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 | |
308f7eeb SR |
948 | /* Reset the head page if it exists */ |
949 | if (cpu_buffer->head_page) | |
950 | rb_set_head_page(cpu_buffer); | |
951 | ||
77ae365e SR |
952 | rb_head_page_deactivate(cpu_buffer); |
953 | ||
3e89c7bb SR |
954 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
955 | return -1; | |
956 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
957 | return -1; | |
7a8e76a3 | 958 | |
77ae365e SR |
959 | if (rb_check_list(cpu_buffer, head)) |
960 | return -1; | |
961 | ||
044fa782 | 962 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 963 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 964 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
965 | return -1; |
966 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 967 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 968 | return -1; |
77ae365e SR |
969 | if (rb_check_list(cpu_buffer, &bpage->list)) |
970 | return -1; | |
7a8e76a3 SR |
971 | } |
972 | ||
77ae365e SR |
973 | rb_head_page_activate(cpu_buffer); |
974 | ||
7a8e76a3 SR |
975 | return 0; |
976 | } | |
977 | ||
438ced17 | 978 | static int __rb_allocate_pages(int nr_pages, struct list_head *pages, int cpu) |
7a8e76a3 | 979 | { |
438ced17 | 980 | int i; |
044fa782 | 981 | struct buffer_page *bpage, *tmp; |
3adc54fa | 982 | |
7a8e76a3 | 983 | for (i = 0; i < nr_pages; i++) { |
7ea59064 | 984 | struct page *page; |
d7ec4bfe VN |
985 | /* |
986 | * __GFP_NORETRY flag makes sure that the allocation fails | |
987 | * gracefully without invoking oom-killer and the system is | |
988 | * not destabilized. | |
989 | */ | |
044fa782 | 990 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
d7ec4bfe | 991 | GFP_KERNEL | __GFP_NORETRY, |
438ced17 | 992 | cpu_to_node(cpu)); |
044fa782 | 993 | if (!bpage) |
e4c2ce82 | 994 | goto free_pages; |
77ae365e | 995 | |
438ced17 | 996 | list_add(&bpage->list, pages); |
77ae365e | 997 | |
438ced17 | 998 | page = alloc_pages_node(cpu_to_node(cpu), |
d7ec4bfe | 999 | GFP_KERNEL | __GFP_NORETRY, 0); |
7ea59064 | 1000 | if (!page) |
7a8e76a3 | 1001 | goto free_pages; |
7ea59064 | 1002 | bpage->page = page_address(page); |
044fa782 | 1003 | rb_init_page(bpage->page); |
7a8e76a3 SR |
1004 | } |
1005 | ||
438ced17 VN |
1006 | return 0; |
1007 | ||
1008 | free_pages: | |
1009 | list_for_each_entry_safe(bpage, tmp, pages, list) { | |
1010 | list_del_init(&bpage->list); | |
1011 | free_buffer_page(bpage); | |
1012 | } | |
1013 | ||
1014 | return -ENOMEM; | |
1015 | } | |
1016 | ||
1017 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
1018 | unsigned nr_pages) | |
1019 | { | |
1020 | LIST_HEAD(pages); | |
1021 | ||
1022 | WARN_ON(!nr_pages); | |
1023 | ||
1024 | if (__rb_allocate_pages(nr_pages, &pages, cpu_buffer->cpu)) | |
1025 | return -ENOMEM; | |
1026 | ||
3adc54fa SR |
1027 | /* |
1028 | * The ring buffer page list is a circular list that does not | |
1029 | * start and end with a list head. All page list items point to | |
1030 | * other pages. | |
1031 | */ | |
1032 | cpu_buffer->pages = pages.next; | |
1033 | list_del(&pages); | |
7a8e76a3 | 1034 | |
438ced17 VN |
1035 | cpu_buffer->nr_pages = nr_pages; |
1036 | ||
7a8e76a3 SR |
1037 | rb_check_pages(cpu_buffer); |
1038 | ||
1039 | return 0; | |
7a8e76a3 SR |
1040 | } |
1041 | ||
1042 | static struct ring_buffer_per_cpu * | |
438ced17 | 1043 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int nr_pages, int cpu) |
7a8e76a3 SR |
1044 | { |
1045 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 1046 | struct buffer_page *bpage; |
7ea59064 | 1047 | struct page *page; |
7a8e76a3 SR |
1048 | int ret; |
1049 | ||
1050 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
1051 | GFP_KERNEL, cpu_to_node(cpu)); | |
1052 | if (!cpu_buffer) | |
1053 | return NULL; | |
1054 | ||
1055 | cpu_buffer->cpu = cpu; | |
1056 | cpu_buffer->buffer = buffer; | |
5389f6fa | 1057 | raw_spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 1058 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
edc35bd7 | 1059 | cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
83f40318 VN |
1060 | INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler); |
1061 | init_completion(&cpu_buffer->update_completion); | |
7a8e76a3 | 1062 | |
044fa782 | 1063 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 1064 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 1065 | if (!bpage) |
e4c2ce82 SR |
1066 | goto fail_free_buffer; |
1067 | ||
77ae365e SR |
1068 | rb_check_bpage(cpu_buffer, bpage); |
1069 | ||
044fa782 | 1070 | cpu_buffer->reader_page = bpage; |
7ea59064 VN |
1071 | page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0); |
1072 | if (!page) | |
e4c2ce82 | 1073 | goto fail_free_reader; |
7ea59064 | 1074 | bpage->page = page_address(page); |
044fa782 | 1075 | rb_init_page(bpage->page); |
e4c2ce82 | 1076 | |
d769041f | 1077 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
d769041f | 1078 | |
438ced17 | 1079 | ret = rb_allocate_pages(cpu_buffer, nr_pages); |
7a8e76a3 | 1080 | if (ret < 0) |
d769041f | 1081 | goto fail_free_reader; |
7a8e76a3 SR |
1082 | |
1083 | cpu_buffer->head_page | |
3adc54fa | 1084 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 1085 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 | 1086 | |
77ae365e SR |
1087 | rb_head_page_activate(cpu_buffer); |
1088 | ||
7a8e76a3 SR |
1089 | return cpu_buffer; |
1090 | ||
d769041f SR |
1091 | fail_free_reader: |
1092 | free_buffer_page(cpu_buffer->reader_page); | |
1093 | ||
7a8e76a3 SR |
1094 | fail_free_buffer: |
1095 | kfree(cpu_buffer); | |
1096 | return NULL; | |
1097 | } | |
1098 | ||
1099 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
1100 | { | |
3adc54fa | 1101 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1102 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1103 | |
d769041f SR |
1104 | free_buffer_page(cpu_buffer->reader_page); |
1105 | ||
77ae365e SR |
1106 | rb_head_page_deactivate(cpu_buffer); |
1107 | ||
3adc54fa SR |
1108 | if (head) { |
1109 | list_for_each_entry_safe(bpage, tmp, head, list) { | |
1110 | list_del_init(&bpage->list); | |
1111 | free_buffer_page(bpage); | |
1112 | } | |
1113 | bpage = list_entry(head, struct buffer_page, list); | |
044fa782 | 1114 | free_buffer_page(bpage); |
7a8e76a3 | 1115 | } |
3adc54fa | 1116 | |
7a8e76a3 SR |
1117 | kfree(cpu_buffer); |
1118 | } | |
1119 | ||
59222efe | 1120 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
1121 | static int rb_cpu_notify(struct notifier_block *self, |
1122 | unsigned long action, void *hcpu); | |
554f786e SR |
1123 | #endif |
1124 | ||
7a8e76a3 SR |
1125 | /** |
1126 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 1127 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
1128 | * @flags: attributes to set for the ring buffer. |
1129 | * | |
1130 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
1131 | * flag. This flag means that the buffer will overwrite old data | |
1132 | * when the buffer wraps. If this flag is not set, the buffer will | |
1133 | * drop data when the tail hits the head. | |
1134 | */ | |
1f8a6a10 PZ |
1135 | struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
1136 | struct lock_class_key *key) | |
7a8e76a3 SR |
1137 | { |
1138 | struct ring_buffer *buffer; | |
1139 | int bsize; | |
438ced17 | 1140 | int cpu, nr_pages; |
7a8e76a3 SR |
1141 | |
1142 | /* keep it in its own cache line */ | |
1143 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
1144 | GFP_KERNEL); | |
1145 | if (!buffer) | |
1146 | return NULL; | |
1147 | ||
9e01c1b7 RR |
1148 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
1149 | goto fail_free_buffer; | |
1150 | ||
438ced17 | 1151 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
7a8e76a3 | 1152 | buffer->flags = flags; |
37886f6a | 1153 | buffer->clock = trace_clock_local; |
1f8a6a10 | 1154 | buffer->reader_lock_key = key; |
7a8e76a3 SR |
1155 | |
1156 | /* need at least two pages */ | |
438ced17 VN |
1157 | if (nr_pages < 2) |
1158 | nr_pages = 2; | |
7a8e76a3 | 1159 | |
3bf832ce FW |
1160 | /* |
1161 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
1162 | * in early initcall, it will not be notified of secondary cpus. | |
1163 | * In that off case, we need to allocate for all possible cpus. | |
1164 | */ | |
1165 | #ifdef CONFIG_HOTPLUG_CPU | |
554f786e SR |
1166 | get_online_cpus(); |
1167 | cpumask_copy(buffer->cpumask, cpu_online_mask); | |
3bf832ce FW |
1168 | #else |
1169 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
1170 | #endif | |
7a8e76a3 SR |
1171 | buffer->cpus = nr_cpu_ids; |
1172 | ||
1173 | bsize = sizeof(void *) * nr_cpu_ids; | |
1174 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
1175 | GFP_KERNEL); | |
1176 | if (!buffer->buffers) | |
9e01c1b7 | 1177 | goto fail_free_cpumask; |
7a8e76a3 SR |
1178 | |
1179 | for_each_buffer_cpu(buffer, cpu) { | |
1180 | buffer->buffers[cpu] = | |
438ced17 | 1181 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
7a8e76a3 SR |
1182 | if (!buffer->buffers[cpu]) |
1183 | goto fail_free_buffers; | |
1184 | } | |
1185 | ||
59222efe | 1186 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1187 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
1188 | buffer->cpu_notify.priority = 0; | |
1189 | register_cpu_notifier(&buffer->cpu_notify); | |
1190 | #endif | |
1191 | ||
1192 | put_online_cpus(); | |
7a8e76a3 SR |
1193 | mutex_init(&buffer->mutex); |
1194 | ||
1195 | return buffer; | |
1196 | ||
1197 | fail_free_buffers: | |
1198 | for_each_buffer_cpu(buffer, cpu) { | |
1199 | if (buffer->buffers[cpu]) | |
1200 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1201 | } | |
1202 | kfree(buffer->buffers); | |
1203 | ||
9e01c1b7 RR |
1204 | fail_free_cpumask: |
1205 | free_cpumask_var(buffer->cpumask); | |
554f786e | 1206 | put_online_cpus(); |
9e01c1b7 | 1207 | |
7a8e76a3 SR |
1208 | fail_free_buffer: |
1209 | kfree(buffer); | |
1210 | return NULL; | |
1211 | } | |
1f8a6a10 | 1212 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
1213 | |
1214 | /** | |
1215 | * ring_buffer_free - free a ring buffer. | |
1216 | * @buffer: the buffer to free. | |
1217 | */ | |
1218 | void | |
1219 | ring_buffer_free(struct ring_buffer *buffer) | |
1220 | { | |
1221 | int cpu; | |
1222 | ||
554f786e SR |
1223 | get_online_cpus(); |
1224 | ||
59222efe | 1225 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1226 | unregister_cpu_notifier(&buffer->cpu_notify); |
1227 | #endif | |
1228 | ||
7a8e76a3 SR |
1229 | for_each_buffer_cpu(buffer, cpu) |
1230 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1231 | ||
554f786e SR |
1232 | put_online_cpus(); |
1233 | ||
bd3f0221 | 1234 | kfree(buffer->buffers); |
9e01c1b7 RR |
1235 | free_cpumask_var(buffer->cpumask); |
1236 | ||
7a8e76a3 SR |
1237 | kfree(buffer); |
1238 | } | |
c4f50183 | 1239 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 1240 | |
37886f6a SR |
1241 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
1242 | u64 (*clock)(void)) | |
1243 | { | |
1244 | buffer->clock = clock; | |
1245 | } | |
1246 | ||
7a8e76a3 SR |
1247 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
1248 | ||
83f40318 VN |
1249 | static inline unsigned long rb_page_entries(struct buffer_page *bpage) |
1250 | { | |
1251 | return local_read(&bpage->entries) & RB_WRITE_MASK; | |
1252 | } | |
1253 | ||
1254 | static inline unsigned long rb_page_write(struct buffer_page *bpage) | |
1255 | { | |
1256 | return local_read(&bpage->write) & RB_WRITE_MASK; | |
1257 | } | |
1258 | ||
5040b4b7 | 1259 | static int |
83f40318 | 1260 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned int nr_pages) |
7a8e76a3 | 1261 | { |
83f40318 VN |
1262 | struct list_head *tail_page, *to_remove, *next_page; |
1263 | struct buffer_page *to_remove_page, *tmp_iter_page; | |
1264 | struct buffer_page *last_page, *first_page; | |
1265 | unsigned int nr_removed; | |
1266 | unsigned long head_bit; | |
1267 | int page_entries; | |
1268 | ||
1269 | head_bit = 0; | |
7a8e76a3 | 1270 | |
5389f6fa | 1271 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1272 | atomic_inc(&cpu_buffer->record_disabled); |
1273 | /* | |
1274 | * We don't race with the readers since we have acquired the reader | |
1275 | * lock. We also don't race with writers after disabling recording. | |
1276 | * This makes it easy to figure out the first and the last page to be | |
1277 | * removed from the list. We unlink all the pages in between including | |
1278 | * the first and last pages. This is done in a busy loop so that we | |
1279 | * lose the least number of traces. | |
1280 | * The pages are freed after we restart recording and unlock readers. | |
1281 | */ | |
1282 | tail_page = &cpu_buffer->tail_page->list; | |
77ae365e | 1283 | |
83f40318 VN |
1284 | /* |
1285 | * tail page might be on reader page, we remove the next page | |
1286 | * from the ring buffer | |
1287 | */ | |
1288 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
1289 | tail_page = rb_list_head(tail_page->next); | |
1290 | to_remove = tail_page; | |
1291 | ||
1292 | /* start of pages to remove */ | |
1293 | first_page = list_entry(rb_list_head(to_remove->next), | |
1294 | struct buffer_page, list); | |
1295 | ||
1296 | for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) { | |
1297 | to_remove = rb_list_head(to_remove)->next; | |
1298 | head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD; | |
7a8e76a3 | 1299 | } |
7a8e76a3 | 1300 | |
83f40318 | 1301 | next_page = rb_list_head(to_remove)->next; |
7a8e76a3 | 1302 | |
83f40318 VN |
1303 | /* |
1304 | * Now we remove all pages between tail_page and next_page. | |
1305 | * Make sure that we have head_bit value preserved for the | |
1306 | * next page | |
1307 | */ | |
1308 | tail_page->next = (struct list_head *)((unsigned long)next_page | | |
1309 | head_bit); | |
1310 | next_page = rb_list_head(next_page); | |
1311 | next_page->prev = tail_page; | |
1312 | ||
1313 | /* make sure pages points to a valid page in the ring buffer */ | |
1314 | cpu_buffer->pages = next_page; | |
1315 | ||
1316 | /* update head page */ | |
1317 | if (head_bit) | |
1318 | cpu_buffer->head_page = list_entry(next_page, | |
1319 | struct buffer_page, list); | |
1320 | ||
1321 | /* | |
1322 | * change read pointer to make sure any read iterators reset | |
1323 | * themselves | |
1324 | */ | |
1325 | cpu_buffer->read = 0; | |
1326 | ||
1327 | /* pages are removed, resume tracing and then free the pages */ | |
1328 | atomic_dec(&cpu_buffer->record_disabled); | |
5389f6fa | 1329 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1330 | |
1331 | RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)); | |
1332 | ||
1333 | /* last buffer page to remove */ | |
1334 | last_page = list_entry(rb_list_head(to_remove), struct buffer_page, | |
1335 | list); | |
1336 | tmp_iter_page = first_page; | |
1337 | ||
1338 | do { | |
1339 | to_remove_page = tmp_iter_page; | |
1340 | rb_inc_page(cpu_buffer, &tmp_iter_page); | |
1341 | ||
1342 | /* update the counters */ | |
1343 | page_entries = rb_page_entries(to_remove_page); | |
1344 | if (page_entries) { | |
1345 | /* | |
1346 | * If something was added to this page, it was full | |
1347 | * since it is not the tail page. So we deduct the | |
1348 | * bytes consumed in ring buffer from here. | |
1349 | * No need to update overruns, since this page is | |
1350 | * deleted from ring buffer and its entries are | |
1351 | * already accounted for. | |
1352 | */ | |
1353 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); | |
1354 | } | |
1355 | ||
1356 | /* | |
1357 | * We have already removed references to this list item, just | |
1358 | * free up the buffer_page and its page | |
1359 | */ | |
1360 | free_buffer_page(to_remove_page); | |
1361 | nr_removed--; | |
1362 | ||
1363 | } while (to_remove_page != last_page); | |
1364 | ||
1365 | RB_WARN_ON(cpu_buffer, nr_removed); | |
5040b4b7 VN |
1366 | |
1367 | return nr_removed == 0; | |
7a8e76a3 SR |
1368 | } |
1369 | ||
5040b4b7 VN |
1370 | static int |
1371 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 1372 | { |
5040b4b7 VN |
1373 | struct list_head *pages = &cpu_buffer->new_pages; |
1374 | int retries, success; | |
7a8e76a3 | 1375 | |
5389f6fa | 1376 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1377 | /* |
1378 | * We are holding the reader lock, so the reader page won't be swapped | |
1379 | * in the ring buffer. Now we are racing with the writer trying to | |
1380 | * move head page and the tail page. | |
1381 | * We are going to adapt the reader page update process where: | |
1382 | * 1. We first splice the start and end of list of new pages between | |
1383 | * the head page and its previous page. | |
1384 | * 2. We cmpxchg the prev_page->next to point from head page to the | |
1385 | * start of new pages list. | |
1386 | * 3. Finally, we update the head->prev to the end of new list. | |
1387 | * | |
1388 | * We will try this process 10 times, to make sure that we don't keep | |
1389 | * spinning. | |
1390 | */ | |
1391 | retries = 10; | |
1392 | success = 0; | |
1393 | while (retries--) { | |
1394 | struct list_head *head_page, *prev_page, *r; | |
1395 | struct list_head *last_page, *first_page; | |
1396 | struct list_head *head_page_with_bit; | |
77ae365e | 1397 | |
5040b4b7 VN |
1398 | head_page = &rb_set_head_page(cpu_buffer)->list; |
1399 | prev_page = head_page->prev; | |
1400 | ||
1401 | first_page = pages->next; | |
1402 | last_page = pages->prev; | |
1403 | ||
1404 | head_page_with_bit = (struct list_head *) | |
1405 | ((unsigned long)head_page | RB_PAGE_HEAD); | |
1406 | ||
1407 | last_page->next = head_page_with_bit; | |
1408 | first_page->prev = prev_page; | |
1409 | ||
1410 | r = cmpxchg(&prev_page->next, head_page_with_bit, first_page); | |
1411 | ||
1412 | if (r == head_page_with_bit) { | |
1413 | /* | |
1414 | * yay, we replaced the page pointer to our new list, | |
1415 | * now, we just have to update to head page's prev | |
1416 | * pointer to point to end of list | |
1417 | */ | |
1418 | head_page->prev = last_page; | |
1419 | success = 1; | |
1420 | break; | |
1421 | } | |
7a8e76a3 | 1422 | } |
7a8e76a3 | 1423 | |
5040b4b7 VN |
1424 | if (success) |
1425 | INIT_LIST_HEAD(pages); | |
1426 | /* | |
1427 | * If we weren't successful in adding in new pages, warn and stop | |
1428 | * tracing | |
1429 | */ | |
1430 | RB_WARN_ON(cpu_buffer, !success); | |
5389f6fa | 1431 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1432 | |
1433 | /* free pages if they weren't inserted */ | |
1434 | if (!success) { | |
1435 | struct buffer_page *bpage, *tmp; | |
1436 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, | |
1437 | list) { | |
1438 | list_del_init(&bpage->list); | |
1439 | free_buffer_page(bpage); | |
1440 | } | |
1441 | } | |
1442 | return success; | |
7a8e76a3 SR |
1443 | } |
1444 | ||
83f40318 | 1445 | static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer) |
438ced17 | 1446 | { |
5040b4b7 VN |
1447 | int success; |
1448 | ||
438ced17 | 1449 | if (cpu_buffer->nr_pages_to_update > 0) |
5040b4b7 | 1450 | success = rb_insert_pages(cpu_buffer); |
438ced17 | 1451 | else |
5040b4b7 VN |
1452 | success = rb_remove_pages(cpu_buffer, |
1453 | -cpu_buffer->nr_pages_to_update); | |
83f40318 | 1454 | |
5040b4b7 VN |
1455 | if (success) |
1456 | cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update; | |
83f40318 VN |
1457 | } |
1458 | ||
1459 | static void update_pages_handler(struct work_struct *work) | |
1460 | { | |
1461 | struct ring_buffer_per_cpu *cpu_buffer = container_of(work, | |
1462 | struct ring_buffer_per_cpu, update_pages_work); | |
1463 | rb_update_pages(cpu_buffer); | |
1464 | complete(&cpu_buffer->update_completion); | |
438ced17 VN |
1465 | } |
1466 | ||
7a8e76a3 SR |
1467 | /** |
1468 | * ring_buffer_resize - resize the ring buffer | |
1469 | * @buffer: the buffer to resize. | |
1470 | * @size: the new size. | |
1471 | * | |
7a8e76a3 SR |
1472 | * Minimum size is 2 * BUF_PAGE_SIZE. |
1473 | * | |
83f40318 | 1474 | * Returns 0 on success and < 0 on failure. |
7a8e76a3 | 1475 | */ |
438ced17 VN |
1476 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size, |
1477 | int cpu_id) | |
7a8e76a3 SR |
1478 | { |
1479 | struct ring_buffer_per_cpu *cpu_buffer; | |
438ced17 | 1480 | unsigned nr_pages; |
83f40318 | 1481 | int cpu, err = 0; |
7a8e76a3 | 1482 | |
ee51a1de IM |
1483 | /* |
1484 | * Always succeed at resizing a non-existent buffer: | |
1485 | */ | |
1486 | if (!buffer) | |
1487 | return size; | |
1488 | ||
7a8e76a3 SR |
1489 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
1490 | size *= BUF_PAGE_SIZE; | |
7a8e76a3 SR |
1491 | |
1492 | /* we need a minimum of two pages */ | |
1493 | if (size < BUF_PAGE_SIZE * 2) | |
1494 | size = BUF_PAGE_SIZE * 2; | |
1495 | ||
83f40318 | 1496 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
18421015 | 1497 | |
83f40318 VN |
1498 | /* |
1499 | * Don't succeed if resizing is disabled, as a reader might be | |
1500 | * manipulating the ring buffer and is expecting a sane state while | |
1501 | * this is true. | |
1502 | */ | |
1503 | if (atomic_read(&buffer->resize_disabled)) | |
1504 | return -EBUSY; | |
18421015 | 1505 | |
83f40318 | 1506 | /* prevent another thread from changing buffer sizes */ |
7a8e76a3 | 1507 | mutex_lock(&buffer->mutex); |
7a8e76a3 | 1508 | |
438ced17 VN |
1509 | if (cpu_id == RING_BUFFER_ALL_CPUS) { |
1510 | /* calculate the pages to update */ | |
7a8e76a3 SR |
1511 | for_each_buffer_cpu(buffer, cpu) { |
1512 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 1513 | |
438ced17 VN |
1514 | cpu_buffer->nr_pages_to_update = nr_pages - |
1515 | cpu_buffer->nr_pages; | |
438ced17 VN |
1516 | /* |
1517 | * nothing more to do for removing pages or no update | |
1518 | */ | |
1519 | if (cpu_buffer->nr_pages_to_update <= 0) | |
1520 | continue; | |
d7ec4bfe | 1521 | /* |
438ced17 VN |
1522 | * to add pages, make sure all new pages can be |
1523 | * allocated without receiving ENOMEM | |
d7ec4bfe | 1524 | */ |
438ced17 VN |
1525 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
1526 | if (__rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 | 1527 | &cpu_buffer->new_pages, cpu)) { |
438ced17 | 1528 | /* not enough memory for new pages */ |
83f40318 VN |
1529 | err = -ENOMEM; |
1530 | goto out_err; | |
1531 | } | |
1532 | } | |
1533 | ||
1534 | get_online_cpus(); | |
1535 | /* | |
1536 | * Fire off all the required work handlers | |
1537 | * Look out for offline CPUs | |
1538 | */ | |
1539 | for_each_buffer_cpu(buffer, cpu) { | |
1540 | cpu_buffer = buffer->buffers[cpu]; | |
1541 | if (!cpu_buffer->nr_pages_to_update || | |
1542 | !cpu_online(cpu)) | |
1543 | continue; | |
1544 | ||
1545 | schedule_work_on(cpu, &cpu_buffer->update_pages_work); | |
1546 | } | |
1547 | /* | |
1548 | * This loop is for the CPUs that are not online. | |
1549 | * We can't schedule anything on them, but it's not necessary | |
1550 | * since we can change their buffer sizes without any race. | |
1551 | */ | |
1552 | for_each_buffer_cpu(buffer, cpu) { | |
1553 | cpu_buffer = buffer->buffers[cpu]; | |
1554 | if (!cpu_buffer->nr_pages_to_update || | |
1555 | cpu_online(cpu)) | |
1556 | continue; | |
1557 | ||
1558 | rb_update_pages(cpu_buffer); | |
7a8e76a3 | 1559 | } |
7a8e76a3 | 1560 | |
438ced17 VN |
1561 | /* wait for all the updates to complete */ |
1562 | for_each_buffer_cpu(buffer, cpu) { | |
1563 | cpu_buffer = buffer->buffers[cpu]; | |
83f40318 VN |
1564 | if (!cpu_buffer->nr_pages_to_update || |
1565 | !cpu_online(cpu)) | |
1566 | continue; | |
1567 | ||
1568 | wait_for_completion(&cpu_buffer->update_completion); | |
1569 | /* reset this value */ | |
1570 | cpu_buffer->nr_pages_to_update = 0; | |
438ced17 | 1571 | } |
83f40318 VN |
1572 | |
1573 | put_online_cpus(); | |
438ced17 VN |
1574 | } else { |
1575 | cpu_buffer = buffer->buffers[cpu_id]; | |
83f40318 | 1576 | |
438ced17 VN |
1577 | if (nr_pages == cpu_buffer->nr_pages) |
1578 | goto out; | |
7a8e76a3 | 1579 | |
438ced17 VN |
1580 | cpu_buffer->nr_pages_to_update = nr_pages - |
1581 | cpu_buffer->nr_pages; | |
1582 | ||
1583 | INIT_LIST_HEAD(&cpu_buffer->new_pages); | |
1584 | if (cpu_buffer->nr_pages_to_update > 0 && | |
1585 | __rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 VN |
1586 | &cpu_buffer->new_pages, cpu_id)) { |
1587 | err = -ENOMEM; | |
1588 | goto out_err; | |
1589 | } | |
438ced17 | 1590 | |
83f40318 VN |
1591 | get_online_cpus(); |
1592 | ||
1593 | if (cpu_online(cpu_id)) { | |
1594 | schedule_work_on(cpu_id, | |
1595 | &cpu_buffer->update_pages_work); | |
1596 | wait_for_completion(&cpu_buffer->update_completion); | |
1597 | } else | |
1598 | rb_update_pages(cpu_buffer); | |
1599 | ||
1600 | put_online_cpus(); | |
1601 | /* reset this value */ | |
1602 | cpu_buffer->nr_pages_to_update = 0; | |
438ced17 | 1603 | } |
7a8e76a3 SR |
1604 | |
1605 | out: | |
659f451f SR |
1606 | /* |
1607 | * The ring buffer resize can happen with the ring buffer | |
1608 | * enabled, so that the update disturbs the tracing as little | |
1609 | * as possible. But if the buffer is disabled, we do not need | |
1610 | * to worry about that, and we can take the time to verify | |
1611 | * that the buffer is not corrupt. | |
1612 | */ | |
1613 | if (atomic_read(&buffer->record_disabled)) { | |
1614 | atomic_inc(&buffer->record_disabled); | |
1615 | /* | |
1616 | * Even though the buffer was disabled, we must make sure | |
1617 | * that it is truly disabled before calling rb_check_pages. | |
1618 | * There could have been a race between checking | |
1619 | * record_disable and incrementing it. | |
1620 | */ | |
1621 | synchronize_sched(); | |
1622 | for_each_buffer_cpu(buffer, cpu) { | |
1623 | cpu_buffer = buffer->buffers[cpu]; | |
1624 | rb_check_pages(cpu_buffer); | |
1625 | } | |
1626 | atomic_dec(&buffer->record_disabled); | |
1627 | } | |
1628 | ||
7a8e76a3 | 1629 | mutex_unlock(&buffer->mutex); |
7a8e76a3 SR |
1630 | return size; |
1631 | ||
83f40318 | 1632 | out_err: |
438ced17 VN |
1633 | for_each_buffer_cpu(buffer, cpu) { |
1634 | struct buffer_page *bpage, *tmp; | |
83f40318 | 1635 | |
438ced17 | 1636 | cpu_buffer = buffer->buffers[cpu]; |
438ced17 | 1637 | cpu_buffer->nr_pages_to_update = 0; |
83f40318 | 1638 | |
438ced17 VN |
1639 | if (list_empty(&cpu_buffer->new_pages)) |
1640 | continue; | |
83f40318 | 1641 | |
438ced17 VN |
1642 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, |
1643 | list) { | |
1644 | list_del_init(&bpage->list); | |
1645 | free_buffer_page(bpage); | |
1646 | } | |
7a8e76a3 | 1647 | } |
641d2f63 | 1648 | mutex_unlock(&buffer->mutex); |
83f40318 | 1649 | return err; |
7a8e76a3 | 1650 | } |
c4f50183 | 1651 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 1652 | |
750912fa DS |
1653 | void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val) |
1654 | { | |
1655 | mutex_lock(&buffer->mutex); | |
1656 | if (val) | |
1657 | buffer->flags |= RB_FL_OVERWRITE; | |
1658 | else | |
1659 | buffer->flags &= ~RB_FL_OVERWRITE; | |
1660 | mutex_unlock(&buffer->mutex); | |
1661 | } | |
1662 | EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite); | |
1663 | ||
8789a9e7 | 1664 | static inline void * |
044fa782 | 1665 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 1666 | { |
044fa782 | 1667 | return bpage->data + index; |
8789a9e7 SR |
1668 | } |
1669 | ||
044fa782 | 1670 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 1671 | { |
044fa782 | 1672 | return bpage->page->data + index; |
7a8e76a3 SR |
1673 | } |
1674 | ||
1675 | static inline struct ring_buffer_event * | |
d769041f | 1676 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1677 | { |
6f807acd SR |
1678 | return __rb_page_index(cpu_buffer->reader_page, |
1679 | cpu_buffer->reader_page->read); | |
1680 | } | |
1681 | ||
7a8e76a3 SR |
1682 | static inline struct ring_buffer_event * |
1683 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
1684 | { | |
6f807acd | 1685 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
1686 | } |
1687 | ||
bf41a158 SR |
1688 | static inline unsigned rb_page_commit(struct buffer_page *bpage) |
1689 | { | |
abc9b56d | 1690 | return local_read(&bpage->page->commit); |
bf41a158 SR |
1691 | } |
1692 | ||
25985edc | 1693 | /* Size is determined by what has been committed */ |
bf41a158 SR |
1694 | static inline unsigned rb_page_size(struct buffer_page *bpage) |
1695 | { | |
1696 | return rb_page_commit(bpage); | |
1697 | } | |
1698 | ||
1699 | static inline unsigned | |
1700 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
1701 | { | |
1702 | return rb_page_commit(cpu_buffer->commit_page); | |
1703 | } | |
1704 | ||
bf41a158 SR |
1705 | static inline unsigned |
1706 | rb_event_index(struct ring_buffer_event *event) | |
1707 | { | |
1708 | unsigned long addr = (unsigned long)event; | |
1709 | ||
22f470f8 | 1710 | return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE; |
bf41a158 SR |
1711 | } |
1712 | ||
0f0c85fc | 1713 | static inline int |
fa743953 SR |
1714 | rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1715 | struct ring_buffer_event *event) | |
bf41a158 SR |
1716 | { |
1717 | unsigned long addr = (unsigned long)event; | |
1718 | unsigned long index; | |
1719 | ||
1720 | index = rb_event_index(event); | |
1721 | addr &= PAGE_MASK; | |
1722 | ||
1723 | return cpu_buffer->commit_page->page == (void *)addr && | |
1724 | rb_commit_index(cpu_buffer) == index; | |
1725 | } | |
1726 | ||
34a148bf | 1727 | static void |
bf41a158 | 1728 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1729 | { |
77ae365e SR |
1730 | unsigned long max_count; |
1731 | ||
bf41a158 SR |
1732 | /* |
1733 | * We only race with interrupts and NMIs on this CPU. | |
1734 | * If we own the commit event, then we can commit | |
1735 | * all others that interrupted us, since the interruptions | |
1736 | * are in stack format (they finish before they come | |
1737 | * back to us). This allows us to do a simple loop to | |
1738 | * assign the commit to the tail. | |
1739 | */ | |
a8ccf1d6 | 1740 | again: |
438ced17 | 1741 | max_count = cpu_buffer->nr_pages * 100; |
77ae365e | 1742 | |
bf41a158 | 1743 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
77ae365e SR |
1744 | if (RB_WARN_ON(cpu_buffer, !(--max_count))) |
1745 | return; | |
1746 | if (RB_WARN_ON(cpu_buffer, | |
1747 | rb_is_reader_page(cpu_buffer->tail_page))) | |
1748 | return; | |
1749 | local_set(&cpu_buffer->commit_page->page->commit, | |
1750 | rb_page_write(cpu_buffer->commit_page)); | |
bf41a158 | 1751 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); |
abc9b56d SR |
1752 | cpu_buffer->write_stamp = |
1753 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1754 | /* add barrier to keep gcc from optimizing too much */ |
1755 | barrier(); | |
1756 | } | |
1757 | while (rb_commit_index(cpu_buffer) != | |
1758 | rb_page_write(cpu_buffer->commit_page)) { | |
77ae365e SR |
1759 | |
1760 | local_set(&cpu_buffer->commit_page->page->commit, | |
1761 | rb_page_write(cpu_buffer->commit_page)); | |
1762 | RB_WARN_ON(cpu_buffer, | |
1763 | local_read(&cpu_buffer->commit_page->page->commit) & | |
1764 | ~RB_WRITE_MASK); | |
bf41a158 SR |
1765 | barrier(); |
1766 | } | |
a8ccf1d6 SR |
1767 | |
1768 | /* again, keep gcc from optimizing */ | |
1769 | barrier(); | |
1770 | ||
1771 | /* | |
1772 | * If an interrupt came in just after the first while loop | |
1773 | * and pushed the tail page forward, we will be left with | |
1774 | * a dangling commit that will never go forward. | |
1775 | */ | |
1776 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1777 | goto again; | |
7a8e76a3 SR |
1778 | } |
1779 | ||
d769041f | 1780 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1781 | { |
abc9b56d | 1782 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1783 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1784 | } |
1785 | ||
34a148bf | 1786 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1787 | { |
1788 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1789 | ||
1790 | /* | |
1791 | * The iterator could be on the reader page (it starts there). | |
1792 | * But the head could have moved, since the reader was | |
1793 | * found. Check for this case and assign the iterator | |
1794 | * to the head page instead of next. | |
1795 | */ | |
1796 | if (iter->head_page == cpu_buffer->reader_page) | |
77ae365e | 1797 | iter->head_page = rb_set_head_page(cpu_buffer); |
d769041f SR |
1798 | else |
1799 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1800 | ||
abc9b56d | 1801 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1802 | iter->head = 0; |
1803 | } | |
1804 | ||
69d1b839 SR |
1805 | /* Slow path, do not inline */ |
1806 | static noinline struct ring_buffer_event * | |
1807 | rb_add_time_stamp(struct ring_buffer_event *event, u64 delta) | |
1808 | { | |
1809 | event->type_len = RINGBUF_TYPE_TIME_EXTEND; | |
1810 | ||
1811 | /* Not the first event on the page? */ | |
1812 | if (rb_event_index(event)) { | |
1813 | event->time_delta = delta & TS_MASK; | |
1814 | event->array[0] = delta >> TS_SHIFT; | |
1815 | } else { | |
1816 | /* nope, just zero it */ | |
1817 | event->time_delta = 0; | |
1818 | event->array[0] = 0; | |
1819 | } | |
1820 | ||
1821 | return skip_time_extend(event); | |
1822 | } | |
1823 | ||
7a8e76a3 SR |
1824 | /** |
1825 | * ring_buffer_update_event - update event type and data | |
1826 | * @event: the even to update | |
1827 | * @type: the type of event | |
1828 | * @length: the size of the event field in the ring buffer | |
1829 | * | |
1830 | * Update the type and data fields of the event. The length | |
1831 | * is the actual size that is written to the ring buffer, | |
1832 | * and with this, we can determine what to place into the | |
1833 | * data field. | |
1834 | */ | |
34a148bf | 1835 | static void |
69d1b839 SR |
1836 | rb_update_event(struct ring_buffer_per_cpu *cpu_buffer, |
1837 | struct ring_buffer_event *event, unsigned length, | |
1838 | int add_timestamp, u64 delta) | |
7a8e76a3 | 1839 | { |
69d1b839 SR |
1840 | /* Only a commit updates the timestamp */ |
1841 | if (unlikely(!rb_event_is_commit(cpu_buffer, event))) | |
1842 | delta = 0; | |
7a8e76a3 | 1843 | |
69d1b839 SR |
1844 | /* |
1845 | * If we need to add a timestamp, then we | |
1846 | * add it to the start of the resevered space. | |
1847 | */ | |
1848 | if (unlikely(add_timestamp)) { | |
1849 | event = rb_add_time_stamp(event, delta); | |
1850 | length -= RB_LEN_TIME_EXTEND; | |
1851 | delta = 0; | |
7a8e76a3 | 1852 | } |
69d1b839 SR |
1853 | |
1854 | event->time_delta = delta; | |
1855 | length -= RB_EVNT_HDR_SIZE; | |
1856 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) { | |
1857 | event->type_len = 0; | |
1858 | event->array[0] = length; | |
1859 | } else | |
1860 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
7a8e76a3 SR |
1861 | } |
1862 | ||
77ae365e SR |
1863 | /* |
1864 | * rb_handle_head_page - writer hit the head page | |
1865 | * | |
1866 | * Returns: +1 to retry page | |
1867 | * 0 to continue | |
1868 | * -1 on error | |
1869 | */ | |
1870 | static int | |
1871 | rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
1872 | struct buffer_page *tail_page, | |
1873 | struct buffer_page *next_page) | |
1874 | { | |
1875 | struct buffer_page *new_head; | |
1876 | int entries; | |
1877 | int type; | |
1878 | int ret; | |
1879 | ||
1880 | entries = rb_page_entries(next_page); | |
1881 | ||
1882 | /* | |
1883 | * The hard part is here. We need to move the head | |
1884 | * forward, and protect against both readers on | |
1885 | * other CPUs and writers coming in via interrupts. | |
1886 | */ | |
1887 | type = rb_head_page_set_update(cpu_buffer, next_page, tail_page, | |
1888 | RB_PAGE_HEAD); | |
1889 | ||
1890 | /* | |
1891 | * type can be one of four: | |
1892 | * NORMAL - an interrupt already moved it for us | |
1893 | * HEAD - we are the first to get here. | |
1894 | * UPDATE - we are the interrupt interrupting | |
1895 | * a current move. | |
1896 | * MOVED - a reader on another CPU moved the next | |
1897 | * pointer to its reader page. Give up | |
1898 | * and try again. | |
1899 | */ | |
1900 | ||
1901 | switch (type) { | |
1902 | case RB_PAGE_HEAD: | |
1903 | /* | |
1904 | * We changed the head to UPDATE, thus | |
1905 | * it is our responsibility to update | |
1906 | * the counters. | |
1907 | */ | |
1908 | local_add(entries, &cpu_buffer->overrun); | |
c64e148a | 1909 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
77ae365e SR |
1910 | |
1911 | /* | |
1912 | * The entries will be zeroed out when we move the | |
1913 | * tail page. | |
1914 | */ | |
1915 | ||
1916 | /* still more to do */ | |
1917 | break; | |
1918 | ||
1919 | case RB_PAGE_UPDATE: | |
1920 | /* | |
1921 | * This is an interrupt that interrupt the | |
1922 | * previous update. Still more to do. | |
1923 | */ | |
1924 | break; | |
1925 | case RB_PAGE_NORMAL: | |
1926 | /* | |
1927 | * An interrupt came in before the update | |
1928 | * and processed this for us. | |
1929 | * Nothing left to do. | |
1930 | */ | |
1931 | return 1; | |
1932 | case RB_PAGE_MOVED: | |
1933 | /* | |
1934 | * The reader is on another CPU and just did | |
1935 | * a swap with our next_page. | |
1936 | * Try again. | |
1937 | */ | |
1938 | return 1; | |
1939 | default: | |
1940 | RB_WARN_ON(cpu_buffer, 1); /* WTF??? */ | |
1941 | return -1; | |
1942 | } | |
1943 | ||
1944 | /* | |
1945 | * Now that we are here, the old head pointer is | |
1946 | * set to UPDATE. This will keep the reader from | |
1947 | * swapping the head page with the reader page. | |
1948 | * The reader (on another CPU) will spin till | |
1949 | * we are finished. | |
1950 | * | |
1951 | * We just need to protect against interrupts | |
1952 | * doing the job. We will set the next pointer | |
1953 | * to HEAD. After that, we set the old pointer | |
1954 | * to NORMAL, but only if it was HEAD before. | |
1955 | * otherwise we are an interrupt, and only | |
1956 | * want the outer most commit to reset it. | |
1957 | */ | |
1958 | new_head = next_page; | |
1959 | rb_inc_page(cpu_buffer, &new_head); | |
1960 | ||
1961 | ret = rb_head_page_set_head(cpu_buffer, new_head, next_page, | |
1962 | RB_PAGE_NORMAL); | |
1963 | ||
1964 | /* | |
1965 | * Valid returns are: | |
1966 | * HEAD - an interrupt came in and already set it. | |
1967 | * NORMAL - One of two things: | |
1968 | * 1) We really set it. | |
1969 | * 2) A bunch of interrupts came in and moved | |
1970 | * the page forward again. | |
1971 | */ | |
1972 | switch (ret) { | |
1973 | case RB_PAGE_HEAD: | |
1974 | case RB_PAGE_NORMAL: | |
1975 | /* OK */ | |
1976 | break; | |
1977 | default: | |
1978 | RB_WARN_ON(cpu_buffer, 1); | |
1979 | return -1; | |
1980 | } | |
1981 | ||
1982 | /* | |
1983 | * It is possible that an interrupt came in, | |
1984 | * set the head up, then more interrupts came in | |
1985 | * and moved it again. When we get back here, | |
1986 | * the page would have been set to NORMAL but we | |
1987 | * just set it back to HEAD. | |
1988 | * | |
1989 | * How do you detect this? Well, if that happened | |
1990 | * the tail page would have moved. | |
1991 | */ | |
1992 | if (ret == RB_PAGE_NORMAL) { | |
1993 | /* | |
1994 | * If the tail had moved passed next, then we need | |
1995 | * to reset the pointer. | |
1996 | */ | |
1997 | if (cpu_buffer->tail_page != tail_page && | |
1998 | cpu_buffer->tail_page != next_page) | |
1999 | rb_head_page_set_normal(cpu_buffer, new_head, | |
2000 | next_page, | |
2001 | RB_PAGE_HEAD); | |
2002 | } | |
2003 | ||
2004 | /* | |
2005 | * If this was the outer most commit (the one that | |
2006 | * changed the original pointer from HEAD to UPDATE), | |
2007 | * then it is up to us to reset it to NORMAL. | |
2008 | */ | |
2009 | if (type == RB_PAGE_HEAD) { | |
2010 | ret = rb_head_page_set_normal(cpu_buffer, next_page, | |
2011 | tail_page, | |
2012 | RB_PAGE_UPDATE); | |
2013 | if (RB_WARN_ON(cpu_buffer, | |
2014 | ret != RB_PAGE_UPDATE)) | |
2015 | return -1; | |
2016 | } | |
2017 | ||
2018 | return 0; | |
2019 | } | |
2020 | ||
34a148bf | 2021 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
2022 | { |
2023 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
2024 | ||
2025 | /* zero length can cause confusions */ | |
2026 | if (!length) | |
2027 | length = 1; | |
2028 | ||
2271048d | 2029 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) |
7a8e76a3 SR |
2030 | length += sizeof(event.array[0]); |
2031 | ||
2032 | length += RB_EVNT_HDR_SIZE; | |
2271048d | 2033 | length = ALIGN(length, RB_ARCH_ALIGNMENT); |
7a8e76a3 SR |
2034 | |
2035 | return length; | |
2036 | } | |
2037 | ||
c7b09308 SR |
2038 | static inline void |
2039 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
2040 | struct buffer_page *tail_page, | |
2041 | unsigned long tail, unsigned long length) | |
2042 | { | |
2043 | struct ring_buffer_event *event; | |
2044 | ||
2045 | /* | |
2046 | * Only the event that crossed the page boundary | |
2047 | * must fill the old tail_page with padding. | |
2048 | */ | |
2049 | if (tail >= BUF_PAGE_SIZE) { | |
b3230c8b SR |
2050 | /* |
2051 | * If the page was filled, then we still need | |
2052 | * to update the real_end. Reset it to zero | |
2053 | * and the reader will ignore it. | |
2054 | */ | |
2055 | if (tail == BUF_PAGE_SIZE) | |
2056 | tail_page->real_end = 0; | |
2057 | ||
c7b09308 SR |
2058 | local_sub(length, &tail_page->write); |
2059 | return; | |
2060 | } | |
2061 | ||
2062 | event = __rb_page_index(tail_page, tail); | |
b0b7065b | 2063 | kmemcheck_annotate_bitfield(event, bitfield); |
c7b09308 | 2064 | |
c64e148a VN |
2065 | /* account for padding bytes */ |
2066 | local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes); | |
2067 | ||
ff0ff84a SR |
2068 | /* |
2069 | * Save the original length to the meta data. | |
2070 | * This will be used by the reader to add lost event | |
2071 | * counter. | |
2072 | */ | |
2073 | tail_page->real_end = tail; | |
2074 | ||
c7b09308 SR |
2075 | /* |
2076 | * If this event is bigger than the minimum size, then | |
2077 | * we need to be careful that we don't subtract the | |
2078 | * write counter enough to allow another writer to slip | |
2079 | * in on this page. | |
2080 | * We put in a discarded commit instead, to make sure | |
2081 | * that this space is not used again. | |
2082 | * | |
2083 | * If we are less than the minimum size, we don't need to | |
2084 | * worry about it. | |
2085 | */ | |
2086 | if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) { | |
2087 | /* No room for any events */ | |
2088 | ||
2089 | /* Mark the rest of the page with padding */ | |
2090 | rb_event_set_padding(event); | |
2091 | ||
2092 | /* Set the write back to the previous setting */ | |
2093 | local_sub(length, &tail_page->write); | |
2094 | return; | |
2095 | } | |
2096 | ||
2097 | /* Put in a discarded event */ | |
2098 | event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE; | |
2099 | event->type_len = RINGBUF_TYPE_PADDING; | |
2100 | /* time delta must be non zero */ | |
2101 | event->time_delta = 1; | |
c7b09308 SR |
2102 | |
2103 | /* Set write to end of buffer */ | |
2104 | length = (tail + length) - BUF_PAGE_SIZE; | |
2105 | local_sub(length, &tail_page->write); | |
2106 | } | |
6634ff26 | 2107 | |
747e94ae SR |
2108 | /* |
2109 | * This is the slow path, force gcc not to inline it. | |
2110 | */ | |
2111 | static noinline struct ring_buffer_event * | |
6634ff26 SR |
2112 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
2113 | unsigned long length, unsigned long tail, | |
e8bc43e8 | 2114 | struct buffer_page *tail_page, u64 ts) |
7a8e76a3 | 2115 | { |
5a50e33c | 2116 | struct buffer_page *commit_page = cpu_buffer->commit_page; |
7a8e76a3 | 2117 | struct ring_buffer *buffer = cpu_buffer->buffer; |
77ae365e SR |
2118 | struct buffer_page *next_page; |
2119 | int ret; | |
aa20ae84 SR |
2120 | |
2121 | next_page = tail_page; | |
2122 | ||
aa20ae84 SR |
2123 | rb_inc_page(cpu_buffer, &next_page); |
2124 | ||
aa20ae84 SR |
2125 | /* |
2126 | * If for some reason, we had an interrupt storm that made | |
2127 | * it all the way around the buffer, bail, and warn | |
2128 | * about it. | |
2129 | */ | |
2130 | if (unlikely(next_page == commit_page)) { | |
77ae365e | 2131 | local_inc(&cpu_buffer->commit_overrun); |
aa20ae84 SR |
2132 | goto out_reset; |
2133 | } | |
2134 | ||
77ae365e SR |
2135 | /* |
2136 | * This is where the fun begins! | |
2137 | * | |
2138 | * We are fighting against races between a reader that | |
2139 | * could be on another CPU trying to swap its reader | |
2140 | * page with the buffer head. | |
2141 | * | |
2142 | * We are also fighting against interrupts coming in and | |
2143 | * moving the head or tail on us as well. | |
2144 | * | |
2145 | * If the next page is the head page then we have filled | |
2146 | * the buffer, unless the commit page is still on the | |
2147 | * reader page. | |
2148 | */ | |
2149 | if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) { | |
aa20ae84 | 2150 | |
77ae365e SR |
2151 | /* |
2152 | * If the commit is not on the reader page, then | |
2153 | * move the header page. | |
2154 | */ | |
2155 | if (!rb_is_reader_page(cpu_buffer->commit_page)) { | |
2156 | /* | |
2157 | * If we are not in overwrite mode, | |
2158 | * this is easy, just stop here. | |
2159 | */ | |
2160 | if (!(buffer->flags & RB_FL_OVERWRITE)) | |
2161 | goto out_reset; | |
2162 | ||
2163 | ret = rb_handle_head_page(cpu_buffer, | |
2164 | tail_page, | |
2165 | next_page); | |
2166 | if (ret < 0) | |
2167 | goto out_reset; | |
2168 | if (ret) | |
2169 | goto out_again; | |
2170 | } else { | |
2171 | /* | |
2172 | * We need to be careful here too. The | |
2173 | * commit page could still be on the reader | |
2174 | * page. We could have a small buffer, and | |
2175 | * have filled up the buffer with events | |
2176 | * from interrupts and such, and wrapped. | |
2177 | * | |
2178 | * Note, if the tail page is also the on the | |
2179 | * reader_page, we let it move out. | |
2180 | */ | |
2181 | if (unlikely((cpu_buffer->commit_page != | |
2182 | cpu_buffer->tail_page) && | |
2183 | (cpu_buffer->commit_page == | |
2184 | cpu_buffer->reader_page))) { | |
2185 | local_inc(&cpu_buffer->commit_overrun); | |
2186 | goto out_reset; | |
2187 | } | |
aa20ae84 SR |
2188 | } |
2189 | } | |
2190 | ||
77ae365e SR |
2191 | ret = rb_tail_page_update(cpu_buffer, tail_page, next_page); |
2192 | if (ret) { | |
2193 | /* | |
2194 | * Nested commits always have zero deltas, so | |
2195 | * just reread the time stamp | |
2196 | */ | |
e8bc43e8 SR |
2197 | ts = rb_time_stamp(buffer); |
2198 | next_page->page->time_stamp = ts; | |
aa20ae84 SR |
2199 | } |
2200 | ||
77ae365e | 2201 | out_again: |
aa20ae84 | 2202 | |
77ae365e | 2203 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
aa20ae84 SR |
2204 | |
2205 | /* fail and let the caller try again */ | |
2206 | return ERR_PTR(-EAGAIN); | |
2207 | ||
45141d46 | 2208 | out_reset: |
6f3b3440 | 2209 | /* reset write */ |
c7b09308 | 2210 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
6f3b3440 | 2211 | |
bf41a158 | 2212 | return NULL; |
7a8e76a3 SR |
2213 | } |
2214 | ||
6634ff26 SR |
2215 | static struct ring_buffer_event * |
2216 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
69d1b839 SR |
2217 | unsigned long length, u64 ts, |
2218 | u64 delta, int add_timestamp) | |
6634ff26 | 2219 | { |
5a50e33c | 2220 | struct buffer_page *tail_page; |
6634ff26 SR |
2221 | struct ring_buffer_event *event; |
2222 | unsigned long tail, write; | |
2223 | ||
69d1b839 SR |
2224 | /* |
2225 | * If the time delta since the last event is too big to | |
2226 | * hold in the time field of the event, then we append a | |
2227 | * TIME EXTEND event ahead of the data event. | |
2228 | */ | |
2229 | if (unlikely(add_timestamp)) | |
2230 | length += RB_LEN_TIME_EXTEND; | |
2231 | ||
6634ff26 SR |
2232 | tail_page = cpu_buffer->tail_page; |
2233 | write = local_add_return(length, &tail_page->write); | |
77ae365e SR |
2234 | |
2235 | /* set write to only the index of the write */ | |
2236 | write &= RB_WRITE_MASK; | |
6634ff26 SR |
2237 | tail = write - length; |
2238 | ||
2239 | /* See if we shot pass the end of this buffer page */ | |
747e94ae | 2240 | if (unlikely(write > BUF_PAGE_SIZE)) |
6634ff26 | 2241 | return rb_move_tail(cpu_buffer, length, tail, |
5a50e33c | 2242 | tail_page, ts); |
6634ff26 SR |
2243 | |
2244 | /* We reserved something on the buffer */ | |
2245 | ||
6634ff26 | 2246 | event = __rb_page_index(tail_page, tail); |
1744a21d | 2247 | kmemcheck_annotate_bitfield(event, bitfield); |
69d1b839 | 2248 | rb_update_event(cpu_buffer, event, length, add_timestamp, delta); |
6634ff26 | 2249 | |
69d1b839 | 2250 | local_inc(&tail_page->entries); |
6634ff26 SR |
2251 | |
2252 | /* | |
fa743953 SR |
2253 | * If this is the first commit on the page, then update |
2254 | * its timestamp. | |
6634ff26 | 2255 | */ |
fa743953 | 2256 | if (!tail) |
e8bc43e8 | 2257 | tail_page->page->time_stamp = ts; |
6634ff26 | 2258 | |
c64e148a VN |
2259 | /* account for these added bytes */ |
2260 | local_add(length, &cpu_buffer->entries_bytes); | |
2261 | ||
6634ff26 SR |
2262 | return event; |
2263 | } | |
2264 | ||
edd813bf SR |
2265 | static inline int |
2266 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
2267 | struct ring_buffer_event *event) | |
2268 | { | |
2269 | unsigned long new_index, old_index; | |
2270 | struct buffer_page *bpage; | |
2271 | unsigned long index; | |
2272 | unsigned long addr; | |
2273 | ||
2274 | new_index = rb_event_index(event); | |
69d1b839 | 2275 | old_index = new_index + rb_event_ts_length(event); |
edd813bf SR |
2276 | addr = (unsigned long)event; |
2277 | addr &= PAGE_MASK; | |
2278 | ||
2279 | bpage = cpu_buffer->tail_page; | |
2280 | ||
2281 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { | |
77ae365e SR |
2282 | unsigned long write_mask = |
2283 | local_read(&bpage->write) & ~RB_WRITE_MASK; | |
c64e148a | 2284 | unsigned long event_length = rb_event_length(event); |
edd813bf SR |
2285 | /* |
2286 | * This is on the tail page. It is possible that | |
2287 | * a write could come in and move the tail page | |
2288 | * and write to the next page. That is fine | |
2289 | * because we just shorten what is on this page. | |
2290 | */ | |
77ae365e SR |
2291 | old_index += write_mask; |
2292 | new_index += write_mask; | |
edd813bf | 2293 | index = local_cmpxchg(&bpage->write, old_index, new_index); |
c64e148a VN |
2294 | if (index == old_index) { |
2295 | /* update counters */ | |
2296 | local_sub(event_length, &cpu_buffer->entries_bytes); | |
edd813bf | 2297 | return 1; |
c64e148a | 2298 | } |
edd813bf SR |
2299 | } |
2300 | ||
2301 | /* could not discard */ | |
2302 | return 0; | |
2303 | } | |
2304 | ||
fa743953 SR |
2305 | static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer) |
2306 | { | |
2307 | local_inc(&cpu_buffer->committing); | |
2308 | local_inc(&cpu_buffer->commits); | |
2309 | } | |
2310 | ||
d9abde21 | 2311 | static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer) |
fa743953 SR |
2312 | { |
2313 | unsigned long commits; | |
2314 | ||
2315 | if (RB_WARN_ON(cpu_buffer, | |
2316 | !local_read(&cpu_buffer->committing))) | |
2317 | return; | |
2318 | ||
2319 | again: | |
2320 | commits = local_read(&cpu_buffer->commits); | |
2321 | /* synchronize with interrupts */ | |
2322 | barrier(); | |
2323 | if (local_read(&cpu_buffer->committing) == 1) | |
2324 | rb_set_commit_to_write(cpu_buffer); | |
2325 | ||
2326 | local_dec(&cpu_buffer->committing); | |
2327 | ||
2328 | /* synchronize with interrupts */ | |
2329 | barrier(); | |
2330 | ||
2331 | /* | |
2332 | * Need to account for interrupts coming in between the | |
2333 | * updating of the commit page and the clearing of the | |
2334 | * committing counter. | |
2335 | */ | |
2336 | if (unlikely(local_read(&cpu_buffer->commits) != commits) && | |
2337 | !local_read(&cpu_buffer->committing)) { | |
2338 | local_inc(&cpu_buffer->committing); | |
2339 | goto again; | |
2340 | } | |
2341 | } | |
2342 | ||
7a8e76a3 | 2343 | static struct ring_buffer_event * |
62f0b3eb SR |
2344 | rb_reserve_next_event(struct ring_buffer *buffer, |
2345 | struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 2346 | unsigned long length) |
7a8e76a3 SR |
2347 | { |
2348 | struct ring_buffer_event *event; | |
69d1b839 | 2349 | u64 ts, delta; |
818e3dd3 | 2350 | int nr_loops = 0; |
69d1b839 | 2351 | int add_timestamp; |
140ff891 | 2352 | u64 diff; |
7a8e76a3 | 2353 | |
fa743953 SR |
2354 | rb_start_commit(cpu_buffer); |
2355 | ||
85bac32c | 2356 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
62f0b3eb SR |
2357 | /* |
2358 | * Due to the ability to swap a cpu buffer from a buffer | |
2359 | * it is possible it was swapped before we committed. | |
2360 | * (committing stops a swap). We check for it here and | |
2361 | * if it happened, we have to fail the write. | |
2362 | */ | |
2363 | barrier(); | |
2364 | if (unlikely(ACCESS_ONCE(cpu_buffer->buffer) != buffer)) { | |
2365 | local_dec(&cpu_buffer->committing); | |
2366 | local_dec(&cpu_buffer->commits); | |
2367 | return NULL; | |
2368 | } | |
85bac32c | 2369 | #endif |
62f0b3eb | 2370 | |
be957c44 | 2371 | length = rb_calculate_event_length(length); |
bf41a158 | 2372 | again: |
69d1b839 SR |
2373 | add_timestamp = 0; |
2374 | delta = 0; | |
2375 | ||
818e3dd3 SR |
2376 | /* |
2377 | * We allow for interrupts to reenter here and do a trace. | |
2378 | * If one does, it will cause this original code to loop | |
2379 | * back here. Even with heavy interrupts happening, this | |
2380 | * should only happen a few times in a row. If this happens | |
2381 | * 1000 times in a row, there must be either an interrupt | |
2382 | * storm or we have something buggy. | |
2383 | * Bail! | |
2384 | */ | |
3e89c7bb | 2385 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
fa743953 | 2386 | goto out_fail; |
818e3dd3 | 2387 | |
6d3f1e12 | 2388 | ts = rb_time_stamp(cpu_buffer->buffer); |
140ff891 | 2389 | diff = ts - cpu_buffer->write_stamp; |
7a8e76a3 | 2390 | |
140ff891 SR |
2391 | /* make sure this diff is calculated here */ |
2392 | barrier(); | |
bf41a158 | 2393 | |
140ff891 SR |
2394 | /* Did the write stamp get updated already? */ |
2395 | if (likely(ts >= cpu_buffer->write_stamp)) { | |
168b6b1d SR |
2396 | delta = diff; |
2397 | if (unlikely(test_time_stamp(delta))) { | |
31274d72 JO |
2398 | int local_clock_stable = 1; |
2399 | #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK | |
2400 | local_clock_stable = sched_clock_stable; | |
2401 | #endif | |
69d1b839 | 2402 | WARN_ONCE(delta > (1ULL << 59), |
31274d72 | 2403 | KERN_WARNING "Delta way too big! %llu ts=%llu write stamp = %llu\n%s", |
69d1b839 SR |
2404 | (unsigned long long)delta, |
2405 | (unsigned long long)ts, | |
31274d72 JO |
2406 | (unsigned long long)cpu_buffer->write_stamp, |
2407 | local_clock_stable ? "" : | |
2408 | "If you just came from a suspend/resume,\n" | |
2409 | "please switch to the trace global clock:\n" | |
2410 | " echo global > /sys/kernel/debug/tracing/trace_clock\n"); | |
69d1b839 | 2411 | add_timestamp = 1; |
7a8e76a3 | 2412 | } |
168b6b1d | 2413 | } |
7a8e76a3 | 2414 | |
69d1b839 SR |
2415 | event = __rb_reserve_next(cpu_buffer, length, ts, |
2416 | delta, add_timestamp); | |
168b6b1d | 2417 | if (unlikely(PTR_ERR(event) == -EAGAIN)) |
bf41a158 SR |
2418 | goto again; |
2419 | ||
fa743953 SR |
2420 | if (!event) |
2421 | goto out_fail; | |
7a8e76a3 | 2422 | |
7a8e76a3 | 2423 | return event; |
fa743953 SR |
2424 | |
2425 | out_fail: | |
2426 | rb_end_commit(cpu_buffer); | |
2427 | return NULL; | |
7a8e76a3 SR |
2428 | } |
2429 | ||
1155de47 PM |
2430 | #ifdef CONFIG_TRACING |
2431 | ||
aa18efb2 | 2432 | #define TRACE_RECURSIVE_DEPTH 16 |
261842b7 | 2433 | |
d9abde21 SR |
2434 | /* Keep this code out of the fast path cache */ |
2435 | static noinline void trace_recursive_fail(void) | |
261842b7 | 2436 | { |
aa18efb2 SR |
2437 | /* Disable all tracing before we do anything else */ |
2438 | tracing_off_permanent(); | |
261842b7 | 2439 | |
7d7d2b80 | 2440 | printk_once(KERN_WARNING "Tracing recursion: depth[%ld]:" |
aa18efb2 | 2441 | "HC[%lu]:SC[%lu]:NMI[%lu]\n", |
b1cff0ad | 2442 | trace_recursion_buffer(), |
aa18efb2 SR |
2443 | hardirq_count() >> HARDIRQ_SHIFT, |
2444 | softirq_count() >> SOFTIRQ_SHIFT, | |
2445 | in_nmi()); | |
261842b7 | 2446 | |
aa18efb2 | 2447 | WARN_ON_ONCE(1); |
d9abde21 SR |
2448 | } |
2449 | ||
2450 | static inline int trace_recursive_lock(void) | |
2451 | { | |
b1cff0ad | 2452 | trace_recursion_inc(); |
d9abde21 | 2453 | |
b1cff0ad | 2454 | if (likely(trace_recursion_buffer() < TRACE_RECURSIVE_DEPTH)) |
d9abde21 SR |
2455 | return 0; |
2456 | ||
2457 | trace_recursive_fail(); | |
2458 | ||
aa18efb2 | 2459 | return -1; |
261842b7 SR |
2460 | } |
2461 | ||
d9abde21 | 2462 | static inline void trace_recursive_unlock(void) |
261842b7 | 2463 | { |
b1cff0ad | 2464 | WARN_ON_ONCE(!trace_recursion_buffer()); |
261842b7 | 2465 | |
b1cff0ad | 2466 | trace_recursion_dec(); |
261842b7 SR |
2467 | } |
2468 | ||
1155de47 PM |
2469 | #else |
2470 | ||
2471 | #define trace_recursive_lock() (0) | |
2472 | #define trace_recursive_unlock() do { } while (0) | |
2473 | ||
2474 | #endif | |
2475 | ||
7a8e76a3 SR |
2476 | /** |
2477 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
2478 | * @buffer: the ring buffer to reserve from | |
2479 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
2480 | * |
2481 | * Returns a reseverd event on the ring buffer to copy directly to. | |
2482 | * The user of this interface will need to get the body to write into | |
2483 | * and can use the ring_buffer_event_data() interface. | |
2484 | * | |
2485 | * The length is the length of the data needed, not the event length | |
2486 | * which also includes the event header. | |
2487 | * | |
2488 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
2489 | * If NULL is returned, then nothing has been allocated or locked. | |
2490 | */ | |
2491 | struct ring_buffer_event * | |
0a987751 | 2492 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
2493 | { |
2494 | struct ring_buffer_per_cpu *cpu_buffer; | |
2495 | struct ring_buffer_event *event; | |
5168ae50 | 2496 | int cpu; |
7a8e76a3 | 2497 | |
033601a3 | 2498 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2499 | return NULL; |
2500 | ||
bf41a158 | 2501 | /* If we are tracing schedule, we don't want to recurse */ |
5168ae50 | 2502 | preempt_disable_notrace(); |
bf41a158 | 2503 | |
52fbe9cd LJ |
2504 | if (atomic_read(&buffer->record_disabled)) |
2505 | goto out_nocheck; | |
2506 | ||
261842b7 SR |
2507 | if (trace_recursive_lock()) |
2508 | goto out_nocheck; | |
2509 | ||
7a8e76a3 SR |
2510 | cpu = raw_smp_processor_id(); |
2511 | ||
9e01c1b7 | 2512 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2513 | goto out; |
7a8e76a3 SR |
2514 | |
2515 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2516 | |
2517 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 2518 | goto out; |
7a8e76a3 | 2519 | |
be957c44 | 2520 | if (length > BUF_MAX_DATA_SIZE) |
bf41a158 | 2521 | goto out; |
7a8e76a3 | 2522 | |
62f0b3eb | 2523 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 2524 | if (!event) |
d769041f | 2525 | goto out; |
7a8e76a3 SR |
2526 | |
2527 | return event; | |
2528 | ||
d769041f | 2529 | out: |
261842b7 SR |
2530 | trace_recursive_unlock(); |
2531 | ||
2532 | out_nocheck: | |
5168ae50 | 2533 | preempt_enable_notrace(); |
7a8e76a3 SR |
2534 | return NULL; |
2535 | } | |
c4f50183 | 2536 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 | 2537 | |
a1863c21 SR |
2538 | static void |
2539 | rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
7a8e76a3 SR |
2540 | struct ring_buffer_event *event) |
2541 | { | |
69d1b839 SR |
2542 | u64 delta; |
2543 | ||
fa743953 SR |
2544 | /* |
2545 | * The event first in the commit queue updates the | |
2546 | * time stamp. | |
2547 | */ | |
69d1b839 SR |
2548 | if (rb_event_is_commit(cpu_buffer, event)) { |
2549 | /* | |
2550 | * A commit event that is first on a page | |
2551 | * updates the write timestamp with the page stamp | |
2552 | */ | |
2553 | if (!rb_event_index(event)) | |
2554 | cpu_buffer->write_stamp = | |
2555 | cpu_buffer->commit_page->page->time_stamp; | |
2556 | else if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
2557 | delta = event->array[0]; | |
2558 | delta <<= TS_SHIFT; | |
2559 | delta += event->time_delta; | |
2560 | cpu_buffer->write_stamp += delta; | |
2561 | } else | |
2562 | cpu_buffer->write_stamp += event->time_delta; | |
2563 | } | |
a1863c21 | 2564 | } |
bf41a158 | 2565 | |
a1863c21 SR |
2566 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, |
2567 | struct ring_buffer_event *event) | |
2568 | { | |
2569 | local_inc(&cpu_buffer->entries); | |
2570 | rb_update_write_stamp(cpu_buffer, event); | |
fa743953 | 2571 | rb_end_commit(cpu_buffer); |
7a8e76a3 SR |
2572 | } |
2573 | ||
2574 | /** | |
2575 | * ring_buffer_unlock_commit - commit a reserved | |
2576 | * @buffer: The buffer to commit to | |
2577 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
2578 | * |
2579 | * This commits the data to the ring buffer, and releases any locks held. | |
2580 | * | |
2581 | * Must be paired with ring_buffer_lock_reserve. | |
2582 | */ | |
2583 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 2584 | struct ring_buffer_event *event) |
7a8e76a3 SR |
2585 | { |
2586 | struct ring_buffer_per_cpu *cpu_buffer; | |
2587 | int cpu = raw_smp_processor_id(); | |
2588 | ||
2589 | cpu_buffer = buffer->buffers[cpu]; | |
2590 | ||
7a8e76a3 SR |
2591 | rb_commit(cpu_buffer, event); |
2592 | ||
261842b7 SR |
2593 | trace_recursive_unlock(); |
2594 | ||
5168ae50 | 2595 | preempt_enable_notrace(); |
7a8e76a3 SR |
2596 | |
2597 | return 0; | |
2598 | } | |
c4f50183 | 2599 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 2600 | |
f3b9aae1 FW |
2601 | static inline void rb_event_discard(struct ring_buffer_event *event) |
2602 | { | |
69d1b839 SR |
2603 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) |
2604 | event = skip_time_extend(event); | |
2605 | ||
334d4169 LJ |
2606 | /* array[0] holds the actual length for the discarded event */ |
2607 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
2608 | event->type_len = RINGBUF_TYPE_PADDING; | |
f3b9aae1 FW |
2609 | /* time delta must be non zero */ |
2610 | if (!event->time_delta) | |
2611 | event->time_delta = 1; | |
2612 | } | |
2613 | ||
a1863c21 SR |
2614 | /* |
2615 | * Decrement the entries to the page that an event is on. | |
2616 | * The event does not even need to exist, only the pointer | |
2617 | * to the page it is on. This may only be called before the commit | |
2618 | * takes place. | |
2619 | */ | |
2620 | static inline void | |
2621 | rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer, | |
2622 | struct ring_buffer_event *event) | |
2623 | { | |
2624 | unsigned long addr = (unsigned long)event; | |
2625 | struct buffer_page *bpage = cpu_buffer->commit_page; | |
2626 | struct buffer_page *start; | |
2627 | ||
2628 | addr &= PAGE_MASK; | |
2629 | ||
2630 | /* Do the likely case first */ | |
2631 | if (likely(bpage->page == (void *)addr)) { | |
2632 | local_dec(&bpage->entries); | |
2633 | return; | |
2634 | } | |
2635 | ||
2636 | /* | |
2637 | * Because the commit page may be on the reader page we | |
2638 | * start with the next page and check the end loop there. | |
2639 | */ | |
2640 | rb_inc_page(cpu_buffer, &bpage); | |
2641 | start = bpage; | |
2642 | do { | |
2643 | if (bpage->page == (void *)addr) { | |
2644 | local_dec(&bpage->entries); | |
2645 | return; | |
2646 | } | |
2647 | rb_inc_page(cpu_buffer, &bpage); | |
2648 | } while (bpage != start); | |
2649 | ||
2650 | /* commit not part of this buffer?? */ | |
2651 | RB_WARN_ON(cpu_buffer, 1); | |
2652 | } | |
2653 | ||
fa1b47dd SR |
2654 | /** |
2655 | * ring_buffer_commit_discard - discard an event that has not been committed | |
2656 | * @buffer: the ring buffer | |
2657 | * @event: non committed event to discard | |
2658 | * | |
dc892f73 SR |
2659 | * Sometimes an event that is in the ring buffer needs to be ignored. |
2660 | * This function lets the user discard an event in the ring buffer | |
2661 | * and then that event will not be read later. | |
2662 | * | |
2663 | * This function only works if it is called before the the item has been | |
2664 | * committed. It will try to free the event from the ring buffer | |
fa1b47dd SR |
2665 | * if another event has not been added behind it. |
2666 | * | |
2667 | * If another event has been added behind it, it will set the event | |
2668 | * up as discarded, and perform the commit. | |
2669 | * | |
2670 | * If this function is called, do not call ring_buffer_unlock_commit on | |
2671 | * the event. | |
2672 | */ | |
2673 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
2674 | struct ring_buffer_event *event) | |
2675 | { | |
2676 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
2677 | int cpu; |
2678 | ||
2679 | /* The event is discarded regardless */ | |
f3b9aae1 | 2680 | rb_event_discard(event); |
fa1b47dd | 2681 | |
fa743953 SR |
2682 | cpu = smp_processor_id(); |
2683 | cpu_buffer = buffer->buffers[cpu]; | |
2684 | ||
fa1b47dd SR |
2685 | /* |
2686 | * This must only be called if the event has not been | |
2687 | * committed yet. Thus we can assume that preemption | |
2688 | * is still disabled. | |
2689 | */ | |
fa743953 | 2690 | RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing)); |
fa1b47dd | 2691 | |
a1863c21 | 2692 | rb_decrement_entry(cpu_buffer, event); |
0f2541d2 | 2693 | if (rb_try_to_discard(cpu_buffer, event)) |
edd813bf | 2694 | goto out; |
fa1b47dd SR |
2695 | |
2696 | /* | |
2697 | * The commit is still visible by the reader, so we | |
a1863c21 | 2698 | * must still update the timestamp. |
fa1b47dd | 2699 | */ |
a1863c21 | 2700 | rb_update_write_stamp(cpu_buffer, event); |
fa1b47dd | 2701 | out: |
fa743953 | 2702 | rb_end_commit(cpu_buffer); |
fa1b47dd | 2703 | |
f3b9aae1 FW |
2704 | trace_recursive_unlock(); |
2705 | ||
5168ae50 | 2706 | preempt_enable_notrace(); |
fa1b47dd SR |
2707 | |
2708 | } | |
2709 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
2710 | ||
7a8e76a3 SR |
2711 | /** |
2712 | * ring_buffer_write - write data to the buffer without reserving | |
2713 | * @buffer: The ring buffer to write to. | |
2714 | * @length: The length of the data being written (excluding the event header) | |
2715 | * @data: The data to write to the buffer. | |
2716 | * | |
2717 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
2718 | * one function. If you already have the data to write to the buffer, it | |
2719 | * may be easier to simply call this function. | |
2720 | * | |
2721 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
2722 | * and not the length of the event which would hold the header. | |
2723 | */ | |
2724 | int ring_buffer_write(struct ring_buffer *buffer, | |
2725 | unsigned long length, | |
2726 | void *data) | |
2727 | { | |
2728 | struct ring_buffer_per_cpu *cpu_buffer; | |
2729 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
2730 | void *body; |
2731 | int ret = -EBUSY; | |
5168ae50 | 2732 | int cpu; |
7a8e76a3 | 2733 | |
033601a3 | 2734 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2735 | return -EBUSY; |
2736 | ||
5168ae50 | 2737 | preempt_disable_notrace(); |
bf41a158 | 2738 | |
52fbe9cd LJ |
2739 | if (atomic_read(&buffer->record_disabled)) |
2740 | goto out; | |
2741 | ||
7a8e76a3 SR |
2742 | cpu = raw_smp_processor_id(); |
2743 | ||
9e01c1b7 | 2744 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2745 | goto out; |
7a8e76a3 SR |
2746 | |
2747 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2748 | |
2749 | if (atomic_read(&cpu_buffer->record_disabled)) | |
2750 | goto out; | |
2751 | ||
be957c44 SR |
2752 | if (length > BUF_MAX_DATA_SIZE) |
2753 | goto out; | |
2754 | ||
62f0b3eb | 2755 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 SR |
2756 | if (!event) |
2757 | goto out; | |
2758 | ||
2759 | body = rb_event_data(event); | |
2760 | ||
2761 | memcpy(body, data, length); | |
2762 | ||
2763 | rb_commit(cpu_buffer, event); | |
2764 | ||
2765 | ret = 0; | |
2766 | out: | |
5168ae50 | 2767 | preempt_enable_notrace(); |
7a8e76a3 SR |
2768 | |
2769 | return ret; | |
2770 | } | |
c4f50183 | 2771 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 2772 | |
34a148bf | 2773 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
2774 | { |
2775 | struct buffer_page *reader = cpu_buffer->reader_page; | |
77ae365e | 2776 | struct buffer_page *head = rb_set_head_page(cpu_buffer); |
bf41a158 SR |
2777 | struct buffer_page *commit = cpu_buffer->commit_page; |
2778 | ||
77ae365e SR |
2779 | /* In case of error, head will be NULL */ |
2780 | if (unlikely(!head)) | |
2781 | return 1; | |
2782 | ||
bf41a158 SR |
2783 | return reader->read == rb_page_commit(reader) && |
2784 | (commit == reader || | |
2785 | (commit == head && | |
2786 | head->read == rb_page_commit(commit))); | |
2787 | } | |
2788 | ||
7a8e76a3 SR |
2789 | /** |
2790 | * ring_buffer_record_disable - stop all writes into the buffer | |
2791 | * @buffer: The ring buffer to stop writes to. | |
2792 | * | |
2793 | * This prevents all writes to the buffer. Any attempt to write | |
2794 | * to the buffer after this will fail and return NULL. | |
2795 | * | |
2796 | * The caller should call synchronize_sched() after this. | |
2797 | */ | |
2798 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
2799 | { | |
2800 | atomic_inc(&buffer->record_disabled); | |
2801 | } | |
c4f50183 | 2802 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
2803 | |
2804 | /** | |
2805 | * ring_buffer_record_enable - enable writes to the buffer | |
2806 | * @buffer: The ring buffer to enable writes | |
2807 | * | |
2808 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 2809 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
2810 | */ |
2811 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
2812 | { | |
2813 | atomic_dec(&buffer->record_disabled); | |
2814 | } | |
c4f50183 | 2815 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 | 2816 | |
499e5470 SR |
2817 | /** |
2818 | * ring_buffer_record_off - stop all writes into the buffer | |
2819 | * @buffer: The ring buffer to stop writes to. | |
2820 | * | |
2821 | * This prevents all writes to the buffer. Any attempt to write | |
2822 | * to the buffer after this will fail and return NULL. | |
2823 | * | |
2824 | * This is different than ring_buffer_record_disable() as | |
2825 | * it works like an on/off switch, where as the disable() verison | |
2826 | * must be paired with a enable(). | |
2827 | */ | |
2828 | void ring_buffer_record_off(struct ring_buffer *buffer) | |
2829 | { | |
2830 | unsigned int rd; | |
2831 | unsigned int new_rd; | |
2832 | ||
2833 | do { | |
2834 | rd = atomic_read(&buffer->record_disabled); | |
2835 | new_rd = rd | RB_BUFFER_OFF; | |
2836 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
2837 | } | |
2838 | EXPORT_SYMBOL_GPL(ring_buffer_record_off); | |
2839 | ||
2840 | /** | |
2841 | * ring_buffer_record_on - restart writes into the buffer | |
2842 | * @buffer: The ring buffer to start writes to. | |
2843 | * | |
2844 | * This enables all writes to the buffer that was disabled by | |
2845 | * ring_buffer_record_off(). | |
2846 | * | |
2847 | * This is different than ring_buffer_record_enable() as | |
2848 | * it works like an on/off switch, where as the enable() verison | |
2849 | * must be paired with a disable(). | |
2850 | */ | |
2851 | void ring_buffer_record_on(struct ring_buffer *buffer) | |
2852 | { | |
2853 | unsigned int rd; | |
2854 | unsigned int new_rd; | |
2855 | ||
2856 | do { | |
2857 | rd = atomic_read(&buffer->record_disabled); | |
2858 | new_rd = rd & ~RB_BUFFER_OFF; | |
2859 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
2860 | } | |
2861 | EXPORT_SYMBOL_GPL(ring_buffer_record_on); | |
2862 | ||
2863 | /** | |
2864 | * ring_buffer_record_is_on - return true if the ring buffer can write | |
2865 | * @buffer: The ring buffer to see if write is enabled | |
2866 | * | |
2867 | * Returns true if the ring buffer is in a state that it accepts writes. | |
2868 | */ | |
2869 | int ring_buffer_record_is_on(struct ring_buffer *buffer) | |
2870 | { | |
2871 | return !atomic_read(&buffer->record_disabled); | |
2872 | } | |
2873 | ||
7a8e76a3 SR |
2874 | /** |
2875 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
2876 | * @buffer: The ring buffer to stop writes to. | |
2877 | * @cpu: The CPU buffer to stop | |
2878 | * | |
2879 | * This prevents all writes to the buffer. Any attempt to write | |
2880 | * to the buffer after this will fail and return NULL. | |
2881 | * | |
2882 | * The caller should call synchronize_sched() after this. | |
2883 | */ | |
2884 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
2885 | { | |
2886 | struct ring_buffer_per_cpu *cpu_buffer; | |
2887 | ||
9e01c1b7 | 2888 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2889 | return; |
7a8e76a3 SR |
2890 | |
2891 | cpu_buffer = buffer->buffers[cpu]; | |
2892 | atomic_inc(&cpu_buffer->record_disabled); | |
2893 | } | |
c4f50183 | 2894 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
2895 | |
2896 | /** | |
2897 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
2898 | * @buffer: The ring buffer to enable writes | |
2899 | * @cpu: The CPU to enable. | |
2900 | * | |
2901 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 2902 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
2903 | */ |
2904 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
2905 | { | |
2906 | struct ring_buffer_per_cpu *cpu_buffer; | |
2907 | ||
9e01c1b7 | 2908 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2909 | return; |
7a8e76a3 SR |
2910 | |
2911 | cpu_buffer = buffer->buffers[cpu]; | |
2912 | atomic_dec(&cpu_buffer->record_disabled); | |
2913 | } | |
c4f50183 | 2914 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 | 2915 | |
f6195aa0 SR |
2916 | /* |
2917 | * The total entries in the ring buffer is the running counter | |
2918 | * of entries entered into the ring buffer, minus the sum of | |
2919 | * the entries read from the ring buffer and the number of | |
2920 | * entries that were overwritten. | |
2921 | */ | |
2922 | static inline unsigned long | |
2923 | rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer) | |
2924 | { | |
2925 | return local_read(&cpu_buffer->entries) - | |
2926 | (local_read(&cpu_buffer->overrun) + cpu_buffer->read); | |
2927 | } | |
2928 | ||
c64e148a VN |
2929 | /** |
2930 | * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer | |
2931 | * @buffer: The ring buffer | |
2932 | * @cpu: The per CPU buffer to read from. | |
2933 | */ | |
2934 | unsigned long ring_buffer_oldest_event_ts(struct ring_buffer *buffer, int cpu) | |
2935 | { | |
2936 | unsigned long flags; | |
2937 | struct ring_buffer_per_cpu *cpu_buffer; | |
2938 | struct buffer_page *bpage; | |
2939 | unsigned long ret; | |
2940 | ||
2941 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
2942 | return 0; | |
2943 | ||
2944 | cpu_buffer = buffer->buffers[cpu]; | |
7115e3fc | 2945 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
2946 | /* |
2947 | * if the tail is on reader_page, oldest time stamp is on the reader | |
2948 | * page | |
2949 | */ | |
2950 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
2951 | bpage = cpu_buffer->reader_page; | |
2952 | else | |
2953 | bpage = rb_set_head_page(cpu_buffer); | |
2954 | ret = bpage->page->time_stamp; | |
7115e3fc | 2955 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
2956 | |
2957 | return ret; | |
2958 | } | |
2959 | EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts); | |
2960 | ||
2961 | /** | |
2962 | * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer | |
2963 | * @buffer: The ring buffer | |
2964 | * @cpu: The per CPU buffer to read from. | |
2965 | */ | |
2966 | unsigned long ring_buffer_bytes_cpu(struct ring_buffer *buffer, int cpu) | |
2967 | { | |
2968 | struct ring_buffer_per_cpu *cpu_buffer; | |
2969 | unsigned long ret; | |
2970 | ||
2971 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
2972 | return 0; | |
2973 | ||
2974 | cpu_buffer = buffer->buffers[cpu]; | |
2975 | ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes; | |
2976 | ||
2977 | return ret; | |
2978 | } | |
2979 | EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu); | |
2980 | ||
7a8e76a3 SR |
2981 | /** |
2982 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
2983 | * @buffer: The ring buffer | |
2984 | * @cpu: The per CPU buffer to get the entries from. | |
2985 | */ | |
2986 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
2987 | { | |
2988 | struct ring_buffer_per_cpu *cpu_buffer; | |
2989 | ||
9e01c1b7 | 2990 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2991 | return 0; |
7a8e76a3 SR |
2992 | |
2993 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 2994 | |
f6195aa0 | 2995 | return rb_num_of_entries(cpu_buffer); |
7a8e76a3 | 2996 | } |
c4f50183 | 2997 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
2998 | |
2999 | /** | |
3000 | * ring_buffer_overrun_cpu - get the number of overruns in a cpu_buffer | |
3001 | * @buffer: The ring buffer | |
3002 | * @cpu: The per CPU buffer to get the number of overruns from | |
3003 | */ | |
3004 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3005 | { | |
3006 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3007 | unsigned long ret; |
7a8e76a3 | 3008 | |
9e01c1b7 | 3009 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3010 | return 0; |
7a8e76a3 SR |
3011 | |
3012 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3013 | ret = local_read(&cpu_buffer->overrun); |
554f786e SR |
3014 | |
3015 | return ret; | |
7a8e76a3 | 3016 | } |
c4f50183 | 3017 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 3018 | |
f0d2c681 SR |
3019 | /** |
3020 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by commits | |
3021 | * @buffer: The ring buffer | |
3022 | * @cpu: The per CPU buffer to get the number of overruns from | |
3023 | */ | |
3024 | unsigned long | |
3025 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3026 | { | |
3027 | struct ring_buffer_per_cpu *cpu_buffer; | |
3028 | unsigned long ret; | |
3029 | ||
3030 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3031 | return 0; | |
3032 | ||
3033 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3034 | ret = local_read(&cpu_buffer->commit_overrun); |
f0d2c681 SR |
3035 | |
3036 | return ret; | |
3037 | } | |
3038 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
3039 | ||
7a8e76a3 SR |
3040 | /** |
3041 | * ring_buffer_entries - get the number of entries in a buffer | |
3042 | * @buffer: The ring buffer | |
3043 | * | |
3044 | * Returns the total number of entries in the ring buffer | |
3045 | * (all CPU entries) | |
3046 | */ | |
3047 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
3048 | { | |
3049 | struct ring_buffer_per_cpu *cpu_buffer; | |
3050 | unsigned long entries = 0; | |
3051 | int cpu; | |
3052 | ||
3053 | /* if you care about this being correct, lock the buffer */ | |
3054 | for_each_buffer_cpu(buffer, cpu) { | |
3055 | cpu_buffer = buffer->buffers[cpu]; | |
f6195aa0 | 3056 | entries += rb_num_of_entries(cpu_buffer); |
7a8e76a3 SR |
3057 | } |
3058 | ||
3059 | return entries; | |
3060 | } | |
c4f50183 | 3061 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
3062 | |
3063 | /** | |
67b394f7 | 3064 | * ring_buffer_overruns - get the number of overruns in buffer |
7a8e76a3 SR |
3065 | * @buffer: The ring buffer |
3066 | * | |
3067 | * Returns the total number of overruns in the ring buffer | |
3068 | * (all CPU entries) | |
3069 | */ | |
3070 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
3071 | { | |
3072 | struct ring_buffer_per_cpu *cpu_buffer; | |
3073 | unsigned long overruns = 0; | |
3074 | int cpu; | |
3075 | ||
3076 | /* if you care about this being correct, lock the buffer */ | |
3077 | for_each_buffer_cpu(buffer, cpu) { | |
3078 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3079 | overruns += local_read(&cpu_buffer->overrun); |
7a8e76a3 SR |
3080 | } |
3081 | ||
3082 | return overruns; | |
3083 | } | |
c4f50183 | 3084 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 3085 | |
642edba5 | 3086 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
3087 | { |
3088 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3089 | ||
d769041f SR |
3090 | /* Iterator usage is expected to have record disabled */ |
3091 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
77ae365e SR |
3092 | iter->head_page = rb_set_head_page(cpu_buffer); |
3093 | if (unlikely(!iter->head_page)) | |
3094 | return; | |
3095 | iter->head = iter->head_page->read; | |
d769041f SR |
3096 | } else { |
3097 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 3098 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
3099 | } |
3100 | if (iter->head) | |
3101 | iter->read_stamp = cpu_buffer->read_stamp; | |
3102 | else | |
abc9b56d | 3103 | iter->read_stamp = iter->head_page->page->time_stamp; |
492a74f4 SR |
3104 | iter->cache_reader_page = cpu_buffer->reader_page; |
3105 | iter->cache_read = cpu_buffer->read; | |
642edba5 | 3106 | } |
f83c9d0f | 3107 | |
642edba5 SR |
3108 | /** |
3109 | * ring_buffer_iter_reset - reset an iterator | |
3110 | * @iter: The iterator to reset | |
3111 | * | |
3112 | * Resets the iterator, so that it will start from the beginning | |
3113 | * again. | |
3114 | */ | |
3115 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
3116 | { | |
554f786e | 3117 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
3118 | unsigned long flags; |
3119 | ||
554f786e SR |
3120 | if (!iter) |
3121 | return; | |
3122 | ||
3123 | cpu_buffer = iter->cpu_buffer; | |
3124 | ||
5389f6fa | 3125 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
642edba5 | 3126 | rb_iter_reset(iter); |
5389f6fa | 3127 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 3128 | } |
c4f50183 | 3129 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
3130 | |
3131 | /** | |
3132 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
3133 | * @iter: The iterator to check | |
3134 | */ | |
3135 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
3136 | { | |
3137 | struct ring_buffer_per_cpu *cpu_buffer; | |
3138 | ||
3139 | cpu_buffer = iter->cpu_buffer; | |
3140 | ||
bf41a158 SR |
3141 | return iter->head_page == cpu_buffer->commit_page && |
3142 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 3143 | } |
c4f50183 | 3144 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
3145 | |
3146 | static void | |
3147 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
3148 | struct ring_buffer_event *event) | |
3149 | { | |
3150 | u64 delta; | |
3151 | ||
334d4169 | 3152 | switch (event->type_len) { |
7a8e76a3 SR |
3153 | case RINGBUF_TYPE_PADDING: |
3154 | return; | |
3155 | ||
3156 | case RINGBUF_TYPE_TIME_EXTEND: | |
3157 | delta = event->array[0]; | |
3158 | delta <<= TS_SHIFT; | |
3159 | delta += event->time_delta; | |
3160 | cpu_buffer->read_stamp += delta; | |
3161 | return; | |
3162 | ||
3163 | case RINGBUF_TYPE_TIME_STAMP: | |
3164 | /* FIXME: not implemented */ | |
3165 | return; | |
3166 | ||
3167 | case RINGBUF_TYPE_DATA: | |
3168 | cpu_buffer->read_stamp += event->time_delta; | |
3169 | return; | |
3170 | ||
3171 | default: | |
3172 | BUG(); | |
3173 | } | |
3174 | return; | |
3175 | } | |
3176 | ||
3177 | static void | |
3178 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
3179 | struct ring_buffer_event *event) | |
3180 | { | |
3181 | u64 delta; | |
3182 | ||
334d4169 | 3183 | switch (event->type_len) { |
7a8e76a3 SR |
3184 | case RINGBUF_TYPE_PADDING: |
3185 | return; | |
3186 | ||
3187 | case RINGBUF_TYPE_TIME_EXTEND: | |
3188 | delta = event->array[0]; | |
3189 | delta <<= TS_SHIFT; | |
3190 | delta += event->time_delta; | |
3191 | iter->read_stamp += delta; | |
3192 | return; | |
3193 | ||
3194 | case RINGBUF_TYPE_TIME_STAMP: | |
3195 | /* FIXME: not implemented */ | |
3196 | return; | |
3197 | ||
3198 | case RINGBUF_TYPE_DATA: | |
3199 | iter->read_stamp += event->time_delta; | |
3200 | return; | |
3201 | ||
3202 | default: | |
3203 | BUG(); | |
3204 | } | |
3205 | return; | |
3206 | } | |
3207 | ||
d769041f SR |
3208 | static struct buffer_page * |
3209 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 3210 | { |
d769041f | 3211 | struct buffer_page *reader = NULL; |
66a8cb95 | 3212 | unsigned long overwrite; |
d769041f | 3213 | unsigned long flags; |
818e3dd3 | 3214 | int nr_loops = 0; |
77ae365e | 3215 | int ret; |
d769041f | 3216 | |
3e03fb7f | 3217 | local_irq_save(flags); |
0199c4e6 | 3218 | arch_spin_lock(&cpu_buffer->lock); |
d769041f SR |
3219 | |
3220 | again: | |
818e3dd3 SR |
3221 | /* |
3222 | * This should normally only loop twice. But because the | |
3223 | * start of the reader inserts an empty page, it causes | |
3224 | * a case where we will loop three times. There should be no | |
3225 | * reason to loop four times (that I know of). | |
3226 | */ | |
3e89c7bb | 3227 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
3228 | reader = NULL; |
3229 | goto out; | |
3230 | } | |
3231 | ||
d769041f SR |
3232 | reader = cpu_buffer->reader_page; |
3233 | ||
3234 | /* If there's more to read, return this page */ | |
bf41a158 | 3235 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
3236 | goto out; |
3237 | ||
3238 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
3239 | if (RB_WARN_ON(cpu_buffer, |
3240 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
3241 | goto out; | |
d769041f SR |
3242 | |
3243 | /* check if we caught up to the tail */ | |
3244 | reader = NULL; | |
bf41a158 | 3245 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 3246 | goto out; |
7a8e76a3 SR |
3247 | |
3248 | /* | |
d769041f | 3249 | * Reset the reader page to size zero. |
7a8e76a3 | 3250 | */ |
77ae365e SR |
3251 | local_set(&cpu_buffer->reader_page->write, 0); |
3252 | local_set(&cpu_buffer->reader_page->entries, 0); | |
3253 | local_set(&cpu_buffer->reader_page->page->commit, 0); | |
ff0ff84a | 3254 | cpu_buffer->reader_page->real_end = 0; |
7a8e76a3 | 3255 | |
77ae365e SR |
3256 | spin: |
3257 | /* | |
3258 | * Splice the empty reader page into the list around the head. | |
3259 | */ | |
3260 | reader = rb_set_head_page(cpu_buffer); | |
0e1ff5d7 | 3261 | cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next); |
d769041f | 3262 | cpu_buffer->reader_page->list.prev = reader->list.prev; |
bf41a158 | 3263 | |
3adc54fa SR |
3264 | /* |
3265 | * cpu_buffer->pages just needs to point to the buffer, it | |
3266 | * has no specific buffer page to point to. Lets move it out | |
25985edc | 3267 | * of our way so we don't accidentally swap it. |
3adc54fa SR |
3268 | */ |
3269 | cpu_buffer->pages = reader->list.prev; | |
3270 | ||
77ae365e SR |
3271 | /* The reader page will be pointing to the new head */ |
3272 | rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list); | |
7a8e76a3 | 3273 | |
66a8cb95 SR |
3274 | /* |
3275 | * We want to make sure we read the overruns after we set up our | |
3276 | * pointers to the next object. The writer side does a | |
3277 | * cmpxchg to cross pages which acts as the mb on the writer | |
3278 | * side. Note, the reader will constantly fail the swap | |
3279 | * while the writer is updating the pointers, so this | |
3280 | * guarantees that the overwrite recorded here is the one we | |
3281 | * want to compare with the last_overrun. | |
3282 | */ | |
3283 | smp_mb(); | |
3284 | overwrite = local_read(&(cpu_buffer->overrun)); | |
3285 | ||
77ae365e SR |
3286 | /* |
3287 | * Here's the tricky part. | |
3288 | * | |
3289 | * We need to move the pointer past the header page. | |
3290 | * But we can only do that if a writer is not currently | |
3291 | * moving it. The page before the header page has the | |
3292 | * flag bit '1' set if it is pointing to the page we want. | |
3293 | * but if the writer is in the process of moving it | |
3294 | * than it will be '2' or already moved '0'. | |
3295 | */ | |
3296 | ||
3297 | ret = rb_head_page_replace(reader, cpu_buffer->reader_page); | |
7a8e76a3 SR |
3298 | |
3299 | /* | |
77ae365e | 3300 | * If we did not convert it, then we must try again. |
7a8e76a3 | 3301 | */ |
77ae365e SR |
3302 | if (!ret) |
3303 | goto spin; | |
7a8e76a3 | 3304 | |
77ae365e SR |
3305 | /* |
3306 | * Yeah! We succeeded in replacing the page. | |
3307 | * | |
3308 | * Now make the new head point back to the reader page. | |
3309 | */ | |
5ded3dc6 | 3310 | rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list; |
77ae365e | 3311 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
d769041f SR |
3312 | |
3313 | /* Finally update the reader page to the new head */ | |
3314 | cpu_buffer->reader_page = reader; | |
3315 | rb_reset_reader_page(cpu_buffer); | |
3316 | ||
66a8cb95 SR |
3317 | if (overwrite != cpu_buffer->last_overrun) { |
3318 | cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun; | |
3319 | cpu_buffer->last_overrun = overwrite; | |
3320 | } | |
3321 | ||
d769041f SR |
3322 | goto again; |
3323 | ||
3324 | out: | |
0199c4e6 | 3325 | arch_spin_unlock(&cpu_buffer->lock); |
3e03fb7f | 3326 | local_irq_restore(flags); |
d769041f SR |
3327 | |
3328 | return reader; | |
3329 | } | |
3330 | ||
3331 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
3332 | { | |
3333 | struct ring_buffer_event *event; | |
3334 | struct buffer_page *reader; | |
3335 | unsigned length; | |
3336 | ||
3337 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 3338 | |
d769041f | 3339 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
3340 | if (RB_WARN_ON(cpu_buffer, !reader)) |
3341 | return; | |
7a8e76a3 | 3342 | |
d769041f SR |
3343 | event = rb_reader_event(cpu_buffer); |
3344 | ||
a1863c21 | 3345 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
e4906eff | 3346 | cpu_buffer->read++; |
d769041f SR |
3347 | |
3348 | rb_update_read_stamp(cpu_buffer, event); | |
3349 | ||
3350 | length = rb_event_length(event); | |
6f807acd | 3351 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
3352 | } |
3353 | ||
3354 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
3355 | { | |
7a8e76a3 SR |
3356 | struct ring_buffer_per_cpu *cpu_buffer; |
3357 | struct ring_buffer_event *event; | |
3358 | unsigned length; | |
3359 | ||
3360 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 SR |
3361 | |
3362 | /* | |
3363 | * Check if we are at the end of the buffer. | |
3364 | */ | |
bf41a158 | 3365 | if (iter->head >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
3366 | /* discarded commits can make the page empty */ |
3367 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 3368 | return; |
d769041f | 3369 | rb_inc_iter(iter); |
7a8e76a3 SR |
3370 | return; |
3371 | } | |
3372 | ||
3373 | event = rb_iter_head_event(iter); | |
3374 | ||
3375 | length = rb_event_length(event); | |
3376 | ||
3377 | /* | |
3378 | * This should not be called to advance the header if we are | |
3379 | * at the tail of the buffer. | |
3380 | */ | |
3e89c7bb | 3381 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 3382 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
3383 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
3384 | return; | |
7a8e76a3 SR |
3385 | |
3386 | rb_update_iter_read_stamp(iter, event); | |
3387 | ||
3388 | iter->head += length; | |
3389 | ||
3390 | /* check for end of page padding */ | |
bf41a158 SR |
3391 | if ((iter->head >= rb_page_size(iter->head_page)) && |
3392 | (iter->head_page != cpu_buffer->commit_page)) | |
7a8e76a3 SR |
3393 | rb_advance_iter(iter); |
3394 | } | |
3395 | ||
66a8cb95 SR |
3396 | static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer) |
3397 | { | |
3398 | return cpu_buffer->lost_events; | |
3399 | } | |
3400 | ||
f83c9d0f | 3401 | static struct ring_buffer_event * |
66a8cb95 SR |
3402 | rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts, |
3403 | unsigned long *lost_events) | |
7a8e76a3 | 3404 | { |
7a8e76a3 | 3405 | struct ring_buffer_event *event; |
d769041f | 3406 | struct buffer_page *reader; |
818e3dd3 | 3407 | int nr_loops = 0; |
7a8e76a3 | 3408 | |
7a8e76a3 | 3409 | again: |
818e3dd3 | 3410 | /* |
69d1b839 SR |
3411 | * We repeat when a time extend is encountered. |
3412 | * Since the time extend is always attached to a data event, | |
3413 | * we should never loop more than once. | |
3414 | * (We never hit the following condition more than twice). | |
818e3dd3 | 3415 | */ |
69d1b839 | 3416 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 3417 | return NULL; |
818e3dd3 | 3418 | |
d769041f SR |
3419 | reader = rb_get_reader_page(cpu_buffer); |
3420 | if (!reader) | |
7a8e76a3 SR |
3421 | return NULL; |
3422 | ||
d769041f | 3423 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 3424 | |
334d4169 | 3425 | switch (event->type_len) { |
7a8e76a3 | 3426 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3427 | if (rb_null_event(event)) |
3428 | RB_WARN_ON(cpu_buffer, 1); | |
3429 | /* | |
3430 | * Because the writer could be discarding every | |
3431 | * event it creates (which would probably be bad) | |
3432 | * if we were to go back to "again" then we may never | |
3433 | * catch up, and will trigger the warn on, or lock | |
3434 | * the box. Return the padding, and we will release | |
3435 | * the current locks, and try again. | |
3436 | */ | |
2d622719 | 3437 | return event; |
7a8e76a3 SR |
3438 | |
3439 | case RINGBUF_TYPE_TIME_EXTEND: | |
3440 | /* Internal data, OK to advance */ | |
d769041f | 3441 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3442 | goto again; |
3443 | ||
3444 | case RINGBUF_TYPE_TIME_STAMP: | |
3445 | /* FIXME: not implemented */ | |
d769041f | 3446 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3447 | goto again; |
3448 | ||
3449 | case RINGBUF_TYPE_DATA: | |
3450 | if (ts) { | |
3451 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
d8eeb2d3 | 3452 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, |
37886f6a | 3453 | cpu_buffer->cpu, ts); |
7a8e76a3 | 3454 | } |
66a8cb95 SR |
3455 | if (lost_events) |
3456 | *lost_events = rb_lost_events(cpu_buffer); | |
7a8e76a3 SR |
3457 | return event; |
3458 | ||
3459 | default: | |
3460 | BUG(); | |
3461 | } | |
3462 | ||
3463 | return NULL; | |
3464 | } | |
c4f50183 | 3465 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 3466 | |
f83c9d0f SR |
3467 | static struct ring_buffer_event * |
3468 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
3469 | { |
3470 | struct ring_buffer *buffer; | |
3471 | struct ring_buffer_per_cpu *cpu_buffer; | |
3472 | struct ring_buffer_event *event; | |
818e3dd3 | 3473 | int nr_loops = 0; |
7a8e76a3 | 3474 | |
7a8e76a3 SR |
3475 | cpu_buffer = iter->cpu_buffer; |
3476 | buffer = cpu_buffer->buffer; | |
3477 | ||
492a74f4 SR |
3478 | /* |
3479 | * Check if someone performed a consuming read to | |
3480 | * the buffer. A consuming read invalidates the iterator | |
3481 | * and we need to reset the iterator in this case. | |
3482 | */ | |
3483 | if (unlikely(iter->cache_read != cpu_buffer->read || | |
3484 | iter->cache_reader_page != cpu_buffer->reader_page)) | |
3485 | rb_iter_reset(iter); | |
3486 | ||
7a8e76a3 | 3487 | again: |
3c05d748 SR |
3488 | if (ring_buffer_iter_empty(iter)) |
3489 | return NULL; | |
3490 | ||
818e3dd3 | 3491 | /* |
69d1b839 SR |
3492 | * We repeat when a time extend is encountered. |
3493 | * Since the time extend is always attached to a data event, | |
3494 | * we should never loop more than once. | |
3495 | * (We never hit the following condition more than twice). | |
818e3dd3 | 3496 | */ |
69d1b839 | 3497 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 3498 | return NULL; |
818e3dd3 | 3499 | |
7a8e76a3 SR |
3500 | if (rb_per_cpu_empty(cpu_buffer)) |
3501 | return NULL; | |
3502 | ||
3c05d748 SR |
3503 | if (iter->head >= local_read(&iter->head_page->page->commit)) { |
3504 | rb_inc_iter(iter); | |
3505 | goto again; | |
3506 | } | |
3507 | ||
7a8e76a3 SR |
3508 | event = rb_iter_head_event(iter); |
3509 | ||
334d4169 | 3510 | switch (event->type_len) { |
7a8e76a3 | 3511 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3512 | if (rb_null_event(event)) { |
3513 | rb_inc_iter(iter); | |
3514 | goto again; | |
3515 | } | |
3516 | rb_advance_iter(iter); | |
3517 | return event; | |
7a8e76a3 SR |
3518 | |
3519 | case RINGBUF_TYPE_TIME_EXTEND: | |
3520 | /* Internal data, OK to advance */ | |
3521 | rb_advance_iter(iter); | |
3522 | goto again; | |
3523 | ||
3524 | case RINGBUF_TYPE_TIME_STAMP: | |
3525 | /* FIXME: not implemented */ | |
3526 | rb_advance_iter(iter); | |
3527 | goto again; | |
3528 | ||
3529 | case RINGBUF_TYPE_DATA: | |
3530 | if (ts) { | |
3531 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
3532 | ring_buffer_normalize_time_stamp(buffer, |
3533 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
3534 | } |
3535 | return event; | |
3536 | ||
3537 | default: | |
3538 | BUG(); | |
3539 | } | |
3540 | ||
3541 | return NULL; | |
3542 | } | |
c4f50183 | 3543 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 3544 | |
8d707e8e SR |
3545 | static inline int rb_ok_to_lock(void) |
3546 | { | |
3547 | /* | |
3548 | * If an NMI die dumps out the content of the ring buffer | |
3549 | * do not grab locks. We also permanently disable the ring | |
3550 | * buffer too. A one time deal is all you get from reading | |
3551 | * the ring buffer from an NMI. | |
3552 | */ | |
464e85eb | 3553 | if (likely(!in_nmi())) |
8d707e8e SR |
3554 | return 1; |
3555 | ||
3556 | tracing_off_permanent(); | |
3557 | return 0; | |
3558 | } | |
3559 | ||
f83c9d0f SR |
3560 | /** |
3561 | * ring_buffer_peek - peek at the next event to be read | |
3562 | * @buffer: The ring buffer to read | |
3563 | * @cpu: The cpu to peak at | |
3564 | * @ts: The timestamp counter of this event. | |
66a8cb95 | 3565 | * @lost_events: a variable to store if events were lost (may be NULL) |
f83c9d0f SR |
3566 | * |
3567 | * This will return the event that will be read next, but does | |
3568 | * not consume the data. | |
3569 | */ | |
3570 | struct ring_buffer_event * | |
66a8cb95 SR |
3571 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts, |
3572 | unsigned long *lost_events) | |
f83c9d0f SR |
3573 | { |
3574 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 3575 | struct ring_buffer_event *event; |
f83c9d0f | 3576 | unsigned long flags; |
8d707e8e | 3577 | int dolock; |
f83c9d0f | 3578 | |
554f786e | 3579 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3580 | return NULL; |
554f786e | 3581 | |
8d707e8e | 3582 | dolock = rb_ok_to_lock(); |
2d622719 | 3583 | again: |
8d707e8e SR |
3584 | local_irq_save(flags); |
3585 | if (dolock) | |
5389f6fa | 3586 | raw_spin_lock(&cpu_buffer->reader_lock); |
66a8cb95 | 3587 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
469535a5 RR |
3588 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
3589 | rb_advance_reader(cpu_buffer); | |
8d707e8e | 3590 | if (dolock) |
5389f6fa | 3591 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3592 | local_irq_restore(flags); |
f83c9d0f | 3593 | |
1b959e18 | 3594 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3595 | goto again; |
2d622719 | 3596 | |
f83c9d0f SR |
3597 | return event; |
3598 | } | |
3599 | ||
3600 | /** | |
3601 | * ring_buffer_iter_peek - peek at the next event to be read | |
3602 | * @iter: The ring buffer iterator | |
3603 | * @ts: The timestamp counter of this event. | |
3604 | * | |
3605 | * This will return the event that will be read next, but does | |
3606 | * not increment the iterator. | |
3607 | */ | |
3608 | struct ring_buffer_event * | |
3609 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
3610 | { | |
3611 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3612 | struct ring_buffer_event *event; | |
3613 | unsigned long flags; | |
3614 | ||
2d622719 | 3615 | again: |
5389f6fa | 3616 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3617 | event = rb_iter_peek(iter, ts); |
5389f6fa | 3618 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3619 | |
1b959e18 | 3620 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3621 | goto again; |
2d622719 | 3622 | |
f83c9d0f SR |
3623 | return event; |
3624 | } | |
3625 | ||
7a8e76a3 SR |
3626 | /** |
3627 | * ring_buffer_consume - return an event and consume it | |
3628 | * @buffer: The ring buffer to get the next event from | |
66a8cb95 SR |
3629 | * @cpu: the cpu to read the buffer from |
3630 | * @ts: a variable to store the timestamp (may be NULL) | |
3631 | * @lost_events: a variable to store if events were lost (may be NULL) | |
7a8e76a3 SR |
3632 | * |
3633 | * Returns the next event in the ring buffer, and that event is consumed. | |
3634 | * Meaning, that sequential reads will keep returning a different event, | |
3635 | * and eventually empty the ring buffer if the producer is slower. | |
3636 | */ | |
3637 | struct ring_buffer_event * | |
66a8cb95 SR |
3638 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts, |
3639 | unsigned long *lost_events) | |
7a8e76a3 | 3640 | { |
554f786e SR |
3641 | struct ring_buffer_per_cpu *cpu_buffer; |
3642 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 3643 | unsigned long flags; |
8d707e8e SR |
3644 | int dolock; |
3645 | ||
3646 | dolock = rb_ok_to_lock(); | |
7a8e76a3 | 3647 | |
2d622719 | 3648 | again: |
554f786e SR |
3649 | /* might be called in atomic */ |
3650 | preempt_disable(); | |
3651 | ||
9e01c1b7 | 3652 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 3653 | goto out; |
7a8e76a3 | 3654 | |
554f786e | 3655 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
3656 | local_irq_save(flags); |
3657 | if (dolock) | |
5389f6fa | 3658 | raw_spin_lock(&cpu_buffer->reader_lock); |
f83c9d0f | 3659 | |
66a8cb95 SR |
3660 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
3661 | if (event) { | |
3662 | cpu_buffer->lost_events = 0; | |
469535a5 | 3663 | rb_advance_reader(cpu_buffer); |
66a8cb95 | 3664 | } |
7a8e76a3 | 3665 | |
8d707e8e | 3666 | if (dolock) |
5389f6fa | 3667 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3668 | local_irq_restore(flags); |
f83c9d0f | 3669 | |
554f786e SR |
3670 | out: |
3671 | preempt_enable(); | |
3672 | ||
1b959e18 | 3673 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3674 | goto again; |
2d622719 | 3675 | |
7a8e76a3 SR |
3676 | return event; |
3677 | } | |
c4f50183 | 3678 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
3679 | |
3680 | /** | |
72c9ddfd | 3681 | * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer |
7a8e76a3 SR |
3682 | * @buffer: The ring buffer to read from |
3683 | * @cpu: The cpu buffer to iterate over | |
3684 | * | |
72c9ddfd DM |
3685 | * This performs the initial preparations necessary to iterate |
3686 | * through the buffer. Memory is allocated, buffer recording | |
3687 | * is disabled, and the iterator pointer is returned to the caller. | |
7a8e76a3 | 3688 | * |
72c9ddfd DM |
3689 | * Disabling buffer recordng prevents the reading from being |
3690 | * corrupted. This is not a consuming read, so a producer is not | |
3691 | * expected. | |
3692 | * | |
3693 | * After a sequence of ring_buffer_read_prepare calls, the user is | |
3694 | * expected to make at least one call to ring_buffer_prepare_sync. | |
3695 | * Afterwards, ring_buffer_read_start is invoked to get things going | |
3696 | * for real. | |
3697 | * | |
3698 | * This overall must be paired with ring_buffer_finish. | |
7a8e76a3 SR |
3699 | */ |
3700 | struct ring_buffer_iter * | |
72c9ddfd | 3701 | ring_buffer_read_prepare(struct ring_buffer *buffer, int cpu) |
7a8e76a3 SR |
3702 | { |
3703 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3704 | struct ring_buffer_iter *iter; |
7a8e76a3 | 3705 | |
9e01c1b7 | 3706 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3707 | return NULL; |
7a8e76a3 SR |
3708 | |
3709 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
3710 | if (!iter) | |
8aabee57 | 3711 | return NULL; |
7a8e76a3 SR |
3712 | |
3713 | cpu_buffer = buffer->buffers[cpu]; | |
3714 | ||
3715 | iter->cpu_buffer = cpu_buffer; | |
3716 | ||
83f40318 | 3717 | atomic_inc(&buffer->resize_disabled); |
7a8e76a3 | 3718 | atomic_inc(&cpu_buffer->record_disabled); |
72c9ddfd DM |
3719 | |
3720 | return iter; | |
3721 | } | |
3722 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare); | |
3723 | ||
3724 | /** | |
3725 | * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls | |
3726 | * | |
3727 | * All previously invoked ring_buffer_read_prepare calls to prepare | |
3728 | * iterators will be synchronized. Afterwards, read_buffer_read_start | |
3729 | * calls on those iterators are allowed. | |
3730 | */ | |
3731 | void | |
3732 | ring_buffer_read_prepare_sync(void) | |
3733 | { | |
7a8e76a3 | 3734 | synchronize_sched(); |
72c9ddfd DM |
3735 | } |
3736 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync); | |
3737 | ||
3738 | /** | |
3739 | * ring_buffer_read_start - start a non consuming read of the buffer | |
3740 | * @iter: The iterator returned by ring_buffer_read_prepare | |
3741 | * | |
3742 | * This finalizes the startup of an iteration through the buffer. | |
3743 | * The iterator comes from a call to ring_buffer_read_prepare and | |
3744 | * an intervening ring_buffer_read_prepare_sync must have been | |
3745 | * performed. | |
3746 | * | |
3747 | * Must be paired with ring_buffer_finish. | |
3748 | */ | |
3749 | void | |
3750 | ring_buffer_read_start(struct ring_buffer_iter *iter) | |
3751 | { | |
3752 | struct ring_buffer_per_cpu *cpu_buffer; | |
3753 | unsigned long flags; | |
3754 | ||
3755 | if (!iter) | |
3756 | return; | |
3757 | ||
3758 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 | 3759 | |
5389f6fa | 3760 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
0199c4e6 | 3761 | arch_spin_lock(&cpu_buffer->lock); |
642edba5 | 3762 | rb_iter_reset(iter); |
0199c4e6 | 3763 | arch_spin_unlock(&cpu_buffer->lock); |
5389f6fa | 3764 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 3765 | } |
c4f50183 | 3766 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
3767 | |
3768 | /** | |
3769 | * ring_buffer_finish - finish reading the iterator of the buffer | |
3770 | * @iter: The iterator retrieved by ring_buffer_start | |
3771 | * | |
3772 | * This re-enables the recording to the buffer, and frees the | |
3773 | * iterator. | |
3774 | */ | |
3775 | void | |
3776 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
3777 | { | |
3778 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3779 | ||
659f451f SR |
3780 | /* |
3781 | * Ring buffer is disabled from recording, here's a good place | |
3782 | * to check the integrity of the ring buffer. | |
3783 | */ | |
3784 | rb_check_pages(cpu_buffer); | |
3785 | ||
7a8e76a3 | 3786 | atomic_dec(&cpu_buffer->record_disabled); |
83f40318 | 3787 | atomic_dec(&cpu_buffer->buffer->resize_disabled); |
7a8e76a3 SR |
3788 | kfree(iter); |
3789 | } | |
c4f50183 | 3790 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
3791 | |
3792 | /** | |
3793 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
3794 | * @iter: The ring buffer iterator | |
3795 | * @ts: The time stamp of the event read. | |
3796 | * | |
3797 | * This reads the next event in the ring buffer and increments the iterator. | |
3798 | */ | |
3799 | struct ring_buffer_event * | |
3800 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
3801 | { | |
3802 | struct ring_buffer_event *event; | |
f83c9d0f SR |
3803 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
3804 | unsigned long flags; | |
7a8e76a3 | 3805 | |
5389f6fa | 3806 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
7e9391cf | 3807 | again: |
f83c9d0f | 3808 | event = rb_iter_peek(iter, ts); |
7a8e76a3 | 3809 | if (!event) |
f83c9d0f | 3810 | goto out; |
7a8e76a3 | 3811 | |
7e9391cf SR |
3812 | if (event->type_len == RINGBUF_TYPE_PADDING) |
3813 | goto again; | |
3814 | ||
7a8e76a3 | 3815 | rb_advance_iter(iter); |
f83c9d0f | 3816 | out: |
5389f6fa | 3817 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
3818 | |
3819 | return event; | |
3820 | } | |
c4f50183 | 3821 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
3822 | |
3823 | /** | |
3824 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
3825 | * @buffer: The ring buffer. | |
3826 | */ | |
438ced17 | 3827 | unsigned long ring_buffer_size(struct ring_buffer *buffer, int cpu) |
7a8e76a3 | 3828 | { |
438ced17 VN |
3829 | /* |
3830 | * Earlier, this method returned | |
3831 | * BUF_PAGE_SIZE * buffer->nr_pages | |
3832 | * Since the nr_pages field is now removed, we have converted this to | |
3833 | * return the per cpu buffer value. | |
3834 | */ | |
3835 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3836 | return 0; | |
3837 | ||
3838 | return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages; | |
7a8e76a3 | 3839 | } |
c4f50183 | 3840 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
3841 | |
3842 | static void | |
3843 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
3844 | { | |
77ae365e SR |
3845 | rb_head_page_deactivate(cpu_buffer); |
3846 | ||
7a8e76a3 | 3847 | cpu_buffer->head_page |
3adc54fa | 3848 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 3849 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 3850 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 3851 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 3852 | |
6f807acd | 3853 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
3854 | |
3855 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
3856 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
3857 | ||
3858 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
5040b4b7 | 3859 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
bf41a158 | 3860 | local_set(&cpu_buffer->reader_page->write, 0); |
778c55d4 | 3861 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 3862 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 3863 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 3864 | |
77ae365e | 3865 | local_set(&cpu_buffer->commit_overrun, 0); |
c64e148a | 3866 | local_set(&cpu_buffer->entries_bytes, 0); |
77ae365e | 3867 | local_set(&cpu_buffer->overrun, 0); |
e4906eff | 3868 | local_set(&cpu_buffer->entries, 0); |
fa743953 SR |
3869 | local_set(&cpu_buffer->committing, 0); |
3870 | local_set(&cpu_buffer->commits, 0); | |
77ae365e | 3871 | cpu_buffer->read = 0; |
c64e148a | 3872 | cpu_buffer->read_bytes = 0; |
69507c06 SR |
3873 | |
3874 | cpu_buffer->write_stamp = 0; | |
3875 | cpu_buffer->read_stamp = 0; | |
77ae365e | 3876 | |
66a8cb95 SR |
3877 | cpu_buffer->lost_events = 0; |
3878 | cpu_buffer->last_overrun = 0; | |
3879 | ||
77ae365e | 3880 | rb_head_page_activate(cpu_buffer); |
7a8e76a3 SR |
3881 | } |
3882 | ||
3883 | /** | |
3884 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
3885 | * @buffer: The ring buffer to reset a per cpu buffer of | |
3886 | * @cpu: The CPU buffer to be reset | |
3887 | */ | |
3888 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
3889 | { | |
3890 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
3891 | unsigned long flags; | |
3892 | ||
9e01c1b7 | 3893 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3894 | return; |
7a8e76a3 | 3895 | |
83f40318 | 3896 | atomic_inc(&buffer->resize_disabled); |
41ede23e SR |
3897 | atomic_inc(&cpu_buffer->record_disabled); |
3898 | ||
83f40318 VN |
3899 | /* Make sure all commits have finished */ |
3900 | synchronize_sched(); | |
3901 | ||
5389f6fa | 3902 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3903 | |
41b6a95d SR |
3904 | if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing))) |
3905 | goto out; | |
3906 | ||
0199c4e6 | 3907 | arch_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
3908 | |
3909 | rb_reset_cpu(cpu_buffer); | |
3910 | ||
0199c4e6 | 3911 | arch_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 3912 | |
41b6a95d | 3913 | out: |
5389f6fa | 3914 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
41ede23e SR |
3915 | |
3916 | atomic_dec(&cpu_buffer->record_disabled); | |
83f40318 | 3917 | atomic_dec(&buffer->resize_disabled); |
7a8e76a3 | 3918 | } |
c4f50183 | 3919 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
3920 | |
3921 | /** | |
3922 | * ring_buffer_reset - reset a ring buffer | |
3923 | * @buffer: The ring buffer to reset all cpu buffers | |
3924 | */ | |
3925 | void ring_buffer_reset(struct ring_buffer *buffer) | |
3926 | { | |
7a8e76a3 SR |
3927 | int cpu; |
3928 | ||
7a8e76a3 | 3929 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 3930 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 3931 | } |
c4f50183 | 3932 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
3933 | |
3934 | /** | |
3935 | * rind_buffer_empty - is the ring buffer empty? | |
3936 | * @buffer: The ring buffer to test | |
3937 | */ | |
3938 | int ring_buffer_empty(struct ring_buffer *buffer) | |
3939 | { | |
3940 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 3941 | unsigned long flags; |
8d707e8e | 3942 | int dolock; |
7a8e76a3 | 3943 | int cpu; |
d4788207 | 3944 | int ret; |
7a8e76a3 | 3945 | |
8d707e8e | 3946 | dolock = rb_ok_to_lock(); |
7a8e76a3 SR |
3947 | |
3948 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
3949 | for_each_buffer_cpu(buffer, cpu) { | |
3950 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e SR |
3951 | local_irq_save(flags); |
3952 | if (dolock) | |
5389f6fa | 3953 | raw_spin_lock(&cpu_buffer->reader_lock); |
d4788207 | 3954 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e | 3955 | if (dolock) |
5389f6fa | 3956 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e SR |
3957 | local_irq_restore(flags); |
3958 | ||
d4788207 | 3959 | if (!ret) |
7a8e76a3 SR |
3960 | return 0; |
3961 | } | |
554f786e | 3962 | |
7a8e76a3 SR |
3963 | return 1; |
3964 | } | |
c4f50183 | 3965 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
3966 | |
3967 | /** | |
3968 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
3969 | * @buffer: The ring buffer | |
3970 | * @cpu: The CPU buffer to test | |
3971 | */ | |
3972 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
3973 | { | |
3974 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 3975 | unsigned long flags; |
8d707e8e | 3976 | int dolock; |
8aabee57 | 3977 | int ret; |
7a8e76a3 | 3978 | |
9e01c1b7 | 3979 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3980 | return 1; |
7a8e76a3 | 3981 | |
8d707e8e SR |
3982 | dolock = rb_ok_to_lock(); |
3983 | ||
7a8e76a3 | 3984 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
3985 | local_irq_save(flags); |
3986 | if (dolock) | |
5389f6fa | 3987 | raw_spin_lock(&cpu_buffer->reader_lock); |
554f786e | 3988 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e | 3989 | if (dolock) |
5389f6fa | 3990 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3991 | local_irq_restore(flags); |
554f786e SR |
3992 | |
3993 | return ret; | |
7a8e76a3 | 3994 | } |
c4f50183 | 3995 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 | 3996 | |
85bac32c | 3997 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
7a8e76a3 SR |
3998 | /** |
3999 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
4000 | * @buffer_a: One buffer to swap with | |
4001 | * @buffer_b: The other buffer to swap with | |
4002 | * | |
4003 | * This function is useful for tracers that want to take a "snapshot" | |
4004 | * of a CPU buffer and has another back up buffer lying around. | |
4005 | * it is expected that the tracer handles the cpu buffer not being | |
4006 | * used at the moment. | |
4007 | */ | |
4008 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
4009 | struct ring_buffer *buffer_b, int cpu) | |
4010 | { | |
4011 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
4012 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
4013 | int ret = -EINVAL; |
4014 | ||
9e01c1b7 RR |
4015 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
4016 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 4017 | goto out; |
7a8e76a3 | 4018 | |
438ced17 VN |
4019 | cpu_buffer_a = buffer_a->buffers[cpu]; |
4020 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
4021 | ||
7a8e76a3 | 4022 | /* At least make sure the two buffers are somewhat the same */ |
438ced17 | 4023 | if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages) |
554f786e SR |
4024 | goto out; |
4025 | ||
4026 | ret = -EAGAIN; | |
7a8e76a3 | 4027 | |
97b17efe | 4028 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 4029 | goto out; |
97b17efe SR |
4030 | |
4031 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 4032 | goto out; |
97b17efe SR |
4033 | |
4034 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 4035 | goto out; |
97b17efe | 4036 | |
97b17efe | 4037 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 4038 | goto out; |
97b17efe SR |
4039 | |
4040 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 4041 | goto out; |
97b17efe | 4042 | |
7a8e76a3 SR |
4043 | /* |
4044 | * We can't do a synchronize_sched here because this | |
4045 | * function can be called in atomic context. | |
4046 | * Normally this will be called from the same CPU as cpu. | |
4047 | * If not it's up to the caller to protect this. | |
4048 | */ | |
4049 | atomic_inc(&cpu_buffer_a->record_disabled); | |
4050 | atomic_inc(&cpu_buffer_b->record_disabled); | |
4051 | ||
98277991 SR |
4052 | ret = -EBUSY; |
4053 | if (local_read(&cpu_buffer_a->committing)) | |
4054 | goto out_dec; | |
4055 | if (local_read(&cpu_buffer_b->committing)) | |
4056 | goto out_dec; | |
4057 | ||
7a8e76a3 SR |
4058 | buffer_a->buffers[cpu] = cpu_buffer_b; |
4059 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
4060 | ||
4061 | cpu_buffer_b->buffer = buffer_a; | |
4062 | cpu_buffer_a->buffer = buffer_b; | |
4063 | ||
98277991 SR |
4064 | ret = 0; |
4065 | ||
4066 | out_dec: | |
7a8e76a3 SR |
4067 | atomic_dec(&cpu_buffer_a->record_disabled); |
4068 | atomic_dec(&cpu_buffer_b->record_disabled); | |
554f786e | 4069 | out: |
554f786e | 4070 | return ret; |
7a8e76a3 | 4071 | } |
c4f50183 | 4072 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
85bac32c | 4073 | #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */ |
7a8e76a3 | 4074 | |
8789a9e7 SR |
4075 | /** |
4076 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
4077 | * @buffer: the buffer to allocate for. | |
4078 | * | |
4079 | * This function is used in conjunction with ring_buffer_read_page. | |
4080 | * When reading a full page from the ring buffer, these functions | |
4081 | * can be used to speed up the process. The calling function should | |
4082 | * allocate a few pages first with this function. Then when it | |
4083 | * needs to get pages from the ring buffer, it passes the result | |
4084 | * of this function into ring_buffer_read_page, which will swap | |
4085 | * the page that was allocated, with the read page of the buffer. | |
4086 | * | |
4087 | * Returns: | |
4088 | * The page allocated, or NULL on error. | |
4089 | */ | |
7ea59064 | 4090 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer, int cpu) |
8789a9e7 | 4091 | { |
044fa782 | 4092 | struct buffer_data_page *bpage; |
7ea59064 | 4093 | struct page *page; |
8789a9e7 | 4094 | |
d7ec4bfe VN |
4095 | page = alloc_pages_node(cpu_to_node(cpu), |
4096 | GFP_KERNEL | __GFP_NORETRY, 0); | |
7ea59064 | 4097 | if (!page) |
8789a9e7 SR |
4098 | return NULL; |
4099 | ||
7ea59064 | 4100 | bpage = page_address(page); |
8789a9e7 | 4101 | |
ef7a4a16 SR |
4102 | rb_init_page(bpage); |
4103 | ||
044fa782 | 4104 | return bpage; |
8789a9e7 | 4105 | } |
d6ce96da | 4106 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
4107 | |
4108 | /** | |
4109 | * ring_buffer_free_read_page - free an allocated read page | |
4110 | * @buffer: the buffer the page was allocate for | |
4111 | * @data: the page to free | |
4112 | * | |
4113 | * Free a page allocated from ring_buffer_alloc_read_page. | |
4114 | */ | |
4115 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
4116 | { | |
4117 | free_page((unsigned long)data); | |
4118 | } | |
d6ce96da | 4119 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
4120 | |
4121 | /** | |
4122 | * ring_buffer_read_page - extract a page from the ring buffer | |
4123 | * @buffer: buffer to extract from | |
4124 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 4125 | * @len: amount to extract |
8789a9e7 SR |
4126 | * @cpu: the cpu of the buffer to extract |
4127 | * @full: should the extraction only happen when the page is full. | |
4128 | * | |
4129 | * This function will pull out a page from the ring buffer and consume it. | |
4130 | * @data_page must be the address of the variable that was returned | |
4131 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
4132 | * to swap with a page in the ring buffer. | |
4133 | * | |
4134 | * for example: | |
b85fa01e | 4135 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
4136 | * if (!rpage) |
4137 | * return error; | |
ef7a4a16 | 4138 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
4139 | * if (ret >= 0) |
4140 | * process_page(rpage, ret); | |
8789a9e7 SR |
4141 | * |
4142 | * When @full is set, the function will not return true unless | |
4143 | * the writer is off the reader page. | |
4144 | * | |
4145 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
4146 | * The ring buffer can be used anywhere in the kernel and can not | |
4147 | * blindly call wake_up. The layer that uses the ring buffer must be | |
4148 | * responsible for that. | |
4149 | * | |
4150 | * Returns: | |
667d2412 LJ |
4151 | * >=0 if data has been transferred, returns the offset of consumed data. |
4152 | * <0 if no data has been transferred. | |
8789a9e7 SR |
4153 | */ |
4154 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 4155 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
4156 | { |
4157 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
4158 | struct ring_buffer_event *event; | |
044fa782 | 4159 | struct buffer_data_page *bpage; |
ef7a4a16 | 4160 | struct buffer_page *reader; |
ff0ff84a | 4161 | unsigned long missed_events; |
8789a9e7 | 4162 | unsigned long flags; |
ef7a4a16 | 4163 | unsigned int commit; |
667d2412 | 4164 | unsigned int read; |
4f3640f8 | 4165 | u64 save_timestamp; |
667d2412 | 4166 | int ret = -1; |
8789a9e7 | 4167 | |
554f786e SR |
4168 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
4169 | goto out; | |
4170 | ||
474d32b6 SR |
4171 | /* |
4172 | * If len is not big enough to hold the page header, then | |
4173 | * we can not copy anything. | |
4174 | */ | |
4175 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 4176 | goto out; |
474d32b6 SR |
4177 | |
4178 | len -= BUF_PAGE_HDR_SIZE; | |
4179 | ||
8789a9e7 | 4180 | if (!data_page) |
554f786e | 4181 | goto out; |
8789a9e7 | 4182 | |
044fa782 SR |
4183 | bpage = *data_page; |
4184 | if (!bpage) | |
554f786e | 4185 | goto out; |
8789a9e7 | 4186 | |
5389f6fa | 4187 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4188 | |
ef7a4a16 SR |
4189 | reader = rb_get_reader_page(cpu_buffer); |
4190 | if (!reader) | |
554f786e | 4191 | goto out_unlock; |
8789a9e7 | 4192 | |
ef7a4a16 SR |
4193 | event = rb_reader_event(cpu_buffer); |
4194 | ||
4195 | read = reader->read; | |
4196 | commit = rb_page_commit(reader); | |
667d2412 | 4197 | |
66a8cb95 | 4198 | /* Check if any events were dropped */ |
ff0ff84a | 4199 | missed_events = cpu_buffer->lost_events; |
66a8cb95 | 4200 | |
8789a9e7 | 4201 | /* |
474d32b6 SR |
4202 | * If this page has been partially read or |
4203 | * if len is not big enough to read the rest of the page or | |
4204 | * a writer is still on the page, then | |
4205 | * we must copy the data from the page to the buffer. | |
4206 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 4207 | */ |
474d32b6 | 4208 | if (read || (len < (commit - read)) || |
ef7a4a16 | 4209 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 4210 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
4211 | unsigned int rpos = read; |
4212 | unsigned int pos = 0; | |
ef7a4a16 | 4213 | unsigned int size; |
8789a9e7 SR |
4214 | |
4215 | if (full) | |
554f786e | 4216 | goto out_unlock; |
8789a9e7 | 4217 | |
ef7a4a16 SR |
4218 | if (len > (commit - read)) |
4219 | len = (commit - read); | |
4220 | ||
69d1b839 SR |
4221 | /* Always keep the time extend and data together */ |
4222 | size = rb_event_ts_length(event); | |
ef7a4a16 SR |
4223 | |
4224 | if (len < size) | |
554f786e | 4225 | goto out_unlock; |
ef7a4a16 | 4226 | |
4f3640f8 SR |
4227 | /* save the current timestamp, since the user will need it */ |
4228 | save_timestamp = cpu_buffer->read_stamp; | |
4229 | ||
ef7a4a16 SR |
4230 | /* Need to copy one event at a time */ |
4231 | do { | |
e1e35927 DS |
4232 | /* We need the size of one event, because |
4233 | * rb_advance_reader only advances by one event, | |
4234 | * whereas rb_event_ts_length may include the size of | |
4235 | * one or two events. | |
4236 | * We have already ensured there's enough space if this | |
4237 | * is a time extend. */ | |
4238 | size = rb_event_length(event); | |
474d32b6 | 4239 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
4240 | |
4241 | len -= size; | |
4242 | ||
4243 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
4244 | rpos = reader->read; |
4245 | pos += size; | |
ef7a4a16 | 4246 | |
18fab912 HY |
4247 | if (rpos >= commit) |
4248 | break; | |
4249 | ||
ef7a4a16 | 4250 | event = rb_reader_event(cpu_buffer); |
69d1b839 SR |
4251 | /* Always keep the time extend and data together */ |
4252 | size = rb_event_ts_length(event); | |
e1e35927 | 4253 | } while (len >= size); |
667d2412 LJ |
4254 | |
4255 | /* update bpage */ | |
ef7a4a16 | 4256 | local_set(&bpage->commit, pos); |
4f3640f8 | 4257 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 4258 | |
474d32b6 SR |
4259 | /* we copied everything to the beginning */ |
4260 | read = 0; | |
8789a9e7 | 4261 | } else { |
afbab76a | 4262 | /* update the entry counter */ |
77ae365e | 4263 | cpu_buffer->read += rb_page_entries(reader); |
c64e148a | 4264 | cpu_buffer->read_bytes += BUF_PAGE_SIZE; |
afbab76a | 4265 | |
8789a9e7 | 4266 | /* swap the pages */ |
044fa782 | 4267 | rb_init_page(bpage); |
ef7a4a16 SR |
4268 | bpage = reader->page; |
4269 | reader->page = *data_page; | |
4270 | local_set(&reader->write, 0); | |
778c55d4 | 4271 | local_set(&reader->entries, 0); |
ef7a4a16 | 4272 | reader->read = 0; |
044fa782 | 4273 | *data_page = bpage; |
ff0ff84a SR |
4274 | |
4275 | /* | |
4276 | * Use the real_end for the data size, | |
4277 | * This gives us a chance to store the lost events | |
4278 | * on the page. | |
4279 | */ | |
4280 | if (reader->real_end) | |
4281 | local_set(&bpage->commit, reader->real_end); | |
8789a9e7 | 4282 | } |
667d2412 | 4283 | ret = read; |
8789a9e7 | 4284 | |
66a8cb95 | 4285 | cpu_buffer->lost_events = 0; |
2711ca23 SR |
4286 | |
4287 | commit = local_read(&bpage->commit); | |
66a8cb95 SR |
4288 | /* |
4289 | * Set a flag in the commit field if we lost events | |
4290 | */ | |
ff0ff84a | 4291 | if (missed_events) { |
ff0ff84a SR |
4292 | /* If there is room at the end of the page to save the |
4293 | * missed events, then record it there. | |
4294 | */ | |
4295 | if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) { | |
4296 | memcpy(&bpage->data[commit], &missed_events, | |
4297 | sizeof(missed_events)); | |
4298 | local_add(RB_MISSED_STORED, &bpage->commit); | |
2711ca23 | 4299 | commit += sizeof(missed_events); |
ff0ff84a | 4300 | } |
66a8cb95 | 4301 | local_add(RB_MISSED_EVENTS, &bpage->commit); |
ff0ff84a | 4302 | } |
66a8cb95 | 4303 | |
2711ca23 SR |
4304 | /* |
4305 | * This page may be off to user land. Zero it out here. | |
4306 | */ | |
4307 | if (commit < BUF_PAGE_SIZE) | |
4308 | memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit); | |
4309 | ||
554f786e | 4310 | out_unlock: |
5389f6fa | 4311 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4312 | |
554f786e | 4313 | out: |
8789a9e7 SR |
4314 | return ret; |
4315 | } | |
d6ce96da | 4316 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 4317 | |
59222efe | 4318 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
4319 | static int rb_cpu_notify(struct notifier_block *self, |
4320 | unsigned long action, void *hcpu) | |
554f786e SR |
4321 | { |
4322 | struct ring_buffer *buffer = | |
4323 | container_of(self, struct ring_buffer, cpu_notify); | |
4324 | long cpu = (long)hcpu; | |
438ced17 VN |
4325 | int cpu_i, nr_pages_same; |
4326 | unsigned int nr_pages; | |
554f786e SR |
4327 | |
4328 | switch (action) { | |
4329 | case CPU_UP_PREPARE: | |
4330 | case CPU_UP_PREPARE_FROZEN: | |
3f237a79 | 4331 | if (cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e SR |
4332 | return NOTIFY_OK; |
4333 | ||
438ced17 VN |
4334 | nr_pages = 0; |
4335 | nr_pages_same = 1; | |
4336 | /* check if all cpu sizes are same */ | |
4337 | for_each_buffer_cpu(buffer, cpu_i) { | |
4338 | /* fill in the size from first enabled cpu */ | |
4339 | if (nr_pages == 0) | |
4340 | nr_pages = buffer->buffers[cpu_i]->nr_pages; | |
4341 | if (nr_pages != buffer->buffers[cpu_i]->nr_pages) { | |
4342 | nr_pages_same = 0; | |
4343 | break; | |
4344 | } | |
4345 | } | |
4346 | /* allocate minimum pages, user can later expand it */ | |
4347 | if (!nr_pages_same) | |
4348 | nr_pages = 2; | |
554f786e | 4349 | buffer->buffers[cpu] = |
438ced17 | 4350 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
554f786e SR |
4351 | if (!buffer->buffers[cpu]) { |
4352 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
4353 | cpu); | |
4354 | return NOTIFY_OK; | |
4355 | } | |
4356 | smp_wmb(); | |
3f237a79 | 4357 | cpumask_set_cpu(cpu, buffer->cpumask); |
554f786e SR |
4358 | break; |
4359 | case CPU_DOWN_PREPARE: | |
4360 | case CPU_DOWN_PREPARE_FROZEN: | |
4361 | /* | |
4362 | * Do nothing. | |
4363 | * If we were to free the buffer, then the user would | |
4364 | * lose any trace that was in the buffer. | |
4365 | */ | |
4366 | break; | |
4367 | default: | |
4368 | break; | |
4369 | } | |
4370 | return NOTIFY_OK; | |
4371 | } | |
4372 | #endif |