTIFY_OK; |
3853 | ||
3854 | buffer->buffers[cpu] = | |
3855 | rb_allocate_cpu_buffer(buffer, cpu); | |
3856 | if (!buffer->buffers[cpu]) { | |
3857 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
3858 | cpu); | |
3859 | return NOTIFY_OK; | |
3860 | } | |
3861 | smp_wmb(); | |
3f237a79 | 3862 | cpumask_set_cpu(cpu, buffer->cpumask); |
554f786e SR |
3863 | break; |
3864 | case CPU_DOWN_PREPARE: | |
3865 | case CPU_DOWN_PREPARE_FROZEN: | |
3866 | /* | |
3867 | * Do nothing. | |
3868 | * If we were to free the buffer, then the user would | |
3869 | * lose any trace that was in the buffer. | |
3870 | */ | |
3871 | break; | |
3872 | default: | |
3873 | break; | |
3874 | } | |
3875 | return NOTIFY_OK; | |
3876 | } | |
3877 | #endif |