Commit | Line | Data |
---|---|---|
7a8e76a3 SR |
1 | /* |
2 | * Generic ring buffer | |
3 | * | |
4 | * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com> | |
5 | */ | |
0b07436d | 6 | #include <linux/ftrace_event.h> |
7a8e76a3 | 7 | #include <linux/ring_buffer.h> |
14131f2f | 8 | #include <linux/trace_clock.h> |
0b07436d | 9 | #include <linux/trace_seq.h> |
7a8e76a3 SR |
10 | #include <linux/spinlock.h> |
11 | #include <linux/debugfs.h> | |
12 | #include <linux/uaccess.h> | |
a81bd80a | 13 | #include <linux/hardirq.h> |
1744a21d | 14 | #include <linux/kmemcheck.h> |
7a8e76a3 SR |
15 | #include <linux/module.h> |
16 | #include <linux/percpu.h> | |
17 | #include <linux/mutex.h> | |
5a0e3ad6 | 18 | #include <linux/slab.h> |
7a8e76a3 SR |
19 | #include <linux/init.h> |
20 | #include <linux/hash.h> | |
21 | #include <linux/list.h> | |
554f786e | 22 | #include <linux/cpu.h> |
7a8e76a3 SR |
23 | #include <linux/fs.h> |
24 | ||
79615760 | 25 | #include <asm/local.h> |
182e9f5f | 26 | |
83f40318 VN |
27 | static void update_pages_handler(struct work_struct *work); |
28 | ||
d1b182a8 SR |
29 | /* |
30 | * The ring buffer header is special. We must manually up keep it. | |
31 | */ | |
32 | int ring_buffer_print_entry_header(struct trace_seq *s) | |
33 | { | |
34 | int ret; | |
35 | ||
334d4169 LJ |
36 | ret = trace_seq_printf(s, "# compressed entry header\n"); |
37 | ret = trace_seq_printf(s, "\ttype_len : 5 bits\n"); | |
d1b182a8 SR |
38 | ret = trace_seq_printf(s, "\ttime_delta : 27 bits\n"); |
39 | ret = trace_seq_printf(s, "\tarray : 32 bits\n"); | |
40 | ret = trace_seq_printf(s, "\n"); | |
41 | ret = trace_seq_printf(s, "\tpadding : type == %d\n", | |
42 | RINGBUF_TYPE_PADDING); | |
43 | ret = trace_seq_printf(s, "\ttime_extend : type == %d\n", | |
44 | RINGBUF_TYPE_TIME_EXTEND); | |
334d4169 LJ |
45 | ret = trace_seq_printf(s, "\tdata max type_len == %d\n", |
46 | RINGBUF_TYPE_DATA_TYPE_LEN_MAX); | |
d1b182a8 SR |
47 | |
48 | return ret; | |
49 | } | |
50 | ||
5cc98548 SR |
51 | /* |
52 | * The ring buffer is made up of a list of pages. A separate list of pages is | |
53 | * allocated for each CPU. A writer may only write to a buffer that is | |
54 | * associated with the CPU it is currently executing on. A reader may read | |
55 | * from any per cpu buffer. | |
56 | * | |
57 | * The reader is special. For each per cpu buffer, the reader has its own | |
58 | * reader page. When a reader has read the entire reader page, this reader | |
59 | * page is swapped with another page in the ring buffer. | |
60 | * | |
61 | * Now, as long as the writer is off the reader page, the reader can do what | |
62 | * ever it wants with that page. The writer will never write to that page | |
63 | * again (as long as it is out of the ring buffer). | |
64 | * | |
65 | * Here's some silly ASCII art. | |
66 | * | |
67 | * +------+ | |
68 | * |reader| RING BUFFER | |
69 | * |page | | |
70 | * +------+ +---+ +---+ +---+ | |
71 | * | |-->| |-->| | | |
72 | * +---+ +---+ +---+ | |
73 | * ^ | | |
74 | * | | | |
75 | * +---------------+ | |
76 | * | |
77 | * | |
78 | * +------+ | |
79 | * |reader| RING BUFFER | |
80 | * |page |------------------v | |
81 | * +------+ +---+ +---+ +---+ | |
82 | * | |-->| |-->| | | |
83 | * +---+ +---+ +---+ | |
84 | * ^ | | |
85 | * | | | |
86 | * +---------------+ | |
87 | * | |
88 | * | |
89 | * +------+ | |
90 | * |reader| RING BUFFER | |
91 | * |page |------------------v | |
92 | * +------+ +---+ +---+ +---+ | |
93 | * ^ | |-->| |-->| | | |
94 | * | +---+ +---+ +---+ | |
95 | * | | | |
96 | * | | | |
97 | * +------------------------------+ | |
98 | * | |
99 | * | |
100 | * +------+ | |
101 | * |buffer| RING BUFFER | |
102 | * |page |------------------v | |
103 | * +------+ +---+ +---+ +---+ | |
104 | * ^ | | | |-->| | | |
105 | * | New +---+ +---+ +---+ | |
106 | * | Reader------^ | | |
107 | * | page | | |
108 | * +------------------------------+ | |
109 | * | |
110 | * | |
111 | * After we make this swap, the reader can hand this page off to the splice | |
112 | * code and be done with it. It can even allocate a new page if it needs to | |
113 | * and swap that into the ring buffer. | |
114 | * | |
115 | * We will be using cmpxchg soon to make all this lockless. | |
116 | * | |
117 | */ | |
118 | ||
033601a3 SR |
119 | /* |
120 | * A fast way to enable or disable all ring buffers is to | |
121 | * call tracing_on or tracing_off. Turning off the ring buffers | |
122 | * prevents all ring buffers from being recorded to. | |
123 | * Turning this switch on, makes it OK to write to the | |
124 | * ring buffer, if the ring buffer is enabled itself. | |
125 | * | |
126 | * There's three layers that must be on in order to write | |
127 | * to the ring buffer. | |
128 | * | |
129 | * 1) This global flag must be set. | |
130 | * 2) The ring buffer must be enabled for recording. | |
131 | * 3) The per cpu buffer must be enabled for recording. | |
132 | * | |
133 | * In case of an anomaly, this global flag has a bit set that | |
134 | * will permantly disable all ring buffers. | |
135 | */ | |
136 | ||
137 | /* | |
138 | * Global flag to disable all recording to ring buffers | |
139 | * This has two bits: ON, DISABLED | |
140 | * | |
141 | * ON DISABLED | |
142 | * ---- ---------- | |
143 | * 0 0 : ring buffers are off | |
144 | * 1 0 : ring buffers are on | |
145 | * X 1 : ring buffers are permanently disabled | |
146 | */ | |
147 | ||
148 | enum { | |
149 | RB_BUFFERS_ON_BIT = 0, | |
150 | RB_BUFFERS_DISABLED_BIT = 1, | |
151 | }; | |
152 | ||
153 | enum { | |
154 | RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT, | |
155 | RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT, | |
156 | }; | |
157 | ||
5e39841c | 158 | static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON; |
a3583244 | 159 | |
499e5470 SR |
160 | /* Used for individual buffers (after the counter) */ |
161 | #define RB_BUFFER_OFF (1 << 20) | |
a3583244 | 162 | |
499e5470 | 163 | #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data) |
033601a3 SR |
164 | |
165 | /** | |
166 | * tracing_off_permanent - permanently disable ring buffers | |
167 | * | |
168 | * This function, once called, will disable all ring buffers | |
c3706f00 | 169 | * permanently. |
033601a3 SR |
170 | */ |
171 | void tracing_off_permanent(void) | |
172 | { | |
173 | set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags); | |
a3583244 SR |
174 | } |
175 | ||
e3d6bf0a | 176 | #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array)) |
67d34724 | 177 | #define RB_ALIGNMENT 4U |
334d4169 | 178 | #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
c7b09308 | 179 | #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */ |
334d4169 | 180 | |
2271048d SR |
181 | #if !defined(CONFIG_64BIT) || defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) |
182 | # define RB_FORCE_8BYTE_ALIGNMENT 0 | |
183 | # define RB_ARCH_ALIGNMENT RB_ALIGNMENT | |
184 | #else | |
185 | # define RB_FORCE_8BYTE_ALIGNMENT 1 | |
186 | # define RB_ARCH_ALIGNMENT 8U | |
187 | #endif | |
188 | ||
334d4169 LJ |
189 | /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */ |
190 | #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX | |
7a8e76a3 SR |
191 | |
192 | enum { | |
193 | RB_LEN_TIME_EXTEND = 8, | |
194 | RB_LEN_TIME_STAMP = 16, | |
195 | }; | |
196 | ||
69d1b839 SR |
197 | #define skip_time_extend(event) \ |
198 | ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND)) | |
199 | ||
2d622719 TZ |
200 | static inline int rb_null_event(struct ring_buffer_event *event) |
201 | { | |
a1863c21 | 202 | return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta; |
2d622719 TZ |
203 | } |
204 | ||
205 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
206 | { | |
a1863c21 | 207 | /* padding has a NULL time_delta */ |
334d4169 | 208 | event->type_len = RINGBUF_TYPE_PADDING; |
2d622719 TZ |
209 | event->time_delta = 0; |
210 | } | |
211 | ||
34a148bf | 212 | static unsigned |
2d622719 | 213 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
214 | { |
215 | unsigned length; | |
216 | ||
334d4169 LJ |
217 | if (event->type_len) |
218 | length = event->type_len * RB_ALIGNMENT; | |
2d622719 TZ |
219 | else |
220 | length = event->array[0]; | |
221 | return length + RB_EVNT_HDR_SIZE; | |
222 | } | |
223 | ||
69d1b839 SR |
224 | /* |
225 | * Return the length of the given event. Will return | |
226 | * the length of the time extend if the event is a | |
227 | * time extend. | |
228 | */ | |
229 | static inline unsigned | |
2d622719 TZ |
230 | rb_event_length(struct ring_buffer_event *event) |
231 | { | |
334d4169 | 232 | switch (event->type_len) { |
7a8e76a3 | 233 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
234 | if (rb_null_event(event)) |
235 | /* undefined */ | |
236 | return -1; | |
334d4169 | 237 | return event->array[0] + RB_EVNT_HDR_SIZE; |
7a8e76a3 SR |
238 | |
239 | case RINGBUF_TYPE_TIME_EXTEND: | |
240 | return RB_LEN_TIME_EXTEND; | |
241 | ||
242 | case RINGBUF_TYPE_TIME_STAMP: | |
243 | return RB_LEN_TIME_STAMP; | |
244 | ||
245 | case RINGBUF_TYPE_DATA: | |
2d622719 | 246 | return rb_event_data_length(event); |
7a8e76a3 SR |
247 | default: |
248 | BUG(); | |
249 | } | |
250 | /* not hit */ | |
251 | return 0; | |
252 | } | |
253 | ||
69d1b839 SR |
254 | /* |
255 | * Return total length of time extend and data, | |
256 | * or just the event length for all other events. | |
257 | */ | |
258 | static inline unsigned | |
259 | rb_event_ts_length(struct ring_buffer_event *event) | |
260 | { | |
261 | unsigned len = 0; | |
262 | ||
263 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
264 | /* time extends include the data event after it */ | |
265 | len = RB_LEN_TIME_EXTEND; | |
266 | event = skip_time_extend(event); | |
267 | } | |
268 | return len + rb_event_length(event); | |
269 | } | |
270 | ||
7a8e76a3 SR |
271 | /** |
272 | * ring_buffer_event_length - return the length of the event | |
273 | * @event: the event to get the length of | |
69d1b839 SR |
274 | * |
275 | * Returns the size of the data load of a data event. | |
276 | * If the event is something other than a data event, it | |
277 | * returns the size of the event itself. With the exception | |
278 | * of a TIME EXTEND, where it still returns the size of the | |
279 | * data load of the data event after it. | |
7a8e76a3 SR |
280 | */ |
281 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
282 | { | |
69d1b839 SR |
283 | unsigned length; |
284 | ||
285 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) | |
286 | event = skip_time_extend(event); | |
287 | ||
288 | length = rb_event_length(event); | |
334d4169 | 289 | if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
465634ad RR |
290 | return length; |
291 | length -= RB_EVNT_HDR_SIZE; | |
292 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
293 | length -= sizeof(event->array[0]); | |
294 | return length; | |
7a8e76a3 | 295 | } |
c4f50183 | 296 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
297 | |
298 | /* inline for ring buffer fast paths */ | |
34a148bf | 299 | static void * |
7a8e76a3 SR |
300 | rb_event_data(struct ring_buffer_event *event) |
301 | { | |
69d1b839 SR |
302 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) |
303 | event = skip_time_extend(event); | |
334d4169 | 304 | BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX); |
7a8e76a3 | 305 | /* If length is in len field, then array[0] has the data */ |
334d4169 | 306 | if (event->type_len) |
7a8e76a3 SR |
307 | return (void *)&event->array[0]; |
308 | /* Otherwise length is in array[0] and array[1] has the data */ | |
309 | return (void *)&event->array[1]; | |
310 | } | |
311 | ||
312 | /** | |
313 | * ring_buffer_event_data - return the data of the event | |
314 | * @event: the event to get the data from | |
315 | */ | |
316 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
317 | { | |
318 | return rb_event_data(event); | |
319 | } | |
c4f50183 | 320 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
321 | |
322 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 323 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
324 | |
325 | #define TS_SHIFT 27 | |
326 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
327 | #define TS_DELTA_TEST (~TS_MASK) | |
328 | ||
66a8cb95 SR |
329 | /* Flag when events were overwritten */ |
330 | #define RB_MISSED_EVENTS (1 << 31) | |
ff0ff84a SR |
331 | /* Missed count stored at end */ |
332 | #define RB_MISSED_STORED (1 << 30) | |
66a8cb95 | 333 | |
abc9b56d | 334 | struct buffer_data_page { |
e4c2ce82 | 335 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 336 | local_t commit; /* write committed index */ |
abc9b56d SR |
337 | unsigned char data[]; /* data of buffer page */ |
338 | }; | |
339 | ||
77ae365e SR |
340 | /* |
341 | * Note, the buffer_page list must be first. The buffer pages | |
342 | * are allocated in cache lines, which means that each buffer | |
343 | * page will be at the beginning of a cache line, and thus | |
344 | * the least significant bits will be zero. We use this to | |
345 | * add flags in the list struct pointers, to make the ring buffer | |
346 | * lockless. | |
347 | */ | |
abc9b56d | 348 | struct buffer_page { |
778c55d4 | 349 | struct list_head list; /* list of buffer pages */ |
abc9b56d | 350 | local_t write; /* index for next write */ |
6f807acd | 351 | unsigned read; /* index for next read */ |
778c55d4 | 352 | local_t entries; /* entries on this page */ |
ff0ff84a | 353 | unsigned long real_end; /* real end of data */ |
abc9b56d | 354 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
355 | }; |
356 | ||
77ae365e SR |
357 | /* |
358 | * The buffer page counters, write and entries, must be reset | |
359 | * atomically when crossing page boundaries. To synchronize this | |
360 | * update, two counters are inserted into the number. One is | |
361 | * the actual counter for the write position or count on the page. | |
362 | * | |
363 | * The other is a counter of updaters. Before an update happens | |
364 | * the update partition of the counter is incremented. This will | |
365 | * allow the updater to update the counter atomically. | |
366 | * | |
367 | * The counter is 20 bits, and the state data is 12. | |
368 | */ | |
369 | #define RB_WRITE_MASK 0xfffff | |
370 | #define RB_WRITE_INTCNT (1 << 20) | |
371 | ||
044fa782 | 372 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 373 | { |
044fa782 | 374 | local_set(&bpage->commit, 0); |
abc9b56d SR |
375 | } |
376 | ||
474d32b6 SR |
377 | /** |
378 | * ring_buffer_page_len - the size of data on the page. | |
379 | * @page: The page to read | |
380 | * | |
381 | * Returns the amount of data on the page, including buffer page header. | |
382 | */ | |
ef7a4a16 SR |
383 | size_t ring_buffer_page_len(void *page) |
384 | { | |
474d32b6 SR |
385 | return local_read(&((struct buffer_data_page *)page)->commit) |
386 | + BUF_PAGE_HDR_SIZE; | |
ef7a4a16 SR |
387 | } |
388 | ||
ed56829c SR |
389 | /* |
390 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
391 | * this issue out. | |
392 | */ | |
34a148bf | 393 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 394 | { |
34a148bf | 395 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 396 | kfree(bpage); |
ed56829c SR |
397 | } |
398 | ||
7a8e76a3 SR |
399 | /* |
400 | * We need to fit the time_stamp delta into 27 bits. | |
401 | */ | |
402 | static inline int test_time_stamp(u64 delta) | |
403 | { | |
404 | if (delta & TS_DELTA_TEST) | |
405 | return 1; | |
406 | return 0; | |
407 | } | |
408 | ||
474d32b6 | 409 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 | 410 | |
be957c44 SR |
411 | /* Max payload is BUF_PAGE_SIZE - header (8bytes) */ |
412 | #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) | |
413 | ||
d1b182a8 SR |
414 | int ring_buffer_print_page_header(struct trace_seq *s) |
415 | { | |
416 | struct buffer_data_page field; | |
417 | int ret; | |
418 | ||
419 | ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t" | |
26a50744 TZ |
420 | "offset:0;\tsize:%u;\tsigned:%u;\n", |
421 | (unsigned int)sizeof(field.time_stamp), | |
422 | (unsigned int)is_signed_type(u64)); | |
d1b182a8 SR |
423 | |
424 | ret = trace_seq_printf(s, "\tfield: local_t commit;\t" | |
26a50744 | 425 | "offset:%u;\tsize:%u;\tsigned:%u;\n", |
d1b182a8 | 426 | (unsigned int)offsetof(typeof(field), commit), |
26a50744 TZ |
427 | (unsigned int)sizeof(field.commit), |
428 | (unsigned int)is_signed_type(long)); | |
d1b182a8 | 429 | |
66a8cb95 SR |
430 | ret = trace_seq_printf(s, "\tfield: int overwrite;\t" |
431 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
432 | (unsigned int)offsetof(typeof(field), commit), | |
433 | 1, | |
434 | (unsigned int)is_signed_type(long)); | |
435 | ||
d1b182a8 | 436 | ret = trace_seq_printf(s, "\tfield: char data;\t" |
26a50744 | 437 | "offset:%u;\tsize:%u;\tsigned:%u;\n", |
d1b182a8 | 438 | (unsigned int)offsetof(typeof(field), data), |
26a50744 TZ |
439 | (unsigned int)BUF_PAGE_SIZE, |
440 | (unsigned int)is_signed_type(char)); | |
d1b182a8 SR |
441 | |
442 | return ret; | |
443 | } | |
444 | ||
7a8e76a3 SR |
445 | /* |
446 | * head_page == tail_page && head == tail then buffer is empty. | |
447 | */ | |
448 | struct ring_buffer_per_cpu { | |
449 | int cpu; | |
985023de | 450 | atomic_t record_disabled; |
7a8e76a3 | 451 | struct ring_buffer *buffer; |
5389f6fa | 452 | raw_spinlock_t reader_lock; /* serialize readers */ |
445c8951 | 453 | arch_spinlock_t lock; |
7a8e76a3 | 454 | struct lock_class_key lock_key; |
438ced17 | 455 | unsigned int nr_pages; |
3adc54fa | 456 | struct list_head *pages; |
6f807acd SR |
457 | struct buffer_page *head_page; /* read from head */ |
458 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 459 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 460 | struct buffer_page *reader_page; |
66a8cb95 SR |
461 | unsigned long lost_events; |
462 | unsigned long last_overrun; | |
c64e148a | 463 | local_t entries_bytes; |
e4906eff | 464 | local_t entries; |
884bfe89 SP |
465 | local_t overrun; |
466 | local_t commit_overrun; | |
467 | local_t dropped_events; | |
fa743953 SR |
468 | local_t committing; |
469 | local_t commits; | |
77ae365e | 470 | unsigned long read; |
c64e148a | 471 | unsigned long read_bytes; |
7a8e76a3 SR |
472 | u64 write_stamp; |
473 | u64 read_stamp; | |
438ced17 VN |
474 | /* ring buffer pages to update, > 0 to add, < 0 to remove */ |
475 | int nr_pages_to_update; | |
476 | struct list_head new_pages; /* new pages to add */ | |
83f40318 | 477 | struct work_struct update_pages_work; |
05fdd70d | 478 | struct completion update_done; |
7a8e76a3 SR |
479 | }; |
480 | ||
481 | struct ring_buffer { | |
7a8e76a3 SR |
482 | unsigned flags; |
483 | int cpus; | |
7a8e76a3 | 484 | atomic_t record_disabled; |
83f40318 | 485 | atomic_t resize_disabled; |
00f62f61 | 486 | cpumask_var_t cpumask; |
7a8e76a3 | 487 | |
1f8a6a10 PZ |
488 | struct lock_class_key *reader_lock_key; |
489 | ||
7a8e76a3 SR |
490 | struct mutex mutex; |
491 | ||
492 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 493 | |
59222efe | 494 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
495 | struct notifier_block cpu_notify; |
496 | #endif | |
37886f6a | 497 | u64 (*clock)(void); |
7a8e76a3 SR |
498 | }; |
499 | ||
500 | struct ring_buffer_iter { | |
501 | struct ring_buffer_per_cpu *cpu_buffer; | |
502 | unsigned long head; | |
503 | struct buffer_page *head_page; | |
492a74f4 SR |
504 | struct buffer_page *cache_reader_page; |
505 | unsigned long cache_read; | |
7a8e76a3 SR |
506 | u64 read_stamp; |
507 | }; | |
508 | ||
f536aafc | 509 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
077c5407 SR |
510 | #define RB_WARN_ON(b, cond) \ |
511 | ({ \ | |
512 | int _____ret = unlikely(cond); \ | |
513 | if (_____ret) { \ | |
514 | if (__same_type(*(b), struct ring_buffer_per_cpu)) { \ | |
515 | struct ring_buffer_per_cpu *__b = \ | |
516 | (void *)b; \ | |
517 | atomic_inc(&__b->buffer->record_disabled); \ | |
518 | } else \ | |
519 | atomic_inc(&b->record_disabled); \ | |
520 | WARN_ON(1); \ | |
521 | } \ | |
522 | _____ret; \ | |
3e89c7bb | 523 | }) |
f536aafc | 524 | |
37886f6a SR |
525 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
526 | #define DEBUG_SHIFT 0 | |
527 | ||
6d3f1e12 | 528 | static inline u64 rb_time_stamp(struct ring_buffer *buffer) |
88eb0125 SR |
529 | { |
530 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
531 | return buffer->clock() << DEBUG_SHIFT; | |
532 | } | |
533 | ||
37886f6a SR |
534 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) |
535 | { | |
536 | u64 time; | |
537 | ||
538 | preempt_disable_notrace(); | |
6d3f1e12 | 539 | time = rb_time_stamp(buffer); |
37886f6a SR |
540 | preempt_enable_no_resched_notrace(); |
541 | ||
542 | return time; | |
543 | } | |
544 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
545 | ||
546 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
547 | int cpu, u64 *ts) | |
548 | { | |
549 | /* Just stupid testing the normalize function and deltas */ | |
550 | *ts >>= DEBUG_SHIFT; | |
551 | } | |
552 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
553 | ||
77ae365e SR |
554 | /* |
555 | * Making the ring buffer lockless makes things tricky. | |
556 | * Although writes only happen on the CPU that they are on, | |
557 | * and they only need to worry about interrupts. Reads can | |
558 | * happen on any CPU. | |
559 | * | |
560 | * The reader page is always off the ring buffer, but when the | |
561 | * reader finishes with a page, it needs to swap its page with | |
562 | * a new one from the buffer. The reader needs to take from | |
563 | * the head (writes go to the tail). But if a writer is in overwrite | |
564 | * mode and wraps, it must push the head page forward. | |
565 | * | |
566 | * Here lies the problem. | |
567 | * | |
568 | * The reader must be careful to replace only the head page, and | |
569 | * not another one. As described at the top of the file in the | |
570 | * ASCII art, the reader sets its old page to point to the next | |
571 | * page after head. It then sets the page after head to point to | |
572 | * the old reader page. But if the writer moves the head page | |
573 | * during this operation, the reader could end up with the tail. | |
574 | * | |
575 | * We use cmpxchg to help prevent this race. We also do something | |
576 | * special with the page before head. We set the LSB to 1. | |
577 | * | |
578 | * When the writer must push the page forward, it will clear the | |
579 | * bit that points to the head page, move the head, and then set | |
580 | * the bit that points to the new head page. | |
581 | * | |
582 | * We also don't want an interrupt coming in and moving the head | |
583 | * page on another writer. Thus we use the second LSB to catch | |
584 | * that too. Thus: | |
585 | * | |
586 | * head->list->prev->next bit 1 bit 0 | |
587 | * ------- ------- | |
588 | * Normal page 0 0 | |
589 | * Points to head page 0 1 | |
590 | * New head page 1 0 | |
591 | * | |
592 | * Note we can not trust the prev pointer of the head page, because: | |
593 | * | |
594 | * +----+ +-----+ +-----+ | |
595 | * | |------>| T |---X--->| N | | |
596 | * | |<------| | | | | |
597 | * +----+ +-----+ +-----+ | |
598 | * ^ ^ | | |
599 | * | +-----+ | | | |
600 | * +----------| R |----------+ | | |
601 | * | |<-----------+ | |
602 | * +-----+ | |
603 | * | |
604 | * Key: ---X--> HEAD flag set in pointer | |
605 | * T Tail page | |
606 | * R Reader page | |
607 | * N Next page | |
608 | * | |
609 | * (see __rb_reserve_next() to see where this happens) | |
610 | * | |
611 | * What the above shows is that the reader just swapped out | |
612 | * the reader page with a page in the buffer, but before it | |
613 | * could make the new header point back to the new page added | |
614 | * it was preempted by a writer. The writer moved forward onto | |
615 | * the new page added by the reader and is about to move forward | |
616 | * again. | |
617 | * | |
618 | * You can see, it is legitimate for the previous pointer of | |
619 | * the head (or any page) not to point back to itself. But only | |
620 | * temporarially. | |
621 | */ | |
622 | ||
623 | #define RB_PAGE_NORMAL 0UL | |
624 | #define RB_PAGE_HEAD 1UL | |
625 | #define RB_PAGE_UPDATE 2UL | |
626 | ||
627 | ||
628 | #define RB_FLAG_MASK 3UL | |
629 | ||
630 | /* PAGE_MOVED is not part of the mask */ | |
631 | #define RB_PAGE_MOVED 4UL | |
632 | ||
633 | /* | |
634 | * rb_list_head - remove any bit | |
635 | */ | |
636 | static struct list_head *rb_list_head(struct list_head *list) | |
637 | { | |
638 | unsigned long val = (unsigned long)list; | |
639 | ||
640 | return (struct list_head *)(val & ~RB_FLAG_MASK); | |
641 | } | |
642 | ||
643 | /* | |
6d3f1e12 | 644 | * rb_is_head_page - test if the given page is the head page |
77ae365e SR |
645 | * |
646 | * Because the reader may move the head_page pointer, we can | |
647 | * not trust what the head page is (it may be pointing to | |
648 | * the reader page). But if the next page is a header page, | |
649 | * its flags will be non zero. | |
650 | */ | |
42b16b3f | 651 | static inline int |
77ae365e SR |
652 | rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer, |
653 | struct buffer_page *page, struct list_head *list) | |
654 | { | |
655 | unsigned long val; | |
656 | ||
657 | val = (unsigned long)list->next; | |
658 | ||
659 | if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list) | |
660 | return RB_PAGE_MOVED; | |
661 | ||
662 | return val & RB_FLAG_MASK; | |
663 | } | |
664 | ||
665 | /* | |
666 | * rb_is_reader_page | |
667 | * | |
668 | * The unique thing about the reader page, is that, if the | |
669 | * writer is ever on it, the previous pointer never points | |
670 | * back to the reader page. | |
671 | */ | |
672 | static int rb_is_reader_page(struct buffer_page *page) | |
673 | { | |
674 | struct list_head *list = page->list.prev; | |
675 | ||
676 | return rb_list_head(list->next) != &page->list; | |
677 | } | |
678 | ||
679 | /* | |
680 | * rb_set_list_to_head - set a list_head to be pointing to head. | |
681 | */ | |
682 | static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer, | |
683 | struct list_head *list) | |
684 | { | |
685 | unsigned long *ptr; | |
686 | ||
687 | ptr = (unsigned long *)&list->next; | |
688 | *ptr |= RB_PAGE_HEAD; | |
689 | *ptr &= ~RB_PAGE_UPDATE; | |
690 | } | |
691 | ||
692 | /* | |
693 | * rb_head_page_activate - sets up head page | |
694 | */ | |
695 | static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer) | |
696 | { | |
697 | struct buffer_page *head; | |
698 | ||
699 | head = cpu_buffer->head_page; | |
700 | if (!head) | |
701 | return; | |
702 | ||
703 | /* | |
704 | * Set the previous list pointer to have the HEAD flag. | |
705 | */ | |
706 | rb_set_list_to_head(cpu_buffer, head->list.prev); | |
707 | } | |
708 | ||
709 | static void rb_list_head_clear(struct list_head *list) | |
710 | { | |
711 | unsigned long *ptr = (unsigned long *)&list->next; | |
712 | ||
713 | *ptr &= ~RB_FLAG_MASK; | |
714 | } | |
715 | ||
716 | /* | |
717 | * rb_head_page_dactivate - clears head page ptr (for free list) | |
718 | */ | |
719 | static void | |
720 | rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer) | |
721 | { | |
722 | struct list_head *hd; | |
723 | ||
724 | /* Go through the whole list and clear any pointers found. */ | |
725 | rb_list_head_clear(cpu_buffer->pages); | |
726 | ||
727 | list_for_each(hd, cpu_buffer->pages) | |
728 | rb_list_head_clear(hd); | |
729 | } | |
730 | ||
731 | static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer, | |
732 | struct buffer_page *head, | |
733 | struct buffer_page *prev, | |
734 | int old_flag, int new_flag) | |
735 | { | |
736 | struct list_head *list; | |
737 | unsigned long val = (unsigned long)&head->list; | |
738 | unsigned long ret; | |
739 | ||
740 | list = &prev->list; | |
741 | ||
742 | val &= ~RB_FLAG_MASK; | |
743 | ||
08a40816 SR |
744 | ret = cmpxchg((unsigned long *)&list->next, |
745 | val | old_flag, val | new_flag); | |
77ae365e SR |
746 | |
747 | /* check if the reader took the page */ | |
748 | if ((ret & ~RB_FLAG_MASK) != val) | |
749 | return RB_PAGE_MOVED; | |
750 | ||
751 | return ret & RB_FLAG_MASK; | |
752 | } | |
753 | ||
754 | static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer, | |
755 | struct buffer_page *head, | |
756 | struct buffer_page *prev, | |
757 | int old_flag) | |
758 | { | |
759 | return rb_head_page_set(cpu_buffer, head, prev, | |
760 | old_flag, RB_PAGE_UPDATE); | |
761 | } | |
762 | ||
763 | static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer, | |
764 | struct buffer_page *head, | |
765 | struct buffer_page *prev, | |
766 | int old_flag) | |
767 | { | |
768 | return rb_head_page_set(cpu_buffer, head, prev, | |
769 | old_flag, RB_PAGE_HEAD); | |
770 | } | |
771 | ||
772 | static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer, | |
773 | struct buffer_page *head, | |
774 | struct buffer_page *prev, | |
775 | int old_flag) | |
776 | { | |
777 | return rb_head_page_set(cpu_buffer, head, prev, | |
778 | old_flag, RB_PAGE_NORMAL); | |
779 | } | |
780 | ||
781 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
782 | struct buffer_page **bpage) | |
783 | { | |
784 | struct list_head *p = rb_list_head((*bpage)->list.next); | |
785 | ||
786 | *bpage = list_entry(p, struct buffer_page, list); | |
787 | } | |
788 | ||
789 | static struct buffer_page * | |
790 | rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer) | |
791 | { | |
792 | struct buffer_page *head; | |
793 | struct buffer_page *page; | |
794 | struct list_head *list; | |
795 | int i; | |
796 | ||
797 | if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page)) | |
798 | return NULL; | |
799 | ||
800 | /* sanity check */ | |
801 | list = cpu_buffer->pages; | |
802 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list)) | |
803 | return NULL; | |
804 | ||
805 | page = head = cpu_buffer->head_page; | |
806 | /* | |
807 | * It is possible that the writer moves the header behind | |
808 | * where we started, and we miss in one loop. | |
809 | * A second loop should grab the header, but we'll do | |
810 | * three loops just because I'm paranoid. | |
811 | */ | |
812 | for (i = 0; i < 3; i++) { | |
813 | do { | |
814 | if (rb_is_head_page(cpu_buffer, page, page->list.prev)) { | |
815 | cpu_buffer->head_page = page; | |
816 | return page; | |
817 | } | |
818 | rb_inc_page(cpu_buffer, &page); | |
819 | } while (page != head); | |
820 | } | |
821 | ||
822 | RB_WARN_ON(cpu_buffer, 1); | |
823 | ||
824 | return NULL; | |
825 | } | |
826 | ||
827 | static int rb_head_page_replace(struct buffer_page *old, | |
828 | struct buffer_page *new) | |
829 | { | |
830 | unsigned long *ptr = (unsigned long *)&old->list.prev->next; | |
831 | unsigned long val; | |
832 | unsigned long ret; | |
833 | ||
834 | val = *ptr & ~RB_FLAG_MASK; | |
835 | val |= RB_PAGE_HEAD; | |
836 | ||
08a40816 | 837 | ret = cmpxchg(ptr, val, (unsigned long)&new->list); |
77ae365e SR |
838 | |
839 | return ret == val; | |
840 | } | |
841 | ||
842 | /* | |
843 | * rb_tail_page_update - move the tail page forward | |
844 | * | |
845 | * Returns 1 if moved tail page, 0 if someone else did. | |
846 | */ | |
847 | static int rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer, | |
848 | struct buffer_page *tail_page, | |
849 | struct buffer_page *next_page) | |
850 | { | |
851 | struct buffer_page *old_tail; | |
852 | unsigned long old_entries; | |
853 | unsigned long old_write; | |
854 | int ret = 0; | |
855 | ||
856 | /* | |
857 | * The tail page now needs to be moved forward. | |
858 | * | |
859 | * We need to reset the tail page, but without messing | |
860 | * with possible erasing of data brought in by interrupts | |
861 | * that have moved the tail page and are currently on it. | |
862 | * | |
863 | * We add a counter to the write field to denote this. | |
864 | */ | |
865 | old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write); | |
866 | old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries); | |
867 | ||
868 | /* | |
869 | * Just make sure we have seen our old_write and synchronize | |
870 | * with any interrupts that come in. | |
871 | */ | |
872 | barrier(); | |
873 | ||
874 | /* | |
875 | * If the tail page is still the same as what we think | |
876 | * it is, then it is up to us to update the tail | |
877 | * pointer. | |
878 | */ | |
879 | if (tail_page == cpu_buffer->tail_page) { | |
880 | /* Zero the write counter */ | |
881 | unsigned long val = old_write & ~RB_WRITE_MASK; | |
882 | unsigned long eval = old_entries & ~RB_WRITE_MASK; | |
883 | ||
884 | /* | |
885 | * This will only succeed if an interrupt did | |
886 | * not come in and change it. In which case, we | |
887 | * do not want to modify it. | |
da706d8b LJ |
888 | * |
889 | * We add (void) to let the compiler know that we do not care | |
890 | * about the return value of these functions. We use the | |
891 | * cmpxchg to only update if an interrupt did not already | |
892 | * do it for us. If the cmpxchg fails, we don't care. | |
77ae365e | 893 | */ |
da706d8b LJ |
894 | (void)local_cmpxchg(&next_page->write, old_write, val); |
895 | (void)local_cmpxchg(&next_page->entries, old_entries, eval); | |
77ae365e SR |
896 | |
897 | /* | |
898 | * No need to worry about races with clearing out the commit. | |
899 | * it only can increment when a commit takes place. But that | |
900 | * only happens in the outer most nested commit. | |
901 | */ | |
902 | local_set(&next_page->page->commit, 0); | |
903 | ||
904 | old_tail = cmpxchg(&cpu_buffer->tail_page, | |
905 | tail_page, next_page); | |
906 | ||
907 | if (old_tail == tail_page) | |
908 | ret = 1; | |
909 | } | |
910 | ||
911 | return ret; | |
912 | } | |
913 | ||
914 | static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer, | |
915 | struct buffer_page *bpage) | |
916 | { | |
917 | unsigned long val = (unsigned long)bpage; | |
918 | ||
919 | if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK)) | |
920 | return 1; | |
921 | ||
922 | return 0; | |
923 | } | |
924 | ||
925 | /** | |
926 | * rb_check_list - make sure a pointer to a list has the last bits zero | |
927 | */ | |
928 | static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer, | |
929 | struct list_head *list) | |
930 | { | |
931 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev)) | |
932 | return 1; | |
933 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next)) | |
934 | return 1; | |
935 | return 0; | |
936 | } | |
937 | ||
7a8e76a3 SR |
938 | /** |
939 | * check_pages - integrity check of buffer pages | |
940 | * @cpu_buffer: CPU buffer with pages to test | |
941 | * | |
c3706f00 | 942 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
943 | * been corrupted. |
944 | */ | |
945 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
946 | { | |
3adc54fa | 947 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 948 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 949 | |
308f7eeb SR |
950 | /* Reset the head page if it exists */ |
951 | if (cpu_buffer->head_page) | |
952 | rb_set_head_page(cpu_buffer); | |
953 | ||
77ae365e SR |
954 | rb_head_page_deactivate(cpu_buffer); |
955 | ||
3e89c7bb SR |
956 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
957 | return -1; | |
958 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
959 | return -1; | |
7a8e76a3 | 960 | |
77ae365e SR |
961 | if (rb_check_list(cpu_buffer, head)) |
962 | return -1; | |
963 | ||
044fa782 | 964 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 965 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 966 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
967 | return -1; |
968 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 969 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 970 | return -1; |
77ae365e SR |
971 | if (rb_check_list(cpu_buffer, &bpage->list)) |
972 | return -1; | |
7a8e76a3 SR |
973 | } |
974 | ||
77ae365e SR |
975 | rb_head_page_activate(cpu_buffer); |
976 | ||
7a8e76a3 SR |
977 | return 0; |
978 | } | |
979 | ||
438ced17 | 980 | static int __rb_allocate_pages(int nr_pages, struct list_head *pages, int cpu) |
7a8e76a3 | 981 | { |
438ced17 | 982 | int i; |
044fa782 | 983 | struct buffer_page *bpage, *tmp; |
3adc54fa | 984 | |
7a8e76a3 | 985 | for (i = 0; i < nr_pages; i++) { |
7ea59064 | 986 | struct page *page; |
d7ec4bfe VN |
987 | /* |
988 | * __GFP_NORETRY flag makes sure that the allocation fails | |
989 | * gracefully without invoking oom-killer and the system is | |
990 | * not destabilized. | |
991 | */ | |
044fa782 | 992 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
d7ec4bfe | 993 | GFP_KERNEL | __GFP_NORETRY, |
438ced17 | 994 | cpu_to_node(cpu)); |
044fa782 | 995 | if (!bpage) |
e4c2ce82 | 996 | goto free_pages; |
77ae365e | 997 | |
438ced17 | 998 | list_add(&bpage->list, pages); |
77ae365e | 999 | |
438ced17 | 1000 | page = alloc_pages_node(cpu_to_node(cpu), |
d7ec4bfe | 1001 | GFP_KERNEL | __GFP_NORETRY, 0); |
7ea59064 | 1002 | if (!page) |
7a8e76a3 | 1003 | goto free_pages; |
7ea59064 | 1004 | bpage->page = page_address(page); |
044fa782 | 1005 | rb_init_page(bpage->page); |
7a8e76a3 SR |
1006 | } |
1007 | ||
438ced17 VN |
1008 | return 0; |
1009 | ||
1010 | free_pages: | |
1011 | list_for_each_entry_safe(bpage, tmp, pages, list) { | |
1012 | list_del_init(&bpage->list); | |
1013 | free_buffer_page(bpage); | |
1014 | } | |
1015 | ||
1016 | return -ENOMEM; | |
1017 | } | |
1018 | ||
1019 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
1020 | unsigned nr_pages) | |
1021 | { | |
1022 | LIST_HEAD(pages); | |
1023 | ||
1024 | WARN_ON(!nr_pages); | |
1025 | ||
1026 | if (__rb_allocate_pages(nr_pages, &pages, cpu_buffer->cpu)) | |
1027 | return -ENOMEM; | |
1028 | ||
3adc54fa SR |
1029 | /* |
1030 | * The ring buffer page list is a circular list that does not | |
1031 | * start and end with a list head. All page list items point to | |
1032 | * other pages. | |
1033 | */ | |
1034 | cpu_buffer->pages = pages.next; | |
1035 | list_del(&pages); | |
7a8e76a3 | 1036 | |
438ced17 VN |
1037 | cpu_buffer->nr_pages = nr_pages; |
1038 | ||
7a8e76a3 SR |
1039 | rb_check_pages(cpu_buffer); |
1040 | ||
1041 | return 0; | |
7a8e76a3 SR |
1042 | } |
1043 | ||
1044 | static struct ring_buffer_per_cpu * | |
438ced17 | 1045 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int nr_pages, int cpu) |
7a8e76a3 SR |
1046 | { |
1047 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 1048 | struct buffer_page *bpage; |
7ea59064 | 1049 | struct page *page; |
7a8e76a3 SR |
1050 | int ret; |
1051 | ||
1052 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
1053 | GFP_KERNEL, cpu_to_node(cpu)); | |
1054 | if (!cpu_buffer) | |
1055 | return NULL; | |
1056 | ||
1057 | cpu_buffer->cpu = cpu; | |
1058 | cpu_buffer->buffer = buffer; | |
5389f6fa | 1059 | raw_spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 1060 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
edc35bd7 | 1061 | cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
83f40318 | 1062 | INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler); |
05fdd70d | 1063 | init_completion(&cpu_buffer->update_done); |
7a8e76a3 | 1064 | |
044fa782 | 1065 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 1066 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 1067 | if (!bpage) |
e4c2ce82 SR |
1068 | goto fail_free_buffer; |
1069 | ||
77ae365e SR |
1070 | rb_check_bpage(cpu_buffer, bpage); |
1071 | ||
044fa782 | 1072 | cpu_buffer->reader_page = bpage; |
7ea59064 VN |
1073 | page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0); |
1074 | if (!page) | |
e4c2ce82 | 1075 | goto fail_free_reader; |
7ea59064 | 1076 | bpage->page = page_address(page); |
044fa782 | 1077 | rb_init_page(bpage->page); |
e4c2ce82 | 1078 | |
d769041f | 1079 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
44b99462 | 1080 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
d769041f | 1081 | |
438ced17 | 1082 | ret = rb_allocate_pages(cpu_buffer, nr_pages); |
7a8e76a3 | 1083 | if (ret < 0) |
d769041f | 1084 | goto fail_free_reader; |
7a8e76a3 SR |
1085 | |
1086 | cpu_buffer->head_page | |
3adc54fa | 1087 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 1088 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 | 1089 | |
77ae365e SR |
1090 | rb_head_page_activate(cpu_buffer); |
1091 | ||
7a8e76a3 SR |
1092 | return cpu_buffer; |
1093 | ||
d769041f SR |
1094 | fail_free_reader: |
1095 | free_buffer_page(cpu_buffer->reader_page); | |
1096 | ||
7a8e76a3 SR |
1097 | fail_free_buffer: |
1098 | kfree(cpu_buffer); | |
1099 | return NULL; | |
1100 | } | |
1101 | ||
1102 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
1103 | { | |
3adc54fa | 1104 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1105 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1106 | |
d769041f SR |
1107 | free_buffer_page(cpu_buffer->reader_page); |
1108 | ||
77ae365e SR |
1109 | rb_head_page_deactivate(cpu_buffer); |
1110 | ||
3adc54fa SR |
1111 | if (head) { |
1112 | list_for_each_entry_safe(bpage, tmp, head, list) { | |
1113 | list_del_init(&bpage->list); | |
1114 | free_buffer_page(bpage); | |
1115 | } | |
1116 | bpage = list_entry(head, struct buffer_page, list); | |
044fa782 | 1117 | free_buffer_page(bpage); |
7a8e76a3 | 1118 | } |
3adc54fa | 1119 | |
7a8e76a3 SR |
1120 | kfree(cpu_buffer); |
1121 | } | |
1122 | ||
59222efe | 1123 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
1124 | static int rb_cpu_notify(struct notifier_block *self, |
1125 | unsigned long action, void *hcpu); | |
554f786e SR |
1126 | #endif |
1127 | ||
7a8e76a3 SR |
1128 | /** |
1129 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 1130 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
1131 | * @flags: attributes to set for the ring buffer. |
1132 | * | |
1133 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
1134 | * flag. This flag means that the buffer will overwrite old data | |
1135 | * when the buffer wraps. If this flag is not set, the buffer will | |
1136 | * drop data when the tail hits the head. | |
1137 | */ | |
1f8a6a10 PZ |
1138 | struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
1139 | struct lock_class_key *key) | |
7a8e76a3 SR |
1140 | { |
1141 | struct ring_buffer *buffer; | |
1142 | int bsize; | |
438ced17 | 1143 | int cpu, nr_pages; |
7a8e76a3 SR |
1144 | |
1145 | /* keep it in its own cache line */ | |
1146 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
1147 | GFP_KERNEL); | |
1148 | if (!buffer) | |
1149 | return NULL; | |
1150 | ||
9e01c1b7 RR |
1151 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
1152 | goto fail_free_buffer; | |
1153 | ||
438ced17 | 1154 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
7a8e76a3 | 1155 | buffer->flags = flags; |
37886f6a | 1156 | buffer->clock = trace_clock_local; |
1f8a6a10 | 1157 | buffer->reader_lock_key = key; |
7a8e76a3 SR |
1158 | |
1159 | /* need at least two pages */ | |
438ced17 VN |
1160 | if (nr_pages < 2) |
1161 | nr_pages = 2; | |
7a8e76a3 | 1162 | |
3bf832ce FW |
1163 | /* |
1164 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
1165 | * in early initcall, it will not be notified of secondary cpus. | |
1166 | * In that off case, we need to allocate for all possible cpus. | |
1167 | */ | |
1168 | #ifdef CONFIG_HOTPLUG_CPU | |
554f786e SR |
1169 | get_online_cpus(); |
1170 | cpumask_copy(buffer->cpumask, cpu_online_mask); | |
3bf832ce FW |
1171 | #else |
1172 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
1173 | #endif | |
7a8e76a3 SR |
1174 | buffer->cpus = nr_cpu_ids; |
1175 | ||
1176 | bsize = sizeof(void *) * nr_cpu_ids; | |
1177 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
1178 | GFP_KERNEL); | |
1179 | if (!buffer->buffers) | |
9e01c1b7 | 1180 | goto fail_free_cpumask; |
7a8e76a3 SR |
1181 | |
1182 | for_each_buffer_cpu(buffer, cpu) { | |
1183 | buffer->buffers[cpu] = | |
438ced17 | 1184 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
7a8e76a3 SR |
1185 | if (!buffer->buffers[cpu]) |
1186 | goto fail_free_buffers; | |
1187 | } | |
1188 | ||
59222efe | 1189 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1190 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
1191 | buffer->cpu_notify.priority = 0; | |
1192 | register_cpu_notifier(&buffer->cpu_notify); | |
1193 | #endif | |
1194 | ||
1195 | put_online_cpus(); | |
7a8e76a3 SR |
1196 | mutex_init(&buffer->mutex); |
1197 | ||
1198 | return buffer; | |
1199 | ||
1200 | fail_free_buffers: | |
1201 | for_each_buffer_cpu(buffer, cpu) { | |
1202 | if (buffer->buffers[cpu]) | |
1203 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1204 | } | |
1205 | kfree(buffer->buffers); | |
1206 | ||
9e01c1b7 RR |
1207 | fail_free_cpumask: |
1208 | free_cpumask_var(buffer->cpumask); | |
554f786e | 1209 | put_online_cpus(); |
9e01c1b7 | 1210 | |
7a8e76a3 SR |
1211 | fail_free_buffer: |
1212 | kfree(buffer); | |
1213 | return NULL; | |
1214 | } | |
1f8a6a10 | 1215 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
1216 | |
1217 | /** | |
1218 | * ring_buffer_free - free a ring buffer. | |
1219 | * @buffer: the buffer to free. | |
1220 | */ | |
1221 | void | |
1222 | ring_buffer_free(struct ring_buffer *buffer) | |
1223 | { | |
1224 | int cpu; | |
1225 | ||
554f786e SR |
1226 | get_online_cpus(); |
1227 | ||
59222efe | 1228 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1229 | unregister_cpu_notifier(&buffer->cpu_notify); |
1230 | #endif | |
1231 | ||
7a8e76a3 SR |
1232 | for_each_buffer_cpu(buffer, cpu) |
1233 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1234 | ||
554f786e SR |
1235 | put_online_cpus(); |
1236 | ||
bd3f0221 | 1237 | kfree(buffer->buffers); |
9e01c1b7 RR |
1238 | free_cpumask_var(buffer->cpumask); |
1239 | ||
7a8e76a3 SR |
1240 | kfree(buffer); |
1241 | } | |
c4f50183 | 1242 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 1243 | |
37886f6a SR |
1244 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
1245 | u64 (*clock)(void)) | |
1246 | { | |
1247 | buffer->clock = clock; | |
1248 | } | |
1249 | ||
7a8e76a3 SR |
1250 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
1251 | ||
83f40318 VN |
1252 | static inline unsigned long rb_page_entries(struct buffer_page *bpage) |
1253 | { | |
1254 | return local_read(&bpage->entries) & RB_WRITE_MASK; | |
1255 | } | |
1256 | ||
1257 | static inline unsigned long rb_page_write(struct buffer_page *bpage) | |
1258 | { | |
1259 | return local_read(&bpage->write) & RB_WRITE_MASK; | |
1260 | } | |
1261 | ||
5040b4b7 | 1262 | static int |
83f40318 | 1263 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned int nr_pages) |
7a8e76a3 | 1264 | { |
83f40318 VN |
1265 | struct list_head *tail_page, *to_remove, *next_page; |
1266 | struct buffer_page *to_remove_page, *tmp_iter_page; | |
1267 | struct buffer_page *last_page, *first_page; | |
1268 | unsigned int nr_removed; | |
1269 | unsigned long head_bit; | |
1270 | int page_entries; | |
1271 | ||
1272 | head_bit = 0; | |
7a8e76a3 | 1273 | |
5389f6fa | 1274 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1275 | atomic_inc(&cpu_buffer->record_disabled); |
1276 | /* | |
1277 | * We don't race with the readers since we have acquired the reader | |
1278 | * lock. We also don't race with writers after disabling recording. | |
1279 | * This makes it easy to figure out the first and the last page to be | |
1280 | * removed from the list. We unlink all the pages in between including | |
1281 | * the first and last pages. This is done in a busy loop so that we | |
1282 | * lose the least number of traces. | |
1283 | * The pages are freed after we restart recording and unlock readers. | |
1284 | */ | |
1285 | tail_page = &cpu_buffer->tail_page->list; | |
77ae365e | 1286 | |
83f40318 VN |
1287 | /* |
1288 | * tail page might be on reader page, we remove the next page | |
1289 | * from the ring buffer | |
1290 | */ | |
1291 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
1292 | tail_page = rb_list_head(tail_page->next); | |
1293 | to_remove = tail_page; | |
1294 | ||
1295 | /* start of pages to remove */ | |
1296 | first_page = list_entry(rb_list_head(to_remove->next), | |
1297 | struct buffer_page, list); | |
1298 | ||
1299 | for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) { | |
1300 | to_remove = rb_list_head(to_remove)->next; | |
1301 | head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD; | |
7a8e76a3 | 1302 | } |
7a8e76a3 | 1303 | |
83f40318 | 1304 | next_page = rb_list_head(to_remove)->next; |
7a8e76a3 | 1305 | |
83f40318 VN |
1306 | /* |
1307 | * Now we remove all pages between tail_page and next_page. | |
1308 | * Make sure that we have head_bit value preserved for the | |
1309 | * next page | |
1310 | */ | |
1311 | tail_page->next = (struct list_head *)((unsigned long)next_page | | |
1312 | head_bit); | |
1313 | next_page = rb_list_head(next_page); | |
1314 | next_page->prev = tail_page; | |
1315 | ||
1316 | /* make sure pages points to a valid page in the ring buffer */ | |
1317 | cpu_buffer->pages = next_page; | |
1318 | ||
1319 | /* update head page */ | |
1320 | if (head_bit) | |
1321 | cpu_buffer->head_page = list_entry(next_page, | |
1322 | struct buffer_page, list); | |
1323 | ||
1324 | /* | |
1325 | * change read pointer to make sure any read iterators reset | |
1326 | * themselves | |
1327 | */ | |
1328 | cpu_buffer->read = 0; | |
1329 | ||
1330 | /* pages are removed, resume tracing and then free the pages */ | |
1331 | atomic_dec(&cpu_buffer->record_disabled); | |
5389f6fa | 1332 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1333 | |
1334 | RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)); | |
1335 | ||
1336 | /* last buffer page to remove */ | |
1337 | last_page = list_entry(rb_list_head(to_remove), struct buffer_page, | |
1338 | list); | |
1339 | tmp_iter_page = first_page; | |
1340 | ||
1341 | do { | |
1342 | to_remove_page = tmp_iter_page; | |
1343 | rb_inc_page(cpu_buffer, &tmp_iter_page); | |
1344 | ||
1345 | /* update the counters */ | |
1346 | page_entries = rb_page_entries(to_remove_page); | |
1347 | if (page_entries) { | |
1348 | /* | |
1349 | * If something was added to this page, it was full | |
1350 | * since it is not the tail page. So we deduct the | |
1351 | * bytes consumed in ring buffer from here. | |
48fdc72f | 1352 | * Increment overrun to account for the lost events. |
83f40318 | 1353 | */ |
48fdc72f | 1354 | local_add(page_entries, &cpu_buffer->overrun); |
83f40318 VN |
1355 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
1356 | } | |
1357 | ||
1358 | /* | |
1359 | * We have already removed references to this list item, just | |
1360 | * free up the buffer_page and its page | |
1361 | */ | |
1362 | free_buffer_page(to_remove_page); | |
1363 | nr_removed--; | |
1364 | ||
1365 | } while (to_remove_page != last_page); | |
1366 | ||
1367 | RB_WARN_ON(cpu_buffer, nr_removed); | |
5040b4b7 VN |
1368 | |
1369 | return nr_removed == 0; | |
7a8e76a3 SR |
1370 | } |
1371 | ||
5040b4b7 VN |
1372 | static int |
1373 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 1374 | { |
5040b4b7 VN |
1375 | struct list_head *pages = &cpu_buffer->new_pages; |
1376 | int retries, success; | |
7a8e76a3 | 1377 | |
5389f6fa | 1378 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1379 | /* |
1380 | * We are holding the reader lock, so the reader page won't be swapped | |
1381 | * in the ring buffer. Now we are racing with the writer trying to | |
1382 | * move head page and the tail page. | |
1383 | * We are going to adapt the reader page update process where: | |
1384 | * 1. We first splice the start and end of list of new pages between | |
1385 | * the head page and its previous page. | |
1386 | * 2. We cmpxchg the prev_page->next to point from head page to the | |
1387 | * start of new pages list. | |
1388 | * 3. Finally, we update the head->prev to the end of new list. | |
1389 | * | |
1390 | * We will try this process 10 times, to make sure that we don't keep | |
1391 | * spinning. | |
1392 | */ | |
1393 | retries = 10; | |
1394 | success = 0; | |
1395 | while (retries--) { | |
1396 | struct list_head *head_page, *prev_page, *r; | |
1397 | struct list_head *last_page, *first_page; | |
1398 | struct list_head *head_page_with_bit; | |
77ae365e | 1399 | |
5040b4b7 | 1400 | head_page = &rb_set_head_page(cpu_buffer)->list; |
54f7be5b SR |
1401 | if (!head_page) |
1402 | break; | |
5040b4b7 VN |
1403 | prev_page = head_page->prev; |
1404 | ||
1405 | first_page = pages->next; | |
1406 | last_page = pages->prev; | |
1407 | ||
1408 | head_page_with_bit = (struct list_head *) | |
1409 | ((unsigned long)head_page | RB_PAGE_HEAD); | |
1410 | ||
1411 | last_page->next = head_page_with_bit; | |
1412 | first_page->prev = prev_page; | |
1413 | ||
1414 | r = cmpxchg(&prev_page->next, head_page_with_bit, first_page); | |
1415 | ||
1416 | if (r == head_page_with_bit) { | |
1417 | /* | |
1418 | * yay, we replaced the page pointer to our new list, | |
1419 | * now, we just have to update to head page's prev | |
1420 | * pointer to point to end of list | |
1421 | */ | |
1422 | head_page->prev = last_page; | |
1423 | success = 1; | |
1424 | break; | |
1425 | } | |
7a8e76a3 | 1426 | } |
7a8e76a3 | 1427 | |
5040b4b7 VN |
1428 | if (success) |
1429 | INIT_LIST_HEAD(pages); | |
1430 | /* | |
1431 | * If we weren't successful in adding in new pages, warn and stop | |
1432 | * tracing | |
1433 | */ | |
1434 | RB_WARN_ON(cpu_buffer, !success); | |
5389f6fa | 1435 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1436 | |
1437 | /* free pages if they weren't inserted */ | |
1438 | if (!success) { | |
1439 | struct buffer_page *bpage, *tmp; | |
1440 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, | |
1441 | list) { | |
1442 | list_del_init(&bpage->list); | |
1443 | free_buffer_page(bpage); | |
1444 | } | |
1445 | } | |
1446 | return success; | |
7a8e76a3 SR |
1447 | } |
1448 | ||
83f40318 | 1449 | static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer) |
438ced17 | 1450 | { |
5040b4b7 VN |
1451 | int success; |
1452 | ||
438ced17 | 1453 | if (cpu_buffer->nr_pages_to_update > 0) |
5040b4b7 | 1454 | success = rb_insert_pages(cpu_buffer); |
438ced17 | 1455 | else |
5040b4b7 VN |
1456 | success = rb_remove_pages(cpu_buffer, |
1457 | -cpu_buffer->nr_pages_to_update); | |
83f40318 | 1458 | |
5040b4b7 VN |
1459 | if (success) |
1460 | cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update; | |
83f40318 VN |
1461 | } |
1462 | ||
1463 | static void update_pages_handler(struct work_struct *work) | |
1464 | { | |
1465 | struct ring_buffer_per_cpu *cpu_buffer = container_of(work, | |
1466 | struct ring_buffer_per_cpu, update_pages_work); | |
1467 | rb_update_pages(cpu_buffer); | |
05fdd70d | 1468 | complete(&cpu_buffer->update_done); |
438ced17 VN |
1469 | } |
1470 | ||
7a8e76a3 SR |
1471 | /** |
1472 | * ring_buffer_resize - resize the ring buffer | |
1473 | * @buffer: the buffer to resize. | |
1474 | * @size: the new size. | |
1475 | * | |
7a8e76a3 SR |
1476 | * Minimum size is 2 * BUF_PAGE_SIZE. |
1477 | * | |
83f40318 | 1478 | * Returns 0 on success and < 0 on failure. |
7a8e76a3 | 1479 | */ |
438ced17 VN |
1480 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size, |
1481 | int cpu_id) | |
7a8e76a3 SR |
1482 | { |
1483 | struct ring_buffer_per_cpu *cpu_buffer; | |
438ced17 | 1484 | unsigned nr_pages; |
83f40318 | 1485 | int cpu, err = 0; |
7a8e76a3 | 1486 | |
ee51a1de IM |
1487 | /* |
1488 | * Always succeed at resizing a non-existent buffer: | |
1489 | */ | |
1490 | if (!buffer) | |
1491 | return size; | |
1492 | ||
6a31e1f1 SR |
1493 | /* Make sure the requested buffer exists */ |
1494 | if (cpu_id != RING_BUFFER_ALL_CPUS && | |
1495 | !cpumask_test_cpu(cpu_id, buffer->cpumask)) | |
1496 | return size; | |
1497 | ||
7a8e76a3 SR |
1498 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
1499 | size *= BUF_PAGE_SIZE; | |
7a8e76a3 SR |
1500 | |
1501 | /* we need a minimum of two pages */ | |
1502 | if (size < BUF_PAGE_SIZE * 2) | |
1503 | size = BUF_PAGE_SIZE * 2; | |
1504 | ||
83f40318 | 1505 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
18421015 | 1506 | |
83f40318 VN |
1507 | /* |
1508 | * Don't succeed if resizing is disabled, as a reader might be | |
1509 | * manipulating the ring buffer and is expecting a sane state while | |
1510 | * this is true. | |
1511 | */ | |
1512 | if (atomic_read(&buffer->resize_disabled)) | |
1513 | return -EBUSY; | |
18421015 | 1514 | |
83f40318 | 1515 | /* prevent another thread from changing buffer sizes */ |
7a8e76a3 | 1516 | mutex_lock(&buffer->mutex); |
7a8e76a3 | 1517 | |
438ced17 VN |
1518 | if (cpu_id == RING_BUFFER_ALL_CPUS) { |
1519 | /* calculate the pages to update */ | |
7a8e76a3 SR |
1520 | for_each_buffer_cpu(buffer, cpu) { |
1521 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 1522 | |
438ced17 VN |
1523 | cpu_buffer->nr_pages_to_update = nr_pages - |
1524 | cpu_buffer->nr_pages; | |
438ced17 VN |
1525 | /* |
1526 | * nothing more to do for removing pages or no update | |
1527 | */ | |
1528 | if (cpu_buffer->nr_pages_to_update <= 0) | |
1529 | continue; | |
d7ec4bfe | 1530 | /* |
438ced17 VN |
1531 | * to add pages, make sure all new pages can be |
1532 | * allocated without receiving ENOMEM | |
d7ec4bfe | 1533 | */ |
438ced17 VN |
1534 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
1535 | if (__rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 | 1536 | &cpu_buffer->new_pages, cpu)) { |
438ced17 | 1537 | /* not enough memory for new pages */ |
83f40318 VN |
1538 | err = -ENOMEM; |
1539 | goto out_err; | |
1540 | } | |
1541 | } | |
1542 | ||
1543 | get_online_cpus(); | |
1544 | /* | |
1545 | * Fire off all the required work handlers | |
05fdd70d | 1546 | * We can't schedule on offline CPUs, but it's not necessary |
83f40318 VN |
1547 | * since we can change their buffer sizes without any race. |
1548 | */ | |
1549 | for_each_buffer_cpu(buffer, cpu) { | |
1550 | cpu_buffer = buffer->buffers[cpu]; | |
05fdd70d | 1551 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
1552 | continue; |
1553 | ||
05fdd70d VN |
1554 | if (cpu_online(cpu)) |
1555 | schedule_work_on(cpu, | |
1556 | &cpu_buffer->update_pages_work); | |
1557 | else | |
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]; | |
05fdd70d | 1564 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
1565 | continue; |
1566 | ||
05fdd70d VN |
1567 | if (cpu_online(cpu)) |
1568 | wait_for_completion(&cpu_buffer->update_done); | |
83f40318 | 1569 | cpu_buffer->nr_pages_to_update = 0; |
438ced17 | 1570 | } |
83f40318 VN |
1571 | |
1572 | put_online_cpus(); | |
438ced17 | 1573 | } else { |
8e49f418 VN |
1574 | /* Make sure this CPU has been intitialized */ |
1575 | if (!cpumask_test_cpu(cpu_id, buffer->cpumask)) | |
1576 | goto out; | |
1577 | ||
438ced17 | 1578 | cpu_buffer = buffer->buffers[cpu_id]; |
83f40318 | 1579 | |
438ced17 VN |
1580 | if (nr_pages == cpu_buffer->nr_pages) |
1581 | goto out; | |
7a8e76a3 | 1582 | |
438ced17 VN |
1583 | cpu_buffer->nr_pages_to_update = nr_pages - |
1584 | cpu_buffer->nr_pages; | |
1585 | ||
1586 | INIT_LIST_HEAD(&cpu_buffer->new_pages); | |
1587 | if (cpu_buffer->nr_pages_to_update > 0 && | |
1588 | __rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 VN |
1589 | &cpu_buffer->new_pages, cpu_id)) { |
1590 | err = -ENOMEM; | |
1591 | goto out_err; | |
1592 | } | |
438ced17 | 1593 | |
83f40318 VN |
1594 | get_online_cpus(); |
1595 | ||
1596 | if (cpu_online(cpu_id)) { | |
1597 | schedule_work_on(cpu_id, | |
1598 | &cpu_buffer->update_pages_work); | |
05fdd70d | 1599 | wait_for_completion(&cpu_buffer->update_done); |
83f40318 VN |
1600 | } else |
1601 | rb_update_pages(cpu_buffer); | |
1602 | ||
83f40318 | 1603 | cpu_buffer->nr_pages_to_update = 0; |
05fdd70d | 1604 | put_online_cpus(); |
438ced17 | 1605 | } |
7a8e76a3 SR |
1606 | |
1607 | out: | |
659f451f SR |
1608 | /* |
1609 | * The ring buffer resize can happen with the ring buffer | |
1610 | * enabled, so that the update disturbs the tracing as little | |
1611 | * as possible. But if the buffer is disabled, we do not need | |
1612 | * to worry about that, and we can take the time to verify | |
1613 | * that the buffer is not corrupt. | |
1614 | */ | |
1615 | if (atomic_read(&buffer->record_disabled)) { | |
1616 | atomic_inc(&buffer->record_disabled); | |
1617 | /* | |
1618 | * Even though the buffer was disabled, we must make sure | |
1619 | * that it is truly disabled before calling rb_check_pages. | |
1620 | * There could have been a race between checking | |
1621 | * record_disable and incrementing it. | |
1622 | */ | |
1623 | synchronize_sched(); | |
1624 | for_each_buffer_cpu(buffer, cpu) { | |
1625 | cpu_buffer = buffer->buffers[cpu]; | |
1626 | rb_check_pages(cpu_buffer); | |
1627 | } | |
1628 | atomic_dec(&buffer->record_disabled); | |
1629 | } | |
1630 | ||
7a8e76a3 | 1631 | mutex_unlock(&buffer->mutex); |
7a8e76a3 SR |
1632 | return size; |
1633 | ||
83f40318 | 1634 | out_err: |
438ced17 VN |
1635 | for_each_buffer_cpu(buffer, cpu) { |
1636 | struct buffer_page *bpage, *tmp; | |
83f40318 | 1637 | |
438ced17 | 1638 | cpu_buffer = buffer->buffers[cpu]; |
438ced17 | 1639 | cpu_buffer->nr_pages_to_update = 0; |
83f40318 | 1640 | |
438ced17 VN |
1641 | if (list_empty(&cpu_buffer->new_pages)) |
1642 | continue; | |
83f40318 | 1643 | |
438ced17 VN |
1644 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, |
1645 | list) { | |
1646 | list_del_init(&bpage->list); | |
1647 | free_buffer_page(bpage); | |
1648 | } | |
7a8e76a3 | 1649 | } |
641d2f63 | 1650 | mutex_unlock(&buffer->mutex); |
83f40318 | 1651 | return err; |
7a8e76a3 | 1652 | } |
c4f50183 | 1653 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 1654 | |
750912fa DS |
1655 | void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val) |
1656 | { | |
1657 | mutex_lock(&buffer->mutex); | |
1658 | if (val) | |
1659 | buffer->flags |= RB_FL_OVERWRITE; | |
1660 | else | |
1661 | buffer->flags &= ~RB_FL_OVERWRITE; | |
1662 | mutex_unlock(&buffer->mutex); | |
1663 | } | |
1664 | EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite); | |
1665 | ||
8789a9e7 | 1666 | static inline void * |
044fa782 | 1667 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 1668 | { |
044fa782 | 1669 | return bpage->data + index; |
8789a9e7 SR |
1670 | } |
1671 | ||
044fa782 | 1672 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 1673 | { |
044fa782 | 1674 | return bpage->page->data + index; |
7a8e76a3 SR |
1675 | } |
1676 | ||
1677 | static inline struct ring_buffer_event * | |
d769041f | 1678 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1679 | { |
6f807acd SR |
1680 | return __rb_page_index(cpu_buffer->reader_page, |
1681 | cpu_buffer->reader_page->read); | |
1682 | } | |
1683 | ||
7a8e76a3 SR |
1684 | static inline struct ring_buffer_event * |
1685 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
1686 | { | |
6f807acd | 1687 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
1688 | } |
1689 | ||
bf41a158 SR |
1690 | static inline unsigned rb_page_commit(struct buffer_page *bpage) |
1691 | { | |
abc9b56d | 1692 | return local_read(&bpage->page->commit); |
bf41a158 SR |
1693 | } |
1694 | ||
25985edc | 1695 | /* Size is determined by what has been committed */ |
bf41a158 SR |
1696 | static inline unsigned rb_page_size(struct buffer_page *bpage) |
1697 | { | |
1698 | return rb_page_commit(bpage); | |
1699 | } | |
1700 | ||
1701 | static inline unsigned | |
1702 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
1703 | { | |
1704 | return rb_page_commit(cpu_buffer->commit_page); | |
1705 | } | |
1706 | ||
bf41a158 SR |
1707 | static inline unsigned |
1708 | rb_event_index(struct ring_buffer_event *event) | |
1709 | { | |
1710 | unsigned long addr = (unsigned long)event; | |
1711 | ||
22f470f8 | 1712 | return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE; |
bf41a158 SR |
1713 | } |
1714 | ||
0f0c85fc | 1715 | static inline int |
fa743953 SR |
1716 | rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1717 | struct ring_buffer_event *event) | |
bf41a158 SR |
1718 | { |
1719 | unsigned long addr = (unsigned long)event; | |
1720 | unsigned long index; | |
1721 | ||
1722 | index = rb_event_index(event); | |
1723 | addr &= PAGE_MASK; | |
1724 | ||
1725 | return cpu_buffer->commit_page->page == (void *)addr && | |
1726 | rb_commit_index(cpu_buffer) == index; | |
1727 | } | |
1728 | ||
34a148bf | 1729 | static void |
bf41a158 | 1730 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1731 | { |
77ae365e SR |
1732 | unsigned long max_count; |
1733 | ||
bf41a158 SR |
1734 | /* |
1735 | * We only race with interrupts and NMIs on this CPU. | |
1736 | * If we own the commit event, then we can commit | |
1737 | * all others that interrupted us, since the interruptions | |
1738 | * are in stack format (they finish before they come | |
1739 | * back to us). This allows us to do a simple loop to | |
1740 | * assign the commit to the tail. | |
1741 | */ | |
a8ccf1d6 | 1742 | again: |
438ced17 | 1743 | max_count = cpu_buffer->nr_pages * 100; |
77ae365e | 1744 | |
bf41a158 | 1745 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
77ae365e SR |
1746 | if (RB_WARN_ON(cpu_buffer, !(--max_count))) |
1747 | return; | |
1748 | if (RB_WARN_ON(cpu_buffer, | |
1749 | rb_is_reader_page(cpu_buffer->tail_page))) | |
1750 | return; | |
1751 | local_set(&cpu_buffer->commit_page->page->commit, | |
1752 | rb_page_write(cpu_buffer->commit_page)); | |
bf41a158 | 1753 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); |
abc9b56d SR |
1754 | cpu_buffer->write_stamp = |
1755 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1756 | /* add barrier to keep gcc from optimizing too much */ |
1757 | barrier(); | |
1758 | } | |
1759 | while (rb_commit_index(cpu_buffer) != | |
1760 | rb_page_write(cpu_buffer->commit_page)) { | |
77ae365e SR |
1761 | |
1762 | local_set(&cpu_buffer->commit_page->page->commit, | |
1763 | rb_page_write(cpu_buffer->commit_page)); | |
1764 | RB_WARN_ON(cpu_buffer, | |
1765 | local_read(&cpu_buffer->commit_page->page->commit) & | |
1766 | ~RB_WRITE_MASK); | |
bf41a158 SR |
1767 | barrier(); |
1768 | } | |
a8ccf1d6 SR |
1769 | |
1770 | /* again, keep gcc from optimizing */ | |
1771 | barrier(); | |
1772 | ||
1773 | /* | |
1774 | * If an interrupt came in just after the first while loop | |
1775 | * and pushed the tail page forward, we will be left with | |
1776 | * a dangling commit that will never go forward. | |
1777 | */ | |
1778 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1779 | goto again; | |
7a8e76a3 SR |
1780 | } |
1781 | ||
d769041f | 1782 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1783 | { |
abc9b56d | 1784 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1785 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1786 | } |
1787 | ||
34a148bf | 1788 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1789 | { |
1790 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1791 | ||
1792 | /* | |
1793 | * The iterator could be on the reader page (it starts there). | |
1794 | * But the head could have moved, since the reader was | |
1795 | * found. Check for this case and assign the iterator | |
1796 | * to the head page instead of next. | |
1797 | */ | |
1798 | if (iter->head_page == cpu_buffer->reader_page) | |
77ae365e | 1799 | iter->head_page = rb_set_head_page(cpu_buffer); |
d769041f SR |
1800 | else |
1801 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1802 | ||
abc9b56d | 1803 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1804 | iter->head = 0; |
1805 | } | |
1806 | ||
69d1b839 SR |
1807 | /* Slow path, do not inline */ |
1808 | static noinline struct ring_buffer_event * | |
1809 | rb_add_time_stamp(struct ring_buffer_event *event, u64 delta) | |
1810 | { | |
1811 | event->type_len = RINGBUF_TYPE_TIME_EXTEND; | |
1812 | ||
1813 | /* Not the first event on the page? */ | |
1814 | if (rb_event_index(event)) { | |
1815 | event->time_delta = delta & TS_MASK; | |
1816 | event->array[0] = delta >> TS_SHIFT; | |
1817 | } else { | |
1818 | /* nope, just zero it */ | |
1819 | event->time_delta = 0; | |
1820 | event->array[0] = 0; | |
1821 | } | |
1822 | ||
1823 | return skip_time_extend(event); | |
1824 | } | |
1825 | ||
7a8e76a3 | 1826 | /** |
01e3e710 | 1827 | * rb_update_event - update event type and data |
7a8e76a3 SR |
1828 | * @event: the even to update |
1829 | * @type: the type of event | |
1830 | * @length: the size of the event field in the ring buffer | |
1831 | * | |
1832 | * Update the type and data fields of the event. The length | |
1833 | * is the actual size that is written to the ring buffer, | |
1834 | * and with this, we can determine what to place into the | |
1835 | * data field. | |
1836 | */ | |
34a148bf | 1837 | static void |
69d1b839 SR |
1838 | rb_update_event(struct ring_buffer_per_cpu *cpu_buffer, |
1839 | struct ring_buffer_event *event, unsigned length, | |
1840 | int add_timestamp, u64 delta) | |
7a8e76a3 | 1841 | { |
69d1b839 SR |
1842 | /* Only a commit updates the timestamp */ |
1843 | if (unlikely(!rb_event_is_commit(cpu_buffer, event))) | |
1844 | delta = 0; | |
7a8e76a3 | 1845 | |
69d1b839 SR |
1846 | /* |
1847 | * If we need to add a timestamp, then we | |
1848 | * add it to the start of the resevered space. | |
1849 | */ | |
1850 | if (unlikely(add_timestamp)) { | |
1851 | event = rb_add_time_stamp(event, delta); | |
1852 | length -= RB_LEN_TIME_EXTEND; | |
1853 | delta = 0; | |
7a8e76a3 | 1854 | } |
69d1b839 SR |
1855 | |
1856 | event->time_delta = delta; | |
1857 | length -= RB_EVNT_HDR_SIZE; | |
1858 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) { | |
1859 | event->type_len = 0; | |
1860 | event->array[0] = length; | |
1861 | } else | |
1862 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
7a8e76a3 SR |
1863 | } |
1864 | ||
77ae365e SR |
1865 | /* |
1866 | * rb_handle_head_page - writer hit the head page | |
1867 | * | |
1868 | * Returns: +1 to retry page | |
1869 | * 0 to continue | |
1870 | * -1 on error | |
1871 | */ | |
1872 | static int | |
1873 | rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
1874 | struct buffer_page *tail_page, | |
1875 | struct buffer_page *next_page) | |
1876 | { | |
1877 | struct buffer_page *new_head; | |
1878 | int entries; | |
1879 | int type; | |
1880 | int ret; | |
1881 | ||
1882 | entries = rb_page_entries(next_page); | |
1883 | ||
1884 | /* | |
1885 | * The hard part is here. We need to move the head | |
1886 | * forward, and protect against both readers on | |
1887 | * other CPUs and writers coming in via interrupts. | |
1888 | */ | |
1889 | type = rb_head_page_set_update(cpu_buffer, next_page, tail_page, | |
1890 | RB_PAGE_HEAD); | |
1891 | ||
1892 | /* | |
1893 | * type can be one of four: | |
1894 | * NORMAL - an interrupt already moved it for us | |
1895 | * HEAD - we are the first to get here. | |
1896 | * UPDATE - we are the interrupt interrupting | |
1897 | * a current move. | |
1898 | * MOVED - a reader on another CPU moved the next | |
1899 | * pointer to its reader page. Give up | |
1900 | * and try again. | |
1901 | */ | |
1902 | ||
1903 | switch (type) { | |
1904 | case RB_PAGE_HEAD: | |
1905 | /* | |
1906 | * We changed the head to UPDATE, thus | |
1907 | * it is our responsibility to update | |
1908 | * the counters. | |
1909 | */ | |
1910 | local_add(entries, &cpu_buffer->overrun); | |
c64e148a | 1911 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
77ae365e SR |
1912 | |
1913 | /* | |
1914 | * The entries will be zeroed out when we move the | |
1915 | * tail page. | |
1916 | */ | |
1917 | ||
1918 | /* still more to do */ | |
1919 | break; | |
1920 | ||
1921 | case RB_PAGE_UPDATE: | |
1922 | /* | |
1923 | * This is an interrupt that interrupt the | |
1924 | * previous update. Still more to do. | |
1925 | */ | |
1926 | break; | |
1927 | case RB_PAGE_NORMAL: | |
1928 | /* | |
1929 | * An interrupt came in before the update | |
1930 | * and processed this for us. | |
1931 | * Nothing left to do. | |
1932 | */ | |
1933 | return 1; | |
1934 | case RB_PAGE_MOVED: | |
1935 | /* | |
1936 | * The reader is on another CPU and just did | |
1937 | * a swap with our next_page. | |
1938 | * Try again. | |
1939 | */ | |
1940 | return 1; | |
1941 | default: | |
1942 | RB_WARN_ON(cpu_buffer, 1); /* WTF??? */ | |
1943 | return -1; | |
1944 | } | |
1945 | ||
1946 | /* | |
1947 | * Now that we are here, the old head pointer is | |
1948 | * set to UPDATE. This will keep the reader from | |
1949 | * swapping the head page with the reader page. | |
1950 | * The reader (on another CPU) will spin till | |
1951 | * we are finished. | |
1952 | * | |
1953 | * We just need to protect against interrupts | |
1954 | * doing the job. We will set the next pointer | |
1955 | * to HEAD. After that, we set the old pointer | |
1956 | * to NORMAL, but only if it was HEAD before. | |
1957 | * otherwise we are an interrupt, and only | |
1958 | * want the outer most commit to reset it. | |
1959 | */ | |
1960 | new_head = next_page; | |
1961 | rb_inc_page(cpu_buffer, &new_head); | |
1962 | ||
1963 | ret = rb_head_page_set_head(cpu_buffer, new_head, next_page, | |
1964 | RB_PAGE_NORMAL); | |
1965 | ||
1966 | /* | |
1967 | * Valid returns are: | |
1968 | * HEAD - an interrupt came in and already set it. | |
1969 | * NORMAL - One of two things: | |
1970 | * 1) We really set it. | |
1971 | * 2) A bunch of interrupts came in and moved | |
1972 | * the page forward again. | |
1973 | */ | |
1974 | switch (ret) { | |
1975 | case RB_PAGE_HEAD: | |
1976 | case RB_PAGE_NORMAL: | |
1977 | /* OK */ | |
1978 | break; | |
1979 | default: | |
1980 | RB_WARN_ON(cpu_buffer, 1); | |
1981 | return -1; | |
1982 | } | |
1983 | ||
1984 | /* | |
1985 | * It is possible that an interrupt came in, | |
1986 | * set the head up, then more interrupts came in | |
1987 | * and moved it again. When we get back here, | |
1988 | * the page would have been set to NORMAL but we | |
1989 | * just set it back to HEAD. | |
1990 | * | |
1991 | * How do you detect this? Well, if that happened | |
1992 | * the tail page would have moved. | |
1993 | */ | |
1994 | if (ret == RB_PAGE_NORMAL) { | |
1995 | /* | |
1996 | * If the tail had moved passed next, then we need | |
1997 | * to reset the pointer. | |
1998 | */ | |
1999 | if (cpu_buffer->tail_page != tail_page && | |
2000 | cpu_buffer->tail_page != next_page) | |
2001 | rb_head_page_set_normal(cpu_buffer, new_head, | |
2002 | next_page, | |
2003 | RB_PAGE_HEAD); | |
2004 | } | |
2005 | ||
2006 | /* | |
2007 | * If this was the outer most commit (the one that | |
2008 | * changed the original pointer from HEAD to UPDATE), | |
2009 | * then it is up to us to reset it to NORMAL. | |
2010 | */ | |
2011 | if (type == RB_PAGE_HEAD) { | |
2012 | ret = rb_head_page_set_normal(cpu_buffer, next_page, | |
2013 | tail_page, | |
2014 | RB_PAGE_UPDATE); | |
2015 | if (RB_WARN_ON(cpu_buffer, | |
2016 | ret != RB_PAGE_UPDATE)) | |
2017 | return -1; | |
2018 | } | |
2019 | ||
2020 | return 0; | |
2021 | } | |
2022 | ||
34a148bf | 2023 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
2024 | { |
2025 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
2026 | ||
2027 | /* zero length can cause confusions */ | |
2028 | if (!length) | |
2029 | length = 1; | |
2030 | ||
2271048d | 2031 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) |
7a8e76a3 SR |
2032 | length += sizeof(event.array[0]); |
2033 | ||
2034 | length += RB_EVNT_HDR_SIZE; | |
2271048d | 2035 | length = ALIGN(length, RB_ARCH_ALIGNMENT); |
7a8e76a3 SR |
2036 | |
2037 | return length; | |
2038 | } | |
2039 | ||
c7b09308 SR |
2040 | static inline void |
2041 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
2042 | struct buffer_page *tail_page, | |
2043 | unsigned long tail, unsigned long length) | |
2044 | { | |
2045 | struct ring_buffer_event *event; | |
2046 | ||
2047 | /* | |
2048 | * Only the event that crossed the page boundary | |
2049 | * must fill the old tail_page with padding. | |
2050 | */ | |
2051 | if (tail >= BUF_PAGE_SIZE) { | |
b3230c8b SR |
2052 | /* |
2053 | * If the page was filled, then we still need | |
2054 | * to update the real_end. Reset it to zero | |
2055 | * and the reader will ignore it. | |
2056 | */ | |
2057 | if (tail == BUF_PAGE_SIZE) | |
2058 | tail_page->real_end = 0; | |
2059 | ||
c7b09308 SR |
2060 | local_sub(length, &tail_page->write); |
2061 | return; | |
2062 | } | |
2063 | ||
2064 | event = __rb_page_index(tail_page, tail); | |
b0b7065b | 2065 | kmemcheck_annotate_bitfield(event, bitfield); |
c7b09308 | 2066 | |
c64e148a VN |
2067 | /* account for padding bytes */ |
2068 | local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes); | |
2069 | ||
ff0ff84a SR |
2070 | /* |
2071 | * Save the original length to the meta data. | |
2072 | * This will be used by the reader to add lost event | |
2073 | * counter. | |
2074 | */ | |
2075 | tail_page->real_end = tail; | |
2076 | ||
c7b09308 SR |
2077 | /* |
2078 | * If this event is bigger than the minimum size, then | |
2079 | * we need to be careful that we don't subtract the | |
2080 | * write counter enough to allow another writer to slip | |
2081 | * in on this page. | |
2082 | * We put in a discarded commit instead, to make sure | |
2083 | * that this space is not used again. | |
2084 | * | |
2085 | * If we are less than the minimum size, we don't need to | |
2086 | * worry about it. | |
2087 | */ | |
2088 | if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) { | |
2089 | /* No room for any events */ | |
2090 | ||
2091 | /* Mark the rest of the page with padding */ | |
2092 | rb_event_set_padding(event); | |
2093 | ||
2094 | /* Set the write back to the previous setting */ | |
2095 | local_sub(length, &tail_page->write); | |
2096 | return; | |
2097 | } | |
2098 | ||
2099 | /* Put in a discarded event */ | |
2100 | event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE; | |
2101 | event->type_len = RINGBUF_TYPE_PADDING; | |
2102 | /* time delta must be non zero */ | |
2103 | event->time_delta = 1; | |
c7b09308 SR |
2104 | |
2105 | /* Set write to end of buffer */ | |
2106 | length = (tail + length) - BUF_PAGE_SIZE; | |
2107 | local_sub(length, &tail_page->write); | |
2108 | } | |
6634ff26 | 2109 | |
747e94ae SR |
2110 | /* |
2111 | * This is the slow path, force gcc not to inline it. | |
2112 | */ | |
2113 | static noinline struct ring_buffer_event * | |
6634ff26 SR |
2114 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
2115 | unsigned long length, unsigned long tail, | |
e8bc43e8 | 2116 | struct buffer_page *tail_page, u64 ts) |
7a8e76a3 | 2117 | { |
5a50e33c | 2118 | struct buffer_page *commit_page = cpu_buffer->commit_page; |
7a8e76a3 | 2119 | struct ring_buffer *buffer = cpu_buffer->buffer; |
77ae365e SR |
2120 | struct buffer_page *next_page; |
2121 | int ret; | |
aa20ae84 SR |
2122 | |
2123 | next_page = tail_page; | |
2124 | ||
aa20ae84 SR |
2125 | rb_inc_page(cpu_buffer, &next_page); |
2126 | ||
aa20ae84 SR |
2127 | /* |
2128 | * If for some reason, we had an interrupt storm that made | |
2129 | * it all the way around the buffer, bail, and warn | |
2130 | * about it. | |
2131 | */ | |
2132 | if (unlikely(next_page == commit_page)) { | |
77ae365e | 2133 | local_inc(&cpu_buffer->commit_overrun); |
aa20ae84 SR |
2134 | goto out_reset; |
2135 | } | |
2136 | ||
77ae365e SR |
2137 | /* |
2138 | * This is where the fun begins! | |
2139 | * | |
2140 | * We are fighting against races between a reader that | |
2141 | * could be on another CPU trying to swap its reader | |
2142 | * page with the buffer head. | |
2143 | * | |
2144 | * We are also fighting against interrupts coming in and | |
2145 | * moving the head or tail on us as well. | |
2146 | * | |
2147 | * If the next page is the head page then we have filled | |
2148 | * the buffer, unless the commit page is still on the | |
2149 | * reader page. | |
2150 | */ | |
2151 | if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) { | |
aa20ae84 | 2152 | |
77ae365e SR |
2153 | /* |
2154 | * If the commit is not on the reader page, then | |
2155 | * move the header page. | |
2156 | */ | |
2157 | if (!rb_is_reader_page(cpu_buffer->commit_page)) { | |
2158 | /* | |
2159 | * If we are not in overwrite mode, | |
2160 | * this is easy, just stop here. | |
2161 | */ | |
884bfe89 SP |
2162 | if (!(buffer->flags & RB_FL_OVERWRITE)) { |
2163 | local_inc(&cpu_buffer->dropped_events); | |
77ae365e | 2164 | goto out_reset; |
884bfe89 | 2165 | } |
77ae365e SR |
2166 | |
2167 | ret = rb_handle_head_page(cpu_buffer, | |
2168 | tail_page, | |
2169 | next_page); | |
2170 | if (ret < 0) | |
2171 | goto out_reset; | |
2172 | if (ret) | |
2173 | goto out_again; | |
2174 | } else { | |
2175 | /* | |
2176 | * We need to be careful here too. The | |
2177 | * commit page could still be on the reader | |
2178 | * page. We could have a small buffer, and | |
2179 | * have filled up the buffer with events | |
2180 | * from interrupts and such, and wrapped. | |
2181 | * | |
2182 | * Note, if the tail page is also the on the | |
2183 | * reader_page, we let it move out. | |
2184 | */ | |
2185 | if (unlikely((cpu_buffer->commit_page != | |
2186 | cpu_buffer->tail_page) && | |
2187 | (cpu_buffer->commit_page == | |
2188 | cpu_buffer->reader_page))) { | |
2189 | local_inc(&cpu_buffer->commit_overrun); | |
2190 | goto out_reset; | |
2191 | } | |
aa20ae84 SR |
2192 | } |
2193 | } | |
2194 | ||
77ae365e SR |
2195 | ret = rb_tail_page_update(cpu_buffer, tail_page, next_page); |
2196 | if (ret) { | |
2197 | /* | |
2198 | * Nested commits always have zero deltas, so | |
2199 | * just reread the time stamp | |
2200 | */ | |
e8bc43e8 SR |
2201 | ts = rb_time_stamp(buffer); |
2202 | next_page->page->time_stamp = ts; | |
aa20ae84 SR |
2203 | } |
2204 | ||
77ae365e | 2205 | out_again: |
aa20ae84 | 2206 | |
77ae365e | 2207 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
aa20ae84 SR |
2208 | |
2209 | /* fail and let the caller try again */ | |
2210 | return ERR_PTR(-EAGAIN); | |
2211 | ||
45141d46 | 2212 | out_reset: |
6f3b3440 | 2213 | /* reset write */ |
c7b09308 | 2214 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
6f3b3440 | 2215 | |
bf41a158 | 2216 | return NULL; |
7a8e76a3 SR |
2217 | } |
2218 | ||
6634ff26 SR |
2219 | static struct ring_buffer_event * |
2220 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
69d1b839 SR |
2221 | unsigned long length, u64 ts, |
2222 | u64 delta, int add_timestamp) | |
6634ff26 | 2223 | { |
5a50e33c | 2224 | struct buffer_page *tail_page; |
6634ff26 SR |
2225 | struct ring_buffer_event *event; |
2226 | unsigned long tail, write; | |
2227 | ||
69d1b839 SR |
2228 | /* |
2229 | * If the time delta since the last event is too big to | |
2230 | * hold in the time field of the event, then we append a | |
2231 | * TIME EXTEND event ahead of the data event. | |
2232 | */ | |
2233 | if (unlikely(add_timestamp)) | |
2234 | length += RB_LEN_TIME_EXTEND; | |
2235 | ||
6634ff26 SR |
2236 | tail_page = cpu_buffer->tail_page; |
2237 | write = local_add_return(length, &tail_page->write); | |
77ae365e SR |
2238 | |
2239 | /* set write to only the index of the write */ | |
2240 | write &= RB_WRITE_MASK; | |
6634ff26 SR |
2241 | tail = write - length; |
2242 | ||
2243 | /* See if we shot pass the end of this buffer page */ | |
747e94ae | 2244 | if (unlikely(write > BUF_PAGE_SIZE)) |
6634ff26 | 2245 | return rb_move_tail(cpu_buffer, length, tail, |
5a50e33c | 2246 | tail_page, ts); |
6634ff26 SR |
2247 | |
2248 | /* We reserved something on the buffer */ | |
2249 | ||
6634ff26 | 2250 | event = __rb_page_index(tail_page, tail); |
1744a21d | 2251 | kmemcheck_annotate_bitfield(event, bitfield); |
69d1b839 | 2252 | rb_update_event(cpu_buffer, event, length, add_timestamp, delta); |
6634ff26 | 2253 | |
69d1b839 | 2254 | local_inc(&tail_page->entries); |
6634ff26 SR |
2255 | |
2256 | /* | |
fa743953 SR |
2257 | * If this is the first commit on the page, then update |
2258 | * its timestamp. | |
6634ff26 | 2259 | */ |
fa743953 | 2260 | if (!tail) |
e8bc43e8 | 2261 | tail_page->page->time_stamp = ts; |
6634ff26 | 2262 | |
c64e148a VN |
2263 | /* account for these added bytes */ |
2264 | local_add(length, &cpu_buffer->entries_bytes); | |
2265 | ||
6634ff26 SR |
2266 | return event; |
2267 | } | |
2268 | ||
edd813bf SR |
2269 | static inline int |
2270 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
2271 | struct ring_buffer_event *event) | |
2272 | { | |
2273 | unsigned long new_index, old_index; | |
2274 | struct buffer_page *bpage; | |
2275 | unsigned long index; | |
2276 | unsigned long addr; | |
2277 | ||
2278 | new_index = rb_event_index(event); | |
69d1b839 | 2279 | old_index = new_index + rb_event_ts_length(event); |
edd813bf SR |
2280 | addr = (unsigned long)event; |
2281 | addr &= PAGE_MASK; | |
2282 | ||
2283 | bpage = cpu_buffer->tail_page; | |
2284 | ||
2285 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { | |
77ae365e SR |
2286 | unsigned long write_mask = |
2287 | local_read(&bpage->write) & ~RB_WRITE_MASK; | |
c64e148a | 2288 | unsigned long event_length = rb_event_length(event); |
edd813bf SR |
2289 | /* |
2290 | * This is on the tail page. It is possible that | |
2291 | * a write could come in and move the tail page | |
2292 | * and write to the next page. That is fine | |
2293 | * because we just shorten what is on this page. | |
2294 | */ | |
77ae365e SR |
2295 | old_index += write_mask; |
2296 | new_index += write_mask; | |
edd813bf | 2297 | index = local_cmpxchg(&bpage->write, old_index, new_index); |
c64e148a VN |
2298 | if (index == old_index) { |
2299 | /* update counters */ | |
2300 | local_sub(event_length, &cpu_buffer->entries_bytes); | |
edd813bf | 2301 | return 1; |
c64e148a | 2302 | } |
edd813bf SR |
2303 | } |
2304 | ||
2305 | /* could not discard */ | |
2306 | return 0; | |
2307 | } | |
2308 | ||
fa743953 SR |
2309 | static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer) |
2310 | { | |
2311 | local_inc(&cpu_buffer->committing); | |
2312 | local_inc(&cpu_buffer->commits); | |
2313 | } | |
2314 | ||
d9abde21 | 2315 | static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer) |
fa743953 SR |
2316 | { |
2317 | unsigned long commits; | |
2318 | ||
2319 | if (RB_WARN_ON(cpu_buffer, | |
2320 | !local_read(&cpu_buffer->committing))) | |
2321 | return; | |
2322 | ||
2323 | again: | |
2324 | commits = local_read(&cpu_buffer->commits); | |
2325 | /* synchronize with interrupts */ | |
2326 | barrier(); | |
2327 | if (local_read(&cpu_buffer->committing) == 1) | |
2328 | rb_set_commit_to_write(cpu_buffer); | |
2329 | ||
2330 | local_dec(&cpu_buffer->committing); | |
2331 | ||
2332 | /* synchronize with interrupts */ | |
2333 | barrier(); | |
2334 | ||
2335 | /* | |
2336 | * Need to account for interrupts coming in between the | |
2337 | * updating of the commit page and the clearing of the | |
2338 | * committing counter. | |
2339 | */ | |
2340 | if (unlikely(local_read(&cpu_buffer->commits) != commits) && | |
2341 | !local_read(&cpu_buffer->committing)) { | |
2342 | local_inc(&cpu_buffer->committing); | |
2343 | goto again; | |
2344 | } | |
2345 | } | |
2346 | ||
7a8e76a3 | 2347 | static struct ring_buffer_event * |
62f0b3eb SR |
2348 | rb_reserve_next_event(struct ring_buffer *buffer, |
2349 | struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 2350 | unsigned long length) |
7a8e76a3 SR |
2351 | { |
2352 | struct ring_buffer_event *event; | |
69d1b839 | 2353 | u64 ts, delta; |
818e3dd3 | 2354 | int nr_loops = 0; |
69d1b839 | 2355 | int add_timestamp; |
140ff891 | 2356 | u64 diff; |
7a8e76a3 | 2357 | |
fa743953 SR |
2358 | rb_start_commit(cpu_buffer); |
2359 | ||
85bac32c | 2360 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
62f0b3eb SR |
2361 | /* |
2362 | * Due to the ability to swap a cpu buffer from a buffer | |
2363 | * it is possible it was swapped before we committed. | |
2364 | * (committing stops a swap). We check for it here and | |
2365 | * if it happened, we have to fail the write. | |
2366 | */ | |
2367 | barrier(); | |
2368 | if (unlikely(ACCESS_ONCE(cpu_buffer->buffer) != buffer)) { | |
2369 | local_dec(&cpu_buffer->committing); | |
2370 | local_dec(&cpu_buffer->commits); | |
2371 | return NULL; | |
2372 | } | |
85bac32c | 2373 | #endif |
62f0b3eb | 2374 | |
be957c44 | 2375 | length = rb_calculate_event_length(length); |
bf41a158 | 2376 | again: |
69d1b839 SR |
2377 | add_timestamp = 0; |
2378 | delta = 0; | |
2379 | ||
818e3dd3 SR |
2380 | /* |
2381 | * We allow for interrupts to reenter here and do a trace. | |
2382 | * If one does, it will cause this original code to loop | |
2383 | * back here. Even with heavy interrupts happening, this | |
2384 | * should only happen a few times in a row. If this happens | |
2385 | * 1000 times in a row, there must be either an interrupt | |
2386 | * storm or we have something buggy. | |
2387 | * Bail! | |
2388 | */ | |
3e89c7bb | 2389 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
fa743953 | 2390 | goto out_fail; |
818e3dd3 | 2391 | |
6d3f1e12 | 2392 | ts = rb_time_stamp(cpu_buffer->buffer); |
140ff891 | 2393 | diff = ts - cpu_buffer->write_stamp; |
7a8e76a3 | 2394 | |
140ff891 SR |
2395 | /* make sure this diff is calculated here */ |
2396 | barrier(); | |
bf41a158 | 2397 | |
140ff891 SR |
2398 | /* Did the write stamp get updated already? */ |
2399 | if (likely(ts >= cpu_buffer->write_stamp)) { | |
168b6b1d SR |
2400 | delta = diff; |
2401 | if (unlikely(test_time_stamp(delta))) { | |
31274d72 JO |
2402 | int local_clock_stable = 1; |
2403 | #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK | |
2404 | local_clock_stable = sched_clock_stable; | |
2405 | #endif | |
69d1b839 | 2406 | WARN_ONCE(delta > (1ULL << 59), |
31274d72 | 2407 | KERN_WARNING "Delta way too big! %llu ts=%llu write stamp = %llu\n%s", |
69d1b839 SR |
2408 | (unsigned long long)delta, |
2409 | (unsigned long long)ts, | |
31274d72 JO |
2410 | (unsigned long long)cpu_buffer->write_stamp, |
2411 | local_clock_stable ? "" : | |
2412 | "If you just came from a suspend/resume,\n" | |
2413 | "please switch to the trace global clock:\n" | |
2414 | " echo global > /sys/kernel/debug/tracing/trace_clock\n"); | |
69d1b839 | 2415 | add_timestamp = 1; |
7a8e76a3 | 2416 | } |
168b6b1d | 2417 | } |
7a8e76a3 | 2418 | |
69d1b839 SR |
2419 | event = __rb_reserve_next(cpu_buffer, length, ts, |
2420 | delta, add_timestamp); | |
168b6b1d | 2421 | if (unlikely(PTR_ERR(event) == -EAGAIN)) |
bf41a158 SR |
2422 | goto again; |
2423 | ||
fa743953 SR |
2424 | if (!event) |
2425 | goto out_fail; | |
7a8e76a3 | 2426 | |
7a8e76a3 | 2427 | return event; |
fa743953 SR |
2428 | |
2429 | out_fail: | |
2430 | rb_end_commit(cpu_buffer); | |
2431 | return NULL; | |
7a8e76a3 SR |
2432 | } |
2433 | ||
1155de47 PM |
2434 | #ifdef CONFIG_TRACING |
2435 | ||
567cd4da SR |
2436 | /* |
2437 | * The lock and unlock are done within a preempt disable section. | |
2438 | * The current_context per_cpu variable can only be modified | |
2439 | * by the current task between lock and unlock. But it can | |
2440 | * be modified more than once via an interrupt. To pass this | |
2441 | * information from the lock to the unlock without having to | |
2442 | * access the 'in_interrupt()' functions again (which do show | |
2443 | * a bit of overhead in something as critical as function tracing, | |
2444 | * we use a bitmask trick. | |
2445 | * | |
2446 | * bit 0 = NMI context | |
2447 | * bit 1 = IRQ context | |
2448 | * bit 2 = SoftIRQ context | |
2449 | * bit 3 = normal context. | |
2450 | * | |
2451 | * This works because this is the order of contexts that can | |
2452 | * preempt other contexts. A SoftIRQ never preempts an IRQ | |
2453 | * context. | |
2454 | * | |
2455 | * When the context is determined, the corresponding bit is | |
2456 | * checked and set (if it was set, then a recursion of that context | |
2457 | * happened). | |
2458 | * | |
2459 | * On unlock, we need to clear this bit. To do so, just subtract | |
2460 | * 1 from the current_context and AND it to itself. | |
2461 | * | |
2462 | * (binary) | |
2463 | * 101 - 1 = 100 | |
2464 | * 101 & 100 = 100 (clearing bit zero) | |
2465 | * | |
2466 | * 1010 - 1 = 1001 | |
2467 | * 1010 & 1001 = 1000 (clearing bit 1) | |
2468 | * | |
2469 | * The least significant bit can be cleared this way, and it | |
2470 | * just so happens that it is the same bit corresponding to | |
2471 | * the current context. | |
2472 | */ | |
2473 | static DEFINE_PER_CPU(unsigned int, current_context); | |
261842b7 | 2474 | |
567cd4da | 2475 | static __always_inline int trace_recursive_lock(void) |
261842b7 | 2476 | { |
567cd4da SR |
2477 | unsigned int val = this_cpu_read(current_context); |
2478 | int bit; | |
d9abde21 | 2479 | |
567cd4da SR |
2480 | if (in_interrupt()) { |
2481 | if (in_nmi()) | |
2482 | bit = 0; | |
2483 | else if (in_irq()) | |
2484 | bit = 1; | |
2485 | else | |
2486 | bit = 2; | |
2487 | } else | |
2488 | bit = 3; | |
d9abde21 | 2489 | |
567cd4da SR |
2490 | if (unlikely(val & (1 << bit))) |
2491 | return 1; | |
d9abde21 | 2492 | |
567cd4da SR |
2493 | val |= (1 << bit); |
2494 | this_cpu_write(current_context, val); | |
d9abde21 | 2495 | |
567cd4da | 2496 | return 0; |
261842b7 SR |
2497 | } |
2498 | ||
567cd4da | 2499 | static __always_inline void trace_recursive_unlock(void) |
261842b7 | 2500 | { |
567cd4da | 2501 | unsigned int val = this_cpu_read(current_context); |
261842b7 | 2502 | |
567cd4da SR |
2503 | val--; |
2504 | val &= this_cpu_read(current_context); | |
2505 | this_cpu_write(current_context, val); | |
261842b7 SR |
2506 | } |
2507 | ||
1155de47 PM |
2508 | #else |
2509 | ||
2510 | #define trace_recursive_lock() (0) | |
2511 | #define trace_recursive_unlock() do { } while (0) | |
2512 | ||
2513 | #endif | |
2514 | ||
7a8e76a3 SR |
2515 | /** |
2516 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
2517 | * @buffer: the ring buffer to reserve from | |
2518 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
2519 | * |
2520 | * Returns a reseverd event on the ring buffer to copy directly to. | |
2521 | * The user of this interface will need to get the body to write into | |
2522 | * and can use the ring_buffer_event_data() interface. | |
2523 | * | |
2524 | * The length is the length of the data needed, not the event length | |
2525 | * which also includes the event header. | |
2526 | * | |
2527 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
2528 | * If NULL is returned, then nothing has been allocated or locked. | |
2529 | */ | |
2530 | struct ring_buffer_event * | |
0a987751 | 2531 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
2532 | { |
2533 | struct ring_buffer_per_cpu *cpu_buffer; | |
2534 | struct ring_buffer_event *event; | |
5168ae50 | 2535 | int cpu; |
7a8e76a3 | 2536 | |
033601a3 | 2537 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2538 | return NULL; |
2539 | ||
bf41a158 | 2540 | /* If we are tracing schedule, we don't want to recurse */ |
5168ae50 | 2541 | preempt_disable_notrace(); |
bf41a158 | 2542 | |
52fbe9cd LJ |
2543 | if (atomic_read(&buffer->record_disabled)) |
2544 | goto out_nocheck; | |
2545 | ||
261842b7 SR |
2546 | if (trace_recursive_lock()) |
2547 | goto out_nocheck; | |
2548 | ||
7a8e76a3 SR |
2549 | cpu = raw_smp_processor_id(); |
2550 | ||
9e01c1b7 | 2551 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2552 | goto out; |
7a8e76a3 SR |
2553 | |
2554 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2555 | |
2556 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 2557 | goto out; |
7a8e76a3 | 2558 | |
be957c44 | 2559 | if (length > BUF_MAX_DATA_SIZE) |
bf41a158 | 2560 | goto out; |
7a8e76a3 | 2561 | |
62f0b3eb | 2562 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 2563 | if (!event) |
d769041f | 2564 | goto out; |
7a8e76a3 SR |
2565 | |
2566 | return event; | |
2567 | ||
d769041f | 2568 | out: |
261842b7 SR |
2569 | trace_recursive_unlock(); |
2570 | ||
2571 | out_nocheck: | |
5168ae50 | 2572 | preempt_enable_notrace(); |
7a8e76a3 SR |
2573 | return NULL; |
2574 | } | |
c4f50183 | 2575 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 | 2576 | |
a1863c21 SR |
2577 | static void |
2578 | rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
7a8e76a3 SR |
2579 | struct ring_buffer_event *event) |
2580 | { | |
69d1b839 SR |
2581 | u64 delta; |
2582 | ||
fa743953 SR |
2583 | /* |
2584 | * The event first in the commit queue updates the | |
2585 | * time stamp. | |
2586 | */ | |
69d1b839 SR |
2587 | if (rb_event_is_commit(cpu_buffer, event)) { |
2588 | /* | |
2589 | * A commit event that is first on a page | |
2590 | * updates the write timestamp with the page stamp | |
2591 | */ | |
2592 | if (!rb_event_index(event)) | |
2593 | cpu_buffer->write_stamp = | |
2594 | cpu_buffer->commit_page->page->time_stamp; | |
2595 | else if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
2596 | delta = event->array[0]; | |
2597 | delta <<= TS_SHIFT; | |
2598 | delta += event->time_delta; | |
2599 | cpu_buffer->write_stamp += delta; | |
2600 | } else | |
2601 | cpu_buffer->write_stamp += event->time_delta; | |
2602 | } | |
a1863c21 | 2603 | } |
bf41a158 | 2604 | |
a1863c21 SR |
2605 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, |
2606 | struct ring_buffer_event *event) | |
2607 | { | |
2608 | local_inc(&cpu_buffer->entries); | |
2609 | rb_update_write_stamp(cpu_buffer, event); | |
fa743953 | 2610 | rb_end_commit(cpu_buffer); |
7a8e76a3 SR |
2611 | } |
2612 | ||
2613 | /** | |
2614 | * ring_buffer_unlock_commit - commit a reserved | |
2615 | * @buffer: The buffer to commit to | |
2616 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
2617 | * |
2618 | * This commits the data to the ring buffer, and releases any locks held. | |
2619 | * | |
2620 | * Must be paired with ring_buffer_lock_reserve. | |
2621 | */ | |
2622 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 2623 | struct ring_buffer_event *event) |
7a8e76a3 SR |
2624 | { |
2625 | struct ring_buffer_per_cpu *cpu_buffer; | |
2626 | int cpu = raw_smp_processor_id(); | |
2627 | ||
2628 | cpu_buffer = buffer->buffers[cpu]; | |
2629 | ||
7a8e76a3 SR |
2630 | rb_commit(cpu_buffer, event); |
2631 | ||
261842b7 SR |
2632 | trace_recursive_unlock(); |
2633 | ||
5168ae50 | 2634 | preempt_enable_notrace(); |
7a8e76a3 SR |
2635 | |
2636 | return 0; | |
2637 | } | |
c4f50183 | 2638 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 2639 | |
f3b9aae1 FW |
2640 | static inline void rb_event_discard(struct ring_buffer_event *event) |
2641 | { | |
69d1b839 SR |
2642 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) |
2643 | event = skip_time_extend(event); | |
2644 | ||
334d4169 LJ |
2645 | /* array[0] holds the actual length for the discarded event */ |
2646 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
2647 | event->type_len = RINGBUF_TYPE_PADDING; | |
f3b9aae1 FW |
2648 | /* time delta must be non zero */ |
2649 | if (!event->time_delta) | |
2650 | event->time_delta = 1; | |
2651 | } | |
2652 | ||
a1863c21 SR |
2653 | /* |
2654 | * Decrement the entries to the page that an event is on. | |
2655 | * The event does not even need to exist, only the pointer | |
2656 | * to the page it is on. This may only be called before the commit | |
2657 | * takes place. | |
2658 | */ | |
2659 | static inline void | |
2660 | rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer, | |
2661 | struct ring_buffer_event *event) | |
2662 | { | |
2663 | unsigned long addr = (unsigned long)event; | |
2664 | struct buffer_page *bpage = cpu_buffer->commit_page; | |
2665 | struct buffer_page *start; | |
2666 | ||
2667 | addr &= PAGE_MASK; | |
2668 | ||
2669 | /* Do the likely case first */ | |
2670 | if (likely(bpage->page == (void *)addr)) { | |
2671 | local_dec(&bpage->entries); | |
2672 | return; | |
2673 | } | |
2674 | ||
2675 | /* | |
2676 | * Because the commit page may be on the reader page we | |
2677 | * start with the next page and check the end loop there. | |
2678 | */ | |
2679 | rb_inc_page(cpu_buffer, &bpage); | |
2680 | start = bpage; | |
2681 | do { | |
2682 | if (bpage->page == (void *)addr) { | |
2683 | local_dec(&bpage->entries); | |
2684 | return; | |
2685 | } | |
2686 | rb_inc_page(cpu_buffer, &bpage); | |
2687 | } while (bpage != start); | |
2688 | ||
2689 | /* commit not part of this buffer?? */ | |
2690 | RB_WARN_ON(cpu_buffer, 1); | |
2691 | } | |
2692 | ||
fa1b47dd SR |
2693 | /** |
2694 | * ring_buffer_commit_discard - discard an event that has not been committed | |
2695 | * @buffer: the ring buffer | |
2696 | * @event: non committed event to discard | |
2697 | * | |
dc892f73 SR |
2698 | * Sometimes an event that is in the ring buffer needs to be ignored. |
2699 | * This function lets the user discard an event in the ring buffer | |
2700 | * and then that event will not be read later. | |
2701 | * | |
2702 | * This function only works if it is called before the the item has been | |
2703 | * committed. It will try to free the event from the ring buffer | |
fa1b47dd SR |
2704 | * if another event has not been added behind it. |
2705 | * | |
2706 | * If another event has been added behind it, it will set the event | |
2707 | * up as discarded, and perform the commit. | |
2708 | * | |
2709 | * If this function is called, do not call ring_buffer_unlock_commit on | |
2710 | * the event. | |
2711 | */ | |
2712 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
2713 | struct ring_buffer_event *event) | |
2714 | { | |
2715 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
2716 | int cpu; |
2717 | ||
2718 | /* The event is discarded regardless */ | |
f3b9aae1 | 2719 | rb_event_discard(event); |
fa1b47dd | 2720 | |
fa743953 SR |
2721 | cpu = smp_processor_id(); |
2722 | cpu_buffer = buffer->buffers[cpu]; | |
2723 | ||
fa1b47dd SR |
2724 | /* |
2725 | * This must only be called if the event has not been | |
2726 | * committed yet. Thus we can assume that preemption | |
2727 | * is still disabled. | |
2728 | */ | |
fa743953 | 2729 | RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing)); |
fa1b47dd | 2730 | |
a1863c21 | 2731 | rb_decrement_entry(cpu_buffer, event); |
0f2541d2 | 2732 | if (rb_try_to_discard(cpu_buffer, event)) |
edd813bf | 2733 | goto out; |
fa1b47dd SR |
2734 | |
2735 | /* | |
2736 | * The commit is still visible by the reader, so we | |
a1863c21 | 2737 | * must still update the timestamp. |
fa1b47dd | 2738 | */ |
a1863c21 | 2739 | rb_update_write_stamp(cpu_buffer, event); |
fa1b47dd | 2740 | out: |
fa743953 | 2741 | rb_end_commit(cpu_buffer); |
fa1b47dd | 2742 | |
f3b9aae1 FW |
2743 | trace_recursive_unlock(); |
2744 | ||
5168ae50 | 2745 | preempt_enable_notrace(); |
fa1b47dd SR |
2746 | |
2747 | } | |
2748 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
2749 | ||
7a8e76a3 SR |
2750 | /** |
2751 | * ring_buffer_write - write data to the buffer without reserving | |
2752 | * @buffer: The ring buffer to write to. | |
2753 | * @length: The length of the data being written (excluding the event header) | |
2754 | * @data: The data to write to the buffer. | |
2755 | * | |
2756 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
2757 | * one function. If you already have the data to write to the buffer, it | |
2758 | * may be easier to simply call this function. | |
2759 | * | |
2760 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
2761 | * and not the length of the event which would hold the header. | |
2762 | */ | |
2763 | int ring_buffer_write(struct ring_buffer *buffer, | |
01e3e710 DS |
2764 | unsigned long length, |
2765 | void *data) | |
7a8e76a3 SR |
2766 | { |
2767 | struct ring_buffer_per_cpu *cpu_buffer; | |
2768 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
2769 | void *body; |
2770 | int ret = -EBUSY; | |
5168ae50 | 2771 | int cpu; |
7a8e76a3 | 2772 | |
033601a3 | 2773 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2774 | return -EBUSY; |
2775 | ||
5168ae50 | 2776 | preempt_disable_notrace(); |
bf41a158 | 2777 | |
52fbe9cd LJ |
2778 | if (atomic_read(&buffer->record_disabled)) |
2779 | goto out; | |
2780 | ||
7a8e76a3 SR |
2781 | cpu = raw_smp_processor_id(); |
2782 | ||
9e01c1b7 | 2783 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2784 | goto out; |
7a8e76a3 SR |
2785 | |
2786 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2787 | |
2788 | if (atomic_read(&cpu_buffer->record_disabled)) | |
2789 | goto out; | |
2790 | ||
be957c44 SR |
2791 | if (length > BUF_MAX_DATA_SIZE) |
2792 | goto out; | |
2793 | ||
62f0b3eb | 2794 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 SR |
2795 | if (!event) |
2796 | goto out; | |
2797 | ||
2798 | body = rb_event_data(event); | |
2799 | ||
2800 | memcpy(body, data, length); | |
2801 | ||
2802 | rb_commit(cpu_buffer, event); | |
2803 | ||
2804 | ret = 0; | |
2805 | out: | |
5168ae50 | 2806 | preempt_enable_notrace(); |
7a8e76a3 SR |
2807 | |
2808 | return ret; | |
2809 | } | |
c4f50183 | 2810 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 2811 | |
34a148bf | 2812 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
2813 | { |
2814 | struct buffer_page *reader = cpu_buffer->reader_page; | |
77ae365e | 2815 | struct buffer_page *head = rb_set_head_page(cpu_buffer); |
bf41a158 SR |
2816 | struct buffer_page *commit = cpu_buffer->commit_page; |
2817 | ||
77ae365e SR |
2818 | /* In case of error, head will be NULL */ |
2819 | if (unlikely(!head)) | |
2820 | return 1; | |
2821 | ||
bf41a158 SR |
2822 | return reader->read == rb_page_commit(reader) && |
2823 | (commit == reader || | |
2824 | (commit == head && | |
2825 | head->read == rb_page_commit(commit))); | |
2826 | } | |
2827 | ||
7a8e76a3 SR |
2828 | /** |
2829 | * ring_buffer_record_disable - stop all writes into the buffer | |
2830 | * @buffer: The ring buffer to stop writes to. | |
2831 | * | |
2832 | * This prevents all writes to the buffer. Any attempt to write | |
2833 | * to the buffer after this will fail and return NULL. | |
2834 | * | |
2835 | * The caller should call synchronize_sched() after this. | |
2836 | */ | |
2837 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
2838 | { | |
2839 | atomic_inc(&buffer->record_disabled); | |
2840 | } | |
c4f50183 | 2841 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
2842 | |
2843 | /** | |
2844 | * ring_buffer_record_enable - enable writes to the buffer | |
2845 | * @buffer: The ring buffer to enable writes | |
2846 | * | |
2847 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 2848 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
2849 | */ |
2850 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
2851 | { | |
2852 | atomic_dec(&buffer->record_disabled); | |
2853 | } | |
c4f50183 | 2854 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 | 2855 | |
499e5470 SR |
2856 | /** |
2857 | * ring_buffer_record_off - stop all writes into the buffer | |
2858 | * @buffer: The ring buffer to stop writes to. | |
2859 | * | |
2860 | * This prevents all writes to the buffer. Any attempt to write | |
2861 | * to the buffer after this will fail and return NULL. | |
2862 | * | |
2863 | * This is different than ring_buffer_record_disable() as | |
87abb3b1 | 2864 | * it works like an on/off switch, where as the disable() version |
499e5470 SR |
2865 | * must be paired with a enable(). |
2866 | */ | |
2867 | void ring_buffer_record_off(struct ring_buffer *buffer) | |
2868 | { | |
2869 | unsigned int rd; | |
2870 | unsigned int new_rd; | |
2871 | ||
2872 | do { | |
2873 | rd = atomic_read(&buffer->record_disabled); | |
2874 | new_rd = rd | RB_BUFFER_OFF; | |
2875 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
2876 | } | |
2877 | EXPORT_SYMBOL_GPL(ring_buffer_record_off); | |
2878 | ||
2879 | /** | |
2880 | * ring_buffer_record_on - restart writes into the buffer | |
2881 | * @buffer: The ring buffer to start writes to. | |
2882 | * | |
2883 | * This enables all writes to the buffer that was disabled by | |
2884 | * ring_buffer_record_off(). | |
2885 | * | |
2886 | * This is different than ring_buffer_record_enable() as | |
87abb3b1 | 2887 | * it works like an on/off switch, where as the enable() version |
499e5470 SR |
2888 | * must be paired with a disable(). |
2889 | */ | |
2890 | void ring_buffer_record_on(struct ring_buffer *buffer) | |
2891 | { | |
2892 | unsigned int rd; | |
2893 | unsigned int new_rd; | |
2894 | ||
2895 | do { | |
2896 | rd = atomic_read(&buffer->record_disabled); | |
2897 | new_rd = rd & ~RB_BUFFER_OFF; | |
2898 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
2899 | } | |
2900 | EXPORT_SYMBOL_GPL(ring_buffer_record_on); | |
2901 | ||
2902 | /** | |
2903 | * ring_buffer_record_is_on - return true if the ring buffer can write | |
2904 | * @buffer: The ring buffer to see if write is enabled | |
2905 | * | |
2906 | * Returns true if the ring buffer is in a state that it accepts writes. | |
2907 | */ | |
2908 | int ring_buffer_record_is_on(struct ring_buffer *buffer) | |
2909 | { | |
2910 | return !atomic_read(&buffer->record_disabled); | |
2911 | } | |
2912 | ||
7a8e76a3 SR |
2913 | /** |
2914 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
2915 | * @buffer: The ring buffer to stop writes to. | |
2916 | * @cpu: The CPU buffer to stop | |
2917 | * | |
2918 | * This prevents all writes to the buffer. Any attempt to write | |
2919 | * to the buffer after this will fail and return NULL. | |
2920 | * | |
2921 | * The caller should call synchronize_sched() after this. | |
2922 | */ | |
2923 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
2924 | { | |
2925 | struct ring_buffer_per_cpu *cpu_buffer; | |
2926 | ||
9e01c1b7 | 2927 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2928 | return; |
7a8e76a3 SR |
2929 | |
2930 | cpu_buffer = buffer->buffers[cpu]; | |
2931 | atomic_inc(&cpu_buffer->record_disabled); | |
2932 | } | |
c4f50183 | 2933 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
2934 | |
2935 | /** | |
2936 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
2937 | * @buffer: The ring buffer to enable writes | |
2938 | * @cpu: The CPU to enable. | |
2939 | * | |
2940 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 2941 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
2942 | */ |
2943 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
2944 | { | |
2945 | struct ring_buffer_per_cpu *cpu_buffer; | |
2946 | ||
9e01c1b7 | 2947 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 2948 | return; |
7a8e76a3 SR |
2949 | |
2950 | cpu_buffer = buffer->buffers[cpu]; | |
2951 | atomic_dec(&cpu_buffer->record_disabled); | |
2952 | } | |
c4f50183 | 2953 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 | 2954 | |
f6195aa0 SR |
2955 | /* |
2956 | * The total entries in the ring buffer is the running counter | |
2957 | * of entries entered into the ring buffer, minus the sum of | |
2958 | * the entries read from the ring buffer and the number of | |
2959 | * entries that were overwritten. | |
2960 | */ | |
2961 | static inline unsigned long | |
2962 | rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer) | |
2963 | { | |
2964 | return local_read(&cpu_buffer->entries) - | |
2965 | (local_read(&cpu_buffer->overrun) + cpu_buffer->read); | |
2966 | } | |
2967 | ||
c64e148a VN |
2968 | /** |
2969 | * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer | |
2970 | * @buffer: The ring buffer | |
2971 | * @cpu: The per CPU buffer to read from. | |
2972 | */ | |
50ecf2c3 | 2973 | u64 ring_buffer_oldest_event_ts(struct ring_buffer *buffer, int cpu) |
c64e148a VN |
2974 | { |
2975 | unsigned long flags; | |
2976 | struct ring_buffer_per_cpu *cpu_buffer; | |
2977 | struct buffer_page *bpage; | |
da830e58 | 2978 | u64 ret = 0; |
c64e148a VN |
2979 | |
2980 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
2981 | return 0; | |
2982 | ||
2983 | cpu_buffer = buffer->buffers[cpu]; | |
7115e3fc | 2984 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
2985 | /* |
2986 | * if the tail is on reader_page, oldest time stamp is on the reader | |
2987 | * page | |
2988 | */ | |
2989 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
2990 | bpage = cpu_buffer->reader_page; | |
2991 | else | |
2992 | bpage = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
2993 | if (bpage) |
2994 | ret = bpage->page->time_stamp; | |
7115e3fc | 2995 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
2996 | |
2997 | return ret; | |
2998 | } | |
2999 | EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts); | |
3000 | ||
3001 | /** | |
3002 | * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer | |
3003 | * @buffer: The ring buffer | |
3004 | * @cpu: The per CPU buffer to read from. | |
3005 | */ | |
3006 | unsigned long ring_buffer_bytes_cpu(struct ring_buffer *buffer, int cpu) | |
3007 | { | |
3008 | struct ring_buffer_per_cpu *cpu_buffer; | |
3009 | unsigned long ret; | |
3010 | ||
3011 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3012 | return 0; | |
3013 | ||
3014 | cpu_buffer = buffer->buffers[cpu]; | |
3015 | ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes; | |
3016 | ||
3017 | return ret; | |
3018 | } | |
3019 | EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu); | |
3020 | ||
7a8e76a3 SR |
3021 | /** |
3022 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
3023 | * @buffer: The ring buffer | |
3024 | * @cpu: The per CPU buffer to get the entries from. | |
3025 | */ | |
3026 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
3027 | { | |
3028 | struct ring_buffer_per_cpu *cpu_buffer; | |
3029 | ||
9e01c1b7 | 3030 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3031 | return 0; |
7a8e76a3 SR |
3032 | |
3033 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 3034 | |
f6195aa0 | 3035 | return rb_num_of_entries(cpu_buffer); |
7a8e76a3 | 3036 | } |
c4f50183 | 3037 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
3038 | |
3039 | /** | |
884bfe89 SP |
3040 | * ring_buffer_overrun_cpu - get the number of overruns caused by the ring |
3041 | * buffer wrapping around (only if RB_FL_OVERWRITE is on). | |
7a8e76a3 SR |
3042 | * @buffer: The ring buffer |
3043 | * @cpu: The per CPU buffer to get the number of overruns from | |
3044 | */ | |
3045 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3046 | { | |
3047 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3048 | unsigned long ret; |
7a8e76a3 | 3049 | |
9e01c1b7 | 3050 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3051 | return 0; |
7a8e76a3 SR |
3052 | |
3053 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3054 | ret = local_read(&cpu_buffer->overrun); |
554f786e SR |
3055 | |
3056 | return ret; | |
7a8e76a3 | 3057 | } |
c4f50183 | 3058 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 3059 | |
f0d2c681 | 3060 | /** |
884bfe89 SP |
3061 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by |
3062 | * commits failing due to the buffer wrapping around while there are uncommitted | |
3063 | * events, such as during an interrupt storm. | |
f0d2c681 SR |
3064 | * @buffer: The ring buffer |
3065 | * @cpu: The per CPU buffer to get the number of overruns from | |
3066 | */ | |
3067 | unsigned long | |
3068 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3069 | { | |
3070 | struct ring_buffer_per_cpu *cpu_buffer; | |
3071 | unsigned long ret; | |
3072 | ||
3073 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3074 | return 0; | |
3075 | ||
3076 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3077 | ret = local_read(&cpu_buffer->commit_overrun); |
f0d2c681 SR |
3078 | |
3079 | return ret; | |
3080 | } | |
3081 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
3082 | ||
884bfe89 SP |
3083 | /** |
3084 | * ring_buffer_dropped_events_cpu - get the number of dropped events caused by | |
3085 | * the ring buffer filling up (only if RB_FL_OVERWRITE is off). | |
3086 | * @buffer: The ring buffer | |
3087 | * @cpu: The per CPU buffer to get the number of overruns from | |
3088 | */ | |
3089 | unsigned long | |
3090 | ring_buffer_dropped_events_cpu(struct ring_buffer *buffer, int cpu) | |
3091 | { | |
3092 | struct ring_buffer_per_cpu *cpu_buffer; | |
3093 | unsigned long ret; | |
3094 | ||
3095 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3096 | return 0; | |
3097 | ||
3098 | cpu_buffer = buffer->buffers[cpu]; | |
3099 | ret = local_read(&cpu_buffer->dropped_events); | |
3100 | ||
3101 | return ret; | |
3102 | } | |
3103 | EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu); | |
3104 | ||
ad964704 SRRH |
3105 | /** |
3106 | * ring_buffer_read_events_cpu - get the number of events successfully read | |
3107 | * @buffer: The ring buffer | |
3108 | * @cpu: The per CPU buffer to get the number of events read | |
3109 | */ | |
3110 | unsigned long | |
3111 | ring_buffer_read_events_cpu(struct ring_buffer *buffer, int cpu) | |
3112 | { | |
3113 | struct ring_buffer_per_cpu *cpu_buffer; | |
3114 | ||
3115 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3116 | return 0; | |
3117 | ||
3118 | cpu_buffer = buffer->buffers[cpu]; | |
3119 | return cpu_buffer->read; | |
3120 | } | |
3121 | EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu); | |
3122 | ||
7a8e76a3 SR |
3123 | /** |
3124 | * ring_buffer_entries - get the number of entries in a buffer | |
3125 | * @buffer: The ring buffer | |
3126 | * | |
3127 | * Returns the total number of entries in the ring buffer | |
3128 | * (all CPU entries) | |
3129 | */ | |
3130 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
3131 | { | |
3132 | struct ring_buffer_per_cpu *cpu_buffer; | |
3133 | unsigned long entries = 0; | |
3134 | int cpu; | |
3135 | ||
3136 | /* if you care about this being correct, lock the buffer */ | |
3137 | for_each_buffer_cpu(buffer, cpu) { | |
3138 | cpu_buffer = buffer->buffers[cpu]; | |
f6195aa0 | 3139 | entries += rb_num_of_entries(cpu_buffer); |
7a8e76a3 SR |
3140 | } |
3141 | ||
3142 | return entries; | |
3143 | } | |
c4f50183 | 3144 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
3145 | |
3146 | /** | |
67b394f7 | 3147 | * ring_buffer_overruns - get the number of overruns in buffer |
7a8e76a3 SR |
3148 | * @buffer: The ring buffer |
3149 | * | |
3150 | * Returns the total number of overruns in the ring buffer | |
3151 | * (all CPU entries) | |
3152 | */ | |
3153 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
3154 | { | |
3155 | struct ring_buffer_per_cpu *cpu_buffer; | |
3156 | unsigned long overruns = 0; | |
3157 | int cpu; | |
3158 | ||
3159 | /* if you care about this being correct, lock the buffer */ | |
3160 | for_each_buffer_cpu(buffer, cpu) { | |
3161 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3162 | overruns += local_read(&cpu_buffer->overrun); |
7a8e76a3 SR |
3163 | } |
3164 | ||
3165 | return overruns; | |
3166 | } | |
c4f50183 | 3167 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 3168 | |
642edba5 | 3169 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
3170 | { |
3171 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3172 | ||
d769041f SR |
3173 | /* Iterator usage is expected to have record disabled */ |
3174 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
77ae365e SR |
3175 | iter->head_page = rb_set_head_page(cpu_buffer); |
3176 | if (unlikely(!iter->head_page)) | |
3177 | return; | |
3178 | iter->head = iter->head_page->read; | |
d769041f SR |
3179 | } else { |
3180 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 3181 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
3182 | } |
3183 | if (iter->head) | |
3184 | iter->read_stamp = cpu_buffer->read_stamp; | |
3185 | else | |
abc9b56d | 3186 | iter->read_stamp = iter->head_page->page->time_stamp; |
492a74f4 SR |
3187 | iter->cache_reader_page = cpu_buffer->reader_page; |
3188 | iter->cache_read = cpu_buffer->read; | |
642edba5 | 3189 | } |
f83c9d0f | 3190 | |
642edba5 SR |
3191 | /** |
3192 | * ring_buffer_iter_reset - reset an iterator | |
3193 | * @iter: The iterator to reset | |
3194 | * | |
3195 | * Resets the iterator, so that it will start from the beginning | |
3196 | * again. | |
3197 | */ | |
3198 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
3199 | { | |
554f786e | 3200 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
3201 | unsigned long flags; |
3202 | ||
554f786e SR |
3203 | if (!iter) |
3204 | return; | |
3205 | ||
3206 | cpu_buffer = iter->cpu_buffer; | |
3207 | ||
5389f6fa | 3208 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
642edba5 | 3209 | rb_iter_reset(iter); |
5389f6fa | 3210 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 3211 | } |
c4f50183 | 3212 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
3213 | |
3214 | /** | |
3215 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
3216 | * @iter: The iterator to check | |
3217 | */ | |
3218 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
3219 | { | |
3220 | struct ring_buffer_per_cpu *cpu_buffer; | |
3221 | ||
3222 | cpu_buffer = iter->cpu_buffer; | |
3223 | ||
bf41a158 SR |
3224 | return iter->head_page == cpu_buffer->commit_page && |
3225 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 3226 | } |
c4f50183 | 3227 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
3228 | |
3229 | static void | |
3230 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
3231 | struct ring_buffer_event *event) | |
3232 | { | |
3233 | u64 delta; | |
3234 | ||
334d4169 | 3235 | switch (event->type_len) { |
7a8e76a3 SR |
3236 | case RINGBUF_TYPE_PADDING: |
3237 | return; | |
3238 | ||
3239 | case RINGBUF_TYPE_TIME_EXTEND: | |
3240 | delta = event->array[0]; | |
3241 | delta <<= TS_SHIFT; | |
3242 | delta += event->time_delta; | |
3243 | cpu_buffer->read_stamp += delta; | |
3244 | return; | |
3245 | ||
3246 | case RINGBUF_TYPE_TIME_STAMP: | |
3247 | /* FIXME: not implemented */ | |
3248 | return; | |
3249 | ||
3250 | case RINGBUF_TYPE_DATA: | |
3251 | cpu_buffer->read_stamp += event->time_delta; | |
3252 | return; | |
3253 | ||
3254 | default: | |
3255 | BUG(); | |
3256 | } | |
3257 | return; | |
3258 | } | |
3259 | ||
3260 | static void | |
3261 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
3262 | struct ring_buffer_event *event) | |
3263 | { | |
3264 | u64 delta; | |
3265 | ||
334d4169 | 3266 | switch (event->type_len) { |
7a8e76a3 SR |
3267 | case RINGBUF_TYPE_PADDING: |
3268 | return; | |
3269 | ||
3270 | case RINGBUF_TYPE_TIME_EXTEND: | |
3271 | delta = event->array[0]; | |
3272 | delta <<= TS_SHIFT; | |
3273 | delta += event->time_delta; | |
3274 | iter->read_stamp += delta; | |
3275 | return; | |
3276 | ||
3277 | case RINGBUF_TYPE_TIME_STAMP: | |
3278 | /* FIXME: not implemented */ | |
3279 | return; | |
3280 | ||
3281 | case RINGBUF_TYPE_DATA: | |
3282 | iter->read_stamp += event->time_delta; | |
3283 | return; | |
3284 | ||
3285 | default: | |
3286 | BUG(); | |
3287 | } | |
3288 | return; | |
3289 | } | |
3290 | ||
d769041f SR |
3291 | static struct buffer_page * |
3292 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 3293 | { |
d769041f | 3294 | struct buffer_page *reader = NULL; |
66a8cb95 | 3295 | unsigned long overwrite; |
d769041f | 3296 | unsigned long flags; |
818e3dd3 | 3297 | int nr_loops = 0; |
77ae365e | 3298 | int ret; |
d769041f | 3299 | |
3e03fb7f | 3300 | local_irq_save(flags); |
0199c4e6 | 3301 | arch_spin_lock(&cpu_buffer->lock); |
d769041f SR |
3302 | |
3303 | again: | |
818e3dd3 SR |
3304 | /* |
3305 | * This should normally only loop twice. But because the | |
3306 | * start of the reader inserts an empty page, it causes | |
3307 | * a case where we will loop three times. There should be no | |
3308 | * reason to loop four times (that I know of). | |
3309 | */ | |
3e89c7bb | 3310 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
3311 | reader = NULL; |
3312 | goto out; | |
3313 | } | |
3314 | ||
d769041f SR |
3315 | reader = cpu_buffer->reader_page; |
3316 | ||
3317 | /* If there's more to read, return this page */ | |
bf41a158 | 3318 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
3319 | goto out; |
3320 | ||
3321 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
3322 | if (RB_WARN_ON(cpu_buffer, |
3323 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
3324 | goto out; | |
d769041f SR |
3325 | |
3326 | /* check if we caught up to the tail */ | |
3327 | reader = NULL; | |
bf41a158 | 3328 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 3329 | goto out; |
7a8e76a3 | 3330 | |
a5fb8331 SR |
3331 | /* Don't bother swapping if the ring buffer is empty */ |
3332 | if (rb_num_of_entries(cpu_buffer) == 0) | |
3333 | goto out; | |
3334 | ||
7a8e76a3 | 3335 | /* |
d769041f | 3336 | * Reset the reader page to size zero. |
7a8e76a3 | 3337 | */ |
77ae365e SR |
3338 | local_set(&cpu_buffer->reader_page->write, 0); |
3339 | local_set(&cpu_buffer->reader_page->entries, 0); | |
3340 | local_set(&cpu_buffer->reader_page->page->commit, 0); | |
ff0ff84a | 3341 | cpu_buffer->reader_page->real_end = 0; |
7a8e76a3 | 3342 | |
77ae365e SR |
3343 | spin: |
3344 | /* | |
3345 | * Splice the empty reader page into the list around the head. | |
3346 | */ | |
3347 | reader = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
3348 | if (!reader) |
3349 | goto out; | |
0e1ff5d7 | 3350 | cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next); |
d769041f | 3351 | cpu_buffer->reader_page->list.prev = reader->list.prev; |
bf41a158 | 3352 | |
3adc54fa SR |
3353 | /* |
3354 | * cpu_buffer->pages just needs to point to the buffer, it | |
3355 | * has no specific buffer page to point to. Lets move it out | |
25985edc | 3356 | * of our way so we don't accidentally swap it. |
3adc54fa SR |
3357 | */ |
3358 | cpu_buffer->pages = reader->list.prev; | |
3359 | ||
77ae365e SR |
3360 | /* The reader page will be pointing to the new head */ |
3361 | rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list); | |
7a8e76a3 | 3362 | |
66a8cb95 SR |
3363 | /* |
3364 | * We want to make sure we read the overruns after we set up our | |
3365 | * pointers to the next object. The writer side does a | |
3366 | * cmpxchg to cross pages which acts as the mb on the writer | |
3367 | * side. Note, the reader will constantly fail the swap | |
3368 | * while the writer is updating the pointers, so this | |
3369 | * guarantees that the overwrite recorded here is the one we | |
3370 | * want to compare with the last_overrun. | |
3371 | */ | |
3372 | smp_mb(); | |
3373 | overwrite = local_read(&(cpu_buffer->overrun)); | |
3374 | ||
77ae365e SR |
3375 | /* |
3376 | * Here's the tricky part. | |
3377 | * | |
3378 | * We need to move the pointer past the header page. | |
3379 | * But we can only do that if a writer is not currently | |
3380 | * moving it. The page before the header page has the | |
3381 | * flag bit '1' set if it is pointing to the page we want. | |
3382 | * but if the writer is in the process of moving it | |
3383 | * than it will be '2' or already moved '0'. | |
3384 | */ | |
3385 | ||
3386 | ret = rb_head_page_replace(reader, cpu_buffer->reader_page); | |
7a8e76a3 SR |
3387 | |
3388 | /* | |
77ae365e | 3389 | * If we did not convert it, then we must try again. |
7a8e76a3 | 3390 | */ |
77ae365e SR |
3391 | if (!ret) |
3392 | goto spin; | |
7a8e76a3 | 3393 | |
77ae365e SR |
3394 | /* |
3395 | * Yeah! We succeeded in replacing the page. | |
3396 | * | |
3397 | * Now make the new head point back to the reader page. | |
3398 | */ | |
5ded3dc6 | 3399 | rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list; |
77ae365e | 3400 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
d769041f SR |
3401 | |
3402 | /* Finally update the reader page to the new head */ | |
3403 | cpu_buffer->reader_page = reader; | |
3404 | rb_reset_reader_page(cpu_buffer); | |
3405 | ||
66a8cb95 SR |
3406 | if (overwrite != cpu_buffer->last_overrun) { |
3407 | cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun; | |
3408 | cpu_buffer->last_overrun = overwrite; | |
3409 | } | |
3410 | ||
d769041f SR |
3411 | goto again; |
3412 | ||
3413 | out: | |
0199c4e6 | 3414 | arch_spin_unlock(&cpu_buffer->lock); |
3e03fb7f | 3415 | local_irq_restore(flags); |
d769041f SR |
3416 | |
3417 | return reader; | |
3418 | } | |
3419 | ||
3420 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
3421 | { | |
3422 | struct ring_buffer_event *event; | |
3423 | struct buffer_page *reader; | |
3424 | unsigned length; | |
3425 | ||
3426 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 3427 | |
d769041f | 3428 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
3429 | if (RB_WARN_ON(cpu_buffer, !reader)) |
3430 | return; | |
7a8e76a3 | 3431 | |
d769041f SR |
3432 | event = rb_reader_event(cpu_buffer); |
3433 | ||
a1863c21 | 3434 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
e4906eff | 3435 | cpu_buffer->read++; |
d769041f SR |
3436 | |
3437 | rb_update_read_stamp(cpu_buffer, event); | |
3438 | ||
3439 | length = rb_event_length(event); | |
6f807acd | 3440 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
3441 | } |
3442 | ||
3443 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
3444 | { | |
7a8e76a3 SR |
3445 | struct ring_buffer_per_cpu *cpu_buffer; |
3446 | struct ring_buffer_event *event; | |
3447 | unsigned length; | |
3448 | ||
3449 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 SR |
3450 | |
3451 | /* | |
3452 | * Check if we are at the end of the buffer. | |
3453 | */ | |
bf41a158 | 3454 | if (iter->head >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
3455 | /* discarded commits can make the page empty */ |
3456 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 3457 | return; |
d769041f | 3458 | rb_inc_iter(iter); |
7a8e76a3 SR |
3459 | return; |
3460 | } | |
3461 | ||
3462 | event = rb_iter_head_event(iter); | |
3463 | ||
3464 | length = rb_event_length(event); | |
3465 | ||
3466 | /* | |
3467 | * This should not be called to advance the header if we are | |
3468 | * at the tail of the buffer. | |
3469 | */ | |
3e89c7bb | 3470 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 3471 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
3472 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
3473 | return; | |
7a8e76a3 SR |
3474 | |
3475 | rb_update_iter_read_stamp(iter, event); | |
3476 | ||
3477 | iter->head += length; | |
3478 | ||
3479 | /* check for end of page padding */ | |
bf41a158 SR |
3480 | if ((iter->head >= rb_page_size(iter->head_page)) && |
3481 | (iter->head_page != cpu_buffer->commit_page)) | |
771e0384 | 3482 | rb_inc_iter(iter); |
7a8e76a3 SR |
3483 | } |
3484 | ||
66a8cb95 SR |
3485 | static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer) |
3486 | { | |
3487 | return cpu_buffer->lost_events; | |
3488 | } | |
3489 | ||
f83c9d0f | 3490 | static struct ring_buffer_event * |
66a8cb95 SR |
3491 | rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts, |
3492 | unsigned long *lost_events) | |
7a8e76a3 | 3493 | { |
7a8e76a3 | 3494 | struct ring_buffer_event *event; |
d769041f | 3495 | struct buffer_page *reader; |
818e3dd3 | 3496 | int nr_loops = 0; |
7a8e76a3 | 3497 | |
7a8e76a3 | 3498 | again: |
818e3dd3 | 3499 | /* |
69d1b839 SR |
3500 | * We repeat when a time extend is encountered. |
3501 | * Since the time extend is always attached to a data event, | |
3502 | * we should never loop more than once. | |
3503 | * (We never hit the following condition more than twice). | |
818e3dd3 | 3504 | */ |
69d1b839 | 3505 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 3506 | return NULL; |
818e3dd3 | 3507 | |
d769041f SR |
3508 | reader = rb_get_reader_page(cpu_buffer); |
3509 | if (!reader) | |
7a8e76a3 SR |
3510 | return NULL; |
3511 | ||
d769041f | 3512 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 3513 | |
334d4169 | 3514 | switch (event->type_len) { |
7a8e76a3 | 3515 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3516 | if (rb_null_event(event)) |
3517 | RB_WARN_ON(cpu_buffer, 1); | |
3518 | /* | |
3519 | * Because the writer could be discarding every | |
3520 | * event it creates (which would probably be bad) | |
3521 | * if we were to go back to "again" then we may never | |
3522 | * catch up, and will trigger the warn on, or lock | |
3523 | * the box. Return the padding, and we will release | |
3524 | * the current locks, and try again. | |
3525 | */ | |
2d622719 | 3526 | return event; |
7a8e76a3 SR |
3527 | |
3528 | case RINGBUF_TYPE_TIME_EXTEND: | |
3529 | /* Internal data, OK to advance */ | |
d769041f | 3530 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3531 | goto again; |
3532 | ||
3533 | case RINGBUF_TYPE_TIME_STAMP: | |
3534 | /* FIXME: not implemented */ | |
d769041f | 3535 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3536 | goto again; |
3537 | ||
3538 | case RINGBUF_TYPE_DATA: | |
3539 | if (ts) { | |
3540 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
d8eeb2d3 | 3541 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, |
37886f6a | 3542 | cpu_buffer->cpu, ts); |
7a8e76a3 | 3543 | } |
66a8cb95 SR |
3544 | if (lost_events) |
3545 | *lost_events = rb_lost_events(cpu_buffer); | |
7a8e76a3 SR |
3546 | return event; |
3547 | ||
3548 | default: | |
3549 | BUG(); | |
3550 | } | |
3551 | ||
3552 | return NULL; | |
3553 | } | |
c4f50183 | 3554 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 3555 | |
f83c9d0f SR |
3556 | static struct ring_buffer_event * |
3557 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
3558 | { |
3559 | struct ring_buffer *buffer; | |
3560 | struct ring_buffer_per_cpu *cpu_buffer; | |
3561 | struct ring_buffer_event *event; | |
818e3dd3 | 3562 | int nr_loops = 0; |
7a8e76a3 | 3563 | |
7a8e76a3 SR |
3564 | cpu_buffer = iter->cpu_buffer; |
3565 | buffer = cpu_buffer->buffer; | |
3566 | ||
492a74f4 SR |
3567 | /* |
3568 | * Check if someone performed a consuming read to | |
3569 | * the buffer. A consuming read invalidates the iterator | |
3570 | * and we need to reset the iterator in this case. | |
3571 | */ | |
3572 | if (unlikely(iter->cache_read != cpu_buffer->read || | |
3573 | iter->cache_reader_page != cpu_buffer->reader_page)) | |
3574 | rb_iter_reset(iter); | |
3575 | ||
7a8e76a3 | 3576 | again: |
3c05d748 SR |
3577 | if (ring_buffer_iter_empty(iter)) |
3578 | return NULL; | |
3579 | ||
818e3dd3 | 3580 | /* |
69d1b839 SR |
3581 | * We repeat when a time extend is encountered. |
3582 | * Since the time extend is always attached to a data event, | |
3583 | * we should never loop more than once. | |
3584 | * (We never hit the following condition more than twice). | |
818e3dd3 | 3585 | */ |
69d1b839 | 3586 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 3587 | return NULL; |
818e3dd3 | 3588 | |
7a8e76a3 SR |
3589 | if (rb_per_cpu_empty(cpu_buffer)) |
3590 | return NULL; | |
3591 | ||
3c05d748 SR |
3592 | if (iter->head >= local_read(&iter->head_page->page->commit)) { |
3593 | rb_inc_iter(iter); | |
3594 | goto again; | |
3595 | } | |
3596 | ||
7a8e76a3 SR |
3597 | event = rb_iter_head_event(iter); |
3598 | ||
334d4169 | 3599 | switch (event->type_len) { |
7a8e76a3 | 3600 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3601 | if (rb_null_event(event)) { |
3602 | rb_inc_iter(iter); | |
3603 | goto again; | |
3604 | } | |
3605 | rb_advance_iter(iter); | |
3606 | return event; | |
7a8e76a3 SR |
3607 | |
3608 | case RINGBUF_TYPE_TIME_EXTEND: | |
3609 | /* Internal data, OK to advance */ | |
3610 | rb_advance_iter(iter); | |
3611 | goto again; | |
3612 | ||
3613 | case RINGBUF_TYPE_TIME_STAMP: | |
3614 | /* FIXME: not implemented */ | |
3615 | rb_advance_iter(iter); | |
3616 | goto again; | |
3617 | ||
3618 | case RINGBUF_TYPE_DATA: | |
3619 | if (ts) { | |
3620 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
3621 | ring_buffer_normalize_time_stamp(buffer, |
3622 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
3623 | } |
3624 | return event; | |
3625 | ||
3626 | default: | |
3627 | BUG(); | |
3628 | } | |
3629 | ||
3630 | return NULL; | |
3631 | } | |
c4f50183 | 3632 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 3633 | |
8d707e8e SR |
3634 | static inline int rb_ok_to_lock(void) |
3635 | { | |
3636 | /* | |
3637 | * If an NMI die dumps out the content of the ring buffer | |
3638 | * do not grab locks. We also permanently disable the ring | |
3639 | * buffer too. A one time deal is all you get from reading | |
3640 | * the ring buffer from an NMI. | |
3641 | */ | |
464e85eb | 3642 | if (likely(!in_nmi())) |
8d707e8e SR |
3643 | return 1; |
3644 | ||
3645 | tracing_off_permanent(); | |
3646 | return 0; | |
3647 | } | |
3648 | ||
f83c9d0f SR |
3649 | /** |
3650 | * ring_buffer_peek - peek at the next event to be read | |
3651 | * @buffer: The ring buffer to read | |
3652 | * @cpu: The cpu to peak at | |
3653 | * @ts: The timestamp counter of this event. | |
66a8cb95 | 3654 | * @lost_events: a variable to store if events were lost (may be NULL) |
f83c9d0f SR |
3655 | * |
3656 | * This will return the event that will be read next, but does | |
3657 | * not consume the data. | |
3658 | */ | |
3659 | struct ring_buffer_event * | |
66a8cb95 SR |
3660 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts, |
3661 | unsigned long *lost_events) | |
f83c9d0f SR |
3662 | { |
3663 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 3664 | struct ring_buffer_event *event; |
f83c9d0f | 3665 | unsigned long flags; |
8d707e8e | 3666 | int dolock; |
f83c9d0f | 3667 | |
554f786e | 3668 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3669 | return NULL; |
554f786e | 3670 | |
8d707e8e | 3671 | dolock = rb_ok_to_lock(); |
2d622719 | 3672 | again: |
8d707e8e SR |
3673 | local_irq_save(flags); |
3674 | if (dolock) | |
5389f6fa | 3675 | raw_spin_lock(&cpu_buffer->reader_lock); |
66a8cb95 | 3676 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
469535a5 RR |
3677 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
3678 | rb_advance_reader(cpu_buffer); | |
8d707e8e | 3679 | if (dolock) |
5389f6fa | 3680 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3681 | local_irq_restore(flags); |
f83c9d0f | 3682 | |
1b959e18 | 3683 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3684 | goto again; |
2d622719 | 3685 | |
f83c9d0f SR |
3686 | return event; |
3687 | } | |
3688 | ||
3689 | /** | |
3690 | * ring_buffer_iter_peek - peek at the next event to be read | |
3691 | * @iter: The ring buffer iterator | |
3692 | * @ts: The timestamp counter of this event. | |
3693 | * | |
3694 | * This will return the event that will be read next, but does | |
3695 | * not increment the iterator. | |
3696 | */ | |
3697 | struct ring_buffer_event * | |
3698 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
3699 | { | |
3700 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3701 | struct ring_buffer_event *event; | |
3702 | unsigned long flags; | |
3703 | ||
2d622719 | 3704 | again: |
5389f6fa | 3705 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3706 | event = rb_iter_peek(iter, ts); |
5389f6fa | 3707 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3708 | |
1b959e18 | 3709 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3710 | goto again; |
2d622719 | 3711 | |
f83c9d0f SR |
3712 | return event; |
3713 | } | |
3714 | ||
7a8e76a3 SR |
3715 | /** |
3716 | * ring_buffer_consume - return an event and consume it | |
3717 | * @buffer: The ring buffer to get the next event from | |
66a8cb95 SR |
3718 | * @cpu: the cpu to read the buffer from |
3719 | * @ts: a variable to store the timestamp (may be NULL) | |
3720 | * @lost_events: a variable to store if events were lost (may be NULL) | |
7a8e76a3 SR |
3721 | * |
3722 | * Returns the next event in the ring buffer, and that event is consumed. | |
3723 | * Meaning, that sequential reads will keep returning a different event, | |
3724 | * and eventually empty the ring buffer if the producer is slower. | |
3725 | */ | |
3726 | struct ring_buffer_event * | |
66a8cb95 SR |
3727 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts, |
3728 | unsigned long *lost_events) | |
7a8e76a3 | 3729 | { |
554f786e SR |
3730 | struct ring_buffer_per_cpu *cpu_buffer; |
3731 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 3732 | unsigned long flags; |
8d707e8e SR |
3733 | int dolock; |
3734 | ||
3735 | dolock = rb_ok_to_lock(); | |
7a8e76a3 | 3736 | |
2d622719 | 3737 | again: |
554f786e SR |
3738 | /* might be called in atomic */ |
3739 | preempt_disable(); | |
3740 | ||
9e01c1b7 | 3741 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 3742 | goto out; |
7a8e76a3 | 3743 | |
554f786e | 3744 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
3745 | local_irq_save(flags); |
3746 | if (dolock) | |
5389f6fa | 3747 | raw_spin_lock(&cpu_buffer->reader_lock); |
f83c9d0f | 3748 | |
66a8cb95 SR |
3749 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
3750 | if (event) { | |
3751 | cpu_buffer->lost_events = 0; | |
469535a5 | 3752 | rb_advance_reader(cpu_buffer); |
66a8cb95 | 3753 | } |
7a8e76a3 | 3754 | |
8d707e8e | 3755 | if (dolock) |
5389f6fa | 3756 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3757 | local_irq_restore(flags); |
f83c9d0f | 3758 | |
554f786e SR |
3759 | out: |
3760 | preempt_enable(); | |
3761 | ||
1b959e18 | 3762 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3763 | goto again; |
2d622719 | 3764 | |
7a8e76a3 SR |
3765 | return event; |
3766 | } | |
c4f50183 | 3767 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
3768 | |
3769 | /** | |
72c9ddfd | 3770 | * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer |
7a8e76a3 SR |
3771 | * @buffer: The ring buffer to read from |
3772 | * @cpu: The cpu buffer to iterate over | |
3773 | * | |
72c9ddfd DM |
3774 | * This performs the initial preparations necessary to iterate |
3775 | * through the buffer. Memory is allocated, buffer recording | |
3776 | * is disabled, and the iterator pointer is returned to the caller. | |
7a8e76a3 | 3777 | * |
72c9ddfd DM |
3778 | * Disabling buffer recordng prevents the reading from being |
3779 | * corrupted. This is not a consuming read, so a producer is not | |
3780 | * expected. | |
3781 | * | |
3782 | * After a sequence of ring_buffer_read_prepare calls, the user is | |
3783 | * expected to make at least one call to ring_buffer_prepare_sync. | |
3784 | * Afterwards, ring_buffer_read_start is invoked to get things going | |
3785 | * for real. | |
3786 | * | |
3787 | * This overall must be paired with ring_buffer_finish. | |
7a8e76a3 SR |
3788 | */ |
3789 | struct ring_buffer_iter * | |
72c9ddfd | 3790 | ring_buffer_read_prepare(struct ring_buffer *buffer, int cpu) |
7a8e76a3 SR |
3791 | { |
3792 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3793 | struct ring_buffer_iter *iter; |
7a8e76a3 | 3794 | |
9e01c1b7 | 3795 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3796 | return NULL; |
7a8e76a3 SR |
3797 | |
3798 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
3799 | if (!iter) | |
8aabee57 | 3800 | return NULL; |
7a8e76a3 SR |
3801 | |
3802 | cpu_buffer = buffer->buffers[cpu]; | |
3803 | ||
3804 | iter->cpu_buffer = cpu_buffer; | |
3805 | ||
83f40318 | 3806 | atomic_inc(&buffer->resize_disabled); |
7a8e76a3 | 3807 | atomic_inc(&cpu_buffer->record_disabled); |
72c9ddfd DM |
3808 | |
3809 | return iter; | |
3810 | } | |
3811 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare); | |
3812 | ||
3813 | /** | |
3814 | * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls | |
3815 | * | |
3816 | * All previously invoked ring_buffer_read_prepare calls to prepare | |
3817 | * iterators will be synchronized. Afterwards, read_buffer_read_start | |
3818 | * calls on those iterators are allowed. | |
3819 | */ | |
3820 | void | |
3821 | ring_buffer_read_prepare_sync(void) | |
3822 | { | |
7a8e76a3 | 3823 | synchronize_sched(); |
72c9ddfd DM |
3824 | } |
3825 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync); | |
3826 | ||
3827 | /** | |
3828 | * ring_buffer_read_start - start a non consuming read of the buffer | |
3829 | * @iter: The iterator returned by ring_buffer_read_prepare | |
3830 | * | |
3831 | * This finalizes the startup of an iteration through the buffer. | |
3832 | * The iterator comes from a call to ring_buffer_read_prepare and | |
3833 | * an intervening ring_buffer_read_prepare_sync must have been | |
3834 | * performed. | |
3835 | * | |
3836 | * Must be paired with ring_buffer_finish. | |
3837 | */ | |
3838 | void | |
3839 | ring_buffer_read_start(struct ring_buffer_iter *iter) | |
3840 | { | |
3841 | struct ring_buffer_per_cpu *cpu_buffer; | |
3842 | unsigned long flags; | |
3843 | ||
3844 | if (!iter) | |
3845 | return; | |
3846 | ||
3847 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 | 3848 | |
5389f6fa | 3849 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
0199c4e6 | 3850 | arch_spin_lock(&cpu_buffer->lock); |
642edba5 | 3851 | rb_iter_reset(iter); |
0199c4e6 | 3852 | arch_spin_unlock(&cpu_buffer->lock); |
5389f6fa | 3853 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 3854 | } |
c4f50183 | 3855 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
3856 | |
3857 | /** | |
3858 | * ring_buffer_finish - finish reading the iterator of the buffer | |
3859 | * @iter: The iterator retrieved by ring_buffer_start | |
3860 | * | |
3861 | * This re-enables the recording to the buffer, and frees the | |
3862 | * iterator. | |
3863 | */ | |
3864 | void | |
3865 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
3866 | { | |
3867 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
9366c1ba | 3868 | unsigned long flags; |
7a8e76a3 | 3869 | |
659f451f SR |
3870 | /* |
3871 | * Ring buffer is disabled from recording, here's a good place | |
9366c1ba SR |
3872 | * to check the integrity of the ring buffer. |
3873 | * Must prevent readers from trying to read, as the check | |
3874 | * clears the HEAD page and readers require it. | |
659f451f | 3875 | */ |
9366c1ba | 3876 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
659f451f | 3877 | rb_check_pages(cpu_buffer); |
9366c1ba | 3878 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
659f451f | 3879 | |
7a8e76a3 | 3880 | atomic_dec(&cpu_buffer->record_disabled); |
83f40318 | 3881 | atomic_dec(&cpu_buffer->buffer->resize_disabled); |
7a8e76a3 SR |
3882 | kfree(iter); |
3883 | } | |
c4f50183 | 3884 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
3885 | |
3886 | /** | |
3887 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
3888 | * @iter: The ring buffer iterator | |
3889 | * @ts: The time stamp of the event read. | |
3890 | * | |
3891 | * This reads the next event in the ring buffer and increments the iterator. | |
3892 | */ | |
3893 | struct ring_buffer_event * | |
3894 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
3895 | { | |
3896 | struct ring_buffer_event *event; | |
f83c9d0f SR |
3897 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
3898 | unsigned long flags; | |
7a8e76a3 | 3899 | |
5389f6fa | 3900 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
7e9391cf | 3901 | again: |
f83c9d0f | 3902 | event = rb_iter_peek(iter, ts); |
7a8e76a3 | 3903 | if (!event) |
f83c9d0f | 3904 | goto out; |
7a8e76a3 | 3905 | |
7e9391cf SR |
3906 | if (event->type_len == RINGBUF_TYPE_PADDING) |
3907 | goto again; | |
3908 | ||
7a8e76a3 | 3909 | rb_advance_iter(iter); |
f83c9d0f | 3910 | out: |
5389f6fa | 3911 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
3912 | |
3913 | return event; | |
3914 | } | |
c4f50183 | 3915 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
3916 | |
3917 | /** | |
3918 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
3919 | * @buffer: The ring buffer. | |
3920 | */ | |
438ced17 | 3921 | unsigned long ring_buffer_size(struct ring_buffer *buffer, int cpu) |
7a8e76a3 | 3922 | { |
438ced17 VN |
3923 | /* |
3924 | * Earlier, this method returned | |
3925 | * BUF_PAGE_SIZE * buffer->nr_pages | |
3926 | * Since the nr_pages field is now removed, we have converted this to | |
3927 | * return the per cpu buffer value. | |
3928 | */ | |
3929 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3930 | return 0; | |
3931 | ||
3932 | return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages; | |
7a8e76a3 | 3933 | } |
c4f50183 | 3934 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
3935 | |
3936 | static void | |
3937 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
3938 | { | |
77ae365e SR |
3939 | rb_head_page_deactivate(cpu_buffer); |
3940 | ||
7a8e76a3 | 3941 | cpu_buffer->head_page |
3adc54fa | 3942 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 3943 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 3944 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 3945 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 3946 | |
6f807acd | 3947 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
3948 | |
3949 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
3950 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
3951 | ||
3952 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
5040b4b7 | 3953 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
bf41a158 | 3954 | local_set(&cpu_buffer->reader_page->write, 0); |
778c55d4 | 3955 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 3956 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 3957 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 3958 | |
c64e148a | 3959 | local_set(&cpu_buffer->entries_bytes, 0); |
77ae365e | 3960 | local_set(&cpu_buffer->overrun, 0); |
884bfe89 SP |
3961 | local_set(&cpu_buffer->commit_overrun, 0); |
3962 | local_set(&cpu_buffer->dropped_events, 0); | |
e4906eff | 3963 | local_set(&cpu_buffer->entries, 0); |
fa743953 SR |
3964 | local_set(&cpu_buffer->committing, 0); |
3965 | local_set(&cpu_buffer->commits, 0); | |
77ae365e | 3966 | cpu_buffer->read = 0; |
c64e148a | 3967 | cpu_buffer->read_bytes = 0; |
69507c06 SR |
3968 | |
3969 | cpu_buffer->write_stamp = 0; | |
3970 | cpu_buffer->read_stamp = 0; | |
77ae365e | 3971 | |
66a8cb95 SR |
3972 | cpu_buffer->lost_events = 0; |
3973 | cpu_buffer->last_overrun = 0; | |
3974 | ||
77ae365e | 3975 | rb_head_page_activate(cpu_buffer); |
7a8e76a3 SR |
3976 | } |
3977 | ||
3978 | /** | |
3979 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
3980 | * @buffer: The ring buffer to reset a per cpu buffer of | |
3981 | * @cpu: The CPU buffer to be reset | |
3982 | */ | |
3983 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
3984 | { | |
3985 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
3986 | unsigned long flags; | |
3987 | ||
9e01c1b7 | 3988 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3989 | return; |
7a8e76a3 | 3990 | |
83f40318 | 3991 | atomic_inc(&buffer->resize_disabled); |
41ede23e SR |
3992 | atomic_inc(&cpu_buffer->record_disabled); |
3993 | ||
83f40318 VN |
3994 | /* Make sure all commits have finished */ |
3995 | synchronize_sched(); | |
3996 | ||
5389f6fa | 3997 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3998 | |
41b6a95d SR |
3999 | if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing))) |
4000 | goto out; | |
4001 | ||
0199c4e6 | 4002 | arch_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
4003 | |
4004 | rb_reset_cpu(cpu_buffer); | |
4005 | ||
0199c4e6 | 4006 | arch_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 4007 | |
41b6a95d | 4008 | out: |
5389f6fa | 4009 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
41ede23e SR |
4010 | |
4011 | atomic_dec(&cpu_buffer->record_disabled); | |
83f40318 | 4012 | atomic_dec(&buffer->resize_disabled); |
7a8e76a3 | 4013 | } |
c4f50183 | 4014 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
4015 | |
4016 | /** | |
4017 | * ring_buffer_reset - reset a ring buffer | |
4018 | * @buffer: The ring buffer to reset all cpu buffers | |
4019 | */ | |
4020 | void ring_buffer_reset(struct ring_buffer *buffer) | |
4021 | { | |
7a8e76a3 SR |
4022 | int cpu; |
4023 | ||
7a8e76a3 | 4024 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 4025 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 4026 | } |
c4f50183 | 4027 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
4028 | |
4029 | /** | |
4030 | * rind_buffer_empty - is the ring buffer empty? | |
4031 | * @buffer: The ring buffer to test | |
4032 | */ | |
4033 | int ring_buffer_empty(struct ring_buffer *buffer) | |
4034 | { | |
4035 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 4036 | unsigned long flags; |
8d707e8e | 4037 | int dolock; |
7a8e76a3 | 4038 | int cpu; |
d4788207 | 4039 | int ret; |
7a8e76a3 | 4040 | |
8d707e8e | 4041 | dolock = rb_ok_to_lock(); |
7a8e76a3 SR |
4042 | |
4043 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
4044 | for_each_buffer_cpu(buffer, cpu) { | |
4045 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e SR |
4046 | local_irq_save(flags); |
4047 | if (dolock) | |
5389f6fa | 4048 | raw_spin_lock(&cpu_buffer->reader_lock); |
d4788207 | 4049 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e | 4050 | if (dolock) |
5389f6fa | 4051 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e SR |
4052 | local_irq_restore(flags); |
4053 | ||
d4788207 | 4054 | if (!ret) |
7a8e76a3 SR |
4055 | return 0; |
4056 | } | |
554f786e | 4057 | |
7a8e76a3 SR |
4058 | return 1; |
4059 | } | |
c4f50183 | 4060 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
4061 | |
4062 | /** | |
4063 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
4064 | * @buffer: The ring buffer | |
4065 | * @cpu: The CPU buffer to test | |
4066 | */ | |
4067 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
4068 | { | |
4069 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 4070 | unsigned long flags; |
8d707e8e | 4071 | int dolock; |
8aabee57 | 4072 | int ret; |
7a8e76a3 | 4073 | |
9e01c1b7 | 4074 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4075 | return 1; |
7a8e76a3 | 4076 | |
8d707e8e SR |
4077 | dolock = rb_ok_to_lock(); |
4078 | ||
7a8e76a3 | 4079 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
4080 | local_irq_save(flags); |
4081 | if (dolock) | |
5389f6fa | 4082 | raw_spin_lock(&cpu_buffer->reader_lock); |
554f786e | 4083 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e | 4084 | if (dolock) |
5389f6fa | 4085 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 4086 | local_irq_restore(flags); |
554f786e SR |
4087 | |
4088 | return ret; | |
7a8e76a3 | 4089 | } |
c4f50183 | 4090 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 | 4091 | |
85bac32c | 4092 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
7a8e76a3 SR |
4093 | /** |
4094 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
4095 | * @buffer_a: One buffer to swap with | |
4096 | * @buffer_b: The other buffer to swap with | |
4097 | * | |
4098 | * This function is useful for tracers that want to take a "snapshot" | |
4099 | * of a CPU buffer and has another back up buffer lying around. | |
4100 | * it is expected that the tracer handles the cpu buffer not being | |
4101 | * used at the moment. | |
4102 | */ | |
4103 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
4104 | struct ring_buffer *buffer_b, int cpu) | |
4105 | { | |
4106 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
4107 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
4108 | int ret = -EINVAL; |
4109 | ||
9e01c1b7 RR |
4110 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
4111 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 4112 | goto out; |
7a8e76a3 | 4113 | |
438ced17 VN |
4114 | cpu_buffer_a = buffer_a->buffers[cpu]; |
4115 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
4116 | ||
7a8e76a3 | 4117 | /* At least make sure the two buffers are somewhat the same */ |
438ced17 | 4118 | if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages) |
554f786e SR |
4119 | goto out; |
4120 | ||
4121 | ret = -EAGAIN; | |
7a8e76a3 | 4122 | |
97b17efe | 4123 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 4124 | goto out; |
97b17efe SR |
4125 | |
4126 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 4127 | goto out; |
97b17efe SR |
4128 | |
4129 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 4130 | goto out; |
97b17efe | 4131 | |
97b17efe | 4132 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 4133 | goto out; |
97b17efe SR |
4134 | |
4135 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 4136 | goto out; |
97b17efe | 4137 | |
7a8e76a3 SR |
4138 | /* |
4139 | * We can't do a synchronize_sched here because this | |
4140 | * function can be called in atomic context. | |
4141 | * Normally this will be called from the same CPU as cpu. | |
4142 | * If not it's up to the caller to protect this. | |
4143 | */ | |
4144 | atomic_inc(&cpu_buffer_a->record_disabled); | |
4145 | atomic_inc(&cpu_buffer_b->record_disabled); | |
4146 | ||
98277991 SR |
4147 | ret = -EBUSY; |
4148 | if (local_read(&cpu_buffer_a->committing)) | |
4149 | goto out_dec; | |
4150 | if (local_read(&cpu_buffer_b->committing)) | |
4151 | goto out_dec; | |
4152 | ||
7a8e76a3 SR |
4153 | buffer_a->buffers[cpu] = cpu_buffer_b; |
4154 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
4155 | ||
4156 | cpu_buffer_b->buffer = buffer_a; | |
4157 | cpu_buffer_a->buffer = buffer_b; | |
4158 | ||
98277991 SR |
4159 | ret = 0; |
4160 | ||
4161 | out_dec: | |
7a8e76a3 SR |
4162 | atomic_dec(&cpu_buffer_a->record_disabled); |
4163 | atomic_dec(&cpu_buffer_b->record_disabled); | |
554f786e | 4164 | out: |
554f786e | 4165 | return ret; |
7a8e76a3 | 4166 | } |
c4f50183 | 4167 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
85bac32c | 4168 | #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */ |
7a8e76a3 | 4169 | |
8789a9e7 SR |
4170 | /** |
4171 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
4172 | * @buffer: the buffer to allocate for. | |
4173 | * | |
4174 | * This function is used in conjunction with ring_buffer_read_page. | |
4175 | * When reading a full page from the ring buffer, these functions | |
4176 | * can be used to speed up the process. The calling function should | |
4177 | * allocate a few pages first with this function. Then when it | |
4178 | * needs to get pages from the ring buffer, it passes the result | |
4179 | * of this function into ring_buffer_read_page, which will swap | |
4180 | * the page that was allocated, with the read page of the buffer. | |
4181 | * | |
4182 | * Returns: | |
4183 | * The page allocated, or NULL on error. | |
4184 | */ | |
7ea59064 | 4185 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer, int cpu) |
8789a9e7 | 4186 | { |
044fa782 | 4187 | struct buffer_data_page *bpage; |
7ea59064 | 4188 | struct page *page; |
8789a9e7 | 4189 | |
d7ec4bfe VN |
4190 | page = alloc_pages_node(cpu_to_node(cpu), |
4191 | GFP_KERNEL | __GFP_NORETRY, 0); | |
7ea59064 | 4192 | if (!page) |
8789a9e7 SR |
4193 | return NULL; |
4194 | ||
7ea59064 | 4195 | bpage = page_address(page); |
8789a9e7 | 4196 | |
ef7a4a16 SR |
4197 | rb_init_page(bpage); |
4198 | ||
044fa782 | 4199 | return bpage; |
8789a9e7 | 4200 | } |
d6ce96da | 4201 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
4202 | |
4203 | /** | |
4204 | * ring_buffer_free_read_page - free an allocated read page | |
4205 | * @buffer: the buffer the page was allocate for | |
4206 | * @data: the page to free | |
4207 | * | |
4208 | * Free a page allocated from ring_buffer_alloc_read_page. | |
4209 | */ | |
4210 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
4211 | { | |
4212 | free_page((unsigned long)data); | |
4213 | } | |
d6ce96da | 4214 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
4215 | |
4216 | /** | |
4217 | * ring_buffer_read_page - extract a page from the ring buffer | |
4218 | * @buffer: buffer to extract from | |
4219 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 4220 | * @len: amount to extract |
8789a9e7 SR |
4221 | * @cpu: the cpu of the buffer to extract |
4222 | * @full: should the extraction only happen when the page is full. | |
4223 | * | |
4224 | * This function will pull out a page from the ring buffer and consume it. | |
4225 | * @data_page must be the address of the variable that was returned | |
4226 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
4227 | * to swap with a page in the ring buffer. | |
4228 | * | |
4229 | * for example: | |
b85fa01e | 4230 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
4231 | * if (!rpage) |
4232 | * return error; | |
ef7a4a16 | 4233 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
4234 | * if (ret >= 0) |
4235 | * process_page(rpage, ret); | |
8789a9e7 SR |
4236 | * |
4237 | * When @full is set, the function will not return true unless | |
4238 | * the writer is off the reader page. | |
4239 | * | |
4240 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
4241 | * The ring buffer can be used anywhere in the kernel and can not | |
4242 | * blindly call wake_up. The layer that uses the ring buffer must be | |
4243 | * responsible for that. | |
4244 | * | |
4245 | * Returns: | |
667d2412 LJ |
4246 | * >=0 if data has been transferred, returns the offset of consumed data. |
4247 | * <0 if no data has been transferred. | |
8789a9e7 SR |
4248 | */ |
4249 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 4250 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
4251 | { |
4252 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
4253 | struct ring_buffer_event *event; | |
044fa782 | 4254 | struct buffer_data_page *bpage; |
ef7a4a16 | 4255 | struct buffer_page *reader; |
ff0ff84a | 4256 | unsigned long missed_events; |
8789a9e7 | 4257 | unsigned long flags; |
ef7a4a16 | 4258 | unsigned int commit; |
667d2412 | 4259 | unsigned int read; |
4f3640f8 | 4260 | u64 save_timestamp; |
667d2412 | 4261 | int ret = -1; |
8789a9e7 | 4262 | |
554f786e SR |
4263 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
4264 | goto out; | |
4265 | ||
474d32b6 SR |
4266 | /* |
4267 | * If len is not big enough to hold the page header, then | |
4268 | * we can not copy anything. | |
4269 | */ | |
4270 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 4271 | goto out; |
474d32b6 SR |
4272 | |
4273 | len -= BUF_PAGE_HDR_SIZE; | |
4274 | ||
8789a9e7 | 4275 | if (!data_page) |
554f786e | 4276 | goto out; |
8789a9e7 | 4277 | |
044fa782 SR |
4278 | bpage = *data_page; |
4279 | if (!bpage) | |
554f786e | 4280 | goto out; |
8789a9e7 | 4281 | |
5389f6fa | 4282 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4283 | |
ef7a4a16 SR |
4284 | reader = rb_get_reader_page(cpu_buffer); |
4285 | if (!reader) | |
554f786e | 4286 | goto out_unlock; |
8789a9e7 | 4287 | |
ef7a4a16 SR |
4288 | event = rb_reader_event(cpu_buffer); |
4289 | ||
4290 | read = reader->read; | |
4291 | commit = rb_page_commit(reader); | |
667d2412 | 4292 | |
66a8cb95 | 4293 | /* Check if any events were dropped */ |
ff0ff84a | 4294 | missed_events = cpu_buffer->lost_events; |
66a8cb95 | 4295 | |
8789a9e7 | 4296 | /* |
474d32b6 SR |
4297 | * If this page has been partially read or |
4298 | * if len is not big enough to read the rest of the page or | |
4299 | * a writer is still on the page, then | |
4300 | * we must copy the data from the page to the buffer. | |
4301 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 4302 | */ |
474d32b6 | 4303 | if (read || (len < (commit - read)) || |
ef7a4a16 | 4304 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 4305 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
4306 | unsigned int rpos = read; |
4307 | unsigned int pos = 0; | |
ef7a4a16 | 4308 | unsigned int size; |
8789a9e7 SR |
4309 | |
4310 | if (full) | |
554f786e | 4311 | goto out_unlock; |
8789a9e7 | 4312 | |
ef7a4a16 SR |
4313 | if (len > (commit - read)) |
4314 | len = (commit - read); | |
4315 | ||
69d1b839 SR |
4316 | /* Always keep the time extend and data together */ |
4317 | size = rb_event_ts_length(event); | |
ef7a4a16 SR |
4318 | |
4319 | if (len < size) | |
554f786e | 4320 | goto out_unlock; |
ef7a4a16 | 4321 | |
4f3640f8 SR |
4322 | /* save the current timestamp, since the user will need it */ |
4323 | save_timestamp = cpu_buffer->read_stamp; | |
4324 | ||
ef7a4a16 SR |
4325 | /* Need to copy one event at a time */ |
4326 | do { | |
e1e35927 DS |
4327 | /* We need the size of one event, because |
4328 | * rb_advance_reader only advances by one event, | |
4329 | * whereas rb_event_ts_length may include the size of | |
4330 | * one or two events. | |
4331 | * We have already ensured there's enough space if this | |
4332 | * is a time extend. */ | |
4333 | size = rb_event_length(event); | |
474d32b6 | 4334 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
4335 | |
4336 | len -= size; | |
4337 | ||
4338 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
4339 | rpos = reader->read; |
4340 | pos += size; | |
ef7a4a16 | 4341 | |
18fab912 HY |
4342 | if (rpos >= commit) |
4343 | break; | |
4344 | ||
ef7a4a16 | 4345 | event = rb_reader_event(cpu_buffer); |
69d1b839 SR |
4346 | /* Always keep the time extend and data together */ |
4347 | size = rb_event_ts_length(event); | |
e1e35927 | 4348 | } while (len >= size); |
667d2412 LJ |
4349 | |
4350 | /* update bpage */ | |
ef7a4a16 | 4351 | local_set(&bpage->commit, pos); |
4f3640f8 | 4352 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 4353 | |
474d32b6 SR |
4354 | /* we copied everything to the beginning */ |
4355 | read = 0; | |
8789a9e7 | 4356 | } else { |
afbab76a | 4357 | /* update the entry counter */ |
77ae365e | 4358 | cpu_buffer->read += rb_page_entries(reader); |
c64e148a | 4359 | cpu_buffer->read_bytes += BUF_PAGE_SIZE; |
afbab76a | 4360 | |
8789a9e7 | 4361 | /* swap the pages */ |
044fa782 | 4362 | rb_init_page(bpage); |
ef7a4a16 SR |
4363 | bpage = reader->page; |
4364 | reader->page = *data_page; | |
4365 | local_set(&reader->write, 0); | |
778c55d4 | 4366 | local_set(&reader->entries, 0); |
ef7a4a16 | 4367 | reader->read = 0; |
044fa782 | 4368 | *data_page = bpage; |
ff0ff84a SR |
4369 | |
4370 | /* | |
4371 | * Use the real_end for the data size, | |
4372 | * This gives us a chance to store the lost events | |
4373 | * on the page. | |
4374 | */ | |
4375 | if (reader->real_end) | |
4376 | local_set(&bpage->commit, reader->real_end); | |
8789a9e7 | 4377 | } |
667d2412 | 4378 | ret = read; |
8789a9e7 | 4379 | |
66a8cb95 | 4380 | cpu_buffer->lost_events = 0; |
2711ca23 SR |
4381 | |
4382 | commit = local_read(&bpage->commit); | |
66a8cb95 SR |
4383 | /* |
4384 | * Set a flag in the commit field if we lost events | |
4385 | */ | |
ff0ff84a | 4386 | if (missed_events) { |
ff0ff84a SR |
4387 | /* If there is room at the end of the page to save the |
4388 | * missed events, then record it there. | |
4389 | */ | |
4390 | if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) { | |
4391 | memcpy(&bpage->data[commit], &missed_events, | |
4392 | sizeof(missed_events)); | |
4393 | local_add(RB_MISSED_STORED, &bpage->commit); | |
2711ca23 | 4394 | commit += sizeof(missed_events); |
ff0ff84a | 4395 | } |
66a8cb95 | 4396 | local_add(RB_MISSED_EVENTS, &bpage->commit); |
ff0ff84a | 4397 | } |
66a8cb95 | 4398 | |
2711ca23 SR |
4399 | /* |
4400 | * This page may be off to user land. Zero it out here. | |
4401 | */ | |
4402 | if (commit < BUF_PAGE_SIZE) | |
4403 | memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit); | |
4404 | ||
554f786e | 4405 | out_unlock: |
5389f6fa | 4406 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4407 | |
554f786e | 4408 | out: |
8789a9e7 SR |
4409 | return ret; |
4410 | } | |
d6ce96da | 4411 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 4412 | |
59222efe | 4413 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
4414 | static int rb_cpu_notify(struct notifier_block *self, |
4415 | unsigned long action, void *hcpu) | |
554f786e SR |
4416 | { |
4417 | struct ring_buffer *buffer = | |
4418 | container_of(self, struct ring_buffer, cpu_notify); | |
4419 | long cpu = (long)hcpu; | |
438ced17 VN |
4420 | int cpu_i, nr_pages_same; |
4421 | unsigned int nr_pages; | |
554f786e SR |
4422 | |
4423 | switch (action) { | |
4424 | case CPU_UP_PREPARE: | |
4425 | case CPU_UP_PREPARE_FROZEN: | |
3f237a79 | 4426 | if (cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e SR |
4427 | return NOTIFY_OK; |
4428 | ||
438ced17 VN |
4429 | nr_pages = 0; |
4430 | nr_pages_same = 1; | |
4431 | /* check if all cpu sizes are same */ | |
4432 | for_each_buffer_cpu(buffer, cpu_i) { | |
4433 | /* fill in the size from first enabled cpu */ | |
4434 | if (nr_pages == 0) | |
4435 | nr_pages = buffer->buffers[cpu_i]->nr_pages; | |
4436 | if (nr_pages != buffer->buffers[cpu_i]->nr_pages) { | |
4437 | nr_pages_same = 0; | |
4438 | break; | |
4439 | } | |
4440 | } | |
4441 | /* allocate minimum pages, user can later expand it */ | |
4442 | if (!nr_pages_same) | |
4443 | nr_pages = 2; | |
554f786e | 4444 | buffer->buffers[cpu] = |
438ced17 | 4445 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
554f786e SR |
4446 | if (!buffer->buffers[cpu]) { |
4447 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
4448 | cpu); | |
4449 | return NOTIFY_OK; | |
4450 | } | |
4451 | smp_wmb(); | |
3f237a79 | 4452 | cpumask_set_cpu(cpu, buffer->cpumask); |
554f786e SR |
4453 | break; |
4454 | case CPU_DOWN_PREPARE: | |
4455 | case CPU_DOWN_PREPARE_FROZEN: | |
4456 | /* | |
4457 | * Do nothing. | |
4458 | * If we were to free the buffer, then the user would | |
4459 | * lose any trace that was in the buffer. | |
4460 | */ | |
4461 | break; | |
4462 | default: | |
4463 | break; | |
4464 | } | |
4465 | return NOTIFY_OK; | |
4466 | } | |
4467 | #endif |