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> |
7a8e76a3 SR |
8 | #include <linux/spinlock.h> |
9 | #include <linux/debugfs.h> | |
10 | #include <linux/uaccess.h> | |
a81bd80a | 11 | #include <linux/hardirq.h> |
1744a21d | 12 | #include <linux/kmemcheck.h> |
7a8e76a3 SR |
13 | #include <linux/module.h> |
14 | #include <linux/percpu.h> | |
15 | #include <linux/mutex.h> | |
5a0e3ad6 | 16 | #include <linux/slab.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 | ||
79615760 | 23 | #include <asm/local.h> |
182e9f5f SR |
24 | #include "trace.h" |
25 | ||
83f40318 VN |
26 | static void update_pages_handler(struct work_struct *work); |
27 | ||
d1b182a8 SR |
28 | /* |
29 | * The ring buffer header is special. We must manually up keep it. | |
30 | */ | |
31 | int ring_buffer_print_entry_header(struct trace_seq *s) | |
32 | { | |
33 | int ret; | |
34 | ||
334d4169 LJ |
35 | ret = trace_seq_printf(s, "# compressed entry header\n"); |
36 | ret = trace_seq_printf(s, "\ttype_len : 5 bits\n"); | |
d1b182a8 SR |
37 | ret = trace_seq_printf(s, "\ttime_delta : 27 bits\n"); |
38 | ret = trace_seq_printf(s, "\tarray : 32 bits\n"); | |
39 | ret = trace_seq_printf(s, "\n"); | |
40 | ret = trace_seq_printf(s, "\tpadding : type == %d\n", | |
41 | RINGBUF_TYPE_PADDING); | |
42 | ret = trace_seq_printf(s, "\ttime_extend : type == %d\n", | |
43 | RINGBUF_TYPE_TIME_EXTEND); | |
334d4169 LJ |
44 | ret = trace_seq_printf(s, "\tdata max type_len == %d\n", |
45 | RINGBUF_TYPE_DATA_TYPE_LEN_MAX); | |
d1b182a8 SR |
46 | |
47 | return ret; | |
48 | } | |
49 | ||
5cc98548 SR |
50 | /* |
51 | * The ring buffer is made up of a list of pages. A separate list of pages is | |
52 | * allocated for each CPU. A writer may only write to a buffer that is | |
53 | * associated with the CPU it is currently executing on. A reader may read | |
54 | * from any per cpu buffer. | |
55 | * | |
56 | * The reader is special. For each per cpu buffer, the reader has its own | |
57 | * reader page. When a reader has read the entire reader page, this reader | |
58 | * page is swapped with another page in the ring buffer. | |
59 | * | |
60 | * Now, as long as the writer is off the reader page, the reader can do what | |
61 | * ever it wants with that page. The writer will never write to that page | |
62 | * again (as long as it is out of the ring buffer). | |
63 | * | |
64 | * Here's some silly ASCII art. | |
65 | * | |
66 | * +------+ | |
67 | * |reader| RING BUFFER | |
68 | * |page | | |
69 | * +------+ +---+ +---+ +---+ | |
70 | * | |-->| |-->| | | |
71 | * +---+ +---+ +---+ | |
72 | * ^ | | |
73 | * | | | |
74 | * +---------------+ | |
75 | * | |
76 | * | |
77 | * +------+ | |
78 | * |reader| RING BUFFER | |
79 | * |page |------------------v | |
80 | * +------+ +---+ +---+ +---+ | |
81 | * | |-->| |-->| | | |
82 | * +---+ +---+ +---+ | |
83 | * ^ | | |
84 | * | | | |
85 | * +---------------+ | |
86 | * | |
87 | * | |
88 | * +------+ | |
89 | * |reader| RING BUFFER | |
90 | * |page |------------------v | |
91 | * +------+ +---+ +---+ +---+ | |
92 | * ^ | |-->| |-->| | | |
93 | * | +---+ +---+ +---+ | |
94 | * | | | |
95 | * | | | |
96 | * +------------------------------+ | |
97 | * | |
98 | * | |
99 | * +------+ | |
100 | * |buffer| RING BUFFER | |
101 | * |page |------------------v | |
102 | * +------+ +---+ +---+ +---+ | |
103 | * ^ | | | |-->| | | |
104 | * | New +---+ +---+ +---+ | |
105 | * | Reader------^ | | |
106 | * | page | | |
107 | * +------------------------------+ | |
108 | * | |
109 | * | |
110 | * After we make this swap, the reader can hand this page off to the splice | |
111 | * code and be done with it. It can even allocate a new page if it needs to | |
112 | * and swap that into the ring buffer. | |
113 | * | |
114 | * We will be using cmpxchg soon to make all this lockless. | |
115 | * | |
116 | */ | |
117 | ||
033601a3 SR |
118 | /* |
119 | * A fast way to enable or disable all ring buffers is to | |
120 | * call tracing_on or tracing_off. Turning off the ring buffers | |
121 | * prevents all ring buffers from being recorded to. | |
122 | * Turning this switch on, makes it OK to write to the | |
123 | * ring buffer, if the ring buffer is enabled itself. | |
124 | * | |
125 | * There's three layers that must be on in order to write | |
126 | * to the ring buffer. | |
127 | * | |
128 | * 1) This global flag must be set. | |
129 | * 2) The ring buffer must be enabled for recording. | |
130 | * 3) The per cpu buffer must be enabled for recording. | |
131 | * | |
132 | * In case of an anomaly, this global flag has a bit set that | |
133 | * will permantly disable all ring buffers. | |
134 | */ | |
135 | ||
136 | /* | |
137 | * Global flag to disable all recording to ring buffers | |
138 | * This has two bits: ON, DISABLED | |
139 | * | |
140 | * ON DISABLED | |
141 | * ---- ---------- | |
142 | * 0 0 : ring buffers are off | |
143 | * 1 0 : ring buffers are on | |
144 | * X 1 : ring buffers are permanently disabled | |
145 | */ | |
146 | ||
147 | enum { | |
148 | RB_BUFFERS_ON_BIT = 0, | |
149 | RB_BUFFERS_DISABLED_BIT = 1, | |
150 | }; | |
151 | ||
152 | enum { | |
153 | RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT, | |
154 | RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT, | |
155 | }; | |
156 | ||
5e39841c | 157 | static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON; |
a3583244 | 158 | |
499e5470 SR |
159 | /* Used for individual buffers (after the counter) */ |
160 | #define RB_BUFFER_OFF (1 << 20) | |
a3583244 | 161 | |
499e5470 | 162 | #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data) |
033601a3 SR |
163 | |
164 | /** | |
165 | * tracing_off_permanent - permanently disable ring buffers | |
166 | * | |
167 | * This function, once called, will disable all ring buffers | |
c3706f00 | 168 | * permanently. |
033601a3 SR |
169 | */ |
170 | void tracing_off_permanent(void) | |
171 | { | |
172 | set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags); | |
a3583244 SR |
173 | } |
174 | ||
e3d6bf0a | 175 | #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array)) |
67d34724 | 176 | #define RB_ALIGNMENT 4U |
334d4169 | 177 | #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
c7b09308 | 178 | #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */ |
334d4169 | 179 | |
2271048d SR |
180 | #if !defined(CONFIG_64BIT) || defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) |
181 | # define RB_FORCE_8BYTE_ALIGNMENT 0 | |
182 | # define RB_ARCH_ALIGNMENT RB_ALIGNMENT | |
183 | #else | |
184 | # define RB_FORCE_8BYTE_ALIGNMENT 1 | |
185 | # define RB_ARCH_ALIGNMENT 8U | |
186 | #endif | |
187 | ||
334d4169 LJ |
188 | /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */ |
189 | #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX | |
7a8e76a3 SR |
190 | |
191 | enum { | |
192 | RB_LEN_TIME_EXTEND = 8, | |
193 | RB_LEN_TIME_STAMP = 16, | |
194 | }; | |
195 | ||
69d1b839 SR |
196 | #define skip_time_extend(event) \ |
197 | ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND)) | |
198 | ||
2d622719 TZ |
199 | static inline int rb_null_event(struct ring_buffer_event *event) |
200 | { | |
a1863c21 | 201 | return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta; |
2d622719 TZ |
202 | } |
203 | ||
204 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
205 | { | |
a1863c21 | 206 | /* padding has a NULL time_delta */ |
334d4169 | 207 | event->type_len = RINGBUF_TYPE_PADDING; |
2d622719 TZ |
208 | event->time_delta = 0; |
209 | } | |
210 | ||
34a148bf | 211 | static unsigned |
2d622719 | 212 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
213 | { |
214 | unsigned length; | |
215 | ||
334d4169 LJ |
216 | if (event->type_len) |
217 | length = event->type_len * RB_ALIGNMENT; | |
2d622719 TZ |
218 | else |
219 | length = event->array[0]; | |
220 | return length + RB_EVNT_HDR_SIZE; | |
221 | } | |
222 | ||
69d1b839 SR |
223 | /* |
224 | * Return the length of the given event. Will return | |
225 | * the length of the time extend if the event is a | |
226 | * time extend. | |
227 | */ | |
228 | static inline unsigned | |
2d622719 TZ |
229 | rb_event_length(struct ring_buffer_event *event) |
230 | { | |
334d4169 | 231 | switch (event->type_len) { |
7a8e76a3 | 232 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
233 | if (rb_null_event(event)) |
234 | /* undefined */ | |
235 | return -1; | |
334d4169 | 236 | return event->array[0] + RB_EVNT_HDR_SIZE; |
7a8e76a3 SR |
237 | |
238 | case RINGBUF_TYPE_TIME_EXTEND: | |
239 | return RB_LEN_TIME_EXTEND; | |
240 | ||
241 | case RINGBUF_TYPE_TIME_STAMP: | |
242 | return RB_LEN_TIME_STAMP; | |
243 | ||
244 | case RINGBUF_TYPE_DATA: | |
2d622719 | 245 | return rb_event_data_length(event); |
7a8e76a3 SR |
246 | default: |
247 | BUG(); | |
248 | } | |
249 | /* not hit */ | |
250 | return 0; | |
251 | } | |
252 | ||
69d1b839 SR |
253 | /* |
254 | * Return total length of time extend and data, | |
255 | * or just the event length for all other events. | |
256 | */ | |
257 | static inline unsigned | |
258 | rb_event_ts_length(struct ring_buffer_event *event) | |
259 | { | |
260 | unsigned len = 0; | |
261 | ||
262 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
263 | /* time extends include the data event after it */ | |
264 | len = RB_LEN_TIME_EXTEND; | |
265 | event = skip_time_extend(event); | |
266 | } | |
267 | return len + rb_event_length(event); | |
268 | } | |
269 | ||
7a8e76a3 SR |
270 | /** |
271 | * ring_buffer_event_length - return the length of the event | |
272 | * @event: the event to get the length of | |
69d1b839 SR |
273 | * |
274 | * Returns the size of the data load of a data event. | |
275 | * If the event is something other than a data event, it | |
276 | * returns the size of the event itself. With the exception | |
277 | * of a TIME EXTEND, where it still returns the size of the | |
278 | * data load of the data event after it. | |
7a8e76a3 SR |
279 | */ |
280 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
281 | { | |
69d1b839 SR |
282 | unsigned length; |
283 | ||
284 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) | |
285 | event = skip_time_extend(event); | |
286 | ||
287 | length = rb_event_length(event); | |
334d4169 | 288 | if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
465634ad RR |
289 | return length; |
290 | length -= RB_EVNT_HDR_SIZE; | |
291 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
292 | length -= sizeof(event->array[0]); | |
293 | return length; | |
7a8e76a3 | 294 | } |
c4f50183 | 295 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
296 | |
297 | /* inline for ring buffer fast paths */ | |
34a148bf | 298 | static void * |
7a8e76a3 SR |
299 | rb_event_data(struct ring_buffer_event *event) |
300 | { | |
69d1b839 SR |
301 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) |
302 | event = skip_time_extend(event); | |
334d4169 | 303 | BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX); |
7a8e76a3 | 304 | /* If length is in len field, then array[0] has the data */ |
334d4169 | 305 | if (event->type_len) |
7a8e76a3 SR |
306 | return (void *)&event->array[0]; |
307 | /* Otherwise length is in array[0] and array[1] has the data */ | |
308 | return (void *)&event->array[1]; | |
309 | } | |
310 | ||
311 | /** | |
312 | * ring_buffer_event_data - return the data of the event | |
313 | * @event: the event to get the data from | |
314 | */ | |
315 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
316 | { | |
317 | return rb_event_data(event); | |
318 | } | |
c4f50183 | 319 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
320 | |
321 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 322 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
323 | |
324 | #define TS_SHIFT 27 | |
325 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
326 | #define TS_DELTA_TEST (~TS_MASK) | |
327 | ||
66a8cb95 SR |
328 | /* Flag when events were overwritten */ |
329 | #define RB_MISSED_EVENTS (1 << 31) | |
ff0ff84a SR |
330 | /* Missed count stored at end */ |
331 | #define RB_MISSED_STORED (1 << 30) | |
66a8cb95 | 332 | |
abc9b56d | 333 | struct buffer_data_page { |
e4c2ce82 | 334 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 335 | local_t commit; /* write committed index */ |
abc9b56d SR |
336 | unsigned char data[]; /* data of buffer page */ |
337 | }; | |
338 | ||
77ae365e SR |
339 | /* |
340 | * Note, the buffer_page list must be first. The buffer pages | |
341 | * are allocated in cache lines, which means that each buffer | |
342 | * page will be at the beginning of a cache line, and thus | |
343 | * the least significant bits will be zero. We use this to | |
344 | * add flags in the list struct pointers, to make the ring buffer | |
345 | * lockless. | |
346 | */ | |
abc9b56d | 347 | struct buffer_page { |
778c55d4 | 348 | struct list_head list; /* list of buffer pages */ |
abc9b56d | 349 | local_t write; /* index for next write */ |
6f807acd | 350 | unsigned read; /* index for next read */ |
778c55d4 | 351 | local_t entries; /* entries on this page */ |
ff0ff84a | 352 | unsigned long real_end; /* real end of data */ |
abc9b56d | 353 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
354 | }; |
355 | ||
77ae365e SR |
356 | /* |
357 | * The buffer page counters, write and entries, must be reset | |
358 | * atomically when crossing page boundaries. To synchronize this | |
359 | * update, two counters are inserted into the number. One is | |
360 | * the actual counter for the write position or count on the page. | |
361 | * | |
362 | * The other is a counter of updaters. Before an update happens | |
363 | * the update partition of the counter is incremented. This will | |
364 | * allow the updater to update the counter atomically. | |
365 | * | |
366 | * The counter is 20 bits, and the state data is 12. | |
367 | */ | |
368 | #define RB_WRITE_MASK 0xfffff | |
369 | #define RB_WRITE_INTCNT (1 << 20) | |
370 | ||
044fa782 | 371 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 372 | { |
044fa782 | 373 | local_set(&bpage->commit, 0); |
abc9b56d SR |
374 | } |
375 | ||
474d32b6 SR |
376 | /** |
377 | * ring_buffer_page_len - the size of data on the page. | |
378 | * @page: The page to read | |
379 | * | |
380 | * Returns the amount of data on the page, including buffer page header. | |
381 | */ | |
ef7a4a16 SR |
382 | size_t ring_buffer_page_len(void *page) |
383 | { | |
474d32b6 SR |
384 | return local_read(&((struct buffer_data_page *)page)->commit) |
385 | + BUF_PAGE_HDR_SIZE; | |
ef7a4a16 SR |
386 | } |
387 | ||
ed56829c SR |
388 | /* |
389 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
390 | * this issue out. | |
391 | */ | |
34a148bf | 392 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 393 | { |
34a148bf | 394 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 395 | kfree(bpage); |
ed56829c SR |
396 | } |
397 | ||
7a8e76a3 SR |
398 | /* |
399 | * We need to fit the time_stamp delta into 27 bits. | |
400 | */ | |
401 | static inline int test_time_stamp(u64 delta) | |
402 | { | |
403 | if (delta & TS_DELTA_TEST) | |
404 | return 1; | |
405 | return 0; | |
406 | } | |
407 | ||
474d32b6 | 408 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 | 409 | |
be957c44 SR |
410 | /* Max payload is BUF_PAGE_SIZE - header (8bytes) */ |
411 | #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) | |
412 | ||
d1b182a8 SR |
413 | int ring_buffer_print_page_header(struct trace_seq *s) |
414 | { | |
415 | struct buffer_data_page field; | |
416 | int ret; | |
417 | ||
418 | ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t" | |
26a50744 TZ |
419 | "offset:0;\tsize:%u;\tsigned:%u;\n", |
420 | (unsigned int)sizeof(field.time_stamp), | |
421 | (unsigned int)is_signed_type(u64)); | |
d1b182a8 SR |
422 | |
423 | ret = trace_seq_printf(s, "\tfield: local_t commit;\t" | |
26a50744 | 424 | "offset:%u;\tsize:%u;\tsigned:%u;\n", |
d1b182a8 | 425 | (unsigned int)offsetof(typeof(field), commit), |
26a50744 TZ |
426 | (unsigned int)sizeof(field.commit), |
427 | (unsigned int)is_signed_type(long)); | |
d1b182a8 | 428 | |
66a8cb95 SR |
429 | ret = trace_seq_printf(s, "\tfield: int overwrite;\t" |
430 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
431 | (unsigned int)offsetof(typeof(field), commit), | |
432 | 1, | |
433 | (unsigned int)is_signed_type(long)); | |
434 | ||
d1b182a8 | 435 | ret = trace_seq_printf(s, "\tfield: char data;\t" |
26a50744 | 436 | "offset:%u;\tsize:%u;\tsigned:%u;\n", |
d1b182a8 | 437 | (unsigned int)offsetof(typeof(field), data), |
26a50744 TZ |
438 | (unsigned int)BUF_PAGE_SIZE, |
439 | (unsigned int)is_signed_type(char)); | |
d1b182a8 SR |
440 | |
441 | return ret; | |
442 | } | |
443 | ||
7a8e76a3 SR |
444 | /* |
445 | * head_page == tail_page && head == tail then buffer is empty. | |
446 | */ | |
447 | struct ring_buffer_per_cpu { | |
448 | int cpu; | |
985023de | 449 | atomic_t record_disabled; |
7a8e76a3 | 450 | struct ring_buffer *buffer; |
5389f6fa | 451 | raw_spinlock_t reader_lock; /* serialize readers */ |
445c8951 | 452 | arch_spinlock_t lock; |
7a8e76a3 | 453 | struct lock_class_key lock_key; |
438ced17 | 454 | unsigned int nr_pages; |
3adc54fa | 455 | struct list_head *pages; |
6f807acd SR |
456 | struct buffer_page *head_page; /* read from head */ |
457 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 458 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 459 | struct buffer_page *reader_page; |
66a8cb95 SR |
460 | unsigned long lost_events; |
461 | unsigned long last_overrun; | |
c64e148a | 462 | local_t entries_bytes; |
77ae365e SR |
463 | local_t commit_overrun; |
464 | local_t overrun; | |
e4906eff | 465 | local_t entries; |
fa743953 SR |
466 | local_t committing; |
467 | local_t commits; | |
77ae365e | 468 | unsigned long read; |
c64e148a | 469 | unsigned long read_bytes; |
7a8e76a3 SR |
470 | u64 write_stamp; |
471 | u64 read_stamp; | |
438ced17 VN |
472 | /* ring buffer pages to update, > 0 to add, < 0 to remove */ |
473 | int nr_pages_to_update; | |
474 | struct list_head new_pages; /* new pages to add */ | |
83f40318 VN |
475 | struct work_struct update_pages_work; |
476 | struct completion update_completion; | |
7a8e76a3 SR |
477 | }; |
478 | ||
479 | struct ring_buffer { | |
7a8e76a3 SR |
480 | unsigned flags; |
481 | int cpus; | |
7a8e76a3 | 482 | atomic_t record_disabled; |
83f40318 | 483 | atomic_t resize_disabled; |
00f62f61 | 484 | cpumask_var_t cpumask; |
7a8e76a3 | 485 | |
1f8a6a10 PZ |
486 | struct lock_class_key *reader_lock_key; |
487 | ||
7a8e76a3 SR |
488 | struct mutex mutex; |
489 | ||
490 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 491 | |
59222efe | 492 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
493 | struct notifier_block cpu_notify; |
494 | #endif | |
37886f6a | 495 | u64 (*clock)(void); |
7a8e76a3 SR |
496 | }; |
497 | ||
498 | struct ring_buffer_iter { | |
499 | struct ring_buffer_per_cpu *cpu_buffer; | |
500 | unsigned long head; | |
501 | struct buffer_page *head_page; | |
492a74f4 SR |
502 | struct buffer_page *cache_reader_page; |
503 | unsigned long cache_read; | |
7a8e76a3 SR |
504 | u64 read_stamp; |
505 | }; | |
506 | ||
f536aafc | 507 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
077c5407 SR |
508 | #define RB_WARN_ON(b, cond) \ |
509 | ({ \ | |
510 | int _____ret = unlikely(cond); \ | |
511 | if (_____ret) { \ | |
512 | if (__same_type(*(b), struct ring_buffer_per_cpu)) { \ | |
513 | struct ring_buffer_per_cpu *__b = \ | |
514 | (void *)b; \ | |
515 | atomic_inc(&__b->buffer->record_disabled); \ | |
516 | } else \ | |
517 | atomic_inc(&b->record_disabled); \ | |
518 | WARN_ON(1); \ | |
519 | } \ | |
520 | _____ret; \ | |
3e89c7bb | 521 | }) |
f536aafc | 522 | |
37886f6a SR |
523 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
524 | #define DEBUG_SHIFT 0 | |
525 | ||
6d3f1e12 | 526 | static inline u64 rb_time_stamp(struct ring_buffer *buffer) |
88eb0125 SR |
527 | { |
528 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
529 | return buffer->clock() << DEBUG_SHIFT; | |
530 | } | |
531 | ||
37886f6a SR |
532 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) |
533 | { | |
534 | u64 time; | |
535 | ||
536 | preempt_disable_notrace(); | |
6d3f1e12 | 537 | time = rb_time_stamp(buffer); |
37886f6a SR |
538 | preempt_enable_no_resched_notrace(); |
539 | ||
540 | return time; | |
541 | } | |
542 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
543 | ||
544 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
545 | int cpu, u64 *ts) | |
546 | { | |
547 | /* Just stupid testing the normalize function and deltas */ | |
548 | *ts >>= DEBUG_SHIFT; | |
549 | } | |
550 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
551 | ||
77ae365e SR |
552 | /* |
553 | * Making the ring buffer lockless makes things tricky. | |
554 | * Although writes only happen on the CPU that they are on, | |
555 | * and they only need to worry about interrupts. Reads can | |
556 | * happen on any CPU. | |
557 | * | |
558 | * The reader page is always off the ring buffer, but when the | |
559 | * reader finishes with a page, it needs to swap its page with | |
560 | * a new one from the buffer. The reader needs to take from | |
561 | * the head (writes go to the tail). But if a writer is in overwrite | |
562 | * mode and wraps, it must push the head page forward. | |
563 | * | |
564 | * Here lies the problem. | |
565 | * | |
566 | * The reader must be careful to replace only the head page, and | |
567 | * not another one. As described at the top of the file in the | |
568 | * ASCII art, the reader sets its old page to point to the next | |
569 | * page after head. It then sets the page after head to point to | |
570 | * the old reader page. But if the writer moves the head page | |
571 | * during this operation, the reader could end up with the tail. | |
572 | * | |
573 | * We use cmpxchg to help prevent this race. We also do something | |
574 | * special with the page before head. We set the LSB to 1. | |
575 | * | |
576 | * When the writer must push the page forward, it will clear the | |
577 | * bit that points to the head page, move the head, and then set | |
578 | * the bit that points to the new head page. | |
579 | * | |
580 | * We also don't want an interrupt coming in and moving the head | |
581 | * page on another writer. Thus we use the second LSB to catch | |
582 | * that too. Thus: | |
583 | * | |
584 | * head->list->prev->next bit 1 bit 0 | |
585 | * ------- ------- | |
586 | * Normal page 0 0 | |
587 | * Points to head page 0 1 | |
588 | * New head page 1 0 | |
589 | * | |
590 | * Note we can not trust the prev pointer of the head page, because: | |
591 | * | |
592 | * +----+ +-----+ +-----+ | |
593 | * | |------>| T |---X--->| N | | |
594 | * | |<------| | | | | |
595 | * +----+ +-----+ +-----+ | |
596 | * ^ ^ | | |
597 | * | +-----+ | | | |
598 | * +----------| R |----------+ | | |
599 | * | |<-----------+ | |
600 | * +-----+ | |
601 | * | |
602 | * Key: ---X--> HEAD flag set in pointer | |
603 | * T Tail page | |
604 | * R Reader page | |
605 | * N Next page | |
606 | * | |
607 | * (see __rb_reserve_next() to see where this happens) | |
608 | * | |
609 | * What the above shows is that the reader just swapped out | |
610 | * the reader page with a page in the buffer, but before it | |
611 | * could make the new header point back to the new page added | |
612 | * it was preempted by a writer. The writer moved forward onto | |
613 | * the new page added by the reader and is about to move forward | |
614 | * again. | |
615 | * | |
616 | * You can see, it is legitimate for the previous pointer of | |
617 | * the head (or any page) not to point back to itself. But only | |
618 | * temporarially. | |
619 | */ | |
620 | ||
621 | #define RB_PAGE_NORMAL 0UL | |
622 | #define RB_PAGE_HEAD 1UL | |
623 | #define RB_PAGE_UPDATE 2UL | |
624 | ||
625 | ||
626 | #define RB_FLAG_MASK 3UL | |
627 | ||
628 | /* PAGE_MOVED is not part of the mask */ | |
629 | #define RB_PAGE_MOVED 4UL | |
630 | ||
631 | /* | |
632 | * rb_list_head - remove any bit | |
633 | */ | |
634 | static struct list_head *rb_list_head(struct list_head *list) | |
635 | { | |
636 | unsigned long val = (unsigned long)list; | |
637 | ||
638 | return (struct list_head *)(val & ~RB_FLAG_MASK); | |
639 | } | |
640 | ||
641 | /* | |
6d3f1e12 | 642 | * rb_is_head_page - test if the given page is the head page |
77ae365e SR |
643 | * |
644 | * Because the reader may move the head_page pointer, we can | |
645 | * not trust what the head page is (it may be pointing to | |
646 | * the reader page). But if the next page is a header page, | |
647 | * its flags will be non zero. | |
648 | */ | |
42b16b3f | 649 | static inline int |
77ae365e SR |
650 | rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer, |
651 | struct buffer_page *page, struct list_head *list) | |
652 | { | |
653 | unsigned long val; | |
654 | ||
655 | val = (unsigned long)list->next; | |
656 | ||
657 | if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list) | |
658 | return RB_PAGE_MOVED; | |
659 | ||
660 | return val & RB_FLAG_MASK; | |
661 | } | |
662 | ||
663 | /* | |
664 | * rb_is_reader_page | |
665 | * | |
666 | * The unique thing about the reader page, is that, if the | |
667 | * writer is ever on it, the previous pointer never points | |
668 | * back to the reader page. | |
669 | */ | |
670 | static int rb_is_reader_page(struct buffer_page *page) | |
671 | { | |
672 | struct list_head *list = page->list.prev; | |
673 | ||
674 | return rb_list_head(list->next) != &page->list; | |
675 | } | |
676 | ||
677 | /* | |
678 | * rb_set_list_to_head - set a list_head to be pointing to head. | |
679 | */ | |
680 | static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer, | |
681 | struct list_head *list) | |
682 | { | |
683 | unsigned long *ptr; | |
684 | ||
685 | ptr = (unsigned long *)&list->next; | |
686 | *ptr |= RB_PAGE_HEAD; | |
687 | *ptr &= ~RB_PAGE_UPDATE; | |
688 | } | |
689 | ||
690 | /* | |
691 | * rb_head_page_activate - sets up head page | |
692 | */ | |
693 | static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer) | |
694 | { | |
695 | struct buffer_page *head; | |
696 | ||
697 | head = cpu_buffer->head_page; | |
698 | if (!head) | |
699 | return; | |
700 | ||
701 | /* | |
702 | * Set the previous list pointer to have the HEAD flag. | |
703 | */ | |
704 | rb_set_list_to_head(cpu_buffer, head->list.prev); | |
705 | } | |
706 | ||
707 | static void rb_list_head_clear(struct list_head *list) | |
708 | { | |
709 | unsigned long *ptr = (unsigned long *)&list->next; | |
710 | ||
711 | *ptr &= ~RB_FLAG_MASK; | |
712 | } | |
713 | ||
714 | /* | |
715 | * rb_head_page_dactivate - clears head page ptr (for free list) | |
716 | */ | |
717 | static void | |
718 | rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer) | |
719 | { | |
720 | struct list_head *hd; | |
721 | ||
722 | /* Go through the whole list and clear any pointers found. */ | |
723 | rb_list_head_clear(cpu_buffer->pages); | |
724 | ||
725 | list_for_each(hd, cpu_buffer->pages) | |
726 | rb_list_head_clear(hd); | |
727 | } | |
728 | ||
729 | static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer, | |
730 | struct buffer_page *head, | |
731 | struct buffer_page *prev, | |
732 | int old_flag, int new_flag) | |
733 | { | |
734 | struct list_head *list; | |
735 | unsigned long val = (unsigned long)&head->list; | |
736 | unsigned long ret; | |
737 | ||
738 | list = &prev->list; | |
739 | ||
740 | val &= ~RB_FLAG_MASK; | |
741 | ||
08a40816 SR |
742 | ret = cmpxchg((unsigned long *)&list->next, |
743 | val | old_flag, val | new_flag); | |
77ae365e SR |
744 | |
745 | /* check if the reader took the page */ | |
746 | if ((ret & ~RB_FLAG_MASK) != val) | |
747 | return RB_PAGE_MOVED; | |
748 | ||
749 | return ret & RB_FLAG_MASK; | |
750 | } | |
751 | ||
752 | static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer, | |
753 | struct buffer_page *head, | |
754 | struct buffer_page *prev, | |
755 | int old_flag) | |
756 | { | |
757 | return rb_head_page_set(cpu_buffer, head, prev, | |
758 | old_flag, RB_PAGE_UPDATE); | |
759 | } | |
760 | ||
761 | static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer, | |
762 | struct buffer_page *head, | |
763 | struct buffer_page *prev, | |
764 | int old_flag) | |
765 | { | |
766 | return rb_head_page_set(cpu_buffer, head, prev, | |
767 | old_flag, RB_PAGE_HEAD); | |
768 | } | |
769 | ||
770 | static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer, | |
771 | struct buffer_page *head, | |
772 | struct buffer_page *prev, | |
773 | int old_flag) | |
774 | { | |
775 | return rb_head_page_set(cpu_buffer, head, prev, | |
776 | old_flag, RB_PAGE_NORMAL); | |
777 | } | |
778 | ||
779 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
780 | struct buffer_page **bpage) | |
781 | { | |
782 | struct list_head *p = rb_list_head((*bpage)->list.next); | |
783 | ||
784 | *bpage = list_entry(p, struct buffer_page, list); | |
785 | } | |
786 | ||
787 | static struct buffer_page * | |
788 | rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer) | |
789 | { | |
790 | struct buffer_page *head; | |
791 | struct buffer_page *page; | |
792 | struct list_head *list; | |
793 | int i; | |
794 | ||
795 | if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page)) | |
796 | return NULL; | |
797 | ||
798 | /* sanity check */ | |
799 | list = cpu_buffer->pages; | |
800 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list)) | |
801 | return NULL; | |
802 | ||
803 | page = head = cpu_buffer->head_page; | |
804 | /* | |
805 | * It is possible that the writer moves the header behind | |
806 | * where we started, and we miss in one loop. | |
807 | * A second loop should grab the header, but we'll do | |
808 | * three loops just because I'm paranoid. | |
809 | */ | |
810 | for (i = 0; i < 3; i++) { | |
811 | do { | |
812 | if (rb_is_head_page(cpu_buffer, page, page->list.prev)) { | |
813 | cpu_buffer->head_page = page; | |
814 | return page; | |
815 | } | |
816 | rb_inc_page(cpu_buffer, &page); | |
817 | } while (page != head); | |
818 | } | |
819 | ||
820 | RB_WARN_ON(cpu_buffer, 1); | |
821 | ||
822 | return NULL; | |
823 | } | |
824 | ||
825 | static int rb_head_page_replace(struct buffer_page *old, | |
826 | struct buffer_page *new) | |
827 | { | |
828 | unsigned long *ptr = (unsigned long *)&old->list.prev->next; | |
829 | unsigned long val; | |
830 | unsigned long ret; | |
831 | ||
832 | val = *ptr & ~RB_FLAG_MASK; | |
833 | val |= RB_PAGE_HEAD; | |
834 | ||
08a40816 | 835 | ret = cmpxchg(ptr, val, (unsigned long)&new->list); |
77ae365e SR |
836 | |
837 | return ret == val; | |
838 | } | |
839 | ||
840 | /* | |
841 | * rb_tail_page_update - move the tail page forward | |
842 | * | |
843 | * Returns 1 if moved tail page, 0 if someone else did. | |
844 | */ | |
845 | static int rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer, | |
846 | struct buffer_page *tail_page, | |
847 | struct buffer_page *next_page) | |
848 | { | |
849 | struct buffer_page *old_tail; | |
850 | unsigned long old_entries; | |
851 | unsigned long old_write; | |
852 | int ret = 0; | |
853 | ||
854 | /* | |
855 | * The tail page now needs to be moved forward. | |
856 | * | |
857 | * We need to reset the tail page, but without messing | |
858 | * with possible erasing of data brought in by interrupts | |
859 | * that have moved the tail page and are currently on it. | |
860 | * | |
861 | * We add a counter to the write field to denote this. | |
862 | */ | |
863 | old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write); | |
864 | old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries); | |
865 | ||
866 | /* | |
867 | * Just make sure we have seen our old_write and synchronize | |
868 | * with any interrupts that come in. | |
869 | */ | |
870 | barrier(); | |
871 | ||
872 | /* | |
873 | * If the tail page is still the same as what we think | |
874 | * it is, then it is up to us to update the tail | |
875 | * pointer. | |
876 | */ | |
877 | if (tail_page == cpu_buffer->tail_page) { | |
878 | /* Zero the write counter */ | |
879 | unsigned long val = old_write & ~RB_WRITE_MASK; | |
880 | unsigned long eval = old_entries & ~RB_WRITE_MASK; | |
881 | ||
882 | /* | |
883 | * This will only succeed if an interrupt did | |
884 | * not come in and change it. In which case, we | |
885 | * do not want to modify it. | |
da706d8b LJ |
886 | * |
887 | * We add (void) to let the compiler know that we do not care | |
888 | * about the return value of these functions. We use the | |
889 | * cmpxchg to only update if an interrupt did not already | |
890 | * do it for us. If the cmpxchg fails, we don't care. | |
77ae365e | 891 | */ |
da706d8b LJ |
892 | (void)local_cmpxchg(&next_page->write, old_write, val); |
893 | (void)local_cmpxchg(&next_page->entries, old_entries, eval); | |
77ae365e SR |
894 | |
895 | /* | |
896 | * No need to worry about races with clearing out the commit. | |
897 | * it only can increment when a commit takes place. But that | |
898 | * only happens in the outer most nested commit. | |
899 | */ | |
900 | local_set(&next_page->page->commit, 0); | |
901 | ||
902 | old_tail = cmpxchg(&cpu_buffer->tail_page, | |
903 | tail_page, next_page); | |
904 | ||
905 | if (old_tail == tail_page) | |
906 | ret = 1; | |
907 | } | |
908 | ||
909 | return ret; | |
910 | } | |
911 | ||
912 | static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer, | |
913 | struct buffer_page *bpage) | |
914 | { | |
915 | unsigned long val = (unsigned long)bpage; | |
916 | ||
917 | if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK)) | |
918 | return 1; | |
919 | ||
920 | return 0; | |
921 | } | |
922 | ||
923 | /** | |
924 | * rb_check_list - make sure a pointer to a list has the last bits zero | |
925 | */ | |
926 | static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer, | |
927 | struct list_head *list) | |
928 | { | |
929 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev)) | |
930 | return 1; | |
931 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next)) | |
932 | return 1; | |
933 | return 0; | |
934 | } | |
935 | ||
7a8e76a3 SR |
936 | /** |
937 | * check_pages - integrity check of buffer pages | |
938 | * @cpu_buffer: CPU buffer with pages to test | |
939 | * | |
c3706f00 | 940 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
941 | * been corrupted. |
942 | */ | |
943 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
944 | { | |
3adc54fa | 945 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 946 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 947 | |
77ae365e SR |
948 | rb_head_page_deactivate(cpu_buffer); |
949 | ||
3e89c7bb SR |
950 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
951 | return -1; | |
952 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
953 | return -1; | |
7a8e76a3 | 954 | |
77ae365e SR |
955 | if (rb_check_list(cpu_buffer, head)) |
956 | return -1; | |
957 | ||
044fa782 | 958 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 959 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 960 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
961 | return -1; |
962 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 963 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 964 | return -1; |
77ae365e SR |
965 | if (rb_check_list(cpu_buffer, &bpage->list)) |
966 | return -1; | |
7a8e76a3 SR |
967 | } |
968 | ||
77ae365e SR |
969 | rb_head_page_activate(cpu_buffer); |
970 | ||
7a8e76a3 SR |
971 | return 0; |
972 | } | |
973 | ||
438ced17 | 974 | static int __rb_allocate_pages(int nr_pages, struct list_head *pages, int cpu) |
7a8e76a3 | 975 | { |
438ced17 | 976 | int i; |
044fa782 | 977 | struct buffer_page *bpage, *tmp; |
3adc54fa | 978 | |
7a8e76a3 | 979 | for (i = 0; i < nr_pages; i++) { |
7ea59064 | 980 | struct page *page; |
d7ec4bfe VN |
981 | /* |
982 | * __GFP_NORETRY flag makes sure that the allocation fails | |
983 | * gracefully without invoking oom-killer and the system is | |
984 | * not destabilized. | |
985 | */ | |
044fa782 | 986 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
d7ec4bfe | 987 | GFP_KERNEL | __GFP_NORETRY, |
438ced17 | 988 | cpu_to_node(cpu)); |
044fa782 | 989 | if (!bpage) |
e4c2ce82 | 990 | goto free_pages; |
77ae365e | 991 | |
438ced17 | 992 | list_add(&bpage->list, pages); |
77ae365e | 993 | |
438ced17 | 994 | page = alloc_pages_node(cpu_to_node(cpu), |
d7ec4bfe | 995 | GFP_KERNEL | __GFP_NORETRY, 0); |
7ea59064 | 996 | if (!page) |
7a8e76a3 | 997 | goto free_pages; |
7ea59064 | 998 | bpage->page = page_address(page); |
044fa782 | 999 | rb_init_page(bpage->page); |
7a8e76a3 SR |
1000 | } |
1001 | ||
438ced17 VN |
1002 | return 0; |
1003 | ||
1004 | free_pages: | |
1005 | list_for_each_entry_safe(bpage, tmp, pages, list) { | |
1006 | list_del_init(&bpage->list); | |
1007 | free_buffer_page(bpage); | |
1008 | } | |
1009 | ||
1010 | return -ENOMEM; | |
1011 | } | |
1012 | ||
1013 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
1014 | unsigned nr_pages) | |
1015 | { | |
1016 | LIST_HEAD(pages); | |
1017 | ||
1018 | WARN_ON(!nr_pages); | |
1019 | ||
1020 | if (__rb_allocate_pages(nr_pages, &pages, cpu_buffer->cpu)) | |
1021 | return -ENOMEM; | |
1022 | ||
3adc54fa SR |
1023 | /* |
1024 | * The ring buffer page list is a circular list that does not | |
1025 | * start and end with a list head. All page list items point to | |
1026 | * other pages. | |
1027 | */ | |
1028 | cpu_buffer->pages = pages.next; | |
1029 | list_del(&pages); | |
7a8e76a3 | 1030 | |
438ced17 VN |
1031 | cpu_buffer->nr_pages = nr_pages; |
1032 | ||
7a8e76a3 SR |
1033 | rb_check_pages(cpu_buffer); |
1034 | ||
1035 | return 0; | |
7a8e76a3 SR |
1036 | } |
1037 | ||
1038 | static struct ring_buffer_per_cpu * | |
438ced17 | 1039 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int nr_pages, int cpu) |
7a8e76a3 SR |
1040 | { |
1041 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 1042 | struct buffer_page *bpage; |
7ea59064 | 1043 | struct page *page; |
7a8e76a3 SR |
1044 | int ret; |
1045 | ||
1046 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
1047 | GFP_KERNEL, cpu_to_node(cpu)); | |
1048 | if (!cpu_buffer) | |
1049 | return NULL; | |
1050 | ||
1051 | cpu_buffer->cpu = cpu; | |
1052 | cpu_buffer->buffer = buffer; | |
5389f6fa | 1053 | raw_spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 1054 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
edc35bd7 | 1055 | cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
83f40318 VN |
1056 | INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler); |
1057 | init_completion(&cpu_buffer->update_completion); | |
7a8e76a3 | 1058 | |
044fa782 | 1059 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 1060 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 1061 | if (!bpage) |
e4c2ce82 SR |
1062 | goto fail_free_buffer; |
1063 | ||
77ae365e SR |
1064 | rb_check_bpage(cpu_buffer, bpage); |
1065 | ||
044fa782 | 1066 | cpu_buffer->reader_page = bpage; |
7ea59064 VN |
1067 | page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0); |
1068 | if (!page) | |
e4c2ce82 | 1069 | goto fail_free_reader; |
7ea59064 | 1070 | bpage->page = page_address(page); |
044fa782 | 1071 | rb_init_page(bpage->page); |
e4c2ce82 | 1072 | |
d769041f | 1073 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 1074 | |
438ced17 | 1075 | ret = rb_allocate_pages(cpu_buffer, nr_pages); |
7a8e76a3 | 1076 | if (ret < 0) |
d769041f | 1077 | goto fail_free_reader; |
7a8e76a3 SR |
1078 | |
1079 | cpu_buffer->head_page | |
3adc54fa | 1080 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 1081 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 | 1082 | |
77ae365e SR |
1083 | rb_head_page_activate(cpu_buffer); |
1084 | ||
7a8e76a3 SR |
1085 | return cpu_buffer; |
1086 | ||
d769041f SR |
1087 | fail_free_reader: |
1088 | free_buffer_page(cpu_buffer->reader_page); | |
1089 | ||
7a8e76a3 SR |
1090 | fail_free_buffer: |
1091 | kfree(cpu_buffer); | |
1092 | return NULL; | |
1093 | } | |
1094 | ||
1095 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
1096 | { | |
3adc54fa | 1097 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1098 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1099 | |
d769041f SR |
1100 | free_buffer_page(cpu_buffer->reader_page); |
1101 | ||
77ae365e SR |
1102 | rb_head_page_deactivate(cpu_buffer); |
1103 | ||
3adc54fa SR |
1104 | if (head) { |
1105 | list_for_each_entry_safe(bpage, tmp, head, list) { | |
1106 | list_del_init(&bpage->list); | |
1107 | free_buffer_page(bpage); | |
1108 | } | |
1109 | bpage = list_entry(head, struct buffer_page, list); | |
044fa782 | 1110 | free_buffer_page(bpage); |
7a8e76a3 | 1111 | } |
3adc54fa | 1112 | |
7a8e76a3 SR |
1113 | kfree(cpu_buffer); |
1114 | } | |
1115 | ||
59222efe | 1116 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
1117 | static int rb_cpu_notify(struct notifier_block *self, |
1118 | unsigned long action, void *hcpu); | |
554f786e SR |
1119 | #endif |
1120 | ||
7a8e76a3 SR |
1121 | /** |
1122 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 1123 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
1124 | * @flags: attributes to set for the ring buffer. |
1125 | * | |
1126 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
1127 | * flag. This flag means that the buffer will overwrite old data | |
1128 | * when the buffer wraps. If this flag is not set, the buffer will | |
1129 | * drop data when the tail hits the head. | |
1130 | */ | |
1f8a6a10 PZ |
1131 | struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
1132 | struct lock_class_key *key) | |
7a8e76a3 SR |
1133 | { |
1134 | struct ring_buffer *buffer; | |
1135 | int bsize; | |
438ced17 | 1136 | int cpu, nr_pages; |
7a8e76a3 SR |
1137 | |
1138 | /* keep it in its own cache line */ | |
1139 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
1140 | GFP_KERNEL); | |
1141 | if (!buffer) | |
1142 | return NULL; | |
1143 | ||
9e01c1b7 RR |
1144 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
1145 | goto fail_free_buffer; | |
1146 | ||
438ced17 | 1147 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
7a8e76a3 | 1148 | buffer->flags = flags; |
37886f6a | 1149 | buffer->clock = trace_clock_local; |
1f8a6a10 | 1150 | buffer->reader_lock_key = key; |
7a8e76a3 SR |
1151 | |
1152 | /* need at least two pages */ | |
438ced17 VN |
1153 | if (nr_pages < 2) |
1154 | nr_pages = 2; | |
7a8e76a3 | 1155 | |
3bf832ce FW |
1156 | /* |
1157 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
1158 | * in early initcall, it will not be notified of secondary cpus. | |
1159 | * In that off case, we need to allocate for all possible cpus. | |
1160 | */ | |
1161 | #ifdef CONFIG_HOTPLUG_CPU | |
554f786e SR |
1162 | get_online_cpus(); |
1163 | cpumask_copy(buffer->cpumask, cpu_online_mask); | |
3bf832ce FW |
1164 | #else |
1165 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
1166 | #endif | |
7a8e76a3 SR |
1167 | buffer->cpus = nr_cpu_ids; |
1168 | ||
1169 | bsize = sizeof(void *) * nr_cpu_ids; | |
1170 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
1171 | GFP_KERNEL); | |
1172 | if (!buffer->buffers) | |
9e01c1b7 | 1173 | goto fail_free_cpumask; |
7a8e76a3 SR |
1174 | |
1175 | for_each_buffer_cpu(buffer, cpu) { | |
1176 | buffer->buffers[cpu] = | |
438ced17 | 1177 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
7a8e76a3 SR |
1178 | if (!buffer->buffers[cpu]) |
1179 | goto fail_free_buffers; | |
1180 | } | |
1181 | ||
59222efe | 1182 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1183 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
1184 | buffer->cpu_notify.priority = 0; | |
1185 | register_cpu_notifier(&buffer->cpu_notify); | |
1186 | #endif | |
1187 | ||
1188 | put_online_cpus(); | |
7a8e76a3 SR |
1189 | mutex_init(&buffer->mutex); |
1190 | ||
1191 | return buffer; | |
1192 | ||
1193 | fail_free_buffers: | |
1194 | for_each_buffer_cpu(buffer, cpu) { | |
1195 | if (buffer->buffers[cpu]) | |
1196 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1197 | } | |
1198 | kfree(buffer->buffers); | |
1199 | ||
9e01c1b7 RR |
1200 | fail_free_cpumask: |
1201 | free_cpumask_var(buffer->cpumask); | |
554f786e | 1202 | put_online_cpus(); |
9e01c1b7 | 1203 | |
7a8e76a3 SR |
1204 | fail_free_buffer: |
1205 | kfree(buffer); | |
1206 | return NULL; | |
1207 | } | |
1f8a6a10 | 1208 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
1209 | |
1210 | /** | |
1211 | * ring_buffer_free - free a ring buffer. | |
1212 | * @buffer: the buffer to free. | |
1213 | */ | |
1214 | void | |
1215 | ring_buffer_free(struct ring_buffer *buffer) | |
1216 | { | |
1217 | int cpu; | |
1218 | ||
554f786e SR |
1219 | get_online_cpus(); |
1220 | ||
59222efe | 1221 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1222 | unregister_cpu_notifier(&buffer->cpu_notify); |
1223 | #endif | |
1224 | ||
7a8e76a3 SR |
1225 | for_each_buffer_cpu(buffer, cpu) |
1226 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1227 | ||
554f786e SR |
1228 | put_online_cpus(); |
1229 | ||
bd3f0221 | 1230 | kfree(buffer->buffers); |
9e01c1b7 RR |
1231 | free_cpumask_var(buffer->cpumask); |
1232 | ||
7a8e76a3 SR |
1233 | kfree(buffer); |
1234 | } | |
c4f50183 | 1235 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 1236 | |
37886f6a SR |
1237 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
1238 | u64 (*clock)(void)) | |
1239 | { | |
1240 | buffer->clock = clock; | |
1241 | } | |
1242 | ||
7a8e76a3 SR |
1243 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
1244 | ||
83f40318 VN |
1245 | static inline unsigned long rb_page_entries(struct buffer_page *bpage) |
1246 | { | |
1247 | return local_read(&bpage->entries) & RB_WRITE_MASK; | |
1248 | } | |
1249 | ||
1250 | static inline unsigned long rb_page_write(struct buffer_page *bpage) | |
1251 | { | |
1252 | return local_read(&bpage->write) & RB_WRITE_MASK; | |
1253 | } | |
1254 | ||
7a8e76a3 | 1255 | static void |
83f40318 | 1256 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned int nr_pages) |
7a8e76a3 | 1257 | { |
83f40318 VN |
1258 | struct list_head *tail_page, *to_remove, *next_page; |
1259 | struct buffer_page *to_remove_page, *tmp_iter_page; | |
1260 | struct buffer_page *last_page, *first_page; | |
1261 | unsigned int nr_removed; | |
1262 | unsigned long head_bit; | |
1263 | int page_entries; | |
1264 | ||
1265 | head_bit = 0; | |
7a8e76a3 | 1266 | |
5389f6fa | 1267 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1268 | atomic_inc(&cpu_buffer->record_disabled); |
1269 | /* | |
1270 | * We don't race with the readers since we have acquired the reader | |
1271 | * lock. We also don't race with writers after disabling recording. | |
1272 | * This makes it easy to figure out the first and the last page to be | |
1273 | * removed from the list. We unlink all the pages in between including | |
1274 | * the first and last pages. This is done in a busy loop so that we | |
1275 | * lose the least number of traces. | |
1276 | * The pages are freed after we restart recording and unlock readers. | |
1277 | */ | |
1278 | tail_page = &cpu_buffer->tail_page->list; | |
77ae365e | 1279 | |
83f40318 VN |
1280 | /* |
1281 | * tail page might be on reader page, we remove the next page | |
1282 | * from the ring buffer | |
1283 | */ | |
1284 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
1285 | tail_page = rb_list_head(tail_page->next); | |
1286 | to_remove = tail_page; | |
1287 | ||
1288 | /* start of pages to remove */ | |
1289 | first_page = list_entry(rb_list_head(to_remove->next), | |
1290 | struct buffer_page, list); | |
1291 | ||
1292 | for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) { | |
1293 | to_remove = rb_list_head(to_remove)->next; | |
1294 | head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD; | |
7a8e76a3 | 1295 | } |
7a8e76a3 | 1296 | |
83f40318 | 1297 | next_page = rb_list_head(to_remove)->next; |
7a8e76a3 | 1298 | |
83f40318 VN |
1299 | /* |
1300 | * Now we remove all pages between tail_page and next_page. | |
1301 | * Make sure that we have head_bit value preserved for the | |
1302 | * next page | |
1303 | */ | |
1304 | tail_page->next = (struct list_head *)((unsigned long)next_page | | |
1305 | head_bit); | |
1306 | next_page = rb_list_head(next_page); | |
1307 | next_page->prev = tail_page; | |
1308 | ||
1309 | /* make sure pages points to a valid page in the ring buffer */ | |
1310 | cpu_buffer->pages = next_page; | |
1311 | ||
1312 | /* update head page */ | |
1313 | if (head_bit) | |
1314 | cpu_buffer->head_page = list_entry(next_page, | |
1315 | struct buffer_page, list); | |
1316 | ||
1317 | /* | |
1318 | * change read pointer to make sure any read iterators reset | |
1319 | * themselves | |
1320 | */ | |
1321 | cpu_buffer->read = 0; | |
1322 | ||
1323 | /* pages are removed, resume tracing and then free the pages */ | |
1324 | atomic_dec(&cpu_buffer->record_disabled); | |
5389f6fa | 1325 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1326 | |
1327 | RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)); | |
1328 | ||
1329 | /* last buffer page to remove */ | |
1330 | last_page = list_entry(rb_list_head(to_remove), struct buffer_page, | |
1331 | list); | |
1332 | tmp_iter_page = first_page; | |
1333 | ||
1334 | do { | |
1335 | to_remove_page = tmp_iter_page; | |
1336 | rb_inc_page(cpu_buffer, &tmp_iter_page); | |
1337 | ||
1338 | /* update the counters */ | |
1339 | page_entries = rb_page_entries(to_remove_page); | |
1340 | if (page_entries) { | |
1341 | /* | |
1342 | * If something was added to this page, it was full | |
1343 | * since it is not the tail page. So we deduct the | |
1344 | * bytes consumed in ring buffer from here. | |
1345 | * No need to update overruns, since this page is | |
1346 | * deleted from ring buffer and its entries are | |
1347 | * already accounted for. | |
1348 | */ | |
1349 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); | |
1350 | } | |
1351 | ||
1352 | /* | |
1353 | * We have already removed references to this list item, just | |
1354 | * free up the buffer_page and its page | |
1355 | */ | |
1356 | free_buffer_page(to_remove_page); | |
1357 | nr_removed--; | |
1358 | ||
1359 | } while (to_remove_page != last_page); | |
1360 | ||
1361 | RB_WARN_ON(cpu_buffer, nr_removed); | |
7a8e76a3 SR |
1362 | } |
1363 | ||
1364 | static void | |
1365 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
1366 | struct list_head *pages, unsigned nr_pages) | |
1367 | { | |
044fa782 | 1368 | struct buffer_page *bpage; |
7a8e76a3 SR |
1369 | struct list_head *p; |
1370 | unsigned i; | |
1371 | ||
5389f6fa | 1372 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1373 | /* stop the writers while inserting pages */ |
1374 | atomic_inc(&cpu_buffer->record_disabled); | |
77ae365e SR |
1375 | rb_head_page_deactivate(cpu_buffer); |
1376 | ||
7a8e76a3 | 1377 | for (i = 0; i < nr_pages; i++) { |
3e89c7bb | 1378 | if (RB_WARN_ON(cpu_buffer, list_empty(pages))) |
292f60c0 | 1379 | goto out; |
7a8e76a3 | 1380 | p = pages->next; |
044fa782 SR |
1381 | bpage = list_entry(p, struct buffer_page, list); |
1382 | list_del_init(&bpage->list); | |
3adc54fa | 1383 | list_add_tail(&bpage->list, cpu_buffer->pages); |
7a8e76a3 SR |
1384 | } |
1385 | rb_reset_cpu(cpu_buffer); | |
7a8e76a3 SR |
1386 | rb_check_pages(cpu_buffer); |
1387 | ||
292f60c0 | 1388 | out: |
83f40318 | 1389 | atomic_dec(&cpu_buffer->record_disabled); |
5389f6fa | 1390 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
7a8e76a3 SR |
1391 | } |
1392 | ||
83f40318 | 1393 | static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer) |
438ced17 VN |
1394 | { |
1395 | if (cpu_buffer->nr_pages_to_update > 0) | |
1396 | rb_insert_pages(cpu_buffer, &cpu_buffer->new_pages, | |
1397 | cpu_buffer->nr_pages_to_update); | |
1398 | else | |
1399 | rb_remove_pages(cpu_buffer, -cpu_buffer->nr_pages_to_update); | |
83f40318 | 1400 | |
438ced17 | 1401 | cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update; |
83f40318 VN |
1402 | } |
1403 | ||
1404 | static void update_pages_handler(struct work_struct *work) | |
1405 | { | |
1406 | struct ring_buffer_per_cpu *cpu_buffer = container_of(work, | |
1407 | struct ring_buffer_per_cpu, update_pages_work); | |
1408 | rb_update_pages(cpu_buffer); | |
1409 | complete(&cpu_buffer->update_completion); | |
438ced17 VN |
1410 | } |
1411 | ||
7a8e76a3 SR |
1412 | /** |
1413 | * ring_buffer_resize - resize the ring buffer | |
1414 | * @buffer: the buffer to resize. | |
1415 | * @size: the new size. | |
1416 | * | |
7a8e76a3 SR |
1417 | * Minimum size is 2 * BUF_PAGE_SIZE. |
1418 | * | |
83f40318 | 1419 | * Returns 0 on success and < 0 on failure. |
7a8e76a3 | 1420 | */ |
438ced17 VN |
1421 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size, |
1422 | int cpu_id) | |
7a8e76a3 SR |
1423 | { |
1424 | struct ring_buffer_per_cpu *cpu_buffer; | |
438ced17 | 1425 | unsigned nr_pages; |
83f40318 | 1426 | int cpu, err = 0; |
7a8e76a3 | 1427 | |
ee51a1de IM |
1428 | /* |
1429 | * Always succeed at resizing a non-existent buffer: | |
1430 | */ | |
1431 | if (!buffer) | |
1432 | return size; | |
1433 | ||
7a8e76a3 SR |
1434 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
1435 | size *= BUF_PAGE_SIZE; | |
7a8e76a3 SR |
1436 | |
1437 | /* we need a minimum of two pages */ | |
1438 | if (size < BUF_PAGE_SIZE * 2) | |
1439 | size = BUF_PAGE_SIZE * 2; | |
1440 | ||
83f40318 | 1441 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
18421015 | 1442 | |
83f40318 VN |
1443 | /* |
1444 | * Don't succeed if resizing is disabled, as a reader might be | |
1445 | * manipulating the ring buffer and is expecting a sane state while | |
1446 | * this is true. | |
1447 | */ | |
1448 | if (atomic_read(&buffer->resize_disabled)) | |
1449 | return -EBUSY; | |
18421015 | 1450 | |
83f40318 | 1451 | /* prevent another thread from changing buffer sizes */ |
7a8e76a3 | 1452 | mutex_lock(&buffer->mutex); |
7a8e76a3 | 1453 | |
438ced17 VN |
1454 | if (cpu_id == RING_BUFFER_ALL_CPUS) { |
1455 | /* calculate the pages to update */ | |
7a8e76a3 SR |
1456 | for_each_buffer_cpu(buffer, cpu) { |
1457 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 1458 | |
438ced17 VN |
1459 | cpu_buffer->nr_pages_to_update = nr_pages - |
1460 | cpu_buffer->nr_pages; | |
438ced17 VN |
1461 | /* |
1462 | * nothing more to do for removing pages or no update | |
1463 | */ | |
1464 | if (cpu_buffer->nr_pages_to_update <= 0) | |
1465 | continue; | |
d7ec4bfe | 1466 | /* |
438ced17 VN |
1467 | * to add pages, make sure all new pages can be |
1468 | * allocated without receiving ENOMEM | |
d7ec4bfe | 1469 | */ |
438ced17 VN |
1470 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
1471 | if (__rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 | 1472 | &cpu_buffer->new_pages, cpu)) { |
438ced17 | 1473 | /* not enough memory for new pages */ |
83f40318 VN |
1474 | err = -ENOMEM; |
1475 | goto out_err; | |
1476 | } | |
1477 | } | |
1478 | ||
1479 | get_online_cpus(); | |
1480 | /* | |
1481 | * Fire off all the required work handlers | |
1482 | * Look out for offline CPUs | |
1483 | */ | |
1484 | for_each_buffer_cpu(buffer, cpu) { | |
1485 | cpu_buffer = buffer->buffers[cpu]; | |
1486 | if (!cpu_buffer->nr_pages_to_update || | |
1487 | !cpu_online(cpu)) | |
1488 | continue; | |
1489 | ||
1490 | schedule_work_on(cpu, &cpu_buffer->update_pages_work); | |
1491 | } | |
1492 | /* | |
1493 | * This loop is for the CPUs that are not online. | |
1494 | * We can't schedule anything on them, but it's not necessary | |
1495 | * since we can change their buffer sizes without any race. | |
1496 | */ | |
1497 | for_each_buffer_cpu(buffer, cpu) { | |
1498 | cpu_buffer = buffer->buffers[cpu]; | |
1499 | if (!cpu_buffer->nr_pages_to_update || | |
1500 | cpu_online(cpu)) | |
1501 | continue; | |
1502 | ||
1503 | rb_update_pages(cpu_buffer); | |
7a8e76a3 | 1504 | } |
7a8e76a3 | 1505 | |
438ced17 VN |
1506 | /* wait for all the updates to complete */ |
1507 | for_each_buffer_cpu(buffer, cpu) { | |
1508 | cpu_buffer = buffer->buffers[cpu]; | |
83f40318 VN |
1509 | if (!cpu_buffer->nr_pages_to_update || |
1510 | !cpu_online(cpu)) | |
1511 | continue; | |
1512 | ||
1513 | wait_for_completion(&cpu_buffer->update_completion); | |
1514 | /* reset this value */ | |
1515 | cpu_buffer->nr_pages_to_update = 0; | |
438ced17 | 1516 | } |
83f40318 VN |
1517 | |
1518 | put_online_cpus(); | |
438ced17 VN |
1519 | } else { |
1520 | cpu_buffer = buffer->buffers[cpu_id]; | |
83f40318 | 1521 | |
438ced17 VN |
1522 | if (nr_pages == cpu_buffer->nr_pages) |
1523 | goto out; | |
7a8e76a3 | 1524 | |
438ced17 VN |
1525 | cpu_buffer->nr_pages_to_update = nr_pages - |
1526 | cpu_buffer->nr_pages; | |
1527 | ||
1528 | INIT_LIST_HEAD(&cpu_buffer->new_pages); | |
1529 | if (cpu_buffer->nr_pages_to_update > 0 && | |
1530 | __rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 VN |
1531 | &cpu_buffer->new_pages, cpu_id)) { |
1532 | err = -ENOMEM; | |
1533 | goto out_err; | |
1534 | } | |
438ced17 | 1535 | |
83f40318 VN |
1536 | get_online_cpus(); |
1537 | ||
1538 | if (cpu_online(cpu_id)) { | |
1539 | schedule_work_on(cpu_id, | |
1540 | &cpu_buffer->update_pages_work); | |
1541 | wait_for_completion(&cpu_buffer->update_completion); | |
1542 | } else | |
1543 | rb_update_pages(cpu_buffer); | |
1544 | ||
1545 | put_online_cpus(); | |
1546 | /* reset this value */ | |
1547 | cpu_buffer->nr_pages_to_update = 0; | |
438ced17 | 1548 | } |
7a8e76a3 SR |
1549 | |
1550 | out: | |
7a8e76a3 | 1551 | mutex_unlock(&buffer->mutex); |
7a8e76a3 SR |
1552 | return size; |
1553 | ||
83f40318 | 1554 | out_err: |
438ced17 VN |
1555 | for_each_buffer_cpu(buffer, cpu) { |
1556 | struct buffer_page *bpage, *tmp; | |
83f40318 | 1557 | |
438ced17 | 1558 | cpu_buffer = buffer->buffers[cpu]; |
438ced17 | 1559 | cpu_buffer->nr_pages_to_update = 0; |
83f40318 | 1560 | |
438ced17 VN |
1561 | if (list_empty(&cpu_buffer->new_pages)) |
1562 | continue; | |
83f40318 | 1563 | |
438ced17 VN |
1564 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, |
1565 | list) { | |
1566 | list_del_init(&bpage->list); | |
1567 | free_buffer_page(bpage); | |
1568 | } | |
7a8e76a3 | 1569 | } |
641d2f63 | 1570 | mutex_unlock(&buffer->mutex); |
83f40318 | 1571 | return err; |
7a8e76a3 | 1572 | } |
c4f50183 | 1573 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 1574 | |
750912fa DS |
1575 | void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val) |
1576 | { | |
1577 | mutex_lock(&buffer->mutex); | |
1578 | if (val) | |
1579 | buffer->flags |= RB_FL_OVERWRITE; | |
1580 | else | |
1581 | buffer->flags &= ~RB_FL_OVERWRITE; | |
1582 | mutex_unlock(&buffer->mutex); | |
1583 | } | |
1584 | EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite); | |
1585 | ||
8789a9e7 | 1586 | static inline void * |
044fa782 | 1587 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 1588 | { |
044fa782 | 1589 | return bpage->data + index; |
8789a9e7 SR |
1590 | } |
1591 | ||
044fa782 | 1592 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 1593 | { |
044fa782 | 1594 | return bpage->page->data + index; |
7a8e76a3 SR |
1595 | } |
1596 | ||
1597 | static inline struct ring_buffer_event * | |
d769041f | 1598 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1599 | { |
6f807acd SR |
1600 | return __rb_page_index(cpu_buffer->reader_page, |
1601 | cpu_buffer->reader_page->read); | |
1602 | } | |
1603 | ||
7a8e76a3 SR |
1604 | static inline struct ring_buffer_event * |
1605 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
1606 | { | |
6f807acd | 1607 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
1608 | } |
1609 | ||
bf41a158 SR |
1610 | static inline unsigned rb_page_commit(struct buffer_page *bpage) |
1611 | { | |
abc9b56d | 1612 | return local_read(&bpage->page->commit); |
bf41a158 SR |
1613 | } |
1614 | ||
25985edc | 1615 | /* Size is determined by what has been committed */ |
bf41a158 SR |
1616 | static inline unsigned rb_page_size(struct buffer_page *bpage) |
1617 | { | |
1618 | return rb_page_commit(bpage); | |
1619 | } | |
1620 | ||
1621 | static inline unsigned | |
1622 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
1623 | { | |
1624 | return rb_page_commit(cpu_buffer->commit_page); | |
1625 | } | |
1626 | ||
bf41a158 SR |
1627 | static inline unsigned |
1628 | rb_event_index(struct ring_buffer_event *event) | |
1629 | { | |
1630 | unsigned long addr = (unsigned long)event; | |
1631 | ||
22f470f8 | 1632 | return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE; |
bf41a158 SR |
1633 | } |
1634 | ||
0f0c85fc | 1635 | static inline int |
fa743953 SR |
1636 | rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1637 | struct ring_buffer_event *event) | |
bf41a158 SR |
1638 | { |
1639 | unsigned long addr = (unsigned long)event; | |
1640 | unsigned long index; | |
1641 | ||
1642 | index = rb_event_index(event); | |
1643 | addr &= PAGE_MASK; | |
1644 | ||
1645 | return cpu_buffer->commit_page->page == (void *)addr && | |
1646 | rb_commit_index(cpu_buffer) == index; | |
1647 | } | |
1648 | ||
34a148bf | 1649 | static void |
bf41a158 | 1650 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1651 | { |
77ae365e SR |
1652 | unsigned long max_count; |
1653 | ||
bf41a158 SR |
1654 | /* |
1655 | * We only race with interrupts and NMIs on this CPU. | |
1656 | * If we own the commit event, then we can commit | |
1657 | * all others that interrupted us, since the interruptions | |
1658 | * are in stack format (they finish before they come | |
1659 | * back to us). This allows us to do a simple loop to | |
1660 | * assign the commit to the tail. | |
1661 | */ | |
a8ccf1d6 | 1662 | again: |
438ced17 | 1663 | max_count = cpu_buffer->nr_pages * 100; |
77ae365e | 1664 | |
bf41a158 | 1665 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
77ae365e SR |
1666 | if (RB_WARN_ON(cpu_buffer, !(--max_count))) |
1667 | return; | |
1668 | if (RB_WARN_ON(cpu_buffer, | |
1669 | rb_is_reader_page(cpu_buffer->tail_page))) | |
1670 | return; | |
1671 | local_set(&cpu_buffer->commit_page->page->commit, | |
1672 | rb_page_write(cpu_buffer->commit_page)); | |
bf41a158 | 1673 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); |
abc9b56d SR |
1674 | cpu_buffer->write_stamp = |
1675 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1676 | /* add barrier to keep gcc from optimizing too much */ |
1677 | barrier(); | |
1678 | } | |
1679 | while (rb_commit_index(cpu_buffer) != | |
1680 | rb_page_write(cpu_buffer->commit_page)) { | |
77ae365e SR |
1681 | |
1682 | local_set(&cpu_buffer->commit_page->page->commit, | |
1683 | rb_page_write(cpu_buffer->commit_page)); | |
1684 | RB_WARN_ON(cpu_buffer, | |
1685 | local_read(&cpu_buffer->commit_page->page->commit) & | |
1686 | ~RB_WRITE_MASK); | |
bf41a158 SR |
1687 | barrier(); |
1688 | } | |
a8ccf1d6 SR |
1689 | |
1690 | /* again, keep gcc from optimizing */ | |
1691 | barrier(); | |
1692 | ||
1693 | /* | |
1694 | * If an interrupt came in just after the first while loop | |
1695 | * and pushed the tail page forward, we will be left with | |
1696 | * a dangling commit that will never go forward. | |
1697 | */ | |
1698 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1699 | goto again; | |
7a8e76a3 SR |
1700 | } |
1701 | ||
d769041f | 1702 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1703 | { |
abc9b56d | 1704 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1705 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1706 | } |
1707 | ||
34a148bf | 1708 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1709 | { |
1710 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1711 | ||
1712 | /* | |
1713 | * The iterator could be on the reader page (it starts there). | |
1714 | * But the head could have moved, since the reader was | |
1715 | * found. Check for this case and assign the iterator | |
1716 | * to the head page instead of next. | |
1717 | */ | |
1718 | if (iter->head_page == cpu_buffer->reader_page) | |
77ae365e | 1719 | iter->head_page = rb_set_head_page(cpu_buffer); |
d769041f SR |
1720 | else |
1721 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1722 | ||
abc9b56d | 1723 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1724 | iter->head = 0; |
1725 | } | |
1726 | ||
69d1b839 SR |
1727 | /* Slow path, do not inline */ |
1728 | static noinline struct ring_buffer_event * | |
1729 | rb_add_time_stamp(struct ring_buffer_event *event, u64 delta) | |
1730 | { | |
1731 | event->type_len = RINGBUF_TYPE_TIME_EXTEND; | |
1732 | ||
1733 | /* Not the first event on the page? */ | |
1734 | if (rb_event_index(event)) { | |
1735 | event->time_delta = delta & TS_MASK; | |
1736 | event->array[0] = delta >> TS_SHIFT; | |
1737 | } else { | |
1738 | /* nope, just zero it */ | |
1739 | event->time_delta = 0; | |
1740 | event->array[0] = 0; | |
1741 | } | |
1742 | ||
1743 | return skip_time_extend(event); | |
1744 | } | |
1745 | ||
7a8e76a3 SR |
1746 | /** |
1747 | * ring_buffer_update_event - update event type and data | |
1748 | * @event: the even to update | |
1749 | * @type: the type of event | |
1750 | * @length: the size of the event field in the ring buffer | |
1751 | * | |
1752 | * Update the type and data fields of the event. The length | |
1753 | * is the actual size that is written to the ring buffer, | |
1754 | * and with this, we can determine what to place into the | |
1755 | * data field. | |
1756 | */ | |
34a148bf | 1757 | static void |
69d1b839 SR |
1758 | rb_update_event(struct ring_buffer_per_cpu *cpu_buffer, |
1759 | struct ring_buffer_event *event, unsigned length, | |
1760 | int add_timestamp, u64 delta) | |
7a8e76a3 | 1761 | { |
69d1b839 SR |
1762 | /* Only a commit updates the timestamp */ |
1763 | if (unlikely(!rb_event_is_commit(cpu_buffer, event))) | |
1764 | delta = 0; | |
7a8e76a3 | 1765 | |
69d1b839 SR |
1766 | /* |
1767 | * If we need to add a timestamp, then we | |
1768 | * add it to the start of the resevered space. | |
1769 | */ | |
1770 | if (unlikely(add_timestamp)) { | |
1771 | event = rb_add_time_stamp(event, delta); | |
1772 | length -= RB_LEN_TIME_EXTEND; | |
1773 | delta = 0; | |
7a8e76a3 | 1774 | } |
69d1b839 SR |
1775 | |
1776 | event->time_delta = delta; | |
1777 | length -= RB_EVNT_HDR_SIZE; | |
1778 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) { | |
1779 | event->type_len = 0; | |
1780 | event->array[0] = length; | |
1781 | } else | |
1782 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
7a8e76a3 SR |
1783 | } |
1784 | ||
77ae365e SR |
1785 | /* |
1786 | * rb_handle_head_page - writer hit the head page | |
1787 | * | |
1788 | * Returns: +1 to retry page | |
1789 | * 0 to continue | |
1790 | * -1 on error | |
1791 | */ | |
1792 | static int | |
1793 | rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
1794 | struct buffer_page *tail_page, | |
1795 | struct buffer_page *next_page) | |
1796 | { | |
1797 | struct buffer_page *new_head; | |
1798 | int entries; | |
1799 | int type; | |
1800 | int ret; | |
1801 | ||
1802 | entries = rb_page_entries(next_page); | |
1803 | ||
1804 | /* | |
1805 | * The hard part is here. We need to move the head | |
1806 | * forward, and protect against both readers on | |
1807 | * other CPUs and writers coming in via interrupts. | |
1808 | */ | |
1809 | type = rb_head_page_set_update(cpu_buffer, next_page, tail_page, | |
1810 | RB_PAGE_HEAD); | |
1811 | ||
1812 | /* | |
1813 | * type can be one of four: | |
1814 | * NORMAL - an interrupt already moved it for us | |
1815 | * HEAD - we are the first to get here. | |
1816 | * UPDATE - we are the interrupt interrupting | |
1817 | * a current move. | |
1818 | * MOVED - a reader on another CPU moved the next | |
1819 | * pointer to its reader page. Give up | |
1820 | * and try again. | |
1821 | */ | |
1822 | ||
1823 | switch (type) { | |
1824 | case RB_PAGE_HEAD: | |
1825 | /* | |
1826 | * We changed the head to UPDATE, thus | |
1827 | * it is our responsibility to update | |
1828 | * the counters. | |
1829 | */ | |
1830 | local_add(entries, &cpu_buffer->overrun); | |
c64e148a | 1831 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
77ae365e SR |
1832 | |
1833 | /* | |
1834 | * The entries will be zeroed out when we move the | |
1835 | * tail page. | |
1836 | */ | |
1837 | ||
1838 | /* still more to do */ | |
1839 | break; | |
1840 | ||
1841 | case RB_PAGE_UPDATE: | |
1842 | /* | |
1843 | * This is an interrupt that interrupt the | |
1844 | * previous update. Still more to do. | |
1845 | */ | |
1846 | break; | |
1847 | case RB_PAGE_NORMAL: | |
1848 | /* | |
1849 | * An interrupt came in before the update | |
1850 | * and processed this for us. | |
1851 | * Nothing left to do. | |
1852 | */ | |
1853 | return 1; | |
1854 | case RB_PAGE_MOVED: | |
1855 | /* | |
1856 | * The reader is on another CPU and just did | |
1857 | * a swap with our next_page. | |
1858 | * Try again. | |
1859 | */ | |
1860 | return 1; | |
1861 | default: | |
1862 | RB_WARN_ON(cpu_buffer, 1); /* WTF??? */ | |
1863 | return -1; | |
1864 | } | |
1865 | ||
1866 | /* | |
1867 | * Now that we are here, the old head pointer is | |
1868 | * set to UPDATE. This will keep the reader from | |
1869 | * swapping the head page with the reader page. | |
1870 | * The reader (on another CPU) will spin till | |
1871 | * we are finished. | |
1872 | * | |
1873 | * We just need to protect against interrupts | |
1874 | * doing the job. We will set the next pointer | |
1875 | * to HEAD. After that, we set the old pointer | |
1876 | * to NORMAL, but only if it was HEAD before. | |
1877 | * otherwise we are an interrupt, and only | |
1878 | * want the outer most commit to reset it. | |
1879 | */ | |
1880 | new_head = next_page; | |
1881 | rb_inc_page(cpu_buffer, &new_head); | |
1882 | ||
1883 | ret = rb_head_page_set_head(cpu_buffer, new_head, next_page, | |
1884 | RB_PAGE_NORMAL); | |
1885 | ||
1886 | /* | |
1887 | * Valid returns are: | |
1888 | * HEAD - an interrupt came in and already set it. | |
1889 | * NORMAL - One of two things: | |
1890 | * 1) We really set it. | |
1891 | * 2) A bunch of interrupts came in and moved | |
1892 | * the page forward again. | |
1893 | */ | |
1894 | switch (ret) { | |
1895 | case RB_PAGE_HEAD: | |
1896 | case RB_PAGE_NORMAL: | |
1897 | /* OK */ | |
1898 | break; | |
1899 | default: | |
1900 | RB_WARN_ON(cpu_buffer, 1); | |
1901 | return -1; | |
1902 | } | |
1903 | ||
1904 | /* | |
1905 | * It is possible that an interrupt came in, | |
1906 | * set the head up, then more interrupts came in | |
1907 | * and moved it again. When we get back here, | |
1908 | * the page would have been set to NORMAL but we | |
1909 | * just set it back to HEAD. | |
1910 | * | |
1911 | * How do you detect this? Well, if that happened | |
1912 | * the tail page would have moved. | |
1913 | */ | |
1914 | if (ret == RB_PAGE_NORMAL) { | |
1915 | /* | |
1916 | * If the tail had moved passed next, then we need | |
1917 | * to reset the pointer. | |
1918 | */ | |
1919 | if (cpu_buffer->tail_page != tail_page && | |
1920 | cpu_buffer->tail_page != next_page) | |
1921 | rb_head_page_set_normal(cpu_buffer, new_head, | |
1922 | next_page, | |
1923 | RB_PAGE_HEAD); | |
1924 | } | |
1925 | ||
1926 | /* | |
1927 | * If this was the outer most commit (the one that | |
1928 | * changed the original pointer from HEAD to UPDATE), | |
1929 | * then it is up to us to reset it to NORMAL. | |
1930 | */ | |
1931 | if (type == RB_PAGE_HEAD) { | |
1932 | ret = rb_head_page_set_normal(cpu_buffer, next_page, | |
1933 | tail_page, | |
1934 | RB_PAGE_UPDATE); | |
1935 | if (RB_WARN_ON(cpu_buffer, | |
1936 | ret != RB_PAGE_UPDATE)) | |
1937 | return -1; | |
1938 | } | |
1939 | ||
1940 | return 0; | |
1941 | } | |
1942 | ||
34a148bf | 1943 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
1944 | { |
1945 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
1946 | ||
1947 | /* zero length can cause confusions */ | |
1948 | if (!length) | |
1949 | length = 1; | |
1950 | ||
2271048d | 1951 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) |
7a8e76a3 SR |
1952 | length += sizeof(event.array[0]); |
1953 | ||
1954 | length += RB_EVNT_HDR_SIZE; | |
2271048d | 1955 | length = ALIGN(length, RB_ARCH_ALIGNMENT); |
7a8e76a3 SR |
1956 | |
1957 | return length; | |
1958 | } | |
1959 | ||
c7b09308 SR |
1960 | static inline void |
1961 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
1962 | struct buffer_page *tail_page, | |
1963 | unsigned long tail, unsigned long length) | |
1964 | { | |
1965 | struct ring_buffer_event *event; | |
1966 | ||
1967 | /* | |
1968 | * Only the event that crossed the page boundary | |
1969 | * must fill the old tail_page with padding. | |
1970 | */ | |
1971 | if (tail >= BUF_PAGE_SIZE) { | |
b3230c8b SR |
1972 | /* |
1973 | * If the page was filled, then we still need | |
1974 | * to update the real_end. Reset it to zero | |
1975 | * and the reader will ignore it. | |
1976 | */ | |
1977 | if (tail == BUF_PAGE_SIZE) | |
1978 | tail_page->real_end = 0; | |
1979 | ||
c7b09308 SR |
1980 | local_sub(length, &tail_page->write); |
1981 | return; | |
1982 | } | |
1983 | ||
1984 | event = __rb_page_index(tail_page, tail); | |
b0b7065b | 1985 | kmemcheck_annotate_bitfield(event, bitfield); |
c7b09308 | 1986 | |
c64e148a VN |
1987 | /* account for padding bytes */ |
1988 | local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes); | |
1989 | ||
ff0ff84a SR |
1990 | /* |
1991 | * Save the original length to the meta data. | |
1992 | * This will be used by the reader to add lost event | |
1993 | * counter. | |
1994 | */ | |
1995 | tail_page->real_end = tail; | |
1996 | ||
c7b09308 SR |
1997 | /* |
1998 | * If this event is bigger than the minimum size, then | |
1999 | * we need to be careful that we don't subtract the | |
2000 | * write counter enough to allow another writer to slip | |
2001 | * in on this page. | |
2002 | * We put in a discarded commit instead, to make sure | |
2003 | * that this space is not used again. | |
2004 | * | |
2005 | * If we are less than the minimum size, we don't need to | |
2006 | * worry about it. | |
2007 | */ | |
2008 | if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) { | |
2009 | /* No room for any events */ | |
2010 | ||
2011 | /* Mark the rest of the page with padding */ | |
2012 | rb_event_set_padding(event); | |
2013 | ||
2014 | /* Set the write back to the previous setting */ | |
2015 | local_sub(length, &tail_page->write); | |
2016 | return; | |
2017 | } | |
2018 | ||
2019 | /* Put in a discarded event */ | |
2020 | event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE; | |
2021 | event->type_len = RINGBUF_TYPE_PADDING; | |
2022 | /* time delta must be non zero */ | |
2023 | event->time_delta = 1; | |
c7b09308 SR |
2024 | |
2025 | /* Set write to end of buffer */ | |
2026 | length = (tail + length) - BUF_PAGE_SIZE; | |
2027 | local_sub(length, &tail_page->write); | |
2028 | } | |
6634ff26 | 2029 | |
747e94ae SR |
2030 | /* |
2031 | * This is the slow path, force gcc not to inline it. | |
2032 | */ | |
2033 | static noinline struct ring_buffer_event * | |
6634ff26 SR |
2034 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
2035 | unsigned long length, unsigned long tail, | |
e8bc43e8 | 2036 | struct buffer_page *tail_page, u64 ts) |
7a8e76a3 | 2037 | { |
5a50e33c | 2038 | struct buffer_page *commit_page = cpu_buffer->commit_page; |
7a8e76a3 | 2039 | struct ring_buffer *buffer = cpu_buffer->buffer; |
77ae365e SR |
2040 | struct buffer_page *next_page; |
2041 | int ret; | |
aa20ae84 SR |
2042 | |
2043 | next_page = tail_page; | |
2044 | ||
aa20ae84 SR |
2045 | rb_inc_page(cpu_buffer, &next_page); |
2046 | ||
aa20ae84 SR |
2047 | /* |
2048 | * If for some reason, we had an interrupt storm that made | |
2049 | * it all the way around the buffer, bail, and warn | |
2050 | * about it. | |
2051 | */ | |
2052 | if (unlikely(next_page == commit_page)) { | |
77ae365e | 2053 | local_inc(&cpu_buffer->commit_overrun); |
aa20ae84 SR |
2054 | goto out_reset; |
2055 | } | |
2056 | ||
77ae365e SR |
2057 | /* |
2058 | * This is where the fun begins! | |
2059 | * | |
2060 | * We are fighting against races between a reader that | |
2061 | * could be on another CPU trying to swap its reader | |
2062 | * page with the buffer head. | |
2063 | * | |
2064 | * We are also fighting against interrupts coming in and | |
2065 | * moving the head or tail on us as well. | |
2066 | * | |
2067 | * If the next page is the head page then we have filled | |
2068 | * the buffer, unless the commit page is still on the | |
2069 | * reader page. | |
2070 | */ | |
2071 | if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) { | |
aa20ae84 | 2072 | |
77ae365e SR |
2073 | /* |
2074 | * If the commit is not on the reader page, then | |
2075 | * move the header page. | |
2076 | */ | |
2077 | if (!rb_is_reader_page(cpu_buffer->commit_page)) { | |
2078 | /* | |
2079 | * If we are not in overwrite mode, | |
2080 | * this is easy, just stop here. | |
2081 | */ | |
2082 | if (!(buffer->flags & RB_FL_OVERWRITE)) | |
2083 | goto out_reset; | |
2084 | ||
2085 | ret = rb_handle_head_page(cpu_buffer, | |
2086 | tail_page, | |
2087 | next_page); | |
2088 | if (ret < 0) | |
2089 | goto out_reset; | |
2090 | if (ret) | |
2091 | goto out_again; | |
2092 | } else { | |
2093 | /* | |
2094 | * We need to be careful here too. The | |
2095 | * commit page could still be on the reader | |
2096 | * page. We could have a small buffer, and | |
2097 | * have filled up the buffer with events | |
2098 | * from interrupts and such, and wrapped. | |
2099 | * | |
2100 | * Note, if the tail page is also the on the | |
2101 | * reader_page, we let it move out. | |
2102 | */ | |
2103 | if (unlikely((cpu_buffer->commit_page != | |
2104 | cpu_buffer->tail_page) && | |
2105 | (cpu_buffer->commit_page == | |
2106 | cpu_buffer->reader_page))) { | |
2107 | local_inc(&cpu_buffer->commit_overrun); | |
2108 | goto out_reset; | |
2109 | } | |
aa20ae84 SR |
2110 | } |
2111 | } | |
2112 | ||
77ae365e SR |
2113 | ret = rb_tail_page_update(cpu_buffer, tail_page, next_page); |
2114 | if (ret) { | |
2115 | /* | |
2116 | * Nested commits always have zero deltas, so | |
2117 | * just reread the time stamp | |
2118 | */ | |
e8bc43e8 SR |
2119 | ts = rb_time_stamp(buffer); |
2120 | next_page->page->time_stamp = ts; | |
aa20ae84 SR |
2121 | } |
2122 | ||
77ae365e | 2123 | out_again: |
aa20ae84 | 2124 | |
77ae365e | 2125 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
aa20ae84 SR |
2126 | |
2127 | /* fail and let the caller try again */ | |
2128 | return ERR_PTR(-EAGAIN); | |
2129 | ||
45141d46 | 2130 | out_reset: |
6f3b3440 | 2131 | /* reset write */ |
c7b09308 | 2132 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
6f3b3440 | 2133 | |
bf41a158 | 2134 | return NULL; |
7a8e76a3 SR |
2135 | } |
2136 | ||
6634ff26 SR |
2137 | static struct ring_buffer_event * |
2138 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
69d1b839 SR |
2139 | unsigned long length, u64 ts, |
2140 | u64 delta, int add_timestamp) | |
6634ff26 | 2141 | { |
5a50e33c | 2142 | struct buffer_page *tail_page; |
6634ff26 SR |
2143 | struct ring_buffer_event *event; |
2144 | unsigned long tail, write; | |
2145 | ||
69d1b839 SR |
2146 | /* |
2147 | * If the time delta since the last event is too big to | |
2148 | * hold in the time field of the event, then we append a | |
2149 | * TIME EXTEND event ahead of the data event. | |
2150 | */ | |
2151 | if (unlikely(add_timestamp)) | |
2152 | length += RB_LEN_TIME_EXTEND; | |
2153 | ||
6634ff26 SR |
2154 | tail_page = cpu_buffer->tail_page; |
2155 | write = local_add_return(length, &tail_page->write); | |
77ae365e SR |
2156 | |
2157 | /* set write to only the index of the write */ | |
2158 | write &= RB_WRITE_MASK; | |
6634ff26 SR |
2159 | tail = write - length; |
2160 | ||
2161 | /* See if we shot pass the end of this buffer page */ | |
747e94ae | 2162 | if (unlikely(write > BUF_PAGE_SIZE)) |
6634ff26 | 2163 | return rb_move_tail(cpu_buffer, length, tail, |
5a50e33c | 2164 | tail_page, ts); |
6634ff26 SR |
2165 | |
2166 | /* We reserved something on the buffer */ | |
2167 | ||
6634ff26 | 2168 | event = __rb_page_index(tail_page, tail); |
1744a21d | 2169 | kmemcheck_annotate_bitfield(event, bitfield); |
69d1b839 | 2170 | rb_update_event(cpu_buffer, event, length, add_timestamp, delta); |
6634ff26 | 2171 | |
69d1b839 | 2172 | local_inc(&tail_page->entries); |
6634ff26 SR |
2173 | |
2174 | /* | |
fa743953 SR |
2175 | * If this is the first commit on the page, then update |
2176 | * its timestamp. | |
6634ff26 | 2177 | */ |
fa743953 | 2178 | if (!tail) |
e8bc43e8 | 2179 | tail_page->page->time_stamp = ts; |
6634ff26 | 2180 | |
c64e148a VN |
2181 | /* account for these added bytes */ |
2182 | local_add(length, &cpu_buffer->entries_bytes); | |
2183 | ||
6634ff26 SR |
2184 | return event; |
2185 | } | |
2186 | ||
edd813bf SR |
2187 | static inline int |
2188 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
2189 | struct ring_buffer_event *event) | |
2190 | { | |
2191 | unsigned long new_index, old_index; | |
2192 | struct buffer_page *bpage; | |
2193 | unsigned long index; | |
2194 | unsigned long addr; | |
2195 | ||
2196 | new_index = rb_event_index(event); | |
69d1b839 | 2197 | old_index = new_index + rb_event_ts_length(event); |
edd813bf SR |
2198 | addr = (unsigned long)event; |
2199 | addr &= PAGE_MASK; | |
2200 | ||
2201 | bpage = cpu_buffer->tail_page; | |
2202 | ||
2203 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { | |
77ae365e SR |
2204 | unsigned long write_mask = |
2205 | local_read(&bpage->write) & ~RB_WRITE_MASK; | |
c64e148a | 2206 | unsigned long event_length = rb_event_length(event); |
edd813bf SR |
2207 | /* |
2208 | * This is on the tail page. It is possible that | |
2209 | * a write could come in and move the tail page | |
2210 | * and write to the next page. That is fine | |
2211 | * because we just shorten what is on this page. | |
2212 | */ | |
77ae365e SR |
2213 | old_index += write_mask; |
2214 | new_index += write_mask; | |
edd813bf | 2215 | index = local_cmpxchg(&bpage->write, old_index, new_index); |
c64e148a VN |
2216 | if (index == old_index) { |
2217 | /* update counters */ | |
2218 | local_sub(event_length, &cpu_buffer->entries_bytes); | |
edd813bf | 2219 | return 1; |
c64e148a | 2220 | } |
edd813bf SR |
2221 | } |
2222 | ||
2223 | /* could not discard */ | |
2224 | return 0; | |
2225 | } | |
2226 | ||
fa743953 SR |
2227 | static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer) |
2228 | { | |
2229 | local_inc(&cpu_buffer->committing); | |
2230 | local_inc(&cpu_buffer->commits); | |
2231 | } | |
2232 | ||
d9abde21 | 2233 | static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer) |
fa743953 SR |
2234 | { |
2235 | unsigned long commits; | |
2236 | ||
2237 | if (RB_WARN_ON(cpu_buffer, | |
2238 | !local_read(&cpu_buffer->committing))) | |
2239 | return; | |
2240 | ||
2241 | again: | |
2242 | commits = local_read(&cpu_buffer->commits); | |
2243 | /* synchronize with interrupts */ | |
2244 | barrier(); | |
2245 | if (local_read(&cpu_buffer->committing) == 1) | |
2246 | rb_set_commit_to_write(cpu_buffer); | |
2247 | ||
2248 | local_dec(&cpu_buffer->committing); | |
2249 | ||
2250 | /* synchronize with interrupts */ | |
2251 | barrier(); | |
2252 | ||
2253 | /* | |
2254 | * Need to account for interrupts coming in between the | |
2255 | * updating of the commit page and the clearing of the | |
2256 | * committing counter. | |
2257 | */ | |
2258 | if (unlikely(local_read(&cpu_buffer->commits) != commits) && | |
2259 | !local_read(&cpu_buffer->committing)) { | |
2260 | local_inc(&cpu_buffer->committing); | |
2261 | goto again; | |
2262 | } | |
2263 | } | |
2264 | ||
7a8e76a3 | 2265 | static struct ring_buffer_event * |
62f0b3eb SR |
2266 | rb_reserve_next_event(struct ring_buffer *buffer, |
2267 | struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 2268 | unsigned long length) |
7a8e76a3 SR |
2269 | { |
2270 | struct ring_buffer_event *event; | |
69d1b839 | 2271 | u64 ts, delta; |
818e3dd3 | 2272 | int nr_loops = 0; |
69d1b839 | 2273 | int add_timestamp; |
140ff891 | 2274 | u64 diff; |
7a8e76a3 | 2275 | |
fa743953 SR |
2276 | rb_start_commit(cpu_buffer); |
2277 | ||
85bac32c | 2278 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
62f0b3eb SR |
2279 | /* |
2280 | * Due to the ability to swap a cpu buffer from a buffer | |
2281 | * it is possible it was swapped before we committed. | |
2282 | * (committing stops a swap). We check for it here and | |
2283 | * if it happened, we have to fail the write. | |
2284 | */ | |
2285 | barrier(); | |
2286 | if (unlikely(ACCESS_ONCE(cpu_buffer->buffer) != buffer)) { | |
2287 | local_dec(&cpu_buffer->committing); | |
2288 | local_dec(&cpu_buffer->commits); | |
2289 | return NULL; | |
2290 | } | |
85bac32c | 2291 | #endif |
62f0b3eb | 2292 | |
be957c44 | 2293 | length = rb_calculate_event_length(length); |
bf41a158 | 2294 | again: |
69d1b839 SR |
2295 | add_timestamp = 0; |
2296 | delta = 0; | |
2297 | ||
818e3dd3 SR |
2298 | /* |
2299 | * We allow for interrupts to reenter here and do a trace. | |
2300 | * If one does, it will cause this original code to loop | |
2301 | * back here. Even with heavy interrupts happening, this | |
2302 | * should only happen a few times in a row. If this happens | |
2303 | * 1000 times in a row, there must be either an interrupt | |
2304 | * storm or we have something buggy. | |
2305 | * Bail! | |
2306 | */ | |
3e89c7bb | 2307 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
fa743953 | 2308 | goto out_fail; |
818e3dd3 | 2309 | |
6d3f1e12 | 2310 | ts = rb_time_stamp(cpu_buffer->buffer); |
140ff891 | 2311 | diff = ts - cpu_buffer->write_stamp; |
7a8e76a3 | 2312 | |
140ff891 SR |
2313 | /* make sure this diff is calculated here */ |
2314 | barrier(); | |
bf41a158 | 2315 | |
140ff891 SR |
2316 | /* Did the write stamp get updated already? */ |
2317 | if (likely(ts >= cpu_buffer->write_stamp)) { | |
168b6b1d SR |
2318 | delta = diff; |
2319 | if (unlikely(test_time_stamp(delta))) { | |
31274d72 JO |
2320 | int local_clock_stable = 1; |
2321 | #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK | |
2322 | local_clock_stable = sched_clock_stable; | |
2323 | #endif | |
69d1b839 | 2324 | WARN_ONCE(delta > (1ULL << 59), |
31274d72 | 2325 | KERN_WARNING "Delta way too big! %llu ts=%llu write stamp = %llu\n%s", |
69d1b839 SR |
2326 | (unsigned long long)delta, |
2327 | (unsigned long long)ts, | |
31274d72 JO |
2328 | (unsigned long long)cpu_buffer->write_stamp, |
2329 | local_clock_stable ? "" : | |
2330 | "If you just came from a suspend/resume,\n" | |
2331 | "please switch to the trace global clock:\n" | |
2332 | " echo global > /sys/kernel/debug/tracing/trace_clock\n"); | |
69d1b839 | 2333 | add_timestamp = 1; |
7a8e76a3 | 2334 | } |
168b6b1d | 2335 | } |
7a8e76a3 | 2336 | |
69d1b839 SR |
2337 | event = __rb_reserve_next(cpu_buffer, length, ts, |
2338 | delta, add_timestamp); | |
168b6b1d | 2339 | if (unlikely(PTR_ERR(event) == -EAGAIN)) |
bf41a158 SR |
2340 | goto again; |
2341 | ||
fa743953 SR |
2342 | if (!event) |
2343 | goto out_fail; | |
7a8e76a3 | 2344 | |
7a8e76a3 | 2345 | return event; |
fa743953 SR |
2346 | |
2347 | out_fail: | |
2348 | rb_end_commit(cpu_buffer); | |
2349 | return NULL; | |
7a8e76a3 SR |
2350 | } |
2351 | ||
1155de47 PM |
2352 | #ifdef CONFIG_TRACING |
2353 | ||
aa18efb2 | 2354 | #define TRACE_RECURSIVE_DEPTH 16 |
261842b7 | 2355 | |
d9abde21 SR |
2356 | /* Keep this code out of the fast path cache */ |
2357 | static noinline void trace_recursive_fail(void) | |
261842b7 | 2358 | { |
aa18efb2 SR |
2359 | /* Disable all tracing before we do anything else */ |
2360 | tracing_off_permanent(); | |
261842b7 | 2361 | |
7d7d2b80 | 2362 | printk_once(KERN_WARNING "Tracing recursion: depth[%ld]:" |
aa18efb2 | 2363 | "HC[%lu]:SC[%lu]:NMI[%lu]\n", |
b1cff0ad | 2364 | trace_recursion_buffer(), |
aa18efb2 SR |
2365 | hardirq_count() >> HARDIRQ_SHIFT, |
2366 | softirq_count() >> SOFTIRQ_SHIFT, | |
2367 | in_nmi()); | |
261842b7 | 2368 | |
aa18efb2 | 2369 | WARN_ON_ONCE(1); |
d9abde21 SR |
2370 | } |
2371 | ||
2372 | static inline int trace_recursive_lock(void) | |
2373 | { | |
b1cff0ad | 2374 | trace_recursion_inc(); |
d9abde21 | 2375 | |
b1cff0ad | 2376 | if (likely(trace_recursion_buffer() < TRACE_RECURSIVE_DEPTH)) |
d9abde21 SR |
2377 | return 0; |
2378 | ||
2379 | trace_recursive_fail(); | |
2380 | ||
aa18efb2 | 2381 | return -1; |
261842b7 SR |
2382 | } |
2383 | ||
d9abde21 | 2384 | static inline void trace_recursive_unlock(void) |
261842b7 | 2385 | { |
b1cff0ad | 2386 | WARN_ON_ONCE(!trace_recursion_buffer()); |
261842b7 | 2387 | |
b1cff0ad | 2388 | trace_recursion_dec(); |
261842b7 SR |
2389 | } |
2390 | ||
1155de47 PM |
2391 | #else |
2392 | ||
2393 | #define trace_recursive_lock() (0) | |
2394 | #define trace_recursive_unlock() do { } while (0) | |
2395 | ||
2396 | #endif | |
2397 | ||
7a8e76a3 SR |
2398 | /** |
2399 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
2400 | * @buffer: the ring buffer to reserve from | |
2401 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
2402 | * |
2403 | * Returns a reseverd event on the ring buffer to copy directly to. | |
2404 | * The user of this interface will need to get the body to write into | |
2405 | * and can use the ring_buffer_event_data() interface. | |
2406 | * | |
2407 | * The length is the length of the data needed, not the event length | |
2408 | * which also includes the event header. | |
2409 | * | |
2410 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
2411 | * If NULL is returned, then nothing has been allocated or locked. | |
2412 | */ | |
2413 | struct ring_buffer_event * | |
0a987751 | 2414 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
2415 | { |
2416 | struct ring_buffer_per_cpu *cpu_buffer; | |
2417 | struct ring_buffer_event *event; | |
5168ae50 | 2418 | int cpu; |
7a8e76a3 | 2419 | |
033601a3 | 2420 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2421 | return NULL; |
2422 | ||
bf41a158 | 2423 | /* If we are tracing schedule, we don't want to recurse */ |
5168ae50 | 2424 | preempt_disable_notrace(); |
bf41a158 | 2425 | |
52fbe9cd LJ |
2426 | if (atomic_read(&buffer->record_disabled)) |
2427 | goto out_nocheck; | |
2428 | ||
261842b7 SR |
2429 | if (trace_recursive_lock()) |
2430 | goto out_nocheck; | |
2431 | ||
7a8e76a3 SR |
2432 | cpu = raw_smp_processor_id(); |
2433 | ||
9e01c1b7 | 2434 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2435 | goto out; |
7a8e76a3 SR |
2436 | |
2437 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2438 | |
2439 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 2440 | goto out; |
7a8e76a3 | 2441 | |
be957c44 | 2442 | if (length > BUF_MAX_DATA_SIZE) |
bf41a158 | 2443 | goto out; |
7a8e76a3 | 2444 | |
62f0b3eb | 2445 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 2446 | if (!event) |
d769041f | 2447 | goto out; |
7a8e76a3 SR |
2448 | |
2449 | return event; | |
2450 | ||
d769041f | 2451 | out: |
261842b7 SR |
2452 | trace_recursive_unlock(); |
2453 | ||
2454 | out_nocheck: | |
5168ae50 | 2455 | preempt_enable_notrace(); |
7a8e76a3 SR |
2456 | return NULL; |
2457 | } | |
c4f50183 | 2458 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 | 2459 | |
a1863c21 SR |
2460 | static void |
2461 | rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
7a8e76a3 SR |
2462 | struct ring_buffer_event *event) |
2463 | { | |
69d1b839 SR |
2464 | u64 delta; |
2465 | ||
fa743953 SR |
2466 | /* |
2467 | * The event first in the commit queue updates the | |
2468 | * time stamp. | |
2469 | */ | |
69d1b839 SR |
2470 | if (rb_event_is_commit(cpu_buffer, event)) { |
2471 | /* | |
2472 | * A commit event that is first on a page | |
2473 | * updates the write timestamp with the page stamp | |
2474 | */ | |
2475 | if (!rb_event_index(event)) | |
2476 | cpu_buffer->write_stamp = | |
2477 | cpu_buffer->commit_page->page->time_stamp; | |
2478 | else if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
2479 | delta = event->array[0]; | |
2480 | delta <<= TS_SHIFT; | |
2481 | delta += event->time_delta; | |
2482 | cpu_buffer->write_stamp += delta; | |
2483 | } else | |
2484 | cpu_buffer->write_stamp += event->time_delta; | |
2485 | } | |
a1863c21 | 2486 | } |
bf41a158 | 2487 | |
a1863c21 SR |
2488 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, |
2489 | struct ring_buffer_event *event) | |
2490 | { | |
2491 | local_inc(&cpu_buffer->entries); | |
2492 | rb_update_write_stamp(cpu_buffer, event); | |
fa743953 | 2493 | rb_end_commit(cpu_buffer); |
7a8e76a3 SR |
2494 | } |
2495 | ||
2496 | /** | |
2497 | * ring_buffer_unlock_commit - commit a reserved | |
2498 | * @buffer: The buffer to commit to | |
2499 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
2500 | * |
2501 | * This commits the data to the ring buffer, and releases any locks held. | |
2502 | * | |
2503 | * Must be paired with ring_buffer_lock_reserve. | |
2504 | */ | |
2505 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 2506 | struct ring_buffer_event *event) |
7a8e76a3 SR |
2507 | { |
2508 | struct ring_buffer_per_cpu *cpu_buffer; | |
2509 | int cpu = raw_smp_processor_id(); | |
2510 | ||
2511 | cpu_buffer = buffer->buffers[cpu]; | |
2512 | ||
7a8e76a3 SR |
2513 | rb_commit(cpu_buffer, event); |
2514 | ||
261842b7 SR |
2515 | trace_recursive_unlock(); |
2516 | ||
5168ae50 | 2517 | preempt_enable_notrace(); |
7a8e76a3 SR |
2518 | |
2519 | return 0; | |
2520 | } | |
c4f50183 | 2521 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 2522 | |
f3b9aae1 FW |
2523 | static inline void rb_event_discard(struct ring_buffer_event *event) |
2524 | { | |
69d1b839 SR |
2525 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) |
2526 | event = skip_time_extend(event); | |
2527 | ||
334d4169 LJ |
2528 | /* array[0] holds the actual length for the discarded event */ |
2529 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
2530 | event->type_len = RINGBUF_TYPE_PADDING; | |
f3b9aae1 FW |
2531 | /* time delta must be non zero */ |
2532 | if (!event->time_delta) | |
2533 | event->time_delta = 1; | |
2534 | } | |
2535 | ||
a1863c21 SR |
2536 | /* |
2537 | * Decrement the entries to the page that an event is on. | |
2538 | * The event does not even need to exist, only the pointer | |
2539 | * to the page it is on. This may only be called before the commit | |
2540 | * takes place. | |
2541 | */ | |
2542 | static inline void | |
2543 | rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer, | |
2544 | struct ring_buffer_event *event) | |
2545 | { | |
2546 | unsigned long addr = (unsigned long)event; | |
2547 | struct buffer_page *bpage = cpu_buffer->commit_page; | |
2548 | struct buffer_page *start; | |
2549 | ||
2550 | addr &= PAGE_MASK; | |
2551 | ||
2552 | /* Do the likely case first */ | |
2553 | if (likely(bpage->page == (void *)addr)) { | |
2554 | local_dec(&bpage->entries); | |
2555 | return; | |
2556 | } | |
2557 | ||
2558 | /* | |
2559 | * Because the commit page may be on the reader page we | |
2560 | * start with the next page and check the end loop there. | |
2561 | */ | |
2562 | rb_inc_page(cpu_buffer, &bpage); | |
2563 | start = bpage; | |
2564 | do { | |
2565 | if (bpage->page == (void *)addr) { | |
2566 | local_dec(&bpage->entries); | |
2567 | return; | |
2568 | } | |
2569 | rb_inc_page(cpu_buffer, &bpage); | |
2570 | } while (bpage != start); | |
2571 | ||
2572 | /* commit not part of this buffer?? */ | |
2573 | RB_WARN_ON(cpu_buffer, 1); | |
2574 | } | |
2575 | ||
fa1b47dd SR |
2576 | /** |
2577 | * ring_buffer_commit_discard - discard an event that has not been committed | |
2578 | * @buffer: the ring buffer | |
2579 | * @event: non committed event to discard | |
2580 | * | |
dc892f73 SR |
2581 | * Sometimes an event that is in the ring buffer needs to be ignored. |
2582 | * This function lets the user discard an event in the ring buffer | |
2583 | * and then that event will not be read later. | |
2584 | * | |
2585 | * This function only works if it is called before the the item has been | |
2586 | * committed. It will try to free the event from the ring buffer | |
fa1b47dd SR |
2587 | * if another event has not been added behind it. |
2588 | * | |
2589 | * If another event has been added behind it, it will set the event | |
2590 | * up as discarded, and perform the commit. | |
2591 | * | |
2592 | * If this function is called, do not call ring_buffer_unlock_commit on | |
2593 | * the event. | |
2594 | */ | |
2595 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
2596 | struct ring_buffer_event *event) | |
2597 | { | |
2598 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
2599 | int cpu; |
2600 | ||
2601 | /* The event is discarded regardless */ | |
f3b9aae1 | 2602 | rb_event_discard(event); |
fa1b47dd | 2603 | |
fa743953 SR |
2604 | cpu = smp_processor_id(); |
2605 | cpu_buffer = buffer->buffers[cpu]; | |
2606 | ||
fa1b47dd SR |
2607 | /* |
2608 | * This must only be called if the event has not been | |
2609 | * committed yet. Thus we can assume that preemption | |
2610 | * is still disabled. | |
2611 | */ | |
fa743953 | 2612 | RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing)); |
fa1b47dd | 2613 | |
a1863c21 | 2614 | rb_decrement_entry(cpu_buffer, event); |
0f2541d2 | 2615 | if (rb_try_to_discard(cpu_buffer, event)) |
edd813bf | 2616 | goto out; |
fa1b47dd SR |
2617 | |
2618 | /* | |
2619 | * The commit is still visible by the reader, so we | |
a1863c21 | 2620 | * must still update the timestamp. |
fa1b47dd | 2621 | */ |
a1863c21 | 2622 | rb_update_write_stamp(cpu_buffer, event); |
fa1b47dd | 2623 | out: |
fa743953 | 2624 | rb_end_commit(cpu_buffer); |
fa1b47dd | 2625 | |
f3b9aae1 FW |
2626 | trace_recursive_unlock(); |
2627 | ||
5168ae50 | 2628 | preempt_enable_notrace(); |
fa1b47dd SR |
2629 | |
2630 | } | |
2631 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
2632 | ||
7a8e76a3 SR |
2633 | /** |
2634 | * ring_buffer_write - write data to the buffer without reserving | |
2635 | * @buffer: The ring buffer to write to. | |
2636 | * @length: The length of the data being written (excluding the event header) | |
2637 | * @data: The data to write to the buffer. | |
2638 | * | |
2639 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
2640 | * one function. If you already have the data to write to the buffer, it | |
2641 | * may be easier to simply call this function. | |
2642 | * | |
2643 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
2644 | * and not the length of the event which would hold the header. | |
2645 | */ | |
2646 | int ring_buffer_write(struct ring_buffer *buffer, | |
2647 | unsigned long length, | |
2648 | void *data) | |
2649 | { | |
2650 | struct ring_buffer_per_cpu *cpu_buffer; | |
2651 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
2652 | void *body; |
2653 | int ret = -EBUSY; | |
5168ae50 | 2654 | int cpu; |
7a8e76a3 | 2655 | |
033601a3 | 2656 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2657 | return -EBUSY; |
2658 | ||
5168ae50 | 2659 | preempt_disable_notrace(); |
bf41a158 | 2660 | |
52fbe9cd LJ |
2661 | if (atomic_read(&buffer->record_disabled)) |
2662 | goto out; | |
2663 | ||
7a8e76a3 SR |
2664 | cpu = raw_smp_processor_id(); |
2665 | ||
9e01c1b7 | 2666 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2667 | goto out; |
7a8e76a3 SR |
2668 | |
2669 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2670 | |
2671 | if (atomic_read(&cpu_buffer->record_disabled)) | |
2672 | goto out; | |
2673 | ||
be957c44 SR |
2674 | if (length > BUF_MAX_DATA_SIZE) |
2675 | goto out; | |
2676 | ||
62f0b3eb | 2677 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 SR |
2678 | if (!event) |
2679 | goto out; | |
2680 | ||
2681 | body = rb_event_data(event); | |
2682 | ||
2683 | memcpy(body, data, length); | |
2684 | ||
2685 | rb_commit(cpu_buffer, event); | |
2686 | ||
2687 | ret = 0; | |
2688 | out: | |
5168ae50 | 2689 | preempt_enable_notrace(); |
7a8e76a3 SR |
2690 | |
2691 | return ret; | |
2692 | } | |
c4f50183 | 2693 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 2694 | |
34a148bf | 2695 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
2696 | { |
2697 | struct buffer_page *reader = cpu_buffer->reader_page; | |
77ae365e | 2698 | struct buffer_page *head = rb_set_head_page(cpu_buffer); |
bf41a158 SR |
2699 | struct buffer_page *commit = cpu_buffer->commit_page; |
2700 | ||
77ae365e SR |
2701 | /* In case of error, head will be NULL */ |
2702 | if (unlikely(!head)) | |
2703 | return 1; | |
2704 | ||
bf41a158 SR |
2705 | return reader->read == rb_page_commit(reader) && |
2706 | (commit == reader || | |
2707 | (commit == head && | |
2708 | head->read == rb_page_commit(commit))); | |
2709 | } | |
2710 | ||
7a8e76a3 SR |
2711 | /** |
2712 | * ring_buffer_record_disable - stop all writes into the buffer | |
2713 | * @buffer: The ring buffer to stop writes to. | |
2714 | * | |
2715 | * This prevents all writes to the buffer. Any attempt to write | |
2716 | * to the buffer after this will fail and return NULL. | |
2717 | * | |
2718 | * The caller should call synchronize_sched() after this. | |
2719 | */ | |
2720 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
2721 | { | |
2722 | atomic_inc(&buffer->record_disabled); | |
2723 | } | |
c4f50183 | 2724 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
2725 | |
2726 | /** | |
2727 | * ring_buffer_record_enable - enable writes to the buffer | |
2728 | * @buffer: The ring buffer to enable writes | |
2729 | * | |
2730 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 2731 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
2732 | */ |
2733 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
2734 | { | |
2735 | atomic_dec(&buffer->record_disabled); | |
2736 | } | |
c4f50183 | 2737 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 | 2738 | |
499e5470 SR |
2739 | /** |
2740 | * ring_buffer_record_off - stop all writes into the buffer | |
2741 | * @buffer: The ring buffer to stop writes to. | |
2742 | * | |
2743 | * This prevents all writes to the buffer. Any attempt to write | |
2744 | * to the buffer after this will fail and return NULL. | |
2745 | * | |
2746 | * This is different than ring_buffer_record_disable() as | |
2747 | * it works like an on/off switch, where as the disable() verison | |
2748 | * must be paired with a enable(). | |
2749 | */ | |
2750 | void ring_buffer_record_off(struct ring_buffer *buffer) | |
2751 | { | |
2752 | unsigned int rd; | |
2753 | unsigned int new_rd; | |
2754 | ||
2755 | do { | |
2756 | rd = atomic_read(&buffer->record_disabled); | |
2757 | new_rd = rd | RB_BUFFER_OFF; | |
2758 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
2759 | } | |
2760 | EXPORT_SYMBOL_GPL(ring_buffer_record_off); | |
2761 | ||
2762 | /** | |
2763 | * ring_buffer_record_on - restart writes into the buffer | |
2764 | * @buffer: The ring buffer to start writes to. | |
2765 | * | |
2766 | * This enables all writes to the buffer that was disabled by | |
2767 | * ring_buffer_record_off(). | |
2768 | * | |
2769 | * This is different than ring_buffer_record_enable() as | |
2770 | * it works like an on/off switch, where as the enable() verison | |
2771 | * must be paired with a disable(). | |
2772 | */ | |
2773 | void ring_buffer_record_on(struct ring_buffer *buffer) | |
2774 | { | |
2775 | unsigned int rd; | |
2776 | unsigned int new_rd; | |
2777 | ||
2778 | do { | |
2779 | rd = atomic_read(&buffer->record_disabled); | |
2780 | new_rd = rd & ~RB_BUFFER_OFF; | |
2781 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
2782 | } | |
2783 | EXPORT_SYMBOL_GPL(ring_buffer_record_on); | |
2784 | ||
2785 | /** | |
2786 | * ring_buffer_record_is_on - return true if the ring buffer can write | |
2787 | * @buffer: The ring buffer to see if write is enabled | |
2788 | * | |
2789 | * Returns true if the ring buffer is in a state that it accepts writes. | |
2790 | */ | |
2791 | int ring_buffer_record_is_on(struct ring_buffer *buffer) | |
2792 | { | |
2793 | return !atomic_read(&buffer->record_disabled); | |
2794 | } | |
2795 | ||
7a8e76a3 SR |
2796 | /** |
2797 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
2798 | * @buffer: The ring buffer to stop writes to. | |
2799 | * @cpu: The CPU buffer to stop | |
2800 | * | |
2801 | * This prevents all writes to the buffer. Any attempt to write | |
2802 | * to the buffer after this will fail and return NULL. | |
2803 | * | |
2804 | * The caller should call synchronize_sched() after this. | |
2805 | */ | |
2806 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
2807 | { | |
2808 | struct ring_buffer_per_cpu *cpu_buffer; | |
2809 | ||
9e01c1b7 | 2810 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2811 | return; |
7a8e76a3 SR |
2812 | |
2813 | cpu_buffer = buffer->buffers[cpu]; | |
2814 | atomic_inc(&cpu_buffer->record_disabled); | |
2815 | } | |
c4f50183 | 2816 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
2817 | |
2818 | /** | |
2819 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
2820 | * @buffer: The ring buffer to enable writes | |
2821 | * @cpu: The CPU to enable. | |
2822 | * | |
2823 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 2824 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
2825 | */ |
2826 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
2827 | { | |
2828 | struct ring_buffer_per_cpu *cpu_buffer; | |
2829 | ||
9e01c1b7 | 2830 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2831 | return; |
7a8e76a3 SR |
2832 | |
2833 | cpu_buffer = buffer->buffers[cpu]; | |
2834 | atomic_dec(&cpu_buffer->record_disabled); | |
2835 | } | |
c4f50183 | 2836 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 | 2837 | |
f6195aa0 SR |
2838 | /* |
2839 | * The total entries in the ring buffer is the running counter | |
2840 | * of entries entered into the ring buffer, minus the sum of | |
2841 | * the entries read from the ring buffer and the number of | |
2842 | * entries that were overwritten. | |
2843 | */ | |
2844 | static inline unsigned long | |
2845 | rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer) | |
2846 | { | |
2847 | return local_read(&cpu_buffer->entries) - | |
2848 | (local_read(&cpu_buffer->overrun) + cpu_buffer->read); | |
2849 | } | |
2850 | ||
c64e148a VN |
2851 | /** |
2852 | * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer | |
2853 | * @buffer: The ring buffer | |
2854 | * @cpu: The per CPU buffer to read from. | |
2855 | */ | |
2856 | unsigned long ring_buffer_oldest_event_ts(struct ring_buffer *buffer, int cpu) | |
2857 | { | |
2858 | unsigned long flags; | |
2859 | struct ring_buffer_per_cpu *cpu_buffer; | |
2860 | struct buffer_page *bpage; | |
2861 | unsigned long ret; | |
2862 | ||
2863 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
2864 | return 0; | |
2865 | ||
2866 | cpu_buffer = buffer->buffers[cpu]; | |
7115e3fc | 2867 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
2868 | /* |
2869 | * if the tail is on reader_page, oldest time stamp is on the reader | |
2870 | * page | |
2871 | */ | |
2872 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
2873 | bpage = cpu_buffer->reader_page; | |
2874 | else | |
2875 | bpage = rb_set_head_page(cpu_buffer); | |
2876 | ret = bpage->page->time_stamp; | |
7115e3fc | 2877 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
2878 | |
2879 | return ret; | |
2880 | } | |
2881 | EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts); | |
2882 | ||
2883 | /** | |
2884 | * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer | |
2885 | * @buffer: The ring buffer | |
2886 | * @cpu: The per CPU buffer to read from. | |
2887 | */ | |
2888 | unsigned long ring_buffer_bytes_cpu(struct ring_buffer *buffer, int cpu) | |
2889 | { | |
2890 | struct ring_buffer_per_cpu *cpu_buffer; | |
2891 | unsigned long ret; | |
2892 | ||
2893 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
2894 | return 0; | |
2895 | ||
2896 | cpu_buffer = buffer->buffers[cpu]; | |
2897 | ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes; | |
2898 | ||
2899 | return ret; | |
2900 | } | |
2901 | EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu); | |
2902 | ||
7a8e76a3 SR |
2903 | /** |
2904 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
2905 | * @buffer: The ring buffer | |
2906 | * @cpu: The per CPU buffer to get the entries from. | |
2907 | */ | |
2908 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
2909 | { | |
2910 | struct ring_buffer_per_cpu *cpu_buffer; | |
2911 | ||
9e01c1b7 | 2912 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2913 | return 0; |
7a8e76a3 SR |
2914 | |
2915 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 2916 | |
f6195aa0 | 2917 | return rb_num_of_entries(cpu_buffer); |
7a8e76a3 | 2918 | } |
c4f50183 | 2919 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
2920 | |
2921 | /** | |
2922 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
2923 | * @buffer: The ring buffer | |
2924 | * @cpu: The per CPU buffer to get the number of overruns from | |
2925 | */ | |
2926 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
2927 | { | |
2928 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 2929 | unsigned long ret; |
7a8e76a3 | 2930 | |
9e01c1b7 | 2931 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2932 | return 0; |
7a8e76a3 SR |
2933 | |
2934 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 2935 | ret = local_read(&cpu_buffer->overrun); |
554f786e SR |
2936 | |
2937 | return ret; | |
7a8e76a3 | 2938 | } |
c4f50183 | 2939 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 2940 | |
f0d2c681 SR |
2941 | /** |
2942 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits | |
2943 | * @buffer: The ring buffer | |
2944 | * @cpu: The per CPU buffer to get the number of overruns from | |
2945 | */ | |
2946 | unsigned long | |
2947 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
2948 | { | |
2949 | struct ring_buffer_per_cpu *cpu_buffer; | |
2950 | unsigned long ret; | |
2951 | ||
2952 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
2953 | return 0; | |
2954 | ||
2955 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 2956 | ret = local_read(&cpu_buffer->commit_overrun); |
f0d2c681 SR |
2957 | |
2958 | return ret; | |
2959 | } | |
2960 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
2961 | ||
7a8e76a3 SR |
2962 | /** |
2963 | * ring_buffer_entries - get the number of entries in a buffer | |
2964 | * @buffer: The ring buffer | |
2965 | * | |
2966 | * Returns the total number of entries in the ring buffer | |
2967 | * (all CPU entries) | |
2968 | */ | |
2969 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
2970 | { | |
2971 | struct ring_buffer_per_cpu *cpu_buffer; | |
2972 | unsigned long entries = 0; | |
2973 | int cpu; | |
2974 | ||
2975 | /* if you care about this being correct, lock the buffer */ | |
2976 | for_each_buffer_cpu(buffer, cpu) { | |
2977 | cpu_buffer = buffer->buffers[cpu]; | |
f6195aa0 | 2978 | entries += rb_num_of_entries(cpu_buffer); |
7a8e76a3 SR |
2979 | } |
2980 | ||
2981 | return entries; | |
2982 | } | |
c4f50183 | 2983 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
2984 | |
2985 | /** | |
67b394f7 | 2986 | * ring_buffer_overruns - get the number of overruns in buffer |
7a8e76a3 SR |
2987 | * @buffer: The ring buffer |
2988 | * | |
2989 | * Returns the total number of overruns in the ring buffer | |
2990 | * (all CPU entries) | |
2991 | */ | |
2992 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
2993 | { | |
2994 | struct ring_buffer_per_cpu *cpu_buffer; | |
2995 | unsigned long overruns = 0; | |
2996 | int cpu; | |
2997 | ||
2998 | /* if you care about this being correct, lock the buffer */ | |
2999 | for_each_buffer_cpu(buffer, cpu) { | |
3000 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3001 | overruns += local_read(&cpu_buffer->overrun); |
7a8e76a3 SR |
3002 | } |
3003 | ||
3004 | return overruns; | |
3005 | } | |
c4f50183 | 3006 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 3007 | |
642edba5 | 3008 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
3009 | { |
3010 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3011 | ||
d769041f SR |
3012 | /* Iterator usage is expected to have record disabled */ |
3013 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
77ae365e SR |
3014 | iter->head_page = rb_set_head_page(cpu_buffer); |
3015 | if (unlikely(!iter->head_page)) | |
3016 | return; | |
3017 | iter->head = iter->head_page->read; | |
d769041f SR |
3018 | } else { |
3019 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 3020 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
3021 | } |
3022 | if (iter->head) | |
3023 | iter->read_stamp = cpu_buffer->read_stamp; | |
3024 | else | |
abc9b56d | 3025 | iter->read_stamp = iter->head_page->page->time_stamp; |
492a74f4 SR |
3026 | iter->cache_reader_page = cpu_buffer->reader_page; |
3027 | iter->cache_read = cpu_buffer->read; | |
642edba5 | 3028 | } |
f83c9d0f | 3029 | |
642edba5 SR |
3030 | /** |
3031 | * ring_buffer_iter_reset - reset an iterator | |
3032 | * @iter: The iterator to reset | |
3033 | * | |
3034 | * Resets the iterator, so that it will start from the beginning | |
3035 | * again. | |
3036 | */ | |
3037 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
3038 | { | |
554f786e | 3039 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
3040 | unsigned long flags; |
3041 | ||
554f786e SR |
3042 | if (!iter) |
3043 | return; | |
3044 | ||
3045 | cpu_buffer = iter->cpu_buffer; | |
3046 | ||
5389f6fa | 3047 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
642edba5 | 3048 | rb_iter_reset(iter); |
5389f6fa | 3049 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 3050 | } |
c4f50183 | 3051 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
3052 | |
3053 | /** | |
3054 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
3055 | * @iter: The iterator to check | |
3056 | */ | |
3057 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
3058 | { | |
3059 | struct ring_buffer_per_cpu *cpu_buffer; | |
3060 | ||
3061 | cpu_buffer = iter->cpu_buffer; | |
3062 | ||
bf41a158 SR |
3063 | return iter->head_page == cpu_buffer->commit_page && |
3064 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 3065 | } |
c4f50183 | 3066 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
3067 | |
3068 | static void | |
3069 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
3070 | struct ring_buffer_event *event) | |
3071 | { | |
3072 | u64 delta; | |
3073 | ||
334d4169 | 3074 | switch (event->type_len) { |
7a8e76a3 SR |
3075 | case RINGBUF_TYPE_PADDING: |
3076 | return; | |
3077 | ||
3078 | case RINGBUF_TYPE_TIME_EXTEND: | |
3079 | delta = event->array[0]; | |
3080 | delta <<= TS_SHIFT; | |
3081 | delta += event->time_delta; | |
3082 | cpu_buffer->read_stamp += delta; | |
3083 | return; | |
3084 | ||
3085 | case RINGBUF_TYPE_TIME_STAMP: | |
3086 | /* FIXME: not implemented */ | |
3087 | return; | |
3088 | ||
3089 | case RINGBUF_TYPE_DATA: | |
3090 | cpu_buffer->read_stamp += event->time_delta; | |
3091 | return; | |
3092 | ||
3093 | default: | |
3094 | BUG(); | |
3095 | } | |
3096 | return; | |
3097 | } | |
3098 | ||
3099 | static void | |
3100 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
3101 | struct ring_buffer_event *event) | |
3102 | { | |
3103 | u64 delta; | |
3104 | ||
334d4169 | 3105 | switch (event->type_len) { |
7a8e76a3 SR |
3106 | case RINGBUF_TYPE_PADDING: |
3107 | return; | |
3108 | ||
3109 | case RINGBUF_TYPE_TIME_EXTEND: | |
3110 | delta = event->array[0]; | |
3111 | delta <<= TS_SHIFT; | |
3112 | delta += event->time_delta; | |
3113 | iter->read_stamp += delta; | |
3114 | return; | |
3115 | ||
3116 | case RINGBUF_TYPE_TIME_STAMP: | |
3117 | /* FIXME: not implemented */ | |
3118 | return; | |
3119 | ||
3120 | case RINGBUF_TYPE_DATA: | |
3121 | iter->read_stamp += event->time_delta; | |
3122 | return; | |
3123 | ||
3124 | default: | |
3125 | BUG(); | |
3126 | } | |
3127 | return; | |
3128 | } | |
3129 | ||
d769041f SR |
3130 | static struct buffer_page * |
3131 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 3132 | { |
d769041f | 3133 | struct buffer_page *reader = NULL; |
66a8cb95 | 3134 | unsigned long overwrite; |
d769041f | 3135 | unsigned long flags; |
818e3dd3 | 3136 | int nr_loops = 0; |
77ae365e | 3137 | int ret; |
d769041f | 3138 | |
3e03fb7f | 3139 | local_irq_save(flags); |
0199c4e6 | 3140 | arch_spin_lock(&cpu_buffer->lock); |
d769041f SR |
3141 | |
3142 | again: | |
818e3dd3 SR |
3143 | /* |
3144 | * This should normally only loop twice. But because the | |
3145 | * start of the reader inserts an empty page, it causes | |
3146 | * a case where we will loop three times. There should be no | |
3147 | * reason to loop four times (that I know of). | |
3148 | */ | |
3e89c7bb | 3149 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
3150 | reader = NULL; |
3151 | goto out; | |
3152 | } | |
3153 | ||
d769041f SR |
3154 | reader = cpu_buffer->reader_page; |
3155 | ||
3156 | /* If there's more to read, return this page */ | |
bf41a158 | 3157 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
3158 | goto out; |
3159 | ||
3160 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
3161 | if (RB_WARN_ON(cpu_buffer, |
3162 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
3163 | goto out; | |
d769041f SR |
3164 | |
3165 | /* check if we caught up to the tail */ | |
3166 | reader = NULL; | |
bf41a158 | 3167 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 3168 | goto out; |
7a8e76a3 SR |
3169 | |
3170 | /* | |
d769041f | 3171 | * Reset the reader page to size zero. |
7a8e76a3 | 3172 | */ |
77ae365e SR |
3173 | local_set(&cpu_buffer->reader_page->write, 0); |
3174 | local_set(&cpu_buffer->reader_page->entries, 0); | |
3175 | local_set(&cpu_buffer->reader_page->page->commit, 0); | |
ff0ff84a | 3176 | cpu_buffer->reader_page->real_end = 0; |
7a8e76a3 | 3177 | |
77ae365e SR |
3178 | spin: |
3179 | /* | |
3180 | * Splice the empty reader page into the list around the head. | |
3181 | */ | |
3182 | reader = rb_set_head_page(cpu_buffer); | |
0e1ff5d7 | 3183 | cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next); |
d769041f | 3184 | cpu_buffer->reader_page->list.prev = reader->list.prev; |
bf41a158 | 3185 | |
3adc54fa SR |
3186 | /* |
3187 | * cpu_buffer->pages just needs to point to the buffer, it | |
3188 | * has no specific buffer page to point to. Lets move it out | |
25985edc | 3189 | * of our way so we don't accidentally swap it. |
3adc54fa SR |
3190 | */ |
3191 | cpu_buffer->pages = reader->list.prev; | |
3192 | ||
77ae365e SR |
3193 | /* The reader page will be pointing to the new head */ |
3194 | rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list); | |
7a8e76a3 | 3195 | |
66a8cb95 SR |
3196 | /* |
3197 | * We want to make sure we read the overruns after we set up our | |
3198 | * pointers to the next object. The writer side does a | |
3199 | * cmpxchg to cross pages which acts as the mb on the writer | |
3200 | * side. Note, the reader will constantly fail the swap | |
3201 | * while the writer is updating the pointers, so this | |
3202 | * guarantees that the overwrite recorded here is the one we | |
3203 | * want to compare with the last_overrun. | |
3204 | */ | |
3205 | smp_mb(); | |
3206 | overwrite = local_read(&(cpu_buffer->overrun)); | |
3207 | ||
77ae365e SR |
3208 | /* |
3209 | * Here's the tricky part. | |
3210 | * | |
3211 | * We need to move the pointer past the header page. | |
3212 | * But we can only do that if a writer is not currently | |
3213 | * moving it. The page before the header page has the | |
3214 | * flag bit '1' set if it is pointing to the page we want. | |
3215 | * but if the writer is in the process of moving it | |
3216 | * than it will be '2' or already moved '0'. | |
3217 | */ | |
3218 | ||
3219 | ret = rb_head_page_replace(reader, cpu_buffer->reader_page); | |
7a8e76a3 SR |
3220 | |
3221 | /* | |
77ae365e | 3222 | * If we did not convert it, then we must try again. |
7a8e76a3 | 3223 | */ |
77ae365e SR |
3224 | if (!ret) |
3225 | goto spin; | |
7a8e76a3 | 3226 | |
77ae365e SR |
3227 | /* |
3228 | * Yeah! We succeeded in replacing the page. | |
3229 | * | |
3230 | * Now make the new head point back to the reader page. | |
3231 | */ | |
5ded3dc6 | 3232 | rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list; |
77ae365e | 3233 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
d769041f SR |
3234 | |
3235 | /* Finally update the reader page to the new head */ | |
3236 | cpu_buffer->reader_page = reader; | |
3237 | rb_reset_reader_page(cpu_buffer); | |
3238 | ||
66a8cb95 SR |
3239 | if (overwrite != cpu_buffer->last_overrun) { |
3240 | cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun; | |
3241 | cpu_buffer->last_overrun = overwrite; | |
3242 | } | |
3243 | ||
d769041f SR |
3244 | goto again; |
3245 | ||
3246 | out: | |
0199c4e6 | 3247 | arch_spin_unlock(&cpu_buffer->lock); |
3e03fb7f | 3248 | local_irq_restore(flags); |
d769041f SR |
3249 | |
3250 | return reader; | |
3251 | } | |
3252 | ||
3253 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
3254 | { | |
3255 | struct ring_buffer_event *event; | |
3256 | struct buffer_page *reader; | |
3257 | unsigned length; | |
3258 | ||
3259 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 3260 | |
d769041f | 3261 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
3262 | if (RB_WARN_ON(cpu_buffer, !reader)) |
3263 | return; | |
7a8e76a3 | 3264 | |
d769041f SR |
3265 | event = rb_reader_event(cpu_buffer); |
3266 | ||
a1863c21 | 3267 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
e4906eff | 3268 | cpu_buffer->read++; |
d769041f SR |
3269 | |
3270 | rb_update_read_stamp(cpu_buffer, event); | |
3271 | ||
3272 | length = rb_event_length(event); | |
6f807acd | 3273 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
3274 | } |
3275 | ||
3276 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
3277 | { | |
7a8e76a3 SR |
3278 | struct ring_buffer_per_cpu *cpu_buffer; |
3279 | struct ring_buffer_event *event; | |
3280 | unsigned length; | |
3281 | ||
3282 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 SR |
3283 | |
3284 | /* | |
3285 | * Check if we are at the end of the buffer. | |
3286 | */ | |
bf41a158 | 3287 | if (iter->head >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
3288 | /* discarded commits can make the page empty */ |
3289 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 3290 | return; |
d769041f | 3291 | rb_inc_iter(iter); |
7a8e76a3 SR |
3292 | return; |
3293 | } | |
3294 | ||
3295 | event = rb_iter_head_event(iter); | |
3296 | ||
3297 | length = rb_event_length(event); | |
3298 | ||
3299 | /* | |
3300 | * This should not be called to advance the header if we are | |
3301 | * at the tail of the buffer. | |
3302 | */ | |
3e89c7bb | 3303 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 3304 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
3305 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
3306 | return; | |
7a8e76a3 SR |
3307 | |
3308 | rb_update_iter_read_stamp(iter, event); | |
3309 | ||
3310 | iter->head += length; | |
3311 | ||
3312 | /* check for end of page padding */ | |
bf41a158 SR |
3313 | if ((iter->head >= rb_page_size(iter->head_page)) && |
3314 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
3315 | rb_advance_iter(iter); |
3316 | } | |
3317 | ||
66a8cb95 SR |
3318 | static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer) |
3319 | { | |
3320 | return cpu_buffer->lost_events; | |
3321 | } | |
3322 | ||
f83c9d0f | 3323 | static struct ring_buffer_event * |
66a8cb95 SR |
3324 | rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts, |
3325 | unsigned long *lost_events) | |
7a8e76a3 | 3326 | { |
7a8e76a3 | 3327 | struct ring_buffer_event *event; |
d769041f | 3328 | struct buffer_page *reader; |
818e3dd3 | 3329 | int nr_loops = 0; |
7a8e76a3 | 3330 | |
7a8e76a3 | 3331 | again: |
818e3dd3 | 3332 | /* |
69d1b839 SR |
3333 | * We repeat when a time extend is encountered. |
3334 | * Since the time extend is always attached to a data event, | |
3335 | * we should never loop more than once. | |
3336 | * (We never hit the following condition more than twice). | |
818e3dd3 | 3337 | */ |
69d1b839 | 3338 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 3339 | return NULL; |
818e3dd3 | 3340 | |
d769041f SR |
3341 | reader = rb_get_reader_page(cpu_buffer); |
3342 | if (!reader) | |
7a8e76a3 SR |
3343 | return NULL; |
3344 | ||
d769041f | 3345 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 3346 | |
334d4169 | 3347 | switch (event->type_len) { |
7a8e76a3 | 3348 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3349 | if (rb_null_event(event)) |
3350 | RB_WARN_ON(cpu_buffer, 1); | |
3351 | /* | |
3352 | * Because the writer could be discarding every | |
3353 | * event it creates (which would probably be bad) | |
3354 | * if we were to go back to "again" then we may never | |
3355 | * catch up, and will trigger the warn on, or lock | |
3356 | * the box. Return the padding, and we will release | |
3357 | * the current locks, and try again. | |
3358 | */ | |
2d622719 | 3359 | return event; |
7a8e76a3 SR |
3360 | |
3361 | case RINGBUF_TYPE_TIME_EXTEND: | |
3362 | /* Internal data, OK to advance */ | |
d769041f | 3363 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3364 | goto again; |
3365 | ||
3366 | case RINGBUF_TYPE_TIME_STAMP: | |
3367 | /* FIXME: not implemented */ | |
d769041f | 3368 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3369 | goto again; |
3370 | ||
3371 | case RINGBUF_TYPE_DATA: | |
3372 | if (ts) { | |
3373 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
d8eeb2d3 | 3374 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, |
37886f6a | 3375 | cpu_buffer->cpu, ts); |
7a8e76a3 | 3376 | } |
66a8cb95 SR |
3377 | if (lost_events) |
3378 | *lost_events = rb_lost_events(cpu_buffer); | |
7a8e76a3 SR |
3379 | return event; |
3380 | ||
3381 | default: | |
3382 | BUG(); | |
3383 | } | |
3384 | ||
3385 | return NULL; | |
3386 | } | |
c4f50183 | 3387 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 3388 | |
f83c9d0f SR |
3389 | static struct ring_buffer_event * |
3390 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
3391 | { |
3392 | struct ring_buffer *buffer; | |
3393 | struct ring_buffer_per_cpu *cpu_buffer; | |
3394 | struct ring_buffer_event *event; | |
818e3dd3 | 3395 | int nr_loops = 0; |
7a8e76a3 | 3396 | |
7a8e76a3 SR |
3397 | cpu_buffer = iter->cpu_buffer; |
3398 | buffer = cpu_buffer->buffer; | |
3399 | ||
492a74f4 SR |
3400 | /* |
3401 | * Check if someone performed a consuming read to | |
3402 | * the buffer. A consuming read invalidates the iterator | |
3403 | * and we need to reset the iterator in this case. | |
3404 | */ | |
3405 | if (unlikely(iter->cache_read != cpu_buffer->read || | |
3406 | iter->cache_reader_page != cpu_buffer->reader_page)) | |
3407 | rb_iter_reset(iter); | |
3408 | ||
7a8e76a3 | 3409 | again: |
3c05d748 SR |
3410 | if (ring_buffer_iter_empty(iter)) |
3411 | return NULL; | |
3412 | ||
818e3dd3 | 3413 | /* |
69d1b839 SR |
3414 | * We repeat when a time extend is encountered. |
3415 | * Since the time extend is always attached to a data event, | |
3416 | * we should never loop more than once. | |
3417 | * (We never hit the following condition more than twice). | |
818e3dd3 | 3418 | */ |
69d1b839 | 3419 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 3420 | return NULL; |
818e3dd3 | 3421 | |
7a8e76a3 SR |
3422 | if (rb_per_cpu_empty(cpu_buffer)) |
3423 | return NULL; | |
3424 | ||
3c05d748 SR |
3425 | if (iter->head >= local_read(&iter->head_page->page->commit)) { |
3426 | rb_inc_iter(iter); | |
3427 | goto again; | |
3428 | } | |
3429 | ||
7a8e76a3 SR |
3430 | event = rb_iter_head_event(iter); |
3431 | ||
334d4169 | 3432 | switch (event->type_len) { |
7a8e76a3 | 3433 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3434 | if (rb_null_event(event)) { |
3435 | rb_inc_iter(iter); | |
3436 | goto again; | |
3437 | } | |
3438 | rb_advance_iter(iter); | |
3439 | return event; | |
7a8e76a3 SR |
3440 | |
3441 | case RINGBUF_TYPE_TIME_EXTEND: | |
3442 | /* Internal data, OK to advance */ | |
3443 | rb_advance_iter(iter); | |
3444 | goto again; | |
3445 | ||
3446 | case RINGBUF_TYPE_TIME_STAMP: | |
3447 | /* FIXME: not implemented */ | |
3448 | rb_advance_iter(iter); | |
3449 | goto again; | |
3450 | ||
3451 | case RINGBUF_TYPE_DATA: | |
3452 | if (ts) { | |
3453 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
3454 | ring_buffer_normalize_time_stamp(buffer, |
3455 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
3456 | } |
3457 | return event; | |
3458 | ||
3459 | default: | |
3460 | BUG(); | |
3461 | } | |
3462 | ||
3463 | return NULL; | |
3464 | } | |
c4f50183 | 3465 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 3466 | |
8d707e8e SR |
3467 | static inline int rb_ok_to_lock(void) |
3468 | { | |
3469 | /* | |
3470 | * If an NMI die dumps out the content of the ring buffer | |
3471 | * do not grab locks. We also permanently disable the ring | |
3472 | * buffer too. A one time deal is all you get from reading | |
3473 | * the ring buffer from an NMI. | |
3474 | */ | |
464e85eb | 3475 | if (likely(!in_nmi())) |
8d707e8e SR |
3476 | return 1; |
3477 | ||
3478 | tracing_off_permanent(); | |
3479 | return 0; | |
3480 | } | |
3481 | ||
f83c9d0f SR |
3482 | /** |
3483 | * ring_buffer_peek - peek at the next event to be read | |
3484 | * @buffer: The ring buffer to read | |
3485 | * @cpu: The cpu to peak at | |
3486 | * @ts: The timestamp counter of this event. | |
66a8cb95 | 3487 | * @lost_events: a variable to store if events were lost (may be NULL) |
f83c9d0f SR |
3488 | * |
3489 | * This will return the event that will be read next, but does | |
3490 | * not consume the data. | |
3491 | */ | |
3492 | struct ring_buffer_event * | |
66a8cb95 SR |
3493 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts, |
3494 | unsigned long *lost_events) | |
f83c9d0f SR |
3495 | { |
3496 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 3497 | struct ring_buffer_event *event; |
f83c9d0f | 3498 | unsigned long flags; |
8d707e8e | 3499 | int dolock; |
f83c9d0f | 3500 | |
554f786e | 3501 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3502 | return NULL; |
554f786e | 3503 | |
8d707e8e | 3504 | dolock = rb_ok_to_lock(); |
2d622719 | 3505 | again: |
8d707e8e SR |
3506 | local_irq_save(flags); |
3507 | if (dolock) | |
5389f6fa | 3508 | raw_spin_lock(&cpu_buffer->reader_lock); |
66a8cb95 | 3509 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
469535a5 RR |
3510 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
3511 | rb_advance_reader(cpu_buffer); | |
8d707e8e | 3512 | if (dolock) |
5389f6fa | 3513 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3514 | local_irq_restore(flags); |
f83c9d0f | 3515 | |
1b959e18 | 3516 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3517 | goto again; |
2d622719 | 3518 | |
f83c9d0f SR |
3519 | return event; |
3520 | } | |
3521 | ||
3522 | /** | |
3523 | * ring_buffer_iter_peek - peek at the next event to be read | |
3524 | * @iter: The ring buffer iterator | |
3525 | * @ts: The timestamp counter of this event. | |
3526 | * | |
3527 | * This will return the event that will be read next, but does | |
3528 | * not increment the iterator. | |
3529 | */ | |
3530 | struct ring_buffer_event * | |
3531 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
3532 | { | |
3533 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3534 | struct ring_buffer_event *event; | |
3535 | unsigned long flags; | |
3536 | ||
2d622719 | 3537 | again: |
5389f6fa | 3538 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3539 | event = rb_iter_peek(iter, ts); |
5389f6fa | 3540 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3541 | |
1b959e18 | 3542 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3543 | goto again; |
2d622719 | 3544 | |
f83c9d0f SR |
3545 | return event; |
3546 | } | |
3547 | ||
7a8e76a3 SR |
3548 | /** |
3549 | * ring_buffer_consume - return an event and consume it | |
3550 | * @buffer: The ring buffer to get the next event from | |
66a8cb95 SR |
3551 | * @cpu: the cpu to read the buffer from |
3552 | * @ts: a variable to store the timestamp (may be NULL) | |
3553 | * @lost_events: a variable to store if events were lost (may be NULL) | |
7a8e76a3 SR |
3554 | * |
3555 | * Returns the next event in the ring buffer, and that event is consumed. | |
3556 | * Meaning, that sequential reads will keep returning a different event, | |
3557 | * and eventually empty the ring buffer if the producer is slower. | |
3558 | */ | |
3559 | struct ring_buffer_event * | |
66a8cb95 SR |
3560 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts, |
3561 | unsigned long *lost_events) | |
7a8e76a3 | 3562 | { |
554f786e SR |
3563 | struct ring_buffer_per_cpu *cpu_buffer; |
3564 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 3565 | unsigned long flags; |
8d707e8e SR |
3566 | int dolock; |
3567 | ||
3568 | dolock = rb_ok_to_lock(); | |
7a8e76a3 | 3569 | |
2d622719 | 3570 | again: |
554f786e SR |
3571 | /* might be called in atomic */ |
3572 | preempt_disable(); | |
3573 | ||
9e01c1b7 | 3574 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 3575 | goto out; |
7a8e76a3 | 3576 | |
554f786e | 3577 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
3578 | local_irq_save(flags); |
3579 | if (dolock) | |
5389f6fa | 3580 | raw_spin_lock(&cpu_buffer->reader_lock); |
f83c9d0f | 3581 | |
66a8cb95 SR |
3582 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
3583 | if (event) { | |
3584 | cpu_buffer->lost_events = 0; | |
469535a5 | 3585 | rb_advance_reader(cpu_buffer); |
66a8cb95 | 3586 | } |
7a8e76a3 | 3587 | |
8d707e8e | 3588 | if (dolock) |
5389f6fa | 3589 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3590 | local_irq_restore(flags); |
f83c9d0f | 3591 | |
554f786e SR |
3592 | out: |
3593 | preempt_enable(); | |
3594 | ||
1b959e18 | 3595 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3596 | goto again; |
2d622719 | 3597 | |
7a8e76a3 SR |
3598 | return event; |
3599 | } | |
c4f50183 | 3600 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
3601 | |
3602 | /** | |
72c9ddfd | 3603 | * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer |
7a8e76a3 SR |
3604 | * @buffer: The ring buffer to read from |
3605 | * @cpu: The cpu buffer to iterate over | |
3606 | * | |
72c9ddfd DM |
3607 | * This performs the initial preparations necessary to iterate |
3608 | * through the buffer. Memory is allocated, buffer recording | |
3609 | * is disabled, and the iterator pointer is returned to the caller. | |
7a8e76a3 | 3610 | * |
72c9ddfd DM |
3611 | * Disabling buffer recordng prevents the reading from being |
3612 | * corrupted. This is not a consuming read, so a producer is not | |
3613 | * expected. | |
3614 | * | |
3615 | * After a sequence of ring_buffer_read_prepare calls, the user is | |
3616 | * expected to make at least one call to ring_buffer_prepare_sync. | |
3617 | * Afterwards, ring_buffer_read_start is invoked to get things going | |
3618 | * for real. | |
3619 | * | |
3620 | * This overall must be paired with ring_buffer_finish. | |
7a8e76a3 SR |
3621 | */ |
3622 | struct ring_buffer_iter * | |
72c9ddfd | 3623 | ring_buffer_read_prepare(struct ring_buffer *buffer, int cpu) |
7a8e76a3 SR |
3624 | { |
3625 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3626 | struct ring_buffer_iter *iter; |
7a8e76a3 | 3627 | |
9e01c1b7 | 3628 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3629 | return NULL; |
7a8e76a3 SR |
3630 | |
3631 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
3632 | if (!iter) | |
8aabee57 | 3633 | return NULL; |
7a8e76a3 SR |
3634 | |
3635 | cpu_buffer = buffer->buffers[cpu]; | |
3636 | ||
3637 | iter->cpu_buffer = cpu_buffer; | |
3638 | ||
83f40318 | 3639 | atomic_inc(&buffer->resize_disabled); |
7a8e76a3 | 3640 | atomic_inc(&cpu_buffer->record_disabled); |
72c9ddfd DM |
3641 | |
3642 | return iter; | |
3643 | } | |
3644 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare); | |
3645 | ||
3646 | /** | |
3647 | * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls | |
3648 | * | |
3649 | * All previously invoked ring_buffer_read_prepare calls to prepare | |
3650 | * iterators will be synchronized. Afterwards, read_buffer_read_start | |
3651 | * calls on those iterators are allowed. | |
3652 | */ | |
3653 | void | |
3654 | ring_buffer_read_prepare_sync(void) | |
3655 | { | |
7a8e76a3 | 3656 | synchronize_sched(); |
72c9ddfd DM |
3657 | } |
3658 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync); | |
3659 | ||
3660 | /** | |
3661 | * ring_buffer_read_start - start a non consuming read of the buffer | |
3662 | * @iter: The iterator returned by ring_buffer_read_prepare | |
3663 | * | |
3664 | * This finalizes the startup of an iteration through the buffer. | |
3665 | * The iterator comes from a call to ring_buffer_read_prepare and | |
3666 | * an intervening ring_buffer_read_prepare_sync must have been | |
3667 | * performed. | |
3668 | * | |
3669 | * Must be paired with ring_buffer_finish. | |
3670 | */ | |
3671 | void | |
3672 | ring_buffer_read_start(struct ring_buffer_iter *iter) | |
3673 | { | |
3674 | struct ring_buffer_per_cpu *cpu_buffer; | |
3675 | unsigned long flags; | |
3676 | ||
3677 | if (!iter) | |
3678 | return; | |
3679 | ||
3680 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 | 3681 | |
5389f6fa | 3682 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
0199c4e6 | 3683 | arch_spin_lock(&cpu_buffer->lock); |
642edba5 | 3684 | rb_iter_reset(iter); |
0199c4e6 | 3685 | arch_spin_unlock(&cpu_buffer->lock); |
5389f6fa | 3686 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 3687 | } |
c4f50183 | 3688 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
3689 | |
3690 | /** | |
3691 | * ring_buffer_finish - finish reading the iterator of the buffer | |
3692 | * @iter: The iterator retrieved by ring_buffer_start | |
3693 | * | |
3694 | * This re-enables the recording to the buffer, and frees the | |
3695 | * iterator. | |
3696 | */ | |
3697 | void | |
3698 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
3699 | { | |
3700 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3701 | ||
3702 | atomic_dec(&cpu_buffer->record_disabled); | |
83f40318 | 3703 | atomic_dec(&cpu_buffer->buffer->resize_disabled); |
7a8e76a3 SR |
3704 | kfree(iter); |
3705 | } | |
c4f50183 | 3706 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
3707 | |
3708 | /** | |
3709 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
3710 | * @iter: The ring buffer iterator | |
3711 | * @ts: The time stamp of the event read. | |
3712 | * | |
3713 | * This reads the next event in the ring buffer and increments the iterator. | |
3714 | */ | |
3715 | struct ring_buffer_event * | |
3716 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
3717 | { | |
3718 | struct ring_buffer_event *event; | |
f83c9d0f SR |
3719 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
3720 | unsigned long flags; | |
7a8e76a3 | 3721 | |
5389f6fa | 3722 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
7e9391cf | 3723 | again: |
f83c9d0f | 3724 | event = rb_iter_peek(iter, ts); |
7a8e76a3 | 3725 | if (!event) |
f83c9d0f | 3726 | goto out; |
7a8e76a3 | 3727 | |
7e9391cf SR |
3728 | if (event->type_len == RINGBUF_TYPE_PADDING) |
3729 | goto again; | |
3730 | ||
7a8e76a3 | 3731 | rb_advance_iter(iter); |
f83c9d0f | 3732 | out: |
5389f6fa | 3733 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
3734 | |
3735 | return event; | |
3736 | } | |
c4f50183 | 3737 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
3738 | |
3739 | /** | |
3740 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
3741 | * @buffer: The ring buffer. | |
3742 | */ | |
438ced17 | 3743 | unsigned long ring_buffer_size(struct ring_buffer *buffer, int cpu) |
7a8e76a3 | 3744 | { |
438ced17 VN |
3745 | /* |
3746 | * Earlier, this method returned | |
3747 | * BUF_PAGE_SIZE * buffer->nr_pages | |
3748 | * Since the nr_pages field is now removed, we have converted this to | |
3749 | * return the per cpu buffer value. | |
3750 | */ | |
3751 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3752 | return 0; | |
3753 | ||
3754 | return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages; | |
7a8e76a3 | 3755 | } |
c4f50183 | 3756 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
3757 | |
3758 | static void | |
3759 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
3760 | { | |
77ae365e SR |
3761 | rb_head_page_deactivate(cpu_buffer); |
3762 | ||
7a8e76a3 | 3763 | cpu_buffer->head_page |
3adc54fa | 3764 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 3765 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 3766 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 3767 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 3768 | |
6f807acd | 3769 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
3770 | |
3771 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
3772 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
3773 | ||
3774 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
3775 | local_set(&cpu_buffer->reader_page->write, 0); | |
778c55d4 | 3776 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 3777 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 3778 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 3779 | |
77ae365e | 3780 | local_set(&cpu_buffer->commit_overrun, 0); |
c64e148a | 3781 | local_set(&cpu_buffer->entries_bytes, 0); |
77ae365e | 3782 | local_set(&cpu_buffer->overrun, 0); |
e4906eff | 3783 | local_set(&cpu_buffer->entries, 0); |
fa743953 SR |
3784 | local_set(&cpu_buffer->committing, 0); |
3785 | local_set(&cpu_buffer->commits, 0); | |
77ae365e | 3786 | cpu_buffer->read = 0; |
c64e148a | 3787 | cpu_buffer->read_bytes = 0; |
69507c06 SR |
3788 | |
3789 | cpu_buffer->write_stamp = 0; | |
3790 | cpu_buffer->read_stamp = 0; | |
77ae365e | 3791 | |
66a8cb95 SR |
3792 | cpu_buffer->lost_events = 0; |
3793 | cpu_buffer->last_overrun = 0; | |
3794 | ||
77ae365e | 3795 | rb_head_page_activate(cpu_buffer); |
7a8e76a3 SR |
3796 | } |
3797 | ||
3798 | /** | |
3799 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
3800 | * @buffer: The ring buffer to reset a per cpu buffer of | |
3801 | * @cpu: The CPU buffer to be reset | |
3802 | */ | |
3803 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
3804 | { | |
3805 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
3806 | unsigned long flags; | |
3807 | ||
9e01c1b7 | 3808 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3809 | return; |
7a8e76a3 | 3810 | |
83f40318 | 3811 | atomic_inc(&buffer->resize_disabled); |
41ede23e SR |
3812 | atomic_inc(&cpu_buffer->record_disabled); |
3813 | ||
83f40318 VN |
3814 | /* Make sure all commits have finished */ |
3815 | synchronize_sched(); | |
3816 | ||
5389f6fa | 3817 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3818 | |
41b6a95d SR |
3819 | if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing))) |
3820 | goto out; | |
3821 | ||
0199c4e6 | 3822 | arch_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
3823 | |
3824 | rb_reset_cpu(cpu_buffer); | |
3825 | ||
0199c4e6 | 3826 | arch_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 3827 | |
41b6a95d | 3828 | out: |
5389f6fa | 3829 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
41ede23e SR |
3830 | |
3831 | atomic_dec(&cpu_buffer->record_disabled); | |
83f40318 | 3832 | atomic_dec(&buffer->resize_disabled); |
7a8e76a3 | 3833 | } |
c4f50183 | 3834 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
3835 | |
3836 | /** | |
3837 | * ring_buffer_reset - reset a ring buffer | |
3838 | * @buffer: The ring buffer to reset all cpu buffers | |
3839 | */ | |
3840 | void ring_buffer_reset(struct ring_buffer *buffer) | |
3841 | { | |
7a8e76a3 SR |
3842 | int cpu; |
3843 | ||
7a8e76a3 | 3844 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 3845 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 3846 | } |
c4f50183 | 3847 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
3848 | |
3849 | /** | |
3850 | * rind_buffer_empty - is the ring buffer empty? | |
3851 | * @buffer: The ring buffer to test | |
3852 | */ | |
3853 | int ring_buffer_empty(struct ring_buffer *buffer) | |
3854 | { | |
3855 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 3856 | unsigned long flags; |
8d707e8e | 3857 | int dolock; |
7a8e76a3 | 3858 | int cpu; |
d4788207 | 3859 | int ret; |
7a8e76a3 | 3860 | |
8d707e8e | 3861 | dolock = rb_ok_to_lock(); |
7a8e76a3 SR |
3862 | |
3863 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
3864 | for_each_buffer_cpu(buffer, cpu) { | |
3865 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e SR |
3866 | local_irq_save(flags); |
3867 | if (dolock) | |
5389f6fa | 3868 | raw_spin_lock(&cpu_buffer->reader_lock); |
d4788207 | 3869 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e | 3870 | if (dolock) |
5389f6fa | 3871 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e SR |
3872 | local_irq_restore(flags); |
3873 | ||
d4788207 | 3874 | if (!ret) |
7a8e76a3 SR |
3875 | return 0; |
3876 | } | |
554f786e | 3877 | |
7a8e76a3 SR |
3878 | return 1; |
3879 | } | |
c4f50183 | 3880 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
3881 | |
3882 | /** | |
3883 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
3884 | * @buffer: The ring buffer | |
3885 | * @cpu: The CPU buffer to test | |
3886 | */ | |
3887 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
3888 | { | |
3889 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 3890 | unsigned long flags; |
8d707e8e | 3891 | int dolock; |
8aabee57 | 3892 | int ret; |
7a8e76a3 | 3893 | |
9e01c1b7 | 3894 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3895 | return 1; |
7a8e76a3 | 3896 | |
8d707e8e SR |
3897 | dolock = rb_ok_to_lock(); |
3898 | ||
7a8e76a3 | 3899 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
3900 | local_irq_save(flags); |
3901 | if (dolock) | |
5389f6fa | 3902 | raw_spin_lock(&cpu_buffer->reader_lock); |
554f786e | 3903 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e | 3904 | if (dolock) |
5389f6fa | 3905 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3906 | local_irq_restore(flags); |
554f786e SR |
3907 | |
3908 | return ret; | |
7a8e76a3 | 3909 | } |
c4f50183 | 3910 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 | 3911 | |
85bac32c | 3912 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
7a8e76a3 SR |
3913 | /** |
3914 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
3915 | * @buffer_a: One buffer to swap with | |
3916 | * @buffer_b: The other buffer to swap with | |
3917 | * | |
3918 | * This function is useful for tracers that want to take a "snapshot" | |
3919 | * of a CPU buffer and has another back up buffer lying around. | |
3920 | * it is expected that the tracer handles the cpu buffer not being | |
3921 | * used at the moment. | |
3922 | */ | |
3923 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
3924 | struct ring_buffer *buffer_b, int cpu) | |
3925 | { | |
3926 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
3927 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
3928 | int ret = -EINVAL; |
3929 | ||
9e01c1b7 RR |
3930 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
3931 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 3932 | goto out; |
7a8e76a3 | 3933 | |
438ced17 VN |
3934 | cpu_buffer_a = buffer_a->buffers[cpu]; |
3935 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
3936 | ||
7a8e76a3 | 3937 | /* At least make sure the two buffers are somewhat the same */ |
438ced17 | 3938 | if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages) |
554f786e SR |
3939 | goto out; |
3940 | ||
3941 | ret = -EAGAIN; | |
7a8e76a3 | 3942 | |
97b17efe | 3943 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 3944 | goto out; |
97b17efe SR |
3945 | |
3946 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 3947 | goto out; |
97b17efe SR |
3948 | |
3949 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 3950 | goto out; |
97b17efe | 3951 | |
97b17efe | 3952 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 3953 | goto out; |
97b17efe SR |
3954 | |
3955 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 3956 | goto out; |
97b17efe | 3957 | |
7a8e76a3 SR |
3958 | /* |
3959 | * We can't do a synchronize_sched here because this | |
3960 | * function can be called in atomic context. | |
3961 | * Normally this will be called from the same CPU as cpu. | |
3962 | * If not it's up to the caller to protect this. | |
3963 | */ | |
3964 | atomic_inc(&cpu_buffer_a->record_disabled); | |
3965 | atomic_inc(&cpu_buffer_b->record_disabled); | |
3966 | ||
98277991 SR |
3967 | ret = -EBUSY; |
3968 | if (local_read(&cpu_buffer_a->committing)) | |
3969 | goto out_dec; | |
3970 | if (local_read(&cpu_buffer_b->committing)) | |
3971 | goto out_dec; | |
3972 | ||
7a8e76a3 SR |
3973 | buffer_a->buffers[cpu] = cpu_buffer_b; |
3974 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
3975 | ||
3976 | cpu_buffer_b->buffer = buffer_a; | |
3977 | cpu_buffer_a->buffer = buffer_b; | |
3978 | ||
98277991 SR |
3979 | ret = 0; |
3980 | ||
3981 | out_dec: | |
7a8e76a3 SR |
3982 | atomic_dec(&cpu_buffer_a->record_disabled); |
3983 | atomic_dec(&cpu_buffer_b->record_disabled); | |
554f786e | 3984 | out: |
554f786e | 3985 | return ret; |
7a8e76a3 | 3986 | } |
c4f50183 | 3987 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
85bac32c | 3988 | #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */ |
7a8e76a3 | 3989 | |
8789a9e7 SR |
3990 | /** |
3991 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
3992 | * @buffer: the buffer to allocate for. | |
3993 | * | |
3994 | * This function is used in conjunction with ring_buffer_read_page. | |
3995 | * When reading a full page from the ring buffer, these functions | |
3996 | * can be used to speed up the process. The calling function should | |
3997 | * allocate a few pages first with this function. Then when it | |
3998 | * needs to get pages from the ring buffer, it passes the result | |
3999 | * of this function into ring_buffer_read_page, which will swap | |
4000 | * the page that was allocated, with the read page of the buffer. | |
4001 | * | |
4002 | * Returns: | |
4003 | * The page allocated, or NULL on error. | |
4004 | */ | |
7ea59064 | 4005 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer, int cpu) |
8789a9e7 | 4006 | { |
044fa782 | 4007 | struct buffer_data_page *bpage; |
7ea59064 | 4008 | struct page *page; |
8789a9e7 | 4009 | |
d7ec4bfe VN |
4010 | page = alloc_pages_node(cpu_to_node(cpu), |
4011 | GFP_KERNEL | __GFP_NORETRY, 0); | |
7ea59064 | 4012 | if (!page) |
8789a9e7 SR |
4013 | return NULL; |
4014 | ||
7ea59064 | 4015 | bpage = page_address(page); |
8789a9e7 | 4016 | |
ef7a4a16 SR |
4017 | rb_init_page(bpage); |
4018 | ||
044fa782 | 4019 | return bpage; |
8789a9e7 | 4020 | } |
d6ce96da | 4021 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
4022 | |
4023 | /** | |
4024 | * ring_buffer_free_read_page - free an allocated read page | |
4025 | * @buffer: the buffer the page was allocate for | |
4026 | * @data: the page to free | |
4027 | * | |
4028 | * Free a page allocated from ring_buffer_alloc_read_page. | |
4029 | */ | |
4030 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
4031 | { | |
4032 | free_page((unsigned long)data); | |
4033 | } | |
d6ce96da | 4034 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
4035 | |
4036 | /** | |
4037 | * ring_buffer_read_page - extract a page from the ring buffer | |
4038 | * @buffer: buffer to extract from | |
4039 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 4040 | * @len: amount to extract |
8789a9e7 SR |
4041 | * @cpu: the cpu of the buffer to extract |
4042 | * @full: should the extraction only happen when the page is full. | |
4043 | * | |
4044 | * This function will pull out a page from the ring buffer and consume it. | |
4045 | * @data_page must be the address of the variable that was returned | |
4046 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
4047 | * to swap with a page in the ring buffer. | |
4048 | * | |
4049 | * for example: | |
b85fa01e | 4050 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
4051 | * if (!rpage) |
4052 | * return error; | |
ef7a4a16 | 4053 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
4054 | * if (ret >= 0) |
4055 | * process_page(rpage, ret); | |
8789a9e7 SR |
4056 | * |
4057 | * When @full is set, the function will not return true unless | |
4058 | * the writer is off the reader page. | |
4059 | * | |
4060 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
4061 | * The ring buffer can be used anywhere in the kernel and can not | |
4062 | * blindly call wake_up. The layer that uses the ring buffer must be | |
4063 | * responsible for that. | |
4064 | * | |
4065 | * Returns: | |
667d2412 LJ |
4066 | * >=0 if data has been transferred, returns the offset of consumed data. |
4067 | * <0 if no data has been transferred. | |
8789a9e7 SR |
4068 | */ |
4069 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 4070 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
4071 | { |
4072 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
4073 | struct ring_buffer_event *event; | |
044fa782 | 4074 | struct buffer_data_page *bpage; |
ef7a4a16 | 4075 | struct buffer_page *reader; |
ff0ff84a | 4076 | unsigned long missed_events; |
8789a9e7 | 4077 | unsigned long flags; |
ef7a4a16 | 4078 | unsigned int commit; |
667d2412 | 4079 | unsigned int read; |
4f3640f8 | 4080 | u64 save_timestamp; |
667d2412 | 4081 | int ret = -1; |
8789a9e7 | 4082 | |
554f786e SR |
4083 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
4084 | goto out; | |
4085 | ||
474d32b6 SR |
4086 | /* |
4087 | * If len is not big enough to hold the page header, then | |
4088 | * we can not copy anything. | |
4089 | */ | |
4090 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 4091 | goto out; |
474d32b6 SR |
4092 | |
4093 | len -= BUF_PAGE_HDR_SIZE; | |
4094 | ||
8789a9e7 | 4095 | if (!data_page) |
554f786e | 4096 | goto out; |
8789a9e7 | 4097 | |
044fa782 SR |
4098 | bpage = *data_page; |
4099 | if (!bpage) | |
554f786e | 4100 | goto out; |
8789a9e7 | 4101 | |
5389f6fa | 4102 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4103 | |
ef7a4a16 SR |
4104 | reader = rb_get_reader_page(cpu_buffer); |
4105 | if (!reader) | |
554f786e | 4106 | goto out_unlock; |
8789a9e7 | 4107 | |
ef7a4a16 SR |
4108 | event = rb_reader_event(cpu_buffer); |
4109 | ||
4110 | read = reader->read; | |
4111 | commit = rb_page_commit(reader); | |
667d2412 | 4112 | |
66a8cb95 | 4113 | /* Check if any events were dropped */ |
ff0ff84a | 4114 | missed_events = cpu_buffer->lost_events; |
66a8cb95 | 4115 | |
8789a9e7 | 4116 | /* |
474d32b6 SR |
4117 | * If this page has been partially read or |
4118 | * if len is not big enough to read the rest of the page or | |
4119 | * a writer is still on the page, then | |
4120 | * we must copy the data from the page to the buffer. | |
4121 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 4122 | */ |
474d32b6 | 4123 | if (read || (len < (commit - read)) || |
ef7a4a16 | 4124 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 4125 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
4126 | unsigned int rpos = read; |
4127 | unsigned int pos = 0; | |
ef7a4a16 | 4128 | unsigned int size; |
8789a9e7 SR |
4129 | |
4130 | if (full) | |
554f786e | 4131 | goto out_unlock; |
8789a9e7 | 4132 | |
ef7a4a16 SR |
4133 | if (len > (commit - read)) |
4134 | len = (commit - read); | |
4135 | ||
69d1b839 SR |
4136 | /* Always keep the time extend and data together */ |
4137 | size = rb_event_ts_length(event); | |
ef7a4a16 SR |
4138 | |
4139 | if (len < size) | |
554f786e | 4140 | goto out_unlock; |
ef7a4a16 | 4141 | |
4f3640f8 SR |
4142 | /* save the current timestamp, since the user will need it */ |
4143 | save_timestamp = cpu_buffer->read_stamp; | |
4144 | ||
ef7a4a16 SR |
4145 | /* Need to copy one event at a time */ |
4146 | do { | |
e1e35927 DS |
4147 | /* We need the size of one event, because |
4148 | * rb_advance_reader only advances by one event, | |
4149 | * whereas rb_event_ts_length may include the size of | |
4150 | * one or two events. | |
4151 | * We have already ensured there's enough space if this | |
4152 | * is a time extend. */ | |
4153 | size = rb_event_length(event); | |
474d32b6 | 4154 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
4155 | |
4156 | len -= size; | |
4157 | ||
4158 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
4159 | rpos = reader->read; |
4160 | pos += size; | |
ef7a4a16 | 4161 | |
18fab912 HY |
4162 | if (rpos >= commit) |
4163 | break; | |
4164 | ||
ef7a4a16 | 4165 | event = rb_reader_event(cpu_buffer); |
69d1b839 SR |
4166 | /* Always keep the time extend and data together */ |
4167 | size = rb_event_ts_length(event); | |
e1e35927 | 4168 | } while (len >= size); |
667d2412 LJ |
4169 | |
4170 | /* update bpage */ | |
ef7a4a16 | 4171 | local_set(&bpage->commit, pos); |
4f3640f8 | 4172 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 4173 | |
474d32b6 SR |
4174 | /* we copied everything to the beginning */ |
4175 | read = 0; | |
8789a9e7 | 4176 | } else { |
afbab76a | 4177 | /* update the entry counter */ |
77ae365e | 4178 | cpu_buffer->read += rb_page_entries(reader); |
c64e148a | 4179 | cpu_buffer->read_bytes += BUF_PAGE_SIZE; |
afbab76a | 4180 | |
8789a9e7 | 4181 | /* swap the pages */ |
044fa782 | 4182 | rb_init_page(bpage); |
ef7a4a16 SR |
4183 | bpage = reader->page; |
4184 | reader->page = *data_page; | |
4185 | local_set(&reader->write, 0); | |
778c55d4 | 4186 | local_set(&reader->entries, 0); |
ef7a4a16 | 4187 | reader->read = 0; |
044fa782 | 4188 | *data_page = bpage; |
ff0ff84a SR |
4189 | |
4190 | /* | |
4191 | * Use the real_end for the data size, | |
4192 | * This gives us a chance to store the lost events | |
4193 | * on the page. | |
4194 | */ | |
4195 | if (reader->real_end) | |
4196 | local_set(&bpage->commit, reader->real_end); | |
8789a9e7 | 4197 | } |
667d2412 | 4198 | ret = read; |
8789a9e7 | 4199 | |
66a8cb95 | 4200 | cpu_buffer->lost_events = 0; |
2711ca23 SR |
4201 | |
4202 | commit = local_read(&bpage->commit); | |
66a8cb95 SR |
4203 | /* |
4204 | * Set a flag in the commit field if we lost events | |
4205 | */ | |
ff0ff84a | 4206 | if (missed_events) { |
ff0ff84a SR |
4207 | /* If there is room at the end of the page to save the |
4208 | * missed events, then record it there. | |
4209 | */ | |
4210 | if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) { | |
4211 | memcpy(&bpage->data[commit], &missed_events, | |
4212 | sizeof(missed_events)); | |
4213 | local_add(RB_MISSED_STORED, &bpage->commit); | |
2711ca23 | 4214 | commit += sizeof(missed_events); |
ff0ff84a | 4215 | } |
66a8cb95 | 4216 | local_add(RB_MISSED_EVENTS, &bpage->commit); |
ff0ff84a | 4217 | } |
66a8cb95 | 4218 | |
2711ca23 SR |
4219 | /* |
4220 | * This page may be off to user land. Zero it out here. | |
4221 | */ | |
4222 | if (commit < BUF_PAGE_SIZE) | |
4223 | memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit); | |
4224 | ||
554f786e | 4225 | out_unlock: |
5389f6fa | 4226 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4227 | |
554f786e | 4228 | out: |
8789a9e7 SR |
4229 | return ret; |
4230 | } | |
d6ce96da | 4231 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 4232 | |
59222efe | 4233 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
4234 | static int rb_cpu_notify(struct notifier_block *self, |
4235 | unsigned long action, void *hcpu) | |
554f786e SR |
4236 | { |
4237 | struct ring_buffer *buffer = | |
4238 | container_of(self, struct ring_buffer, cpu_notify); | |
4239 | long cpu = (long)hcpu; | |
438ced17 VN |
4240 | int cpu_i, nr_pages_same; |
4241 | unsigned int nr_pages; | |
554f786e SR |
4242 | |
4243 | switch (action) { | |
4244 | case CPU_UP_PREPARE: | |
4245 | case CPU_UP_PREPARE_FROZEN: | |
3f237a79 | 4246 | if (cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e SR |
4247 | return NOTIFY_OK; |
4248 | ||
438ced17 VN |
4249 | nr_pages = 0; |
4250 | nr_pages_same = 1; | |
4251 | /* check if all cpu sizes are same */ | |
4252 | for_each_buffer_cpu(buffer, cpu_i) { | |
4253 | /* fill in the size from first enabled cpu */ | |
4254 | if (nr_pages == 0) | |
4255 | nr_pages = buffer->buffers[cpu_i]->nr_pages; | |
4256 | if (nr_pages != buffer->buffers[cpu_i]->nr_pages) { | |
4257 | nr_pages_same = 0; | |
4258 | break; | |
4259 | } | |
4260 | } | |
4261 | /* allocate minimum pages, user can later expand it */ | |
4262 | if (!nr_pages_same) | |
4263 | nr_pages = 2; | |
554f786e | 4264 | buffer->buffers[cpu] = |
438ced17 | 4265 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
554f786e SR |
4266 | if (!buffer->buffers[cpu]) { |
4267 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
4268 | cpu); | |
4269 | return NOTIFY_OK; | |
4270 | } | |
4271 | smp_wmb(); | |
3f237a79 | 4272 | cpumask_set_cpu(cpu, buffer->cpumask); |
554f786e SR |
4273 | break; |
4274 | case CPU_DOWN_PREPARE: | |
4275 | case CPU_DOWN_PREPARE_FROZEN: | |
4276 | /* | |
4277 | * Do nothing. | |
4278 | * If we were to free the buffer, then the user would | |
4279 | * lose any trace that was in the buffer. | |
4280 | */ | |
4281 | break; | |
4282 | default: | |
4283 | break; | |
4284 | } | |
4285 | return NOTIFY_OK; | |
4286 | } | |
4287 | #endif |