]>
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 | 10 | #include <linux/spinlock.h> |
15693458 | 11 | #include <linux/irq_work.h> |
7a8e76a3 SR |
12 | #include <linux/debugfs.h> |
13 | #include <linux/uaccess.h> | |
a81bd80a | 14 | #include <linux/hardirq.h> |
1744a21d | 15 | #include <linux/kmemcheck.h> |
7a8e76a3 SR |
16 | #include <linux/module.h> |
17 | #include <linux/percpu.h> | |
18 | #include <linux/mutex.h> | |
5a0e3ad6 | 19 | #include <linux/slab.h> |
7a8e76a3 SR |
20 | #include <linux/init.h> |
21 | #include <linux/hash.h> | |
22 | #include <linux/list.h> | |
554f786e | 23 | #include <linux/cpu.h> |
7a8e76a3 SR |
24 | #include <linux/fs.h> |
25 | ||
79615760 | 26 | #include <asm/local.h> |
182e9f5f | 27 | |
83f40318 VN |
28 | static void update_pages_handler(struct work_struct *work); |
29 | ||
d1b182a8 SR |
30 | /* |
31 | * The ring buffer header is special. We must manually up keep it. | |
32 | */ | |
33 | int ring_buffer_print_entry_header(struct trace_seq *s) | |
34 | { | |
35 | int ret; | |
36 | ||
334d4169 LJ |
37 | ret = trace_seq_printf(s, "# compressed entry header\n"); |
38 | ret = trace_seq_printf(s, "\ttype_len : 5 bits\n"); | |
d1b182a8 SR |
39 | ret = trace_seq_printf(s, "\ttime_delta : 27 bits\n"); |
40 | ret = trace_seq_printf(s, "\tarray : 32 bits\n"); | |
41 | ret = trace_seq_printf(s, "\n"); | |
42 | ret = trace_seq_printf(s, "\tpadding : type == %d\n", | |
43 | RINGBUF_TYPE_PADDING); | |
44 | ret = trace_seq_printf(s, "\ttime_extend : type == %d\n", | |
45 | RINGBUF_TYPE_TIME_EXTEND); | |
334d4169 LJ |
46 | ret = trace_seq_printf(s, "\tdata max type_len == %d\n", |
47 | RINGBUF_TYPE_DATA_TYPE_LEN_MAX); | |
d1b182a8 SR |
48 | |
49 | return ret; | |
50 | } | |
51 | ||
5cc98548 SR |
52 | /* |
53 | * The ring buffer is made up of a list of pages. A separate list of pages is | |
54 | * allocated for each CPU. A writer may only write to a buffer that is | |
55 | * associated with the CPU it is currently executing on. A reader may read | |
56 | * from any per cpu buffer. | |
57 | * | |
58 | * The reader is special. For each per cpu buffer, the reader has its own | |
59 | * reader page. When a reader has read the entire reader page, this reader | |
60 | * page is swapped with another page in the ring buffer. | |
61 | * | |
62 | * Now, as long as the writer is off the reader page, the reader can do what | |
63 | * ever it wants with that page. The writer will never write to that page | |
64 | * again (as long as it is out of the ring buffer). | |
65 | * | |
66 | * Here's some silly ASCII art. | |
67 | * | |
68 | * +------+ | |
69 | * |reader| RING BUFFER | |
70 | * |page | | |
71 | * +------+ +---+ +---+ +---+ | |
72 | * | |-->| |-->| | | |
73 | * +---+ +---+ +---+ | |
74 | * ^ | | |
75 | * | | | |
76 | * +---------------+ | |
77 | * | |
78 | * | |
79 | * +------+ | |
80 | * |reader| RING BUFFER | |
81 | * |page |------------------v | |
82 | * +------+ +---+ +---+ +---+ | |
83 | * | |-->| |-->| | | |
84 | * +---+ +---+ +---+ | |
85 | * ^ | | |
86 | * | | | |
87 | * +---------------+ | |
88 | * | |
89 | * | |
90 | * +------+ | |
91 | * |reader| RING BUFFER | |
92 | * |page |------------------v | |
93 | * +------+ +---+ +---+ +---+ | |
94 | * ^ | |-->| |-->| | | |
95 | * | +---+ +---+ +---+ | |
96 | * | | | |
97 | * | | | |
98 | * +------------------------------+ | |
99 | * | |
100 | * | |
101 | * +------+ | |
102 | * |buffer| RING BUFFER | |
103 | * |page |------------------v | |
104 | * +------+ +---+ +---+ +---+ | |
105 | * ^ | | | |-->| | | |
106 | * | New +---+ +---+ +---+ | |
107 | * | Reader------^ | | |
108 | * | page | | |
109 | * +------------------------------+ | |
110 | * | |
111 | * | |
112 | * After we make this swap, the reader can hand this page off to the splice | |
113 | * code and be done with it. It can even allocate a new page if it needs to | |
114 | * and swap that into the ring buffer. | |
115 | * | |
116 | * We will be using cmpxchg soon to make all this lockless. | |
117 | * | |
118 | */ | |
119 | ||
033601a3 SR |
120 | /* |
121 | * A fast way to enable or disable all ring buffers is to | |
122 | * call tracing_on or tracing_off. Turning off the ring buffers | |
123 | * prevents all ring buffers from being recorded to. | |
124 | * Turning this switch on, makes it OK to write to the | |
125 | * ring buffer, if the ring buffer is enabled itself. | |
126 | * | |
127 | * There's three layers that must be on in order to write | |
128 | * to the ring buffer. | |
129 | * | |
130 | * 1) This global flag must be set. | |
131 | * 2) The ring buffer must be enabled for recording. | |
132 | * 3) The per cpu buffer must be enabled for recording. | |
133 | * | |
134 | * In case of an anomaly, this global flag has a bit set that | |
135 | * will permantly disable all ring buffers. | |
136 | */ | |
137 | ||
138 | /* | |
139 | * Global flag to disable all recording to ring buffers | |
140 | * This has two bits: ON, DISABLED | |
141 | * | |
142 | * ON DISABLED | |
143 | * ---- ---------- | |
144 | * 0 0 : ring buffers are off | |
145 | * 1 0 : ring buffers are on | |
146 | * X 1 : ring buffers are permanently disabled | |
147 | */ | |
148 | ||
149 | enum { | |
150 | RB_BUFFERS_ON_BIT = 0, | |
151 | RB_BUFFERS_DISABLED_BIT = 1, | |
152 | }; | |
153 | ||
154 | enum { | |
155 | RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT, | |
156 | RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT, | |
157 | }; | |
158 | ||
5e39841c | 159 | static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON; |
a3583244 | 160 | |
499e5470 SR |
161 | /* Used for individual buffers (after the counter) */ |
162 | #define RB_BUFFER_OFF (1 << 20) | |
a3583244 | 163 | |
499e5470 | 164 | #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data) |
033601a3 SR |
165 | |
166 | /** | |
167 | * tracing_off_permanent - permanently disable ring buffers | |
168 | * | |
169 | * This function, once called, will disable all ring buffers | |
c3706f00 | 170 | * permanently. |
033601a3 SR |
171 | */ |
172 | void tracing_off_permanent(void) | |
173 | { | |
174 | set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags); | |
a3583244 SR |
175 | } |
176 | ||
e3d6bf0a | 177 | #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array)) |
67d34724 | 178 | #define RB_ALIGNMENT 4U |
334d4169 | 179 | #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
c7b09308 | 180 | #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */ |
334d4169 | 181 | |
2271048d SR |
182 | #if !defined(CONFIG_64BIT) || defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) |
183 | # define RB_FORCE_8BYTE_ALIGNMENT 0 | |
184 | # define RB_ARCH_ALIGNMENT RB_ALIGNMENT | |
185 | #else | |
186 | # define RB_FORCE_8BYTE_ALIGNMENT 1 | |
187 | # define RB_ARCH_ALIGNMENT 8U | |
188 | #endif | |
189 | ||
334d4169 LJ |
190 | /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */ |
191 | #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX | |
7a8e76a3 SR |
192 | |
193 | enum { | |
194 | RB_LEN_TIME_EXTEND = 8, | |
195 | RB_LEN_TIME_STAMP = 16, | |
196 | }; | |
197 | ||
69d1b839 SR |
198 | #define skip_time_extend(event) \ |
199 | ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND)) | |
200 | ||
2d622719 TZ |
201 | static inline int rb_null_event(struct ring_buffer_event *event) |
202 | { | |
a1863c21 | 203 | return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta; |
2d622719 TZ |
204 | } |
205 | ||
206 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
207 | { | |
a1863c21 | 208 | /* padding has a NULL time_delta */ |
334d4169 | 209 | event->type_len = RINGBUF_TYPE_PADDING; |
2d622719 TZ |
210 | event->time_delta = 0; |
211 | } | |
212 | ||
34a148bf | 213 | static unsigned |
2d622719 | 214 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
215 | { |
216 | unsigned length; | |
217 | ||
334d4169 LJ |
218 | if (event->type_len) |
219 | length = event->type_len * RB_ALIGNMENT; | |
2d622719 TZ |
220 | else |
221 | length = event->array[0]; | |
222 | return length + RB_EVNT_HDR_SIZE; | |
223 | } | |
224 | ||
69d1b839 SR |
225 | /* |
226 | * Return the length of the given event. Will return | |
227 | * the length of the time extend if the event is a | |
228 | * time extend. | |
229 | */ | |
230 | static inline unsigned | |
2d622719 TZ |
231 | rb_event_length(struct ring_buffer_event *event) |
232 | { | |
334d4169 | 233 | switch (event->type_len) { |
7a8e76a3 | 234 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
235 | if (rb_null_event(event)) |
236 | /* undefined */ | |
237 | return -1; | |
334d4169 | 238 | return event->array[0] + RB_EVNT_HDR_SIZE; |
7a8e76a3 SR |
239 | |
240 | case RINGBUF_TYPE_TIME_EXTEND: | |
241 | return RB_LEN_TIME_EXTEND; | |
242 | ||
243 | case RINGBUF_TYPE_TIME_STAMP: | |
244 | return RB_LEN_TIME_STAMP; | |
245 | ||
246 | case RINGBUF_TYPE_DATA: | |
2d622719 | 247 | return rb_event_data_length(event); |
7a8e76a3 SR |
248 | default: |
249 | BUG(); | |
250 | } | |
251 | /* not hit */ | |
252 | return 0; | |
253 | } | |
254 | ||
69d1b839 SR |
255 | /* |
256 | * Return total length of time extend and data, | |
257 | * or just the event length for all other events. | |
258 | */ | |
259 | static inline unsigned | |
260 | rb_event_ts_length(struct ring_buffer_event *event) | |
261 | { | |
262 | unsigned len = 0; | |
263 | ||
264 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
265 | /* time extends include the data event after it */ | |
266 | len = RB_LEN_TIME_EXTEND; | |
267 | event = skip_time_extend(event); | |
268 | } | |
269 | return len + rb_event_length(event); | |
270 | } | |
271 | ||
7a8e76a3 SR |
272 | /** |
273 | * ring_buffer_event_length - return the length of the event | |
274 | * @event: the event to get the length of | |
69d1b839 SR |
275 | * |
276 | * Returns the size of the data load of a data event. | |
277 | * If the event is something other than a data event, it | |
278 | * returns the size of the event itself. With the exception | |
279 | * of a TIME EXTEND, where it still returns the size of the | |
280 | * data load of the data event after it. | |
7a8e76a3 SR |
281 | */ |
282 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
283 | { | |
69d1b839 SR |
284 | unsigned length; |
285 | ||
286 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) | |
287 | event = skip_time_extend(event); | |
288 | ||
289 | length = rb_event_length(event); | |
334d4169 | 290 | if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
465634ad RR |
291 | return length; |
292 | length -= RB_EVNT_HDR_SIZE; | |
293 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
294 | length -= sizeof(event->array[0]); | |
295 | return length; | |
7a8e76a3 | 296 | } |
c4f50183 | 297 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
298 | |
299 | /* inline for ring buffer fast paths */ | |
34a148bf | 300 | static void * |
7a8e76a3 SR |
301 | rb_event_data(struct ring_buffer_event *event) |
302 | { | |
69d1b839 SR |
303 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) |
304 | event = skip_time_extend(event); | |
334d4169 | 305 | BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX); |
7a8e76a3 | 306 | /* If length is in len field, then array[0] has the data */ |
334d4169 | 307 | if (event->type_len) |
7a8e76a3 SR |
308 | return (void *)&event->array[0]; |
309 | /* Otherwise length is in array[0] and array[1] has the data */ | |
310 | return (void *)&event->array[1]; | |
311 | } | |
312 | ||
313 | /** | |
314 | * ring_buffer_event_data - return the data of the event | |
315 | * @event: the event to get the data from | |
316 | */ | |
317 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
318 | { | |
319 | return rb_event_data(event); | |
320 | } | |
c4f50183 | 321 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
322 | |
323 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 324 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
325 | |
326 | #define TS_SHIFT 27 | |
327 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
328 | #define TS_DELTA_TEST (~TS_MASK) | |
329 | ||
66a8cb95 SR |
330 | /* Flag when events were overwritten */ |
331 | #define RB_MISSED_EVENTS (1 << 31) | |
ff0ff84a SR |
332 | /* Missed count stored at end */ |
333 | #define RB_MISSED_STORED (1 << 30) | |
66a8cb95 | 334 | |
abc9b56d | 335 | struct buffer_data_page { |
e4c2ce82 | 336 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 337 | local_t commit; /* write committed index */ |
abc9b56d SR |
338 | unsigned char data[]; /* data of buffer page */ |
339 | }; | |
340 | ||
77ae365e SR |
341 | /* |
342 | * Note, the buffer_page list must be first. The buffer pages | |
343 | * are allocated in cache lines, which means that each buffer | |
344 | * page will be at the beginning of a cache line, and thus | |
345 | * the least significant bits will be zero. We use this to | |
346 | * add flags in the list struct pointers, to make the ring buffer | |
347 | * lockless. | |
348 | */ | |
abc9b56d | 349 | struct buffer_page { |
778c55d4 | 350 | struct list_head list; /* list of buffer pages */ |
abc9b56d | 351 | local_t write; /* index for next write */ |
6f807acd | 352 | unsigned read; /* index for next read */ |
778c55d4 | 353 | local_t entries; /* entries on this page */ |
ff0ff84a | 354 | unsigned long real_end; /* real end of data */ |
abc9b56d | 355 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
356 | }; |
357 | ||
77ae365e SR |
358 | /* |
359 | * The buffer page counters, write and entries, must be reset | |
360 | * atomically when crossing page boundaries. To synchronize this | |
361 | * update, two counters are inserted into the number. One is | |
362 | * the actual counter for the write position or count on the page. | |
363 | * | |
364 | * The other is a counter of updaters. Before an update happens | |
365 | * the update partition of the counter is incremented. This will | |
366 | * allow the updater to update the counter atomically. | |
367 | * | |
368 | * The counter is 20 bits, and the state data is 12. | |
369 | */ | |
370 | #define RB_WRITE_MASK 0xfffff | |
371 | #define RB_WRITE_INTCNT (1 << 20) | |
372 | ||
044fa782 | 373 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 374 | { |
044fa782 | 375 | local_set(&bpage->commit, 0); |
abc9b56d SR |
376 | } |
377 | ||
474d32b6 SR |
378 | /** |
379 | * ring_buffer_page_len - the size of data on the page. | |
380 | * @page: The page to read | |
381 | * | |
382 | * Returns the amount of data on the page, including buffer page header. | |
383 | */ | |
ef7a4a16 SR |
384 | size_t ring_buffer_page_len(void *page) |
385 | { | |
474d32b6 SR |
386 | return local_read(&((struct buffer_data_page *)page)->commit) |
387 | + BUF_PAGE_HDR_SIZE; | |
ef7a4a16 SR |
388 | } |
389 | ||
ed56829c SR |
390 | /* |
391 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
392 | * this issue out. | |
393 | */ | |
34a148bf | 394 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 395 | { |
34a148bf | 396 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 397 | kfree(bpage); |
ed56829c SR |
398 | } |
399 | ||
7a8e76a3 SR |
400 | /* |
401 | * We need to fit the time_stamp delta into 27 bits. | |
402 | */ | |
403 | static inline int test_time_stamp(u64 delta) | |
404 | { | |
405 | if (delta & TS_DELTA_TEST) | |
406 | return 1; | |
407 | return 0; | |
408 | } | |
409 | ||
474d32b6 | 410 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 | 411 | |
be957c44 SR |
412 | /* Max payload is BUF_PAGE_SIZE - header (8bytes) */ |
413 | #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) | |
414 | ||
d1b182a8 SR |
415 | int ring_buffer_print_page_header(struct trace_seq *s) |
416 | { | |
417 | struct buffer_data_page field; | |
418 | int ret; | |
419 | ||
420 | ret = trace_seq_printf(s, "\tfield: u64 timestamp;\t" | |
26a50744 TZ |
421 | "offset:0;\tsize:%u;\tsigned:%u;\n", |
422 | (unsigned int)sizeof(field.time_stamp), | |
423 | (unsigned int)is_signed_type(u64)); | |
d1b182a8 SR |
424 | |
425 | ret = trace_seq_printf(s, "\tfield: local_t commit;\t" | |
26a50744 | 426 | "offset:%u;\tsize:%u;\tsigned:%u;\n", |
d1b182a8 | 427 | (unsigned int)offsetof(typeof(field), commit), |
26a50744 TZ |
428 | (unsigned int)sizeof(field.commit), |
429 | (unsigned int)is_signed_type(long)); | |
d1b182a8 | 430 | |
66a8cb95 SR |
431 | ret = trace_seq_printf(s, "\tfield: int overwrite;\t" |
432 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
433 | (unsigned int)offsetof(typeof(field), commit), | |
434 | 1, | |
435 | (unsigned int)is_signed_type(long)); | |
436 | ||
d1b182a8 | 437 | ret = trace_seq_printf(s, "\tfield: char data;\t" |
26a50744 | 438 | "offset:%u;\tsize:%u;\tsigned:%u;\n", |
d1b182a8 | 439 | (unsigned int)offsetof(typeof(field), data), |
26a50744 TZ |
440 | (unsigned int)BUF_PAGE_SIZE, |
441 | (unsigned int)is_signed_type(char)); | |
d1b182a8 SR |
442 | |
443 | return ret; | |
444 | } | |
445 | ||
15693458 SRRH |
446 | struct rb_irq_work { |
447 | struct irq_work work; | |
448 | wait_queue_head_t waiters; | |
449 | bool waiters_pending; | |
450 | }; | |
451 | ||
7a8e76a3 SR |
452 | /* |
453 | * head_page == tail_page && head == tail then buffer is empty. | |
454 | */ | |
455 | struct ring_buffer_per_cpu { | |
456 | int cpu; | |
985023de | 457 | atomic_t record_disabled; |
7a8e76a3 | 458 | struct ring_buffer *buffer; |
5389f6fa | 459 | raw_spinlock_t reader_lock; /* serialize readers */ |
445c8951 | 460 | arch_spinlock_t lock; |
7a8e76a3 | 461 | struct lock_class_key lock_key; |
438ced17 | 462 | unsigned int nr_pages; |
3adc54fa | 463 | struct list_head *pages; |
6f807acd SR |
464 | struct buffer_page *head_page; /* read from head */ |
465 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 466 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 467 | struct buffer_page *reader_page; |
66a8cb95 SR |
468 | unsigned long lost_events; |
469 | unsigned long last_overrun; | |
c64e148a | 470 | local_t entries_bytes; |
e4906eff | 471 | local_t entries; |
884bfe89 SP |
472 | local_t overrun; |
473 | local_t commit_overrun; | |
474 | local_t dropped_events; | |
fa743953 SR |
475 | local_t committing; |
476 | local_t commits; | |
77ae365e | 477 | unsigned long read; |
c64e148a | 478 | unsigned long read_bytes; |
7a8e76a3 SR |
479 | u64 write_stamp; |
480 | u64 read_stamp; | |
438ced17 VN |
481 | /* ring buffer pages to update, > 0 to add, < 0 to remove */ |
482 | int nr_pages_to_update; | |
483 | struct list_head new_pages; /* new pages to add */ | |
83f40318 | 484 | struct work_struct update_pages_work; |
05fdd70d | 485 | struct completion update_done; |
15693458 SRRH |
486 | |
487 | struct rb_irq_work irq_work; | |
7a8e76a3 SR |
488 | }; |
489 | ||
490 | struct ring_buffer { | |
7a8e76a3 SR |
491 | unsigned flags; |
492 | int cpus; | |
7a8e76a3 | 493 | atomic_t record_disabled; |
83f40318 | 494 | atomic_t resize_disabled; |
00f62f61 | 495 | cpumask_var_t cpumask; |
7a8e76a3 | 496 | |
1f8a6a10 PZ |
497 | struct lock_class_key *reader_lock_key; |
498 | ||
7a8e76a3 SR |
499 | struct mutex mutex; |
500 | ||
501 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 502 | |
59222efe | 503 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
504 | struct notifier_block cpu_notify; |
505 | #endif | |
37886f6a | 506 | u64 (*clock)(void); |
15693458 SRRH |
507 | |
508 | struct rb_irq_work irq_work; | |
7a8e76a3 SR |
509 | }; |
510 | ||
511 | struct ring_buffer_iter { | |
512 | struct ring_buffer_per_cpu *cpu_buffer; | |
513 | unsigned long head; | |
514 | struct buffer_page *head_page; | |
492a74f4 SR |
515 | struct buffer_page *cache_reader_page; |
516 | unsigned long cache_read; | |
7a8e76a3 SR |
517 | u64 read_stamp; |
518 | }; | |
519 | ||
15693458 SRRH |
520 | /* |
521 | * rb_wake_up_waiters - wake up tasks waiting for ring buffer input | |
522 | * | |
523 | * Schedules a delayed work to wake up any task that is blocked on the | |
524 | * ring buffer waiters queue. | |
525 | */ | |
526 | static void rb_wake_up_waiters(struct irq_work *work) | |
527 | { | |
528 | struct rb_irq_work *rbwork = container_of(work, struct rb_irq_work, work); | |
529 | ||
530 | wake_up_all(&rbwork->waiters); | |
531 | } | |
532 | ||
533 | /** | |
534 | * ring_buffer_wait - wait for input to the ring buffer | |
535 | * @buffer: buffer to wait on | |
536 | * @cpu: the cpu buffer to wait on | |
537 | * | |
538 | * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon | |
539 | * as data is added to any of the @buffer's cpu buffers. Otherwise | |
540 | * it will wait for data to be added to a specific cpu buffer. | |
541 | */ | |
542 | void ring_buffer_wait(struct ring_buffer *buffer, int cpu) | |
543 | { | |
544 | struct ring_buffer_per_cpu *cpu_buffer; | |
545 | DEFINE_WAIT(wait); | |
546 | struct rb_irq_work *work; | |
547 | ||
548 | /* | |
549 | * Depending on what the caller is waiting for, either any | |
550 | * data in any cpu buffer, or a specific buffer, put the | |
551 | * caller on the appropriate wait queue. | |
552 | */ | |
553 | if (cpu == RING_BUFFER_ALL_CPUS) | |
554 | work = &buffer->irq_work; | |
555 | else { | |
556 | cpu_buffer = buffer->buffers[cpu]; | |
557 | work = &cpu_buffer->irq_work; | |
558 | } | |
559 | ||
560 | ||
561 | prepare_to_wait(&work->waiters, &wait, TASK_INTERRUPTIBLE); | |
562 | ||
563 | /* | |
564 | * The events can happen in critical sections where | |
565 | * checking a work queue can cause deadlocks. | |
566 | * After adding a task to the queue, this flag is set | |
567 | * only to notify events to try to wake up the queue | |
568 | * using irq_work. | |
569 | * | |
570 | * We don't clear it even if the buffer is no longer | |
571 | * empty. The flag only causes the next event to run | |
572 | * irq_work to do the work queue wake up. The worse | |
573 | * that can happen if we race with !trace_empty() is that | |
574 | * an event will cause an irq_work to try to wake up | |
575 | * an empty queue. | |
576 | * | |
577 | * There's no reason to protect this flag either, as | |
578 | * the work queue and irq_work logic will do the necessary | |
579 | * synchronization for the wake ups. The only thing | |
580 | * that is necessary is that the wake up happens after | |
581 | * a task has been queued. It's OK for spurious wake ups. | |
582 | */ | |
583 | work->waiters_pending = true; | |
584 | ||
585 | if ((cpu == RING_BUFFER_ALL_CPUS && ring_buffer_empty(buffer)) || | |
586 | (cpu != RING_BUFFER_ALL_CPUS && ring_buffer_empty_cpu(buffer, cpu))) | |
587 | schedule(); | |
588 | ||
589 | finish_wait(&work->waiters, &wait); | |
590 | } | |
591 | ||
592 | /** | |
593 | * ring_buffer_poll_wait - poll on buffer input | |
594 | * @buffer: buffer to wait on | |
595 | * @cpu: the cpu buffer to wait on | |
596 | * @filp: the file descriptor | |
597 | * @poll_table: The poll descriptor | |
598 | * | |
599 | * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon | |
600 | * as data is added to any of the @buffer's cpu buffers. Otherwise | |
601 | * it will wait for data to be added to a specific cpu buffer. | |
602 | * | |
603 | * Returns POLLIN | POLLRDNORM if data exists in the buffers, | |
604 | * zero otherwise. | |
605 | */ | |
606 | int ring_buffer_poll_wait(struct ring_buffer *buffer, int cpu, | |
607 | struct file *filp, poll_table *poll_table) | |
608 | { | |
609 | struct ring_buffer_per_cpu *cpu_buffer; | |
610 | struct rb_irq_work *work; | |
611 | ||
612 | if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) || | |
613 | (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu))) | |
614 | return POLLIN | POLLRDNORM; | |
615 | ||
616 | if (cpu == RING_BUFFER_ALL_CPUS) | |
617 | work = &buffer->irq_work; | |
618 | else { | |
619 | cpu_buffer = buffer->buffers[cpu]; | |
620 | work = &cpu_buffer->irq_work; | |
621 | } | |
622 | ||
623 | work->waiters_pending = true; | |
624 | poll_wait(filp, &work->waiters, poll_table); | |
625 | ||
626 | if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) || | |
627 | (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu))) | |
628 | return POLLIN | POLLRDNORM; | |
629 | return 0; | |
630 | } | |
631 | ||
f536aafc | 632 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
077c5407 SR |
633 | #define RB_WARN_ON(b, cond) \ |
634 | ({ \ | |
635 | int _____ret = unlikely(cond); \ | |
636 | if (_____ret) { \ | |
637 | if (__same_type(*(b), struct ring_buffer_per_cpu)) { \ | |
638 | struct ring_buffer_per_cpu *__b = \ | |
639 | (void *)b; \ | |
640 | atomic_inc(&__b->buffer->record_disabled); \ | |
641 | } else \ | |
642 | atomic_inc(&b->record_disabled); \ | |
643 | WARN_ON(1); \ | |
644 | } \ | |
645 | _____ret; \ | |
3e89c7bb | 646 | }) |
f536aafc | 647 | |
37886f6a SR |
648 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
649 | #define DEBUG_SHIFT 0 | |
650 | ||
6d3f1e12 | 651 | static inline u64 rb_time_stamp(struct ring_buffer *buffer) |
88eb0125 SR |
652 | { |
653 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
654 | return buffer->clock() << DEBUG_SHIFT; | |
655 | } | |
656 | ||
37886f6a SR |
657 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) |
658 | { | |
659 | u64 time; | |
660 | ||
661 | preempt_disable_notrace(); | |
6d3f1e12 | 662 | time = rb_time_stamp(buffer); |
37886f6a SR |
663 | preempt_enable_no_resched_notrace(); |
664 | ||
665 | return time; | |
666 | } | |
667 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
668 | ||
669 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
670 | int cpu, u64 *ts) | |
671 | { | |
672 | /* Just stupid testing the normalize function and deltas */ | |
673 | *ts >>= DEBUG_SHIFT; | |
674 | } | |
675 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
676 | ||
77ae365e SR |
677 | /* |
678 | * Making the ring buffer lockless makes things tricky. | |
679 | * Although writes only happen on the CPU that they are on, | |
680 | * and they only need to worry about interrupts. Reads can | |
681 | * happen on any CPU. | |
682 | * | |
683 | * The reader page is always off the ring buffer, but when the | |
684 | * reader finishes with a page, it needs to swap its page with | |
685 | * a new one from the buffer. The reader needs to take from | |
686 | * the head (writes go to the tail). But if a writer is in overwrite | |
687 | * mode and wraps, it must push the head page forward. | |
688 | * | |
689 | * Here lies the problem. | |
690 | * | |
691 | * The reader must be careful to replace only the head page, and | |
692 | * not another one. As described at the top of the file in the | |
693 | * ASCII art, the reader sets its old page to point to the next | |
694 | * page after head. It then sets the page after head to point to | |
695 | * the old reader page. But if the writer moves the head page | |
696 | * during this operation, the reader could end up with the tail. | |
697 | * | |
698 | * We use cmpxchg to help prevent this race. We also do something | |
699 | * special with the page before head. We set the LSB to 1. | |
700 | * | |
701 | * When the writer must push the page forward, it will clear the | |
702 | * bit that points to the head page, move the head, and then set | |
703 | * the bit that points to the new head page. | |
704 | * | |
705 | * We also don't want an interrupt coming in and moving the head | |
706 | * page on another writer. Thus we use the second LSB to catch | |
707 | * that too. Thus: | |
708 | * | |
709 | * head->list->prev->next bit 1 bit 0 | |
710 | * ------- ------- | |
711 | * Normal page 0 0 | |
712 | * Points to head page 0 1 | |
713 | * New head page 1 0 | |
714 | * | |
715 | * Note we can not trust the prev pointer of the head page, because: | |
716 | * | |
717 | * +----+ +-----+ +-----+ | |
718 | * | |------>| T |---X--->| N | | |
719 | * | |<------| | | | | |
720 | * +----+ +-----+ +-----+ | |
721 | * ^ ^ | | |
722 | * | +-----+ | | | |
723 | * +----------| R |----------+ | | |
724 | * | |<-----------+ | |
725 | * +-----+ | |
726 | * | |
727 | * Key: ---X--> HEAD flag set in pointer | |
728 | * T Tail page | |
729 | * R Reader page | |
730 | * N Next page | |
731 | * | |
732 | * (see __rb_reserve_next() to see where this happens) | |
733 | * | |
734 | * What the above shows is that the reader just swapped out | |
735 | * the reader page with a page in the buffer, but before it | |
736 | * could make the new header point back to the new page added | |
737 | * it was preempted by a writer. The writer moved forward onto | |
738 | * the new page added by the reader and is about to move forward | |
739 | * again. | |
740 | * | |
741 | * You can see, it is legitimate for the previous pointer of | |
742 | * the head (or any page) not to point back to itself. But only | |
743 | * temporarially. | |
744 | */ | |
745 | ||
746 | #define RB_PAGE_NORMAL 0UL | |
747 | #define RB_PAGE_HEAD 1UL | |
748 | #define RB_PAGE_UPDATE 2UL | |
749 | ||
750 | ||
751 | #define RB_FLAG_MASK 3UL | |
752 | ||
753 | /* PAGE_MOVED is not part of the mask */ | |
754 | #define RB_PAGE_MOVED 4UL | |
755 | ||
756 | /* | |
757 | * rb_list_head - remove any bit | |
758 | */ | |
759 | static struct list_head *rb_list_head(struct list_head *list) | |
760 | { | |
761 | unsigned long val = (unsigned long)list; | |
762 | ||
763 | return (struct list_head *)(val & ~RB_FLAG_MASK); | |
764 | } | |
765 | ||
766 | /* | |
6d3f1e12 | 767 | * rb_is_head_page - test if the given page is the head page |
77ae365e SR |
768 | * |
769 | * Because the reader may move the head_page pointer, we can | |
770 | * not trust what the head page is (it may be pointing to | |
771 | * the reader page). But if the next page is a header page, | |
772 | * its flags will be non zero. | |
773 | */ | |
42b16b3f | 774 | static inline int |
77ae365e SR |
775 | rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer, |
776 | struct buffer_page *page, struct list_head *list) | |
777 | { | |
778 | unsigned long val; | |
779 | ||
780 | val = (unsigned long)list->next; | |
781 | ||
782 | if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list) | |
783 | return RB_PAGE_MOVED; | |
784 | ||
785 | return val & RB_FLAG_MASK; | |
786 | } | |
787 | ||
788 | /* | |
789 | * rb_is_reader_page | |
790 | * | |
791 | * The unique thing about the reader page, is that, if the | |
792 | * writer is ever on it, the previous pointer never points | |
793 | * back to the reader page. | |
794 | */ | |
795 | static int rb_is_reader_page(struct buffer_page *page) | |
796 | { | |
797 | struct list_head *list = page->list.prev; | |
798 | ||
799 | return rb_list_head(list->next) != &page->list; | |
800 | } | |
801 | ||
802 | /* | |
803 | * rb_set_list_to_head - set a list_head to be pointing to head. | |
804 | */ | |
805 | static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer, | |
806 | struct list_head *list) | |
807 | { | |
808 | unsigned long *ptr; | |
809 | ||
810 | ptr = (unsigned long *)&list->next; | |
811 | *ptr |= RB_PAGE_HEAD; | |
812 | *ptr &= ~RB_PAGE_UPDATE; | |
813 | } | |
814 | ||
815 | /* | |
816 | * rb_head_page_activate - sets up head page | |
817 | */ | |
818 | static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer) | |
819 | { | |
820 | struct buffer_page *head; | |
821 | ||
822 | head = cpu_buffer->head_page; | |
823 | if (!head) | |
824 | return; | |
825 | ||
826 | /* | |
827 | * Set the previous list pointer to have the HEAD flag. | |
828 | */ | |
829 | rb_set_list_to_head(cpu_buffer, head->list.prev); | |
830 | } | |
831 | ||
832 | static void rb_list_head_clear(struct list_head *list) | |
833 | { | |
834 | unsigned long *ptr = (unsigned long *)&list->next; | |
835 | ||
836 | *ptr &= ~RB_FLAG_MASK; | |
837 | } | |
838 | ||
839 | /* | |
840 | * rb_head_page_dactivate - clears head page ptr (for free list) | |
841 | */ | |
842 | static void | |
843 | rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer) | |
844 | { | |
845 | struct list_head *hd; | |
846 | ||
847 | /* Go through the whole list and clear any pointers found. */ | |
848 | rb_list_head_clear(cpu_buffer->pages); | |
849 | ||
850 | list_for_each(hd, cpu_buffer->pages) | |
851 | rb_list_head_clear(hd); | |
852 | } | |
853 | ||
854 | static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer, | |
855 | struct buffer_page *head, | |
856 | struct buffer_page *prev, | |
857 | int old_flag, int new_flag) | |
858 | { | |
859 | struct list_head *list; | |
860 | unsigned long val = (unsigned long)&head->list; | |
861 | unsigned long ret; | |
862 | ||
863 | list = &prev->list; | |
864 | ||
865 | val &= ~RB_FLAG_MASK; | |
866 | ||
08a40816 SR |
867 | ret = cmpxchg((unsigned long *)&list->next, |
868 | val | old_flag, val | new_flag); | |
77ae365e SR |
869 | |
870 | /* check if the reader took the page */ | |
871 | if ((ret & ~RB_FLAG_MASK) != val) | |
872 | return RB_PAGE_MOVED; | |
873 | ||
874 | return ret & RB_FLAG_MASK; | |
875 | } | |
876 | ||
877 | static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer, | |
878 | struct buffer_page *head, | |
879 | struct buffer_page *prev, | |
880 | int old_flag) | |
881 | { | |
882 | return rb_head_page_set(cpu_buffer, head, prev, | |
883 | old_flag, RB_PAGE_UPDATE); | |
884 | } | |
885 | ||
886 | static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer, | |
887 | struct buffer_page *head, | |
888 | struct buffer_page *prev, | |
889 | int old_flag) | |
890 | { | |
891 | return rb_head_page_set(cpu_buffer, head, prev, | |
892 | old_flag, RB_PAGE_HEAD); | |
893 | } | |
894 | ||
895 | static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer, | |
896 | struct buffer_page *head, | |
897 | struct buffer_page *prev, | |
898 | int old_flag) | |
899 | { | |
900 | return rb_head_page_set(cpu_buffer, head, prev, | |
901 | old_flag, RB_PAGE_NORMAL); | |
902 | } | |
903 | ||
904 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
905 | struct buffer_page **bpage) | |
906 | { | |
907 | struct list_head *p = rb_list_head((*bpage)->list.next); | |
908 | ||
909 | *bpage = list_entry(p, struct buffer_page, list); | |
910 | } | |
911 | ||
912 | static struct buffer_page * | |
913 | rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer) | |
914 | { | |
915 | struct buffer_page *head; | |
916 | struct buffer_page *page; | |
917 | struct list_head *list; | |
918 | int i; | |
919 | ||
920 | if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page)) | |
921 | return NULL; | |
922 | ||
923 | /* sanity check */ | |
924 | list = cpu_buffer->pages; | |
925 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list)) | |
926 | return NULL; | |
927 | ||
928 | page = head = cpu_buffer->head_page; | |
929 | /* | |
930 | * It is possible that the writer moves the header behind | |
931 | * where we started, and we miss in one loop. | |
932 | * A second loop should grab the header, but we'll do | |
933 | * three loops just because I'm paranoid. | |
934 | */ | |
935 | for (i = 0; i < 3; i++) { | |
936 | do { | |
937 | if (rb_is_head_page(cpu_buffer, page, page->list.prev)) { | |
938 | cpu_buffer->head_page = page; | |
939 | return page; | |
940 | } | |
941 | rb_inc_page(cpu_buffer, &page); | |
942 | } while (page != head); | |
943 | } | |
944 | ||
945 | RB_WARN_ON(cpu_buffer, 1); | |
946 | ||
947 | return NULL; | |
948 | } | |
949 | ||
950 | static int rb_head_page_replace(struct buffer_page *old, | |
951 | struct buffer_page *new) | |
952 | { | |
953 | unsigned long *ptr = (unsigned long *)&old->list.prev->next; | |
954 | unsigned long val; | |
955 | unsigned long ret; | |
956 | ||
957 | val = *ptr & ~RB_FLAG_MASK; | |
958 | val |= RB_PAGE_HEAD; | |
959 | ||
08a40816 | 960 | ret = cmpxchg(ptr, val, (unsigned long)&new->list); |
77ae365e SR |
961 | |
962 | return ret == val; | |
963 | } | |
964 | ||
965 | /* | |
966 | * rb_tail_page_update - move the tail page forward | |
967 | * | |
968 | * Returns 1 if moved tail page, 0 if someone else did. | |
969 | */ | |
970 | static int rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer, | |
971 | struct buffer_page *tail_page, | |
972 | struct buffer_page *next_page) | |
973 | { | |
974 | struct buffer_page *old_tail; | |
975 | unsigned long old_entries; | |
976 | unsigned long old_write; | |
977 | int ret = 0; | |
978 | ||
979 | /* | |
980 | * The tail page now needs to be moved forward. | |
981 | * | |
982 | * We need to reset the tail page, but without messing | |
983 | * with possible erasing of data brought in by interrupts | |
984 | * that have moved the tail page and are currently on it. | |
985 | * | |
986 | * We add a counter to the write field to denote this. | |
987 | */ | |
988 | old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write); | |
989 | old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries); | |
990 | ||
991 | /* | |
992 | * Just make sure we have seen our old_write and synchronize | |
993 | * with any interrupts that come in. | |
994 | */ | |
995 | barrier(); | |
996 | ||
997 | /* | |
998 | * If the tail page is still the same as what we think | |
999 | * it is, then it is up to us to update the tail | |
1000 | * pointer. | |
1001 | */ | |
1002 | if (tail_page == cpu_buffer->tail_page) { | |
1003 | /* Zero the write counter */ | |
1004 | unsigned long val = old_write & ~RB_WRITE_MASK; | |
1005 | unsigned long eval = old_entries & ~RB_WRITE_MASK; | |
1006 | ||
1007 | /* | |
1008 | * This will only succeed if an interrupt did | |
1009 | * not come in and change it. In which case, we | |
1010 | * do not want to modify it. | |
da706d8b LJ |
1011 | * |
1012 | * We add (void) to let the compiler know that we do not care | |
1013 | * about the return value of these functions. We use the | |
1014 | * cmpxchg to only update if an interrupt did not already | |
1015 | * do it for us. If the cmpxchg fails, we don't care. | |
77ae365e | 1016 | */ |
da706d8b LJ |
1017 | (void)local_cmpxchg(&next_page->write, old_write, val); |
1018 | (void)local_cmpxchg(&next_page->entries, old_entries, eval); | |
77ae365e SR |
1019 | |
1020 | /* | |
1021 | * No need to worry about races with clearing out the commit. | |
1022 | * it only can increment when a commit takes place. But that | |
1023 | * only happens in the outer most nested commit. | |
1024 | */ | |
1025 | local_set(&next_page->page->commit, 0); | |
1026 | ||
1027 | old_tail = cmpxchg(&cpu_buffer->tail_page, | |
1028 | tail_page, next_page); | |
1029 | ||
1030 | if (old_tail == tail_page) | |
1031 | ret = 1; | |
1032 | } | |
1033 | ||
1034 | return ret; | |
1035 | } | |
1036 | ||
1037 | static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer, | |
1038 | struct buffer_page *bpage) | |
1039 | { | |
1040 | unsigned long val = (unsigned long)bpage; | |
1041 | ||
1042 | if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK)) | |
1043 | return 1; | |
1044 | ||
1045 | return 0; | |
1046 | } | |
1047 | ||
1048 | /** | |
1049 | * rb_check_list - make sure a pointer to a list has the last bits zero | |
1050 | */ | |
1051 | static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer, | |
1052 | struct list_head *list) | |
1053 | { | |
1054 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev)) | |
1055 | return 1; | |
1056 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next)) | |
1057 | return 1; | |
1058 | return 0; | |
1059 | } | |
1060 | ||
7a8e76a3 SR |
1061 | /** |
1062 | * check_pages - integrity check of buffer pages | |
1063 | * @cpu_buffer: CPU buffer with pages to test | |
1064 | * | |
c3706f00 | 1065 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
1066 | * been corrupted. |
1067 | */ | |
1068 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
1069 | { | |
3adc54fa | 1070 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1071 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1072 | |
308f7eeb SR |
1073 | /* Reset the head page if it exists */ |
1074 | if (cpu_buffer->head_page) | |
1075 | rb_set_head_page(cpu_buffer); | |
1076 | ||
77ae365e SR |
1077 | rb_head_page_deactivate(cpu_buffer); |
1078 | ||
3e89c7bb SR |
1079 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
1080 | return -1; | |
1081 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
1082 | return -1; | |
7a8e76a3 | 1083 | |
77ae365e SR |
1084 | if (rb_check_list(cpu_buffer, head)) |
1085 | return -1; | |
1086 | ||
044fa782 | 1087 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 1088 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 1089 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
1090 | return -1; |
1091 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 1092 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 1093 | return -1; |
77ae365e SR |
1094 | if (rb_check_list(cpu_buffer, &bpage->list)) |
1095 | return -1; | |
7a8e76a3 SR |
1096 | } |
1097 | ||
77ae365e SR |
1098 | rb_head_page_activate(cpu_buffer); |
1099 | ||
7a8e76a3 SR |
1100 | return 0; |
1101 | } | |
1102 | ||
438ced17 | 1103 | static int __rb_allocate_pages(int nr_pages, struct list_head *pages, int cpu) |
7a8e76a3 | 1104 | { |
438ced17 | 1105 | int i; |
044fa782 | 1106 | struct buffer_page *bpage, *tmp; |
3adc54fa | 1107 | |
7a8e76a3 | 1108 | for (i = 0; i < nr_pages; i++) { |
7ea59064 | 1109 | struct page *page; |
d7ec4bfe VN |
1110 | /* |
1111 | * __GFP_NORETRY flag makes sure that the allocation fails | |
1112 | * gracefully without invoking oom-killer and the system is | |
1113 | * not destabilized. | |
1114 | */ | |
044fa782 | 1115 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
d7ec4bfe | 1116 | GFP_KERNEL | __GFP_NORETRY, |
438ced17 | 1117 | cpu_to_node(cpu)); |
044fa782 | 1118 | if (!bpage) |
e4c2ce82 | 1119 | goto free_pages; |
77ae365e | 1120 | |
438ced17 | 1121 | list_add(&bpage->list, pages); |
77ae365e | 1122 | |
438ced17 | 1123 | page = alloc_pages_node(cpu_to_node(cpu), |
d7ec4bfe | 1124 | GFP_KERNEL | __GFP_NORETRY, 0); |
7ea59064 | 1125 | if (!page) |
7a8e76a3 | 1126 | goto free_pages; |
7ea59064 | 1127 | bpage->page = page_address(page); |
044fa782 | 1128 | rb_init_page(bpage->page); |
7a8e76a3 SR |
1129 | } |
1130 | ||
438ced17 VN |
1131 | return 0; |
1132 | ||
1133 | free_pages: | |
1134 | list_for_each_entry_safe(bpage, tmp, pages, list) { | |
1135 | list_del_init(&bpage->list); | |
1136 | free_buffer_page(bpage); | |
1137 | } | |
1138 | ||
1139 | return -ENOMEM; | |
1140 | } | |
1141 | ||
1142 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
1143 | unsigned nr_pages) | |
1144 | { | |
1145 | LIST_HEAD(pages); | |
1146 | ||
1147 | WARN_ON(!nr_pages); | |
1148 | ||
1149 | if (__rb_allocate_pages(nr_pages, &pages, cpu_buffer->cpu)) | |
1150 | return -ENOMEM; | |
1151 | ||
3adc54fa SR |
1152 | /* |
1153 | * The ring buffer page list is a circular list that does not | |
1154 | * start and end with a list head. All page list items point to | |
1155 | * other pages. | |
1156 | */ | |
1157 | cpu_buffer->pages = pages.next; | |
1158 | list_del(&pages); | |
7a8e76a3 | 1159 | |
438ced17 VN |
1160 | cpu_buffer->nr_pages = nr_pages; |
1161 | ||
7a8e76a3 SR |
1162 | rb_check_pages(cpu_buffer); |
1163 | ||
1164 | return 0; | |
7a8e76a3 SR |
1165 | } |
1166 | ||
1167 | static struct ring_buffer_per_cpu * | |
438ced17 | 1168 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int nr_pages, int cpu) |
7a8e76a3 SR |
1169 | { |
1170 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 1171 | struct buffer_page *bpage; |
7ea59064 | 1172 | struct page *page; |
7a8e76a3 SR |
1173 | int ret; |
1174 | ||
1175 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
1176 | GFP_KERNEL, cpu_to_node(cpu)); | |
1177 | if (!cpu_buffer) | |
1178 | return NULL; | |
1179 | ||
1180 | cpu_buffer->cpu = cpu; | |
1181 | cpu_buffer->buffer = buffer; | |
5389f6fa | 1182 | raw_spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 1183 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
edc35bd7 | 1184 | cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
83f40318 | 1185 | INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler); |
05fdd70d | 1186 | init_completion(&cpu_buffer->update_done); |
15693458 | 1187 | init_irq_work(&cpu_buffer->irq_work.work, rb_wake_up_waiters); |
f1dc6725 | 1188 | init_waitqueue_head(&cpu_buffer->irq_work.waiters); |
7a8e76a3 | 1189 | |
044fa782 | 1190 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 1191 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 1192 | if (!bpage) |
e4c2ce82 SR |
1193 | goto fail_free_buffer; |
1194 | ||
77ae365e SR |
1195 | rb_check_bpage(cpu_buffer, bpage); |
1196 | ||
044fa782 | 1197 | cpu_buffer->reader_page = bpage; |
7ea59064 VN |
1198 | page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0); |
1199 | if (!page) | |
e4c2ce82 | 1200 | goto fail_free_reader; |
7ea59064 | 1201 | bpage->page = page_address(page); |
044fa782 | 1202 | rb_init_page(bpage->page); |
e4c2ce82 | 1203 | |
d769041f | 1204 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
44b99462 | 1205 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
d769041f | 1206 | |
438ced17 | 1207 | ret = rb_allocate_pages(cpu_buffer, nr_pages); |
7a8e76a3 | 1208 | if (ret < 0) |
d769041f | 1209 | goto fail_free_reader; |
7a8e76a3 SR |
1210 | |
1211 | cpu_buffer->head_page | |
3adc54fa | 1212 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 1213 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 | 1214 | |
77ae365e SR |
1215 | rb_head_page_activate(cpu_buffer); |
1216 | ||
7a8e76a3 SR |
1217 | return cpu_buffer; |
1218 | ||
d769041f SR |
1219 | fail_free_reader: |
1220 | free_buffer_page(cpu_buffer->reader_page); | |
1221 | ||
7a8e76a3 SR |
1222 | fail_free_buffer: |
1223 | kfree(cpu_buffer); | |
1224 | return NULL; | |
1225 | } | |
1226 | ||
1227 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
1228 | { | |
3adc54fa | 1229 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1230 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1231 | |
d769041f SR |
1232 | free_buffer_page(cpu_buffer->reader_page); |
1233 | ||
77ae365e SR |
1234 | rb_head_page_deactivate(cpu_buffer); |
1235 | ||
3adc54fa SR |
1236 | if (head) { |
1237 | list_for_each_entry_safe(bpage, tmp, head, list) { | |
1238 | list_del_init(&bpage->list); | |
1239 | free_buffer_page(bpage); | |
1240 | } | |
1241 | bpage = list_entry(head, struct buffer_page, list); | |
044fa782 | 1242 | free_buffer_page(bpage); |
7a8e76a3 | 1243 | } |
3adc54fa | 1244 | |
7a8e76a3 SR |
1245 | kfree(cpu_buffer); |
1246 | } | |
1247 | ||
59222efe | 1248 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
1249 | static int rb_cpu_notify(struct notifier_block *self, |
1250 | unsigned long action, void *hcpu); | |
554f786e SR |
1251 | #endif |
1252 | ||
7a8e76a3 SR |
1253 | /** |
1254 | * ring_buffer_alloc - allocate a new ring_buffer | |
68814b58 | 1255 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
1256 | * @flags: attributes to set for the ring buffer. |
1257 | * | |
1258 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
1259 | * flag. This flag means that the buffer will overwrite old data | |
1260 | * when the buffer wraps. If this flag is not set, the buffer will | |
1261 | * drop data when the tail hits the head. | |
1262 | */ | |
1f8a6a10 PZ |
1263 | struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
1264 | struct lock_class_key *key) | |
7a8e76a3 SR |
1265 | { |
1266 | struct ring_buffer *buffer; | |
1267 | int bsize; | |
438ced17 | 1268 | int cpu, nr_pages; |
7a8e76a3 SR |
1269 | |
1270 | /* keep it in its own cache line */ | |
1271 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
1272 | GFP_KERNEL); | |
1273 | if (!buffer) | |
1274 | return NULL; | |
1275 | ||
9e01c1b7 RR |
1276 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
1277 | goto fail_free_buffer; | |
1278 | ||
438ced17 | 1279 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
7a8e76a3 | 1280 | buffer->flags = flags; |
37886f6a | 1281 | buffer->clock = trace_clock_local; |
1f8a6a10 | 1282 | buffer->reader_lock_key = key; |
7a8e76a3 | 1283 | |
15693458 | 1284 | init_irq_work(&buffer->irq_work.work, rb_wake_up_waiters); |
f1dc6725 | 1285 | init_waitqueue_head(&buffer->irq_work.waiters); |
15693458 | 1286 | |
7a8e76a3 | 1287 | /* need at least two pages */ |
438ced17 VN |
1288 | if (nr_pages < 2) |
1289 | nr_pages = 2; | |
7a8e76a3 | 1290 | |
3bf832ce FW |
1291 | /* |
1292 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
1293 | * in early initcall, it will not be notified of secondary cpus. | |
1294 | * In that off case, we need to allocate for all possible cpus. | |
1295 | */ | |
1296 | #ifdef CONFIG_HOTPLUG_CPU | |
554f786e SR |
1297 | get_online_cpus(); |
1298 | cpumask_copy(buffer->cpumask, cpu_online_mask); | |
3bf832ce FW |
1299 | #else |
1300 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
1301 | #endif | |
7a8e76a3 SR |
1302 | buffer->cpus = nr_cpu_ids; |
1303 | ||
1304 | bsize = sizeof(void *) * nr_cpu_ids; | |
1305 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
1306 | GFP_KERNEL); | |
1307 | if (!buffer->buffers) | |
9e01c1b7 | 1308 | goto fail_free_cpumask; |
7a8e76a3 SR |
1309 | |
1310 | for_each_buffer_cpu(buffer, cpu) { | |
1311 | buffer->buffers[cpu] = | |
438ced17 | 1312 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
7a8e76a3 SR |
1313 | if (!buffer->buffers[cpu]) |
1314 | goto fail_free_buffers; | |
1315 | } | |
1316 | ||
59222efe | 1317 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1318 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
1319 | buffer->cpu_notify.priority = 0; | |
1320 | register_cpu_notifier(&buffer->cpu_notify); | |
1321 | #endif | |
1322 | ||
1323 | put_online_cpus(); | |
7a8e76a3 SR |
1324 | mutex_init(&buffer->mutex); |
1325 | ||
1326 | return buffer; | |
1327 | ||
1328 | fail_free_buffers: | |
1329 | for_each_buffer_cpu(buffer, cpu) { | |
1330 | if (buffer->buffers[cpu]) | |
1331 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1332 | } | |
1333 | kfree(buffer->buffers); | |
1334 | ||
9e01c1b7 RR |
1335 | fail_free_cpumask: |
1336 | free_cpumask_var(buffer->cpumask); | |
554f786e | 1337 | put_online_cpus(); |
9e01c1b7 | 1338 | |
7a8e76a3 SR |
1339 | fail_free_buffer: |
1340 | kfree(buffer); | |
1341 | return NULL; | |
1342 | } | |
1f8a6a10 | 1343 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
1344 | |
1345 | /** | |
1346 | * ring_buffer_free - free a ring buffer. | |
1347 | * @buffer: the buffer to free. | |
1348 | */ | |
1349 | void | |
1350 | ring_buffer_free(struct ring_buffer *buffer) | |
1351 | { | |
1352 | int cpu; | |
1353 | ||
554f786e SR |
1354 | get_online_cpus(); |
1355 | ||
59222efe | 1356 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1357 | unregister_cpu_notifier(&buffer->cpu_notify); |
1358 | #endif | |
1359 | ||
7a8e76a3 SR |
1360 | for_each_buffer_cpu(buffer, cpu) |
1361 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1362 | ||
554f786e SR |
1363 | put_online_cpus(); |
1364 | ||
bd3f0221 | 1365 | kfree(buffer->buffers); |
9e01c1b7 RR |
1366 | free_cpumask_var(buffer->cpumask); |
1367 | ||
7a8e76a3 SR |
1368 | kfree(buffer); |
1369 | } | |
c4f50183 | 1370 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 1371 | |
37886f6a SR |
1372 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
1373 | u64 (*clock)(void)) | |
1374 | { | |
1375 | buffer->clock = clock; | |
1376 | } | |
1377 | ||
7a8e76a3 SR |
1378 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
1379 | ||
83f40318 VN |
1380 | static inline unsigned long rb_page_entries(struct buffer_page *bpage) |
1381 | { | |
1382 | return local_read(&bpage->entries) & RB_WRITE_MASK; | |
1383 | } | |
1384 | ||
1385 | static inline unsigned long rb_page_write(struct buffer_page *bpage) | |
1386 | { | |
1387 | return local_read(&bpage->write) & RB_WRITE_MASK; | |
1388 | } | |
1389 | ||
5040b4b7 | 1390 | static int |
83f40318 | 1391 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned int nr_pages) |
7a8e76a3 | 1392 | { |
83f40318 VN |
1393 | struct list_head *tail_page, *to_remove, *next_page; |
1394 | struct buffer_page *to_remove_page, *tmp_iter_page; | |
1395 | struct buffer_page *last_page, *first_page; | |
1396 | unsigned int nr_removed; | |
1397 | unsigned long head_bit; | |
1398 | int page_entries; | |
1399 | ||
1400 | head_bit = 0; | |
7a8e76a3 | 1401 | |
5389f6fa | 1402 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1403 | atomic_inc(&cpu_buffer->record_disabled); |
1404 | /* | |
1405 | * We don't race with the readers since we have acquired the reader | |
1406 | * lock. We also don't race with writers after disabling recording. | |
1407 | * This makes it easy to figure out the first and the last page to be | |
1408 | * removed from the list. We unlink all the pages in between including | |
1409 | * the first and last pages. This is done in a busy loop so that we | |
1410 | * lose the least number of traces. | |
1411 | * The pages are freed after we restart recording and unlock readers. | |
1412 | */ | |
1413 | tail_page = &cpu_buffer->tail_page->list; | |
77ae365e | 1414 | |
83f40318 VN |
1415 | /* |
1416 | * tail page might be on reader page, we remove the next page | |
1417 | * from the ring buffer | |
1418 | */ | |
1419 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
1420 | tail_page = rb_list_head(tail_page->next); | |
1421 | to_remove = tail_page; | |
1422 | ||
1423 | /* start of pages to remove */ | |
1424 | first_page = list_entry(rb_list_head(to_remove->next), | |
1425 | struct buffer_page, list); | |
1426 | ||
1427 | for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) { | |
1428 | to_remove = rb_list_head(to_remove)->next; | |
1429 | head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD; | |
7a8e76a3 | 1430 | } |
7a8e76a3 | 1431 | |
83f40318 | 1432 | next_page = rb_list_head(to_remove)->next; |
7a8e76a3 | 1433 | |
83f40318 VN |
1434 | /* |
1435 | * Now we remove all pages between tail_page and next_page. | |
1436 | * Make sure that we have head_bit value preserved for the | |
1437 | * next page | |
1438 | */ | |
1439 | tail_page->next = (struct list_head *)((unsigned long)next_page | | |
1440 | head_bit); | |
1441 | next_page = rb_list_head(next_page); | |
1442 | next_page->prev = tail_page; | |
1443 | ||
1444 | /* make sure pages points to a valid page in the ring buffer */ | |
1445 | cpu_buffer->pages = next_page; | |
1446 | ||
1447 | /* update head page */ | |
1448 | if (head_bit) | |
1449 | cpu_buffer->head_page = list_entry(next_page, | |
1450 | struct buffer_page, list); | |
1451 | ||
1452 | /* | |
1453 | * change read pointer to make sure any read iterators reset | |
1454 | * themselves | |
1455 | */ | |
1456 | cpu_buffer->read = 0; | |
1457 | ||
1458 | /* pages are removed, resume tracing and then free the pages */ | |
1459 | atomic_dec(&cpu_buffer->record_disabled); | |
5389f6fa | 1460 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1461 | |
1462 | RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)); | |
1463 | ||
1464 | /* last buffer page to remove */ | |
1465 | last_page = list_entry(rb_list_head(to_remove), struct buffer_page, | |
1466 | list); | |
1467 | tmp_iter_page = first_page; | |
1468 | ||
1469 | do { | |
1470 | to_remove_page = tmp_iter_page; | |
1471 | rb_inc_page(cpu_buffer, &tmp_iter_page); | |
1472 | ||
1473 | /* update the counters */ | |
1474 | page_entries = rb_page_entries(to_remove_page); | |
1475 | if (page_entries) { | |
1476 | /* | |
1477 | * If something was added to this page, it was full | |
1478 | * since it is not the tail page. So we deduct the | |
1479 | * bytes consumed in ring buffer from here. | |
48fdc72f | 1480 | * Increment overrun to account for the lost events. |
83f40318 | 1481 | */ |
48fdc72f | 1482 | local_add(page_entries, &cpu_buffer->overrun); |
83f40318 VN |
1483 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
1484 | } | |
1485 | ||
1486 | /* | |
1487 | * We have already removed references to this list item, just | |
1488 | * free up the buffer_page and its page | |
1489 | */ | |
1490 | free_buffer_page(to_remove_page); | |
1491 | nr_removed--; | |
1492 | ||
1493 | } while (to_remove_page != last_page); | |
1494 | ||
1495 | RB_WARN_ON(cpu_buffer, nr_removed); | |
5040b4b7 VN |
1496 | |
1497 | return nr_removed == 0; | |
7a8e76a3 SR |
1498 | } |
1499 | ||
5040b4b7 VN |
1500 | static int |
1501 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 1502 | { |
5040b4b7 VN |
1503 | struct list_head *pages = &cpu_buffer->new_pages; |
1504 | int retries, success; | |
7a8e76a3 | 1505 | |
5389f6fa | 1506 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1507 | /* |
1508 | * We are holding the reader lock, so the reader page won't be swapped | |
1509 | * in the ring buffer. Now we are racing with the writer trying to | |
1510 | * move head page and the tail page. | |
1511 | * We are going to adapt the reader page update process where: | |
1512 | * 1. We first splice the start and end of list of new pages between | |
1513 | * the head page and its previous page. | |
1514 | * 2. We cmpxchg the prev_page->next to point from head page to the | |
1515 | * start of new pages list. | |
1516 | * 3. Finally, we update the head->prev to the end of new list. | |
1517 | * | |
1518 | * We will try this process 10 times, to make sure that we don't keep | |
1519 | * spinning. | |
1520 | */ | |
1521 | retries = 10; | |
1522 | success = 0; | |
1523 | while (retries--) { | |
1524 | struct list_head *head_page, *prev_page, *r; | |
1525 | struct list_head *last_page, *first_page; | |
1526 | struct list_head *head_page_with_bit; | |
77ae365e | 1527 | |
5040b4b7 | 1528 | head_page = &rb_set_head_page(cpu_buffer)->list; |
54f7be5b SR |
1529 | if (!head_page) |
1530 | break; | |
5040b4b7 VN |
1531 | prev_page = head_page->prev; |
1532 | ||
1533 | first_page = pages->next; | |
1534 | last_page = pages->prev; | |
1535 | ||
1536 | head_page_with_bit = (struct list_head *) | |
1537 | ((unsigned long)head_page | RB_PAGE_HEAD); | |
1538 | ||
1539 | last_page->next = head_page_with_bit; | |
1540 | first_page->prev = prev_page; | |
1541 | ||
1542 | r = cmpxchg(&prev_page->next, head_page_with_bit, first_page); | |
1543 | ||
1544 | if (r == head_page_with_bit) { | |
1545 | /* | |
1546 | * yay, we replaced the page pointer to our new list, | |
1547 | * now, we just have to update to head page's prev | |
1548 | * pointer to point to end of list | |
1549 | */ | |
1550 | head_page->prev = last_page; | |
1551 | success = 1; | |
1552 | break; | |
1553 | } | |
7a8e76a3 | 1554 | } |
7a8e76a3 | 1555 | |
5040b4b7 VN |
1556 | if (success) |
1557 | INIT_LIST_HEAD(pages); | |
1558 | /* | |
1559 | * If we weren't successful in adding in new pages, warn and stop | |
1560 | * tracing | |
1561 | */ | |
1562 | RB_WARN_ON(cpu_buffer, !success); | |
5389f6fa | 1563 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1564 | |
1565 | /* free pages if they weren't inserted */ | |
1566 | if (!success) { | |
1567 | struct buffer_page *bpage, *tmp; | |
1568 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, | |
1569 | list) { | |
1570 | list_del_init(&bpage->list); | |
1571 | free_buffer_page(bpage); | |
1572 | } | |
1573 | } | |
1574 | return success; | |
7a8e76a3 SR |
1575 | } |
1576 | ||
83f40318 | 1577 | static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer) |
438ced17 | 1578 | { |
5040b4b7 VN |
1579 | int success; |
1580 | ||
438ced17 | 1581 | if (cpu_buffer->nr_pages_to_update > 0) |
5040b4b7 | 1582 | success = rb_insert_pages(cpu_buffer); |
438ced17 | 1583 | else |
5040b4b7 VN |
1584 | success = rb_remove_pages(cpu_buffer, |
1585 | -cpu_buffer->nr_pages_to_update); | |
83f40318 | 1586 | |
5040b4b7 VN |
1587 | if (success) |
1588 | cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update; | |
83f40318 VN |
1589 | } |
1590 | ||
1591 | static void update_pages_handler(struct work_struct *work) | |
1592 | { | |
1593 | struct ring_buffer_per_cpu *cpu_buffer = container_of(work, | |
1594 | struct ring_buffer_per_cpu, update_pages_work); | |
1595 | rb_update_pages(cpu_buffer); | |
05fdd70d | 1596 | complete(&cpu_buffer->update_done); |
438ced17 VN |
1597 | } |
1598 | ||
7a8e76a3 SR |
1599 | /** |
1600 | * ring_buffer_resize - resize the ring buffer | |
1601 | * @buffer: the buffer to resize. | |
1602 | * @size: the new size. | |
1603 | * | |
7a8e76a3 SR |
1604 | * Minimum size is 2 * BUF_PAGE_SIZE. |
1605 | * | |
83f40318 | 1606 | * Returns 0 on success and < 0 on failure. |
7a8e76a3 | 1607 | */ |
438ced17 VN |
1608 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size, |
1609 | int cpu_id) | |
7a8e76a3 SR |
1610 | { |
1611 | struct ring_buffer_per_cpu *cpu_buffer; | |
438ced17 | 1612 | unsigned nr_pages; |
83f40318 | 1613 | int cpu, err = 0; |
7a8e76a3 | 1614 | |
ee51a1de IM |
1615 | /* |
1616 | * Always succeed at resizing a non-existent buffer: | |
1617 | */ | |
1618 | if (!buffer) | |
1619 | return size; | |
1620 | ||
6a31e1f1 SR |
1621 | /* Make sure the requested buffer exists */ |
1622 | if (cpu_id != RING_BUFFER_ALL_CPUS && | |
1623 | !cpumask_test_cpu(cpu_id, buffer->cpumask)) | |
1624 | return size; | |
1625 | ||
7a8e76a3 SR |
1626 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
1627 | size *= BUF_PAGE_SIZE; | |
7a8e76a3 SR |
1628 | |
1629 | /* we need a minimum of two pages */ | |
1630 | if (size < BUF_PAGE_SIZE * 2) | |
1631 | size = BUF_PAGE_SIZE * 2; | |
1632 | ||
83f40318 | 1633 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
18421015 | 1634 | |
83f40318 VN |
1635 | /* |
1636 | * Don't succeed if resizing is disabled, as a reader might be | |
1637 | * manipulating the ring buffer and is expecting a sane state while | |
1638 | * this is true. | |
1639 | */ | |
1640 | if (atomic_read(&buffer->resize_disabled)) | |
1641 | return -EBUSY; | |
18421015 | 1642 | |
83f40318 | 1643 | /* prevent another thread from changing buffer sizes */ |
7a8e76a3 | 1644 | mutex_lock(&buffer->mutex); |
7a8e76a3 | 1645 | |
438ced17 VN |
1646 | if (cpu_id == RING_BUFFER_ALL_CPUS) { |
1647 | /* calculate the pages to update */ | |
7a8e76a3 SR |
1648 | for_each_buffer_cpu(buffer, cpu) { |
1649 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 1650 | |
438ced17 VN |
1651 | cpu_buffer->nr_pages_to_update = nr_pages - |
1652 | cpu_buffer->nr_pages; | |
438ced17 VN |
1653 | /* |
1654 | * nothing more to do for removing pages or no update | |
1655 | */ | |
1656 | if (cpu_buffer->nr_pages_to_update <= 0) | |
1657 | continue; | |
d7ec4bfe | 1658 | /* |
438ced17 VN |
1659 | * to add pages, make sure all new pages can be |
1660 | * allocated without receiving ENOMEM | |
d7ec4bfe | 1661 | */ |
438ced17 VN |
1662 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
1663 | if (__rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 | 1664 | &cpu_buffer->new_pages, cpu)) { |
438ced17 | 1665 | /* not enough memory for new pages */ |
83f40318 VN |
1666 | err = -ENOMEM; |
1667 | goto out_err; | |
1668 | } | |
1669 | } | |
1670 | ||
1671 | get_online_cpus(); | |
1672 | /* | |
1673 | * Fire off all the required work handlers | |
05fdd70d | 1674 | * We can't schedule on offline CPUs, but it's not necessary |
83f40318 VN |
1675 | * since we can change their buffer sizes without any race. |
1676 | */ | |
1677 | for_each_buffer_cpu(buffer, cpu) { | |
1678 | cpu_buffer = buffer->buffers[cpu]; | |
05fdd70d | 1679 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
1680 | continue; |
1681 | ||
05fdd70d VN |
1682 | if (cpu_online(cpu)) |
1683 | schedule_work_on(cpu, | |
1684 | &cpu_buffer->update_pages_work); | |
1685 | else | |
1686 | rb_update_pages(cpu_buffer); | |
7a8e76a3 | 1687 | } |
7a8e76a3 | 1688 | |
438ced17 VN |
1689 | /* wait for all the updates to complete */ |
1690 | for_each_buffer_cpu(buffer, cpu) { | |
1691 | cpu_buffer = buffer->buffers[cpu]; | |
05fdd70d | 1692 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
1693 | continue; |
1694 | ||
05fdd70d VN |
1695 | if (cpu_online(cpu)) |
1696 | wait_for_completion(&cpu_buffer->update_done); | |
83f40318 | 1697 | cpu_buffer->nr_pages_to_update = 0; |
438ced17 | 1698 | } |
83f40318 VN |
1699 | |
1700 | put_online_cpus(); | |
438ced17 | 1701 | } else { |
8e49f418 VN |
1702 | /* Make sure this CPU has been intitialized */ |
1703 | if (!cpumask_test_cpu(cpu_id, buffer->cpumask)) | |
1704 | goto out; | |
1705 | ||
438ced17 | 1706 | cpu_buffer = buffer->buffers[cpu_id]; |
83f40318 | 1707 | |
438ced17 VN |
1708 | if (nr_pages == cpu_buffer->nr_pages) |
1709 | goto out; | |
7a8e76a3 | 1710 | |
438ced17 VN |
1711 | cpu_buffer->nr_pages_to_update = nr_pages - |
1712 | cpu_buffer->nr_pages; | |
1713 | ||
1714 | INIT_LIST_HEAD(&cpu_buffer->new_pages); | |
1715 | if (cpu_buffer->nr_pages_to_update > 0 && | |
1716 | __rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 VN |
1717 | &cpu_buffer->new_pages, cpu_id)) { |
1718 | err = -ENOMEM; | |
1719 | goto out_err; | |
1720 | } | |
438ced17 | 1721 | |
83f40318 VN |
1722 | get_online_cpus(); |
1723 | ||
1724 | if (cpu_online(cpu_id)) { | |
1725 | schedule_work_on(cpu_id, | |
1726 | &cpu_buffer->update_pages_work); | |
05fdd70d | 1727 | wait_for_completion(&cpu_buffer->update_done); |
83f40318 VN |
1728 | } else |
1729 | rb_update_pages(cpu_buffer); | |
1730 | ||
83f40318 | 1731 | cpu_buffer->nr_pages_to_update = 0; |
05fdd70d | 1732 | put_online_cpus(); |
438ced17 | 1733 | } |
7a8e76a3 SR |
1734 | |
1735 | out: | |
659f451f SR |
1736 | /* |
1737 | * The ring buffer resize can happen with the ring buffer | |
1738 | * enabled, so that the update disturbs the tracing as little | |
1739 | * as possible. But if the buffer is disabled, we do not need | |
1740 | * to worry about that, and we can take the time to verify | |
1741 | * that the buffer is not corrupt. | |
1742 | */ | |
1743 | if (atomic_read(&buffer->record_disabled)) { | |
1744 | atomic_inc(&buffer->record_disabled); | |
1745 | /* | |
1746 | * Even though the buffer was disabled, we must make sure | |
1747 | * that it is truly disabled before calling rb_check_pages. | |
1748 | * There could have been a race between checking | |
1749 | * record_disable and incrementing it. | |
1750 | */ | |
1751 | synchronize_sched(); | |
1752 | for_each_buffer_cpu(buffer, cpu) { | |
1753 | cpu_buffer = buffer->buffers[cpu]; | |
1754 | rb_check_pages(cpu_buffer); | |
1755 | } | |
1756 | atomic_dec(&buffer->record_disabled); | |
1757 | } | |
1758 | ||
7a8e76a3 | 1759 | mutex_unlock(&buffer->mutex); |
7a8e76a3 SR |
1760 | return size; |
1761 | ||
83f40318 | 1762 | out_err: |
438ced17 VN |
1763 | for_each_buffer_cpu(buffer, cpu) { |
1764 | struct buffer_page *bpage, *tmp; | |
83f40318 | 1765 | |
438ced17 | 1766 | cpu_buffer = buffer->buffers[cpu]; |
438ced17 | 1767 | cpu_buffer->nr_pages_to_update = 0; |
83f40318 | 1768 | |
438ced17 VN |
1769 | if (list_empty(&cpu_buffer->new_pages)) |
1770 | continue; | |
83f40318 | 1771 | |
438ced17 VN |
1772 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, |
1773 | list) { | |
1774 | list_del_init(&bpage->list); | |
1775 | free_buffer_page(bpage); | |
1776 | } | |
7a8e76a3 | 1777 | } |
641d2f63 | 1778 | mutex_unlock(&buffer->mutex); |
83f40318 | 1779 | return err; |
7a8e76a3 | 1780 | } |
c4f50183 | 1781 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 1782 | |
750912fa DS |
1783 | void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val) |
1784 | { | |
1785 | mutex_lock(&buffer->mutex); | |
1786 | if (val) | |
1787 | buffer->flags |= RB_FL_OVERWRITE; | |
1788 | else | |
1789 | buffer->flags &= ~RB_FL_OVERWRITE; | |
1790 | mutex_unlock(&buffer->mutex); | |
1791 | } | |
1792 | EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite); | |
1793 | ||
8789a9e7 | 1794 | static inline void * |
044fa782 | 1795 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 1796 | { |
044fa782 | 1797 | return bpage->data + index; |
8789a9e7 SR |
1798 | } |
1799 | ||
044fa782 | 1800 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 1801 | { |
044fa782 | 1802 | return bpage->page->data + index; |
7a8e76a3 SR |
1803 | } |
1804 | ||
1805 | static inline struct ring_buffer_event * | |
d769041f | 1806 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1807 | { |
6f807acd SR |
1808 | return __rb_page_index(cpu_buffer->reader_page, |
1809 | cpu_buffer->reader_page->read); | |
1810 | } | |
1811 | ||
7a8e76a3 SR |
1812 | static inline struct ring_buffer_event * |
1813 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
1814 | { | |
6f807acd | 1815 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
1816 | } |
1817 | ||
bf41a158 SR |
1818 | static inline unsigned rb_page_commit(struct buffer_page *bpage) |
1819 | { | |
abc9b56d | 1820 | return local_read(&bpage->page->commit); |
bf41a158 SR |
1821 | } |
1822 | ||
25985edc | 1823 | /* Size is determined by what has been committed */ |
bf41a158 SR |
1824 | static inline unsigned rb_page_size(struct buffer_page *bpage) |
1825 | { | |
1826 | return rb_page_commit(bpage); | |
1827 | } | |
1828 | ||
1829 | static inline unsigned | |
1830 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
1831 | { | |
1832 | return rb_page_commit(cpu_buffer->commit_page); | |
1833 | } | |
1834 | ||
bf41a158 SR |
1835 | static inline unsigned |
1836 | rb_event_index(struct ring_buffer_event *event) | |
1837 | { | |
1838 | unsigned long addr = (unsigned long)event; | |
1839 | ||
22f470f8 | 1840 | return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE; |
bf41a158 SR |
1841 | } |
1842 | ||
0f0c85fc | 1843 | static inline int |
fa743953 SR |
1844 | rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1845 | struct ring_buffer_event *event) | |
bf41a158 SR |
1846 | { |
1847 | unsigned long addr = (unsigned long)event; | |
1848 | unsigned long index; | |
1849 | ||
1850 | index = rb_event_index(event); | |
1851 | addr &= PAGE_MASK; | |
1852 | ||
1853 | return cpu_buffer->commit_page->page == (void *)addr && | |
1854 | rb_commit_index(cpu_buffer) == index; | |
1855 | } | |
1856 | ||
34a148bf | 1857 | static void |
bf41a158 | 1858 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1859 | { |
77ae365e SR |
1860 | unsigned long max_count; |
1861 | ||
bf41a158 SR |
1862 | /* |
1863 | * We only race with interrupts and NMIs on this CPU. | |
1864 | * If we own the commit event, then we can commit | |
1865 | * all others that interrupted us, since the interruptions | |
1866 | * are in stack format (they finish before they come | |
1867 | * back to us). This allows us to do a simple loop to | |
1868 | * assign the commit to the tail. | |
1869 | */ | |
a8ccf1d6 | 1870 | again: |
438ced17 | 1871 | max_count = cpu_buffer->nr_pages * 100; |
77ae365e | 1872 | |
bf41a158 | 1873 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
77ae365e SR |
1874 | if (RB_WARN_ON(cpu_buffer, !(--max_count))) |
1875 | return; | |
1876 | if (RB_WARN_ON(cpu_buffer, | |
1877 | rb_is_reader_page(cpu_buffer->tail_page))) | |
1878 | return; | |
1879 | local_set(&cpu_buffer->commit_page->page->commit, | |
1880 | rb_page_write(cpu_buffer->commit_page)); | |
bf41a158 | 1881 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); |
abc9b56d SR |
1882 | cpu_buffer->write_stamp = |
1883 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1884 | /* add barrier to keep gcc from optimizing too much */ |
1885 | barrier(); | |
1886 | } | |
1887 | while (rb_commit_index(cpu_buffer) != | |
1888 | rb_page_write(cpu_buffer->commit_page)) { | |
77ae365e SR |
1889 | |
1890 | local_set(&cpu_buffer->commit_page->page->commit, | |
1891 | rb_page_write(cpu_buffer->commit_page)); | |
1892 | RB_WARN_ON(cpu_buffer, | |
1893 | local_read(&cpu_buffer->commit_page->page->commit) & | |
1894 | ~RB_WRITE_MASK); | |
bf41a158 SR |
1895 | barrier(); |
1896 | } | |
a8ccf1d6 SR |
1897 | |
1898 | /* again, keep gcc from optimizing */ | |
1899 | barrier(); | |
1900 | ||
1901 | /* | |
1902 | * If an interrupt came in just after the first while loop | |
1903 | * and pushed the tail page forward, we will be left with | |
1904 | * a dangling commit that will never go forward. | |
1905 | */ | |
1906 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1907 | goto again; | |
7a8e76a3 SR |
1908 | } |
1909 | ||
d769041f | 1910 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1911 | { |
abc9b56d | 1912 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1913 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1914 | } |
1915 | ||
34a148bf | 1916 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1917 | { |
1918 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1919 | ||
1920 | /* | |
1921 | * The iterator could be on the reader page (it starts there). | |
1922 | * But the head could have moved, since the reader was | |
1923 | * found. Check for this case and assign the iterator | |
1924 | * to the head page instead of next. | |
1925 | */ | |
1926 | if (iter->head_page == cpu_buffer->reader_page) | |
77ae365e | 1927 | iter->head_page = rb_set_head_page(cpu_buffer); |
d769041f SR |
1928 | else |
1929 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1930 | ||
abc9b56d | 1931 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1932 | iter->head = 0; |
1933 | } | |
1934 | ||
69d1b839 SR |
1935 | /* Slow path, do not inline */ |
1936 | static noinline struct ring_buffer_event * | |
1937 | rb_add_time_stamp(struct ring_buffer_event *event, u64 delta) | |
1938 | { | |
1939 | event->type_len = RINGBUF_TYPE_TIME_EXTEND; | |
1940 | ||
1941 | /* Not the first event on the page? */ | |
1942 | if (rb_event_index(event)) { | |
1943 | event->time_delta = delta & TS_MASK; | |
1944 | event->array[0] = delta >> TS_SHIFT; | |
1945 | } else { | |
1946 | /* nope, just zero it */ | |
1947 | event->time_delta = 0; | |
1948 | event->array[0] = 0; | |
1949 | } | |
1950 | ||
1951 | return skip_time_extend(event); | |
1952 | } | |
1953 | ||
7a8e76a3 | 1954 | /** |
01e3e710 | 1955 | * rb_update_event - update event type and data |
7a8e76a3 SR |
1956 | * @event: the even to update |
1957 | * @type: the type of event | |
1958 | * @length: the size of the event field in the ring buffer | |
1959 | * | |
1960 | * Update the type and data fields of the event. The length | |
1961 | * is the actual size that is written to the ring buffer, | |
1962 | * and with this, we can determine what to place into the | |
1963 | * data field. | |
1964 | */ | |
34a148bf | 1965 | static void |
69d1b839 SR |
1966 | rb_update_event(struct ring_buffer_per_cpu *cpu_buffer, |
1967 | struct ring_buffer_event *event, unsigned length, | |
1968 | int add_timestamp, u64 delta) | |
7a8e76a3 | 1969 | { |
69d1b839 SR |
1970 | /* Only a commit updates the timestamp */ |
1971 | if (unlikely(!rb_event_is_commit(cpu_buffer, event))) | |
1972 | delta = 0; | |
7a8e76a3 | 1973 | |
69d1b839 SR |
1974 | /* |
1975 | * If we need to add a timestamp, then we | |
1976 | * add it to the start of the resevered space. | |
1977 | */ | |
1978 | if (unlikely(add_timestamp)) { | |
1979 | event = rb_add_time_stamp(event, delta); | |
1980 | length -= RB_LEN_TIME_EXTEND; | |
1981 | delta = 0; | |
7a8e76a3 | 1982 | } |
69d1b839 SR |
1983 | |
1984 | event->time_delta = delta; | |
1985 | length -= RB_EVNT_HDR_SIZE; | |
1986 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) { | |
1987 | event->type_len = 0; | |
1988 | event->array[0] = length; | |
1989 | } else | |
1990 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
7a8e76a3 SR |
1991 | } |
1992 | ||
77ae365e SR |
1993 | /* |
1994 | * rb_handle_head_page - writer hit the head page | |
1995 | * | |
1996 | * Returns: +1 to retry page | |
1997 | * 0 to continue | |
1998 | * -1 on error | |
1999 | */ | |
2000 | static int | |
2001 | rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
2002 | struct buffer_page *tail_page, | |
2003 | struct buffer_page *next_page) | |
2004 | { | |
2005 | struct buffer_page *new_head; | |
2006 | int entries; | |
2007 | int type; | |
2008 | int ret; | |
2009 | ||
2010 | entries = rb_page_entries(next_page); | |
2011 | ||
2012 | /* | |
2013 | * The hard part is here. We need to move the head | |
2014 | * forward, and protect against both readers on | |
2015 | * other CPUs and writers coming in via interrupts. | |
2016 | */ | |
2017 | type = rb_head_page_set_update(cpu_buffer, next_page, tail_page, | |
2018 | RB_PAGE_HEAD); | |
2019 | ||
2020 | /* | |
2021 | * type can be one of four: | |
2022 | * NORMAL - an interrupt already moved it for us | |
2023 | * HEAD - we are the first to get here. | |
2024 | * UPDATE - we are the interrupt interrupting | |
2025 | * a current move. | |
2026 | * MOVED - a reader on another CPU moved the next | |
2027 | * pointer to its reader page. Give up | |
2028 | * and try again. | |
2029 | */ | |
2030 | ||
2031 | switch (type) { | |
2032 | case RB_PAGE_HEAD: | |
2033 | /* | |
2034 | * We changed the head to UPDATE, thus | |
2035 | * it is our responsibility to update | |
2036 | * the counters. | |
2037 | */ | |
2038 | local_add(entries, &cpu_buffer->overrun); | |
c64e148a | 2039 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
77ae365e SR |
2040 | |
2041 | /* | |
2042 | * The entries will be zeroed out when we move the | |
2043 | * tail page. | |
2044 | */ | |
2045 | ||
2046 | /* still more to do */ | |
2047 | break; | |
2048 | ||
2049 | case RB_PAGE_UPDATE: | |
2050 | /* | |
2051 | * This is an interrupt that interrupt the | |
2052 | * previous update. Still more to do. | |
2053 | */ | |
2054 | break; | |
2055 | case RB_PAGE_NORMAL: | |
2056 | /* | |
2057 | * An interrupt came in before the update | |
2058 | * and processed this for us. | |
2059 | * Nothing left to do. | |
2060 | */ | |
2061 | return 1; | |
2062 | case RB_PAGE_MOVED: | |
2063 | /* | |
2064 | * The reader is on another CPU and just did | |
2065 | * a swap with our next_page. | |
2066 | * Try again. | |
2067 | */ | |
2068 | return 1; | |
2069 | default: | |
2070 | RB_WARN_ON(cpu_buffer, 1); /* WTF??? */ | |
2071 | return -1; | |
2072 | } | |
2073 | ||
2074 | /* | |
2075 | * Now that we are here, the old head pointer is | |
2076 | * set to UPDATE. This will keep the reader from | |
2077 | * swapping the head page with the reader page. | |
2078 | * The reader (on another CPU) will spin till | |
2079 | * we are finished. | |
2080 | * | |
2081 | * We just need to protect against interrupts | |
2082 | * doing the job. We will set the next pointer | |
2083 | * to HEAD. After that, we set the old pointer | |
2084 | * to NORMAL, but only if it was HEAD before. | |
2085 | * otherwise we are an interrupt, and only | |
2086 | * want the outer most commit to reset it. | |
2087 | */ | |
2088 | new_head = next_page; | |
2089 | rb_inc_page(cpu_buffer, &new_head); | |
2090 | ||
2091 | ret = rb_head_page_set_head(cpu_buffer, new_head, next_page, | |
2092 | RB_PAGE_NORMAL); | |
2093 | ||
2094 | /* | |
2095 | * Valid returns are: | |
2096 | * HEAD - an interrupt came in and already set it. | |
2097 | * NORMAL - One of two things: | |
2098 | * 1) We really set it. | |
2099 | * 2) A bunch of interrupts came in and moved | |
2100 | * the page forward again. | |
2101 | */ | |
2102 | switch (ret) { | |
2103 | case RB_PAGE_HEAD: | |
2104 | case RB_PAGE_NORMAL: | |
2105 | /* OK */ | |
2106 | break; | |
2107 | default: | |
2108 | RB_WARN_ON(cpu_buffer, 1); | |
2109 | return -1; | |
2110 | } | |
2111 | ||
2112 | /* | |
2113 | * It is possible that an interrupt came in, | |
2114 | * set the head up, then more interrupts came in | |
2115 | * and moved it again. When we get back here, | |
2116 | * the page would have been set to NORMAL but we | |
2117 | * just set it back to HEAD. | |
2118 | * | |
2119 | * How do you detect this? Well, if that happened | |
2120 | * the tail page would have moved. | |
2121 | */ | |
2122 | if (ret == RB_PAGE_NORMAL) { | |
2123 | /* | |
2124 | * If the tail had moved passed next, then we need | |
2125 | * to reset the pointer. | |
2126 | */ | |
2127 | if (cpu_buffer->tail_page != tail_page && | |
2128 | cpu_buffer->tail_page != next_page) | |
2129 | rb_head_page_set_normal(cpu_buffer, new_head, | |
2130 | next_page, | |
2131 | RB_PAGE_HEAD); | |
2132 | } | |
2133 | ||
2134 | /* | |
2135 | * If this was the outer most commit (the one that | |
2136 | * changed the original pointer from HEAD to UPDATE), | |
2137 | * then it is up to us to reset it to NORMAL. | |
2138 | */ | |
2139 | if (type == RB_PAGE_HEAD) { | |
2140 | ret = rb_head_page_set_normal(cpu_buffer, next_page, | |
2141 | tail_page, | |
2142 | RB_PAGE_UPDATE); | |
2143 | if (RB_WARN_ON(cpu_buffer, | |
2144 | ret != RB_PAGE_UPDATE)) | |
2145 | return -1; | |
2146 | } | |
2147 | ||
2148 | return 0; | |
2149 | } | |
2150 | ||
34a148bf | 2151 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
2152 | { |
2153 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
2154 | ||
2155 | /* zero length can cause confusions */ | |
2156 | if (!length) | |
2157 | length = 1; | |
2158 | ||
2271048d | 2159 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) |
7a8e76a3 SR |
2160 | length += sizeof(event.array[0]); |
2161 | ||
2162 | length += RB_EVNT_HDR_SIZE; | |
2271048d | 2163 | length = ALIGN(length, RB_ARCH_ALIGNMENT); |
7a8e76a3 SR |
2164 | |
2165 | return length; | |
2166 | } | |
2167 | ||
c7b09308 SR |
2168 | static inline void |
2169 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
2170 | struct buffer_page *tail_page, | |
2171 | unsigned long tail, unsigned long length) | |
2172 | { | |
2173 | struct ring_buffer_event *event; | |
2174 | ||
2175 | /* | |
2176 | * Only the event that crossed the page boundary | |
2177 | * must fill the old tail_page with padding. | |
2178 | */ | |
2179 | if (tail >= BUF_PAGE_SIZE) { | |
b3230c8b SR |
2180 | /* |
2181 | * If the page was filled, then we still need | |
2182 | * to update the real_end. Reset it to zero | |
2183 | * and the reader will ignore it. | |
2184 | */ | |
2185 | if (tail == BUF_PAGE_SIZE) | |
2186 | tail_page->real_end = 0; | |
2187 | ||
c7b09308 SR |
2188 | local_sub(length, &tail_page->write); |
2189 | return; | |
2190 | } | |
2191 | ||
2192 | event = __rb_page_index(tail_page, tail); | |
b0b7065b | 2193 | kmemcheck_annotate_bitfield(event, bitfield); |
c7b09308 | 2194 | |
c64e148a VN |
2195 | /* account for padding bytes */ |
2196 | local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes); | |
2197 | ||
ff0ff84a SR |
2198 | /* |
2199 | * Save the original length to the meta data. | |
2200 | * This will be used by the reader to add lost event | |
2201 | * counter. | |
2202 | */ | |
2203 | tail_page->real_end = tail; | |
2204 | ||
c7b09308 SR |
2205 | /* |
2206 | * If this event is bigger than the minimum size, then | |
2207 | * we need to be careful that we don't subtract the | |
2208 | * write counter enough to allow another writer to slip | |
2209 | * in on this page. | |
2210 | * We put in a discarded commit instead, to make sure | |
2211 | * that this space is not used again. | |
2212 | * | |
2213 | * If we are less than the minimum size, we don't need to | |
2214 | * worry about it. | |
2215 | */ | |
2216 | if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) { | |
2217 | /* No room for any events */ | |
2218 | ||
2219 | /* Mark the rest of the page with padding */ | |
2220 | rb_event_set_padding(event); | |
2221 | ||
2222 | /* Set the write back to the previous setting */ | |
2223 | local_sub(length, &tail_page->write); | |
2224 | return; | |
2225 | } | |
2226 | ||
2227 | /* Put in a discarded event */ | |
2228 | event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE; | |
2229 | event->type_len = RINGBUF_TYPE_PADDING; | |
2230 | /* time delta must be non zero */ | |
2231 | event->time_delta = 1; | |
c7b09308 SR |
2232 | |
2233 | /* Set write to end of buffer */ | |
2234 | length = (tail + length) - BUF_PAGE_SIZE; | |
2235 | local_sub(length, &tail_page->write); | |
2236 | } | |
6634ff26 | 2237 | |
747e94ae SR |
2238 | /* |
2239 | * This is the slow path, force gcc not to inline it. | |
2240 | */ | |
2241 | static noinline struct ring_buffer_event * | |
6634ff26 SR |
2242 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
2243 | unsigned long length, unsigned long tail, | |
e8bc43e8 | 2244 | struct buffer_page *tail_page, u64 ts) |
7a8e76a3 | 2245 | { |
5a50e33c | 2246 | struct buffer_page *commit_page = cpu_buffer->commit_page; |
7a8e76a3 | 2247 | struct ring_buffer *buffer = cpu_buffer->buffer; |
77ae365e SR |
2248 | struct buffer_page *next_page; |
2249 | int ret; | |
aa20ae84 SR |
2250 | |
2251 | next_page = tail_page; | |
2252 | ||
aa20ae84 SR |
2253 | rb_inc_page(cpu_buffer, &next_page); |
2254 | ||
aa20ae84 SR |
2255 | /* |
2256 | * If for some reason, we had an interrupt storm that made | |
2257 | * it all the way around the buffer, bail, and warn | |
2258 | * about it. | |
2259 | */ | |
2260 | if (unlikely(next_page == commit_page)) { | |
77ae365e | 2261 | local_inc(&cpu_buffer->commit_overrun); |
aa20ae84 SR |
2262 | goto out_reset; |
2263 | } | |
2264 | ||
77ae365e SR |
2265 | /* |
2266 | * This is where the fun begins! | |
2267 | * | |
2268 | * We are fighting against races between a reader that | |
2269 | * could be on another CPU trying to swap its reader | |
2270 | * page with the buffer head. | |
2271 | * | |
2272 | * We are also fighting against interrupts coming in and | |
2273 | * moving the head or tail on us as well. | |
2274 | * | |
2275 | * If the next page is the head page then we have filled | |
2276 | * the buffer, unless the commit page is still on the | |
2277 | * reader page. | |
2278 | */ | |
2279 | if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) { | |
aa20ae84 | 2280 | |
77ae365e SR |
2281 | /* |
2282 | * If the commit is not on the reader page, then | |
2283 | * move the header page. | |
2284 | */ | |
2285 | if (!rb_is_reader_page(cpu_buffer->commit_page)) { | |
2286 | /* | |
2287 | * If we are not in overwrite mode, | |
2288 | * this is easy, just stop here. | |
2289 | */ | |
884bfe89 SP |
2290 | if (!(buffer->flags & RB_FL_OVERWRITE)) { |
2291 | local_inc(&cpu_buffer->dropped_events); | |
77ae365e | 2292 | goto out_reset; |
884bfe89 | 2293 | } |
77ae365e SR |
2294 | |
2295 | ret = rb_handle_head_page(cpu_buffer, | |
2296 | tail_page, | |
2297 | next_page); | |
2298 | if (ret < 0) | |
2299 | goto out_reset; | |
2300 | if (ret) | |
2301 | goto out_again; | |
2302 | } else { | |
2303 | /* | |
2304 | * We need to be careful here too. The | |
2305 | * commit page could still be on the reader | |
2306 | * page. We could have a small buffer, and | |
2307 | * have filled up the buffer with events | |
2308 | * from interrupts and such, and wrapped. | |
2309 | * | |
2310 | * Note, if the tail page is also the on the | |
2311 | * reader_page, we let it move out. | |
2312 | */ | |
2313 | if (unlikely((cpu_buffer->commit_page != | |
2314 | cpu_buffer->tail_page) && | |
2315 | (cpu_buffer->commit_page == | |
2316 | cpu_buffer->reader_page))) { | |
2317 | local_inc(&cpu_buffer->commit_overrun); | |
2318 | goto out_reset; | |
2319 | } | |
aa20ae84 SR |
2320 | } |
2321 | } | |
2322 | ||
77ae365e SR |
2323 | ret = rb_tail_page_update(cpu_buffer, tail_page, next_page); |
2324 | if (ret) { | |
2325 | /* | |
2326 | * Nested commits always have zero deltas, so | |
2327 | * just reread the time stamp | |
2328 | */ | |
e8bc43e8 SR |
2329 | ts = rb_time_stamp(buffer); |
2330 | next_page->page->time_stamp = ts; | |
aa20ae84 SR |
2331 | } |
2332 | ||
77ae365e | 2333 | out_again: |
aa20ae84 | 2334 | |
77ae365e | 2335 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
aa20ae84 SR |
2336 | |
2337 | /* fail and let the caller try again */ | |
2338 | return ERR_PTR(-EAGAIN); | |
2339 | ||
45141d46 | 2340 | out_reset: |
6f3b3440 | 2341 | /* reset write */ |
c7b09308 | 2342 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
6f3b3440 | 2343 | |
bf41a158 | 2344 | return NULL; |
7a8e76a3 SR |
2345 | } |
2346 | ||
6634ff26 SR |
2347 | static struct ring_buffer_event * |
2348 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
69d1b839 SR |
2349 | unsigned long length, u64 ts, |
2350 | u64 delta, int add_timestamp) | |
6634ff26 | 2351 | { |
5a50e33c | 2352 | struct buffer_page *tail_page; |
6634ff26 SR |
2353 | struct ring_buffer_event *event; |
2354 | unsigned long tail, write; | |
2355 | ||
69d1b839 SR |
2356 | /* |
2357 | * If the time delta since the last event is too big to | |
2358 | * hold in the time field of the event, then we append a | |
2359 | * TIME EXTEND event ahead of the data event. | |
2360 | */ | |
2361 | if (unlikely(add_timestamp)) | |
2362 | length += RB_LEN_TIME_EXTEND; | |
2363 | ||
6634ff26 SR |
2364 | tail_page = cpu_buffer->tail_page; |
2365 | write = local_add_return(length, &tail_page->write); | |
77ae365e SR |
2366 | |
2367 | /* set write to only the index of the write */ | |
2368 | write &= RB_WRITE_MASK; | |
6634ff26 SR |
2369 | tail = write - length; |
2370 | ||
2371 | /* See if we shot pass the end of this buffer page */ | |
747e94ae | 2372 | if (unlikely(write > BUF_PAGE_SIZE)) |
6634ff26 | 2373 | return rb_move_tail(cpu_buffer, length, tail, |
5a50e33c | 2374 | tail_page, ts); |
6634ff26 SR |
2375 | |
2376 | /* We reserved something on the buffer */ | |
2377 | ||
6634ff26 | 2378 | event = __rb_page_index(tail_page, tail); |
1744a21d | 2379 | kmemcheck_annotate_bitfield(event, bitfield); |
69d1b839 | 2380 | rb_update_event(cpu_buffer, event, length, add_timestamp, delta); |
6634ff26 | 2381 | |
69d1b839 | 2382 | local_inc(&tail_page->entries); |
6634ff26 SR |
2383 | |
2384 | /* | |
fa743953 SR |
2385 | * If this is the first commit on the page, then update |
2386 | * its timestamp. | |
6634ff26 | 2387 | */ |
fa743953 | 2388 | if (!tail) |
e8bc43e8 | 2389 | tail_page->page->time_stamp = ts; |
6634ff26 | 2390 | |
c64e148a VN |
2391 | /* account for these added bytes */ |
2392 | local_add(length, &cpu_buffer->entries_bytes); | |
2393 | ||
6634ff26 SR |
2394 | return event; |
2395 | } | |
2396 | ||
edd813bf SR |
2397 | static inline int |
2398 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
2399 | struct ring_buffer_event *event) | |
2400 | { | |
2401 | unsigned long new_index, old_index; | |
2402 | struct buffer_page *bpage; | |
2403 | unsigned long index; | |
2404 | unsigned long addr; | |
2405 | ||
2406 | new_index = rb_event_index(event); | |
69d1b839 | 2407 | old_index = new_index + rb_event_ts_length(event); |
edd813bf SR |
2408 | addr = (unsigned long)event; |
2409 | addr &= PAGE_MASK; | |
2410 | ||
2411 | bpage = cpu_buffer->tail_page; | |
2412 | ||
2413 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { | |
77ae365e SR |
2414 | unsigned long write_mask = |
2415 | local_read(&bpage->write) & ~RB_WRITE_MASK; | |
c64e148a | 2416 | unsigned long event_length = rb_event_length(event); |
edd813bf SR |
2417 | /* |
2418 | * This is on the tail page. It is possible that | |
2419 | * a write could come in and move the tail page | |
2420 | * and write to the next page. That is fine | |
2421 | * because we just shorten what is on this page. | |
2422 | */ | |
77ae365e SR |
2423 | old_index += write_mask; |
2424 | new_index += write_mask; | |
edd813bf | 2425 | index = local_cmpxchg(&bpage->write, old_index, new_index); |
c64e148a VN |
2426 | if (index == old_index) { |
2427 | /* update counters */ | |
2428 | local_sub(event_length, &cpu_buffer->entries_bytes); | |
edd813bf | 2429 | return 1; |
c64e148a | 2430 | } |
edd813bf SR |
2431 | } |
2432 | ||
2433 | /* could not discard */ | |
2434 | return 0; | |
2435 | } | |
2436 | ||
fa743953 SR |
2437 | static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer) |
2438 | { | |
2439 | local_inc(&cpu_buffer->committing); | |
2440 | local_inc(&cpu_buffer->commits); | |
2441 | } | |
2442 | ||
d9abde21 | 2443 | static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer) |
fa743953 SR |
2444 | { |
2445 | unsigned long commits; | |
2446 | ||
2447 | if (RB_WARN_ON(cpu_buffer, | |
2448 | !local_read(&cpu_buffer->committing))) | |
2449 | return; | |
2450 | ||
2451 | again: | |
2452 | commits = local_read(&cpu_buffer->commits); | |
2453 | /* synchronize with interrupts */ | |
2454 | barrier(); | |
2455 | if (local_read(&cpu_buffer->committing) == 1) | |
2456 | rb_set_commit_to_write(cpu_buffer); | |
2457 | ||
2458 | local_dec(&cpu_buffer->committing); | |
2459 | ||
2460 | /* synchronize with interrupts */ | |
2461 | barrier(); | |
2462 | ||
2463 | /* | |
2464 | * Need to account for interrupts coming in between the | |
2465 | * updating of the commit page and the clearing of the | |
2466 | * committing counter. | |
2467 | */ | |
2468 | if (unlikely(local_read(&cpu_buffer->commits) != commits) && | |
2469 | !local_read(&cpu_buffer->committing)) { | |
2470 | local_inc(&cpu_buffer->committing); | |
2471 | goto again; | |
2472 | } | |
2473 | } | |
2474 | ||
7a8e76a3 | 2475 | static struct ring_buffer_event * |
62f0b3eb SR |
2476 | rb_reserve_next_event(struct ring_buffer *buffer, |
2477 | struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 2478 | unsigned long length) |
7a8e76a3 SR |
2479 | { |
2480 | struct ring_buffer_event *event; | |
69d1b839 | 2481 | u64 ts, delta; |
818e3dd3 | 2482 | int nr_loops = 0; |
69d1b839 | 2483 | int add_timestamp; |
140ff891 | 2484 | u64 diff; |
7a8e76a3 | 2485 | |
fa743953 SR |
2486 | rb_start_commit(cpu_buffer); |
2487 | ||
85bac32c | 2488 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
62f0b3eb SR |
2489 | /* |
2490 | * Due to the ability to swap a cpu buffer from a buffer | |
2491 | * it is possible it was swapped before we committed. | |
2492 | * (committing stops a swap). We check for it here and | |
2493 | * if it happened, we have to fail the write. | |
2494 | */ | |
2495 | barrier(); | |
2496 | if (unlikely(ACCESS_ONCE(cpu_buffer->buffer) != buffer)) { | |
2497 | local_dec(&cpu_buffer->committing); | |
2498 | local_dec(&cpu_buffer->commits); | |
2499 | return NULL; | |
2500 | } | |
85bac32c | 2501 | #endif |
62f0b3eb | 2502 | |
be957c44 | 2503 | length = rb_calculate_event_length(length); |
bf41a158 | 2504 | again: |
69d1b839 SR |
2505 | add_timestamp = 0; |
2506 | delta = 0; | |
2507 | ||
818e3dd3 SR |
2508 | /* |
2509 | * We allow for interrupts to reenter here and do a trace. | |
2510 | * If one does, it will cause this original code to loop | |
2511 | * back here. Even with heavy interrupts happening, this | |
2512 | * should only happen a few times in a row. If this happens | |
2513 | * 1000 times in a row, there must be either an interrupt | |
2514 | * storm or we have something buggy. | |
2515 | * Bail! | |
2516 | */ | |
3e89c7bb | 2517 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
fa743953 | 2518 | goto out_fail; |
818e3dd3 | 2519 | |
6d3f1e12 | 2520 | ts = rb_time_stamp(cpu_buffer->buffer); |
140ff891 | 2521 | diff = ts - cpu_buffer->write_stamp; |
7a8e76a3 | 2522 | |
140ff891 SR |
2523 | /* make sure this diff is calculated here */ |
2524 | barrier(); | |
bf41a158 | 2525 | |
140ff891 SR |
2526 | /* Did the write stamp get updated already? */ |
2527 | if (likely(ts >= cpu_buffer->write_stamp)) { | |
168b6b1d SR |
2528 | delta = diff; |
2529 | if (unlikely(test_time_stamp(delta))) { | |
31274d72 JO |
2530 | int local_clock_stable = 1; |
2531 | #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK | |
2532 | local_clock_stable = sched_clock_stable; | |
2533 | #endif | |
69d1b839 | 2534 | WARN_ONCE(delta > (1ULL << 59), |
31274d72 | 2535 | KERN_WARNING "Delta way too big! %llu ts=%llu write stamp = %llu\n%s", |
69d1b839 SR |
2536 | (unsigned long long)delta, |
2537 | (unsigned long long)ts, | |
31274d72 JO |
2538 | (unsigned long long)cpu_buffer->write_stamp, |
2539 | local_clock_stable ? "" : | |
2540 | "If you just came from a suspend/resume,\n" | |
2541 | "please switch to the trace global clock:\n" | |
2542 | " echo global > /sys/kernel/debug/tracing/trace_clock\n"); | |
69d1b839 | 2543 | add_timestamp = 1; |
7a8e76a3 | 2544 | } |
168b6b1d | 2545 | } |
7a8e76a3 | 2546 | |
69d1b839 SR |
2547 | event = __rb_reserve_next(cpu_buffer, length, ts, |
2548 | delta, add_timestamp); | |
168b6b1d | 2549 | if (unlikely(PTR_ERR(event) == -EAGAIN)) |
bf41a158 SR |
2550 | goto again; |
2551 | ||
fa743953 SR |
2552 | if (!event) |
2553 | goto out_fail; | |
7a8e76a3 | 2554 | |
7a8e76a3 | 2555 | return event; |
fa743953 SR |
2556 | |
2557 | out_fail: | |
2558 | rb_end_commit(cpu_buffer); | |
2559 | return NULL; | |
7a8e76a3 SR |
2560 | } |
2561 | ||
1155de47 PM |
2562 | #ifdef CONFIG_TRACING |
2563 | ||
567cd4da SR |
2564 | /* |
2565 | * The lock and unlock are done within a preempt disable section. | |
2566 | * The current_context per_cpu variable can only be modified | |
2567 | * by the current task between lock and unlock. But it can | |
2568 | * be modified more than once via an interrupt. To pass this | |
2569 | * information from the lock to the unlock without having to | |
2570 | * access the 'in_interrupt()' functions again (which do show | |
2571 | * a bit of overhead in something as critical as function tracing, | |
2572 | * we use a bitmask trick. | |
2573 | * | |
2574 | * bit 0 = NMI context | |
2575 | * bit 1 = IRQ context | |
2576 | * bit 2 = SoftIRQ context | |
2577 | * bit 3 = normal context. | |
2578 | * | |
2579 | * This works because this is the order of contexts that can | |
2580 | * preempt other contexts. A SoftIRQ never preempts an IRQ | |
2581 | * context. | |
2582 | * | |
2583 | * When the context is determined, the corresponding bit is | |
2584 | * checked and set (if it was set, then a recursion of that context | |
2585 | * happened). | |
2586 | * | |
2587 | * On unlock, we need to clear this bit. To do so, just subtract | |
2588 | * 1 from the current_context and AND it to itself. | |
2589 | * | |
2590 | * (binary) | |
2591 | * 101 - 1 = 100 | |
2592 | * 101 & 100 = 100 (clearing bit zero) | |
2593 | * | |
2594 | * 1010 - 1 = 1001 | |
2595 | * 1010 & 1001 = 1000 (clearing bit 1) | |
2596 | * | |
2597 | * The least significant bit can be cleared this way, and it | |
2598 | * just so happens that it is the same bit corresponding to | |
2599 | * the current context. | |
2600 | */ | |
2601 | static DEFINE_PER_CPU(unsigned int, current_context); | |
261842b7 | 2602 | |
567cd4da | 2603 | static __always_inline int trace_recursive_lock(void) |
261842b7 | 2604 | { |
567cd4da SR |
2605 | unsigned int val = this_cpu_read(current_context); |
2606 | int bit; | |
d9abde21 | 2607 | |
567cd4da SR |
2608 | if (in_interrupt()) { |
2609 | if (in_nmi()) | |
2610 | bit = 0; | |
2611 | else if (in_irq()) | |
2612 | bit = 1; | |
2613 | else | |
2614 | bit = 2; | |
2615 | } else | |
2616 | bit = 3; | |
d9abde21 | 2617 | |
567cd4da SR |
2618 | if (unlikely(val & (1 << bit))) |
2619 | return 1; | |
d9abde21 | 2620 | |
567cd4da SR |
2621 | val |= (1 << bit); |
2622 | this_cpu_write(current_context, val); | |
d9abde21 | 2623 | |
567cd4da | 2624 | return 0; |
261842b7 SR |
2625 | } |
2626 | ||
567cd4da | 2627 | static __always_inline void trace_recursive_unlock(void) |
261842b7 | 2628 | { |
567cd4da | 2629 | unsigned int val = this_cpu_read(current_context); |
261842b7 | 2630 | |
567cd4da SR |
2631 | val--; |
2632 | val &= this_cpu_read(current_context); | |
2633 | this_cpu_write(current_context, val); | |
261842b7 SR |
2634 | } |
2635 | ||
1155de47 PM |
2636 | #else |
2637 | ||
2638 | #define trace_recursive_lock() (0) | |
2639 | #define trace_recursive_unlock() do { } while (0) | |
2640 | ||
2641 | #endif | |
2642 | ||
7a8e76a3 SR |
2643 | /** |
2644 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
2645 | * @buffer: the ring buffer to reserve from | |
2646 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
2647 | * |
2648 | * Returns a reseverd event on the ring buffer to copy directly to. | |
2649 | * The user of this interface will need to get the body to write into | |
2650 | * and can use the ring_buffer_event_data() interface. | |
2651 | * | |
2652 | * The length is the length of the data needed, not the event length | |
2653 | * which also includes the event header. | |
2654 | * | |
2655 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
2656 | * If NULL is returned, then nothing has been allocated or locked. | |
2657 | */ | |
2658 | struct ring_buffer_event * | |
0a987751 | 2659 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
2660 | { |
2661 | struct ring_buffer_per_cpu *cpu_buffer; | |
2662 | struct ring_buffer_event *event; | |
5168ae50 | 2663 | int cpu; |
7a8e76a3 | 2664 | |
033601a3 | 2665 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2666 | return NULL; |
2667 | ||
bf41a158 | 2668 | /* If we are tracing schedule, we don't want to recurse */ |
5168ae50 | 2669 | preempt_disable_notrace(); |
bf41a158 | 2670 | |
52fbe9cd LJ |
2671 | if (atomic_read(&buffer->record_disabled)) |
2672 | goto out_nocheck; | |
2673 | ||
261842b7 SR |
2674 | if (trace_recursive_lock()) |
2675 | goto out_nocheck; | |
2676 | ||
7a8e76a3 SR |
2677 | cpu = raw_smp_processor_id(); |
2678 | ||
9e01c1b7 | 2679 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2680 | goto out; |
7a8e76a3 SR |
2681 | |
2682 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2683 | |
2684 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 2685 | goto out; |
7a8e76a3 | 2686 | |
be957c44 | 2687 | if (length > BUF_MAX_DATA_SIZE) |
bf41a158 | 2688 | goto out; |
7a8e76a3 | 2689 | |
62f0b3eb | 2690 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 2691 | if (!event) |
d769041f | 2692 | goto out; |
7a8e76a3 SR |
2693 | |
2694 | return event; | |
2695 | ||
d769041f | 2696 | out: |
261842b7 SR |
2697 | trace_recursive_unlock(); |
2698 | ||
2699 | out_nocheck: | |
5168ae50 | 2700 | preempt_enable_notrace(); |
7a8e76a3 SR |
2701 | return NULL; |
2702 | } | |
c4f50183 | 2703 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 | 2704 | |
a1863c21 SR |
2705 | static void |
2706 | rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
7a8e76a3 SR |
2707 | struct ring_buffer_event *event) |
2708 | { | |
69d1b839 SR |
2709 | u64 delta; |
2710 | ||
fa743953 SR |
2711 | /* |
2712 | * The event first in the commit queue updates the | |
2713 | * time stamp. | |
2714 | */ | |
69d1b839 SR |
2715 | if (rb_event_is_commit(cpu_buffer, event)) { |
2716 | /* | |
2717 | * A commit event that is first on a page | |
2718 | * updates the write timestamp with the page stamp | |
2719 | */ | |
2720 | if (!rb_event_index(event)) | |
2721 | cpu_buffer->write_stamp = | |
2722 | cpu_buffer->commit_page->page->time_stamp; | |
2723 | else if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
2724 | delta = event->array[0]; | |
2725 | delta <<= TS_SHIFT; | |
2726 | delta += event->time_delta; | |
2727 | cpu_buffer->write_stamp += delta; | |
2728 | } else | |
2729 | cpu_buffer->write_stamp += event->time_delta; | |
2730 | } | |
a1863c21 | 2731 | } |
bf41a158 | 2732 | |
a1863c21 SR |
2733 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, |
2734 | struct ring_buffer_event *event) | |
2735 | { | |
2736 | local_inc(&cpu_buffer->entries); | |
2737 | rb_update_write_stamp(cpu_buffer, event); | |
fa743953 | 2738 | rb_end_commit(cpu_buffer); |
7a8e76a3 SR |
2739 | } |
2740 | ||
15693458 SRRH |
2741 | static __always_inline void |
2742 | rb_wakeups(struct ring_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer) | |
2743 | { | |
2744 | if (buffer->irq_work.waiters_pending) { | |
2745 | buffer->irq_work.waiters_pending = false; | |
2746 | /* irq_work_queue() supplies it's own memory barriers */ | |
2747 | irq_work_queue(&buffer->irq_work.work); | |
2748 | } | |
2749 | ||
2750 | if (cpu_buffer->irq_work.waiters_pending) { | |
2751 | cpu_buffer->irq_work.waiters_pending = false; | |
2752 | /* irq_work_queue() supplies it's own memory barriers */ | |
2753 | irq_work_queue(&cpu_buffer->irq_work.work); | |
2754 | } | |
2755 | } | |
2756 | ||
7a8e76a3 SR |
2757 | /** |
2758 | * ring_buffer_unlock_commit - commit a reserved | |
2759 | * @buffer: The buffer to commit to | |
2760 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
2761 | * |
2762 | * This commits the data to the ring buffer, and releases any locks held. | |
2763 | * | |
2764 | * Must be paired with ring_buffer_lock_reserve. | |
2765 | */ | |
2766 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 2767 | struct ring_buffer_event *event) |
7a8e76a3 SR |
2768 | { |
2769 | struct ring_buffer_per_cpu *cpu_buffer; | |
2770 | int cpu = raw_smp_processor_id(); | |
2771 | ||
2772 | cpu_buffer = buffer->buffers[cpu]; | |
2773 | ||
7a8e76a3 SR |
2774 | rb_commit(cpu_buffer, event); |
2775 | ||
15693458 SRRH |
2776 | rb_wakeups(buffer, cpu_buffer); |
2777 | ||
261842b7 SR |
2778 | trace_recursive_unlock(); |
2779 | ||
5168ae50 | 2780 | preempt_enable_notrace(); |
7a8e76a3 SR |
2781 | |
2782 | return 0; | |
2783 | } | |
c4f50183 | 2784 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 2785 | |
f3b9aae1 FW |
2786 | static inline void rb_event_discard(struct ring_buffer_event *event) |
2787 | { | |
69d1b839 SR |
2788 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) |
2789 | event = skip_time_extend(event); | |
2790 | ||
334d4169 LJ |
2791 | /* array[0] holds the actual length for the discarded event */ |
2792 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
2793 | event->type_len = RINGBUF_TYPE_PADDING; | |
f3b9aae1 FW |
2794 | /* time delta must be non zero */ |
2795 | if (!event->time_delta) | |
2796 | event->time_delta = 1; | |
2797 | } | |
2798 | ||
a1863c21 SR |
2799 | /* |
2800 | * Decrement the entries to the page that an event is on. | |
2801 | * The event does not even need to exist, only the pointer | |
2802 | * to the page it is on. This may only be called before the commit | |
2803 | * takes place. | |
2804 | */ | |
2805 | static inline void | |
2806 | rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer, | |
2807 | struct ring_buffer_event *event) | |
2808 | { | |
2809 | unsigned long addr = (unsigned long)event; | |
2810 | struct buffer_page *bpage = cpu_buffer->commit_page; | |
2811 | struct buffer_page *start; | |
2812 | ||
2813 | addr &= PAGE_MASK; | |
2814 | ||
2815 | /* Do the likely case first */ | |
2816 | if (likely(bpage->page == (void *)addr)) { | |
2817 | local_dec(&bpage->entries); | |
2818 | return; | |
2819 | } | |
2820 | ||
2821 | /* | |
2822 | * Because the commit page may be on the reader page we | |
2823 | * start with the next page and check the end loop there. | |
2824 | */ | |
2825 | rb_inc_page(cpu_buffer, &bpage); | |
2826 | start = bpage; | |
2827 | do { | |
2828 | if (bpage->page == (void *)addr) { | |
2829 | local_dec(&bpage->entries); | |
2830 | return; | |
2831 | } | |
2832 | rb_inc_page(cpu_buffer, &bpage); | |
2833 | } while (bpage != start); | |
2834 | ||
2835 | /* commit not part of this buffer?? */ | |
2836 | RB_WARN_ON(cpu_buffer, 1); | |
2837 | } | |
2838 | ||
fa1b47dd SR |
2839 | /** |
2840 | * ring_buffer_commit_discard - discard an event that has not been committed | |
2841 | * @buffer: the ring buffer | |
2842 | * @event: non committed event to discard | |
2843 | * | |
dc892f73 SR |
2844 | * Sometimes an event that is in the ring buffer needs to be ignored. |
2845 | * This function lets the user discard an event in the ring buffer | |
2846 | * and then that event will not be read later. | |
2847 | * | |
2848 | * This function only works if it is called before the the item has been | |
2849 | * committed. It will try to free the event from the ring buffer | |
fa1b47dd SR |
2850 | * if another event has not been added behind it. |
2851 | * | |
2852 | * If another event has been added behind it, it will set the event | |
2853 | * up as discarded, and perform the commit. | |
2854 | * | |
2855 | * If this function is called, do not call ring_buffer_unlock_commit on | |
2856 | * the event. | |
2857 | */ | |
2858 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
2859 | struct ring_buffer_event *event) | |
2860 | { | |
2861 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
2862 | int cpu; |
2863 | ||
2864 | /* The event is discarded regardless */ | |
f3b9aae1 | 2865 | rb_event_discard(event); |
fa1b47dd | 2866 | |
fa743953 SR |
2867 | cpu = smp_processor_id(); |
2868 | cpu_buffer = buffer->buffers[cpu]; | |
2869 | ||
fa1b47dd SR |
2870 | /* |
2871 | * This must only be called if the event has not been | |
2872 | * committed yet. Thus we can assume that preemption | |
2873 | * is still disabled. | |
2874 | */ | |
fa743953 | 2875 | RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing)); |
fa1b47dd | 2876 | |
a1863c21 | 2877 | rb_decrement_entry(cpu_buffer, event); |
0f2541d2 | 2878 | if (rb_try_to_discard(cpu_buffer, event)) |
edd813bf | 2879 | goto out; |
fa1b47dd SR |
2880 | |
2881 | /* | |
2882 | * The commit is still visible by the reader, so we | |
a1863c21 | 2883 | * must still update the timestamp. |
fa1b47dd | 2884 | */ |
a1863c21 | 2885 | rb_update_write_stamp(cpu_buffer, event); |
fa1b47dd | 2886 | out: |
fa743953 | 2887 | rb_end_commit(cpu_buffer); |
fa1b47dd | 2888 | |
f3b9aae1 FW |
2889 | trace_recursive_unlock(); |
2890 | ||
5168ae50 | 2891 | preempt_enable_notrace(); |
fa1b47dd SR |
2892 | |
2893 | } | |
2894 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
2895 | ||
7a8e76a3 SR |
2896 | /** |
2897 | * ring_buffer_write - write data to the buffer without reserving | |
2898 | * @buffer: The ring buffer to write to. | |
2899 | * @length: The length of the data being written (excluding the event header) | |
2900 | * @data: The data to write to the buffer. | |
2901 | * | |
2902 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
2903 | * one function. If you already have the data to write to the buffer, it | |
2904 | * may be easier to simply call this function. | |
2905 | * | |
2906 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
2907 | * and not the length of the event which would hold the header. | |
2908 | */ | |
2909 | int ring_buffer_write(struct ring_buffer *buffer, | |
01e3e710 DS |
2910 | unsigned long length, |
2911 | void *data) | |
7a8e76a3 SR |
2912 | { |
2913 | struct ring_buffer_per_cpu *cpu_buffer; | |
2914 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
2915 | void *body; |
2916 | int ret = -EBUSY; | |
5168ae50 | 2917 | int cpu; |
7a8e76a3 | 2918 | |
033601a3 | 2919 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2920 | return -EBUSY; |
2921 | ||
5168ae50 | 2922 | preempt_disable_notrace(); |
bf41a158 | 2923 | |
52fbe9cd LJ |
2924 | if (atomic_read(&buffer->record_disabled)) |
2925 | goto out; | |
2926 | ||
7a8e76a3 SR |
2927 | cpu = raw_smp_processor_id(); |
2928 | ||
9e01c1b7 | 2929 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2930 | goto out; |
7a8e76a3 SR |
2931 | |
2932 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2933 | |
2934 | if (atomic_read(&cpu_buffer->record_disabled)) | |
2935 | goto out; | |
2936 | ||
be957c44 SR |
2937 | if (length > BUF_MAX_DATA_SIZE) |
2938 | goto out; | |
2939 | ||
62f0b3eb | 2940 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 SR |
2941 | if (!event) |
2942 | goto out; | |
2943 | ||
2944 | body = rb_event_data(event); | |
2945 | ||
2946 | memcpy(body, data, length); | |
2947 | ||
2948 | rb_commit(cpu_buffer, event); | |
2949 | ||
15693458 SRRH |
2950 | rb_wakeups(buffer, cpu_buffer); |
2951 | ||
7a8e76a3 SR |
2952 | ret = 0; |
2953 | out: | |
5168ae50 | 2954 | preempt_enable_notrace(); |
7a8e76a3 SR |
2955 | |
2956 | return ret; | |
2957 | } | |
c4f50183 | 2958 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 2959 | |
34a148bf | 2960 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
2961 | { |
2962 | struct buffer_page *reader = cpu_buffer->reader_page; | |
77ae365e | 2963 | struct buffer_page *head = rb_set_head_page(cpu_buffer); |
bf41a158 SR |
2964 | struct buffer_page *commit = cpu_buffer->commit_page; |
2965 | ||
77ae365e SR |
2966 | /* In case of error, head will be NULL */ |
2967 | if (unlikely(!head)) | |
2968 | return 1; | |
2969 | ||
bf41a158 SR |
2970 | return reader->read == rb_page_commit(reader) && |
2971 | (commit == reader || | |
2972 | (commit == head && | |
2973 | head->read == rb_page_commit(commit))); | |
2974 | } | |
2975 | ||
7a8e76a3 SR |
2976 | /** |
2977 | * ring_buffer_record_disable - stop all writes into the buffer | |
2978 | * @buffer: The ring buffer to stop writes to. | |
2979 | * | |
2980 | * This prevents all writes to the buffer. Any attempt to write | |
2981 | * to the buffer after this will fail and return NULL. | |
2982 | * | |
2983 | * The caller should call synchronize_sched() after this. | |
2984 | */ | |
2985 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
2986 | { | |
2987 | atomic_inc(&buffer->record_disabled); | |
2988 | } | |
c4f50183 | 2989 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
2990 | |
2991 | /** | |
2992 | * ring_buffer_record_enable - enable writes to the buffer | |
2993 | * @buffer: The ring buffer to enable writes | |
2994 | * | |
2995 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 2996 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
2997 | */ |
2998 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
2999 | { | |
3000 | atomic_dec(&buffer->record_disabled); | |
3001 | } | |
c4f50183 | 3002 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 | 3003 | |
499e5470 SR |
3004 | /** |
3005 | * ring_buffer_record_off - stop all writes into the buffer | |
3006 | * @buffer: The ring buffer to stop writes to. | |
3007 | * | |
3008 | * This prevents all writes to the buffer. Any attempt to write | |
3009 | * to the buffer after this will fail and return NULL. | |
3010 | * | |
3011 | * This is different than ring_buffer_record_disable() as | |
87abb3b1 | 3012 | * it works like an on/off switch, where as the disable() version |
499e5470 SR |
3013 | * must be paired with a enable(). |
3014 | */ | |
3015 | void ring_buffer_record_off(struct ring_buffer *buffer) | |
3016 | { | |
3017 | unsigned int rd; | |
3018 | unsigned int new_rd; | |
3019 | ||
3020 | do { | |
3021 | rd = atomic_read(&buffer->record_disabled); | |
3022 | new_rd = rd | RB_BUFFER_OFF; | |
3023 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
3024 | } | |
3025 | EXPORT_SYMBOL_GPL(ring_buffer_record_off); | |
3026 | ||
3027 | /** | |
3028 | * ring_buffer_record_on - restart writes into the buffer | |
3029 | * @buffer: The ring buffer to start writes to. | |
3030 | * | |
3031 | * This enables all writes to the buffer that was disabled by | |
3032 | * ring_buffer_record_off(). | |
3033 | * | |
3034 | * This is different than ring_buffer_record_enable() as | |
87abb3b1 | 3035 | * it works like an on/off switch, where as the enable() version |
499e5470 SR |
3036 | * must be paired with a disable(). |
3037 | */ | |
3038 | void ring_buffer_record_on(struct ring_buffer *buffer) | |
3039 | { | |
3040 | unsigned int rd; | |
3041 | unsigned int new_rd; | |
3042 | ||
3043 | do { | |
3044 | rd = atomic_read(&buffer->record_disabled); | |
3045 | new_rd = rd & ~RB_BUFFER_OFF; | |
3046 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
3047 | } | |
3048 | EXPORT_SYMBOL_GPL(ring_buffer_record_on); | |
3049 | ||
3050 | /** | |
3051 | * ring_buffer_record_is_on - return true if the ring buffer can write | |
3052 | * @buffer: The ring buffer to see if write is enabled | |
3053 | * | |
3054 | * Returns true if the ring buffer is in a state that it accepts writes. | |
3055 | */ | |
3056 | int ring_buffer_record_is_on(struct ring_buffer *buffer) | |
3057 | { | |
3058 | return !atomic_read(&buffer->record_disabled); | |
3059 | } | |
3060 | ||
7a8e76a3 SR |
3061 | /** |
3062 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
3063 | * @buffer: The ring buffer to stop writes to. | |
3064 | * @cpu: The CPU buffer to stop | |
3065 | * | |
3066 | * This prevents all writes to the buffer. Any attempt to write | |
3067 | * to the buffer after this will fail and return NULL. | |
3068 | * | |
3069 | * The caller should call synchronize_sched() after this. | |
3070 | */ | |
3071 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
3072 | { | |
3073 | struct ring_buffer_per_cpu *cpu_buffer; | |
3074 | ||
9e01c1b7 | 3075 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3076 | return; |
7a8e76a3 SR |
3077 | |
3078 | cpu_buffer = buffer->buffers[cpu]; | |
3079 | atomic_inc(&cpu_buffer->record_disabled); | |
3080 | } | |
c4f50183 | 3081 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
3082 | |
3083 | /** | |
3084 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
3085 | * @buffer: The ring buffer to enable writes | |
3086 | * @cpu: The CPU to enable. | |
3087 | * | |
3088 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 3089 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
3090 | */ |
3091 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
3092 | { | |
3093 | struct ring_buffer_per_cpu *cpu_buffer; | |
3094 | ||
9e01c1b7 | 3095 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3096 | return; |
7a8e76a3 SR |
3097 | |
3098 | cpu_buffer = buffer->buffers[cpu]; | |
3099 | atomic_dec(&cpu_buffer->record_disabled); | |
3100 | } | |
c4f50183 | 3101 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 | 3102 | |
f6195aa0 SR |
3103 | /* |
3104 | * The total entries in the ring buffer is the running counter | |
3105 | * of entries entered into the ring buffer, minus the sum of | |
3106 | * the entries read from the ring buffer and the number of | |
3107 | * entries that were overwritten. | |
3108 | */ | |
3109 | static inline unsigned long | |
3110 | rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer) | |
3111 | { | |
3112 | return local_read(&cpu_buffer->entries) - | |
3113 | (local_read(&cpu_buffer->overrun) + cpu_buffer->read); | |
3114 | } | |
3115 | ||
c64e148a VN |
3116 | /** |
3117 | * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer | |
3118 | * @buffer: The ring buffer | |
3119 | * @cpu: The per CPU buffer to read from. | |
3120 | */ | |
50ecf2c3 | 3121 | u64 ring_buffer_oldest_event_ts(struct ring_buffer *buffer, int cpu) |
c64e148a VN |
3122 | { |
3123 | unsigned long flags; | |
3124 | struct ring_buffer_per_cpu *cpu_buffer; | |
3125 | struct buffer_page *bpage; | |
da830e58 | 3126 | u64 ret = 0; |
c64e148a VN |
3127 | |
3128 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3129 | return 0; | |
3130 | ||
3131 | cpu_buffer = buffer->buffers[cpu]; | |
7115e3fc | 3132 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
3133 | /* |
3134 | * if the tail is on reader_page, oldest time stamp is on the reader | |
3135 | * page | |
3136 | */ | |
3137 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
3138 | bpage = cpu_buffer->reader_page; | |
3139 | else | |
3140 | bpage = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
3141 | if (bpage) |
3142 | ret = bpage->page->time_stamp; | |
7115e3fc | 3143 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
3144 | |
3145 | return ret; | |
3146 | } | |
3147 | EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts); | |
3148 | ||
3149 | /** | |
3150 | * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer | |
3151 | * @buffer: The ring buffer | |
3152 | * @cpu: The per CPU buffer to read from. | |
3153 | */ | |
3154 | unsigned long ring_buffer_bytes_cpu(struct ring_buffer *buffer, int cpu) | |
3155 | { | |
3156 | struct ring_buffer_per_cpu *cpu_buffer; | |
3157 | unsigned long ret; | |
3158 | ||
3159 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3160 | return 0; | |
3161 | ||
3162 | cpu_buffer = buffer->buffers[cpu]; | |
3163 | ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes; | |
3164 | ||
3165 | return ret; | |
3166 | } | |
3167 | EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu); | |
3168 | ||
7a8e76a3 SR |
3169 | /** |
3170 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
3171 | * @buffer: The ring buffer | |
3172 | * @cpu: The per CPU buffer to get the entries from. | |
3173 | */ | |
3174 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
3175 | { | |
3176 | struct ring_buffer_per_cpu *cpu_buffer; | |
3177 | ||
9e01c1b7 | 3178 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3179 | return 0; |
7a8e76a3 SR |
3180 | |
3181 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 3182 | |
f6195aa0 | 3183 | return rb_num_of_entries(cpu_buffer); |
7a8e76a3 | 3184 | } |
c4f50183 | 3185 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
3186 | |
3187 | /** | |
884bfe89 SP |
3188 | * ring_buffer_overrun_cpu - get the number of overruns caused by the ring |
3189 | * buffer wrapping around (only if RB_FL_OVERWRITE is on). | |
7a8e76a3 SR |
3190 | * @buffer: The ring buffer |
3191 | * @cpu: The per CPU buffer to get the number of overruns from | |
3192 | */ | |
3193 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3194 | { | |
3195 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3196 | unsigned long ret; |
7a8e76a3 | 3197 | |
9e01c1b7 | 3198 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3199 | return 0; |
7a8e76a3 SR |
3200 | |
3201 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3202 | ret = local_read(&cpu_buffer->overrun); |
554f786e SR |
3203 | |
3204 | return ret; | |
7a8e76a3 | 3205 | } |
c4f50183 | 3206 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 3207 | |
f0d2c681 | 3208 | /** |
884bfe89 SP |
3209 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by |
3210 | * commits failing due to the buffer wrapping around while there are uncommitted | |
3211 | * events, such as during an interrupt storm. | |
f0d2c681 SR |
3212 | * @buffer: The ring buffer |
3213 | * @cpu: The per CPU buffer to get the number of overruns from | |
3214 | */ | |
3215 | unsigned long | |
3216 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3217 | { | |
3218 | struct ring_buffer_per_cpu *cpu_buffer; | |
3219 | unsigned long ret; | |
3220 | ||
3221 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3222 | return 0; | |
3223 | ||
3224 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3225 | ret = local_read(&cpu_buffer->commit_overrun); |
f0d2c681 SR |
3226 | |
3227 | return ret; | |
3228 | } | |
3229 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
3230 | ||
884bfe89 SP |
3231 | /** |
3232 | * ring_buffer_dropped_events_cpu - get the number of dropped events caused by | |
3233 | * the ring buffer filling up (only if RB_FL_OVERWRITE is off). | |
3234 | * @buffer: The ring buffer | |
3235 | * @cpu: The per CPU buffer to get the number of overruns from | |
3236 | */ | |
3237 | unsigned long | |
3238 | ring_buffer_dropped_events_cpu(struct ring_buffer *buffer, int cpu) | |
3239 | { | |
3240 | struct ring_buffer_per_cpu *cpu_buffer; | |
3241 | unsigned long ret; | |
3242 | ||
3243 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3244 | return 0; | |
3245 | ||
3246 | cpu_buffer = buffer->buffers[cpu]; | |
3247 | ret = local_read(&cpu_buffer->dropped_events); | |
3248 | ||
3249 | return ret; | |
3250 | } | |
3251 | EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu); | |
3252 | ||
ad964704 SRRH |
3253 | /** |
3254 | * ring_buffer_read_events_cpu - get the number of events successfully read | |
3255 | * @buffer: The ring buffer | |
3256 | * @cpu: The per CPU buffer to get the number of events read | |
3257 | */ | |
3258 | unsigned long | |
3259 | ring_buffer_read_events_cpu(struct ring_buffer *buffer, int cpu) | |
3260 | { | |
3261 | struct ring_buffer_per_cpu *cpu_buffer; | |
3262 | ||
3263 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3264 | return 0; | |
3265 | ||
3266 | cpu_buffer = buffer->buffers[cpu]; | |
3267 | return cpu_buffer->read; | |
3268 | } | |
3269 | EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu); | |
3270 | ||
7a8e76a3 SR |
3271 | /** |
3272 | * ring_buffer_entries - get the number of entries in a buffer | |
3273 | * @buffer: The ring buffer | |
3274 | * | |
3275 | * Returns the total number of entries in the ring buffer | |
3276 | * (all CPU entries) | |
3277 | */ | |
3278 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
3279 | { | |
3280 | struct ring_buffer_per_cpu *cpu_buffer; | |
3281 | unsigned long entries = 0; | |
3282 | int cpu; | |
3283 | ||
3284 | /* if you care about this being correct, lock the buffer */ | |
3285 | for_each_buffer_cpu(buffer, cpu) { | |
3286 | cpu_buffer = buffer->buffers[cpu]; | |
f6195aa0 | 3287 | entries += rb_num_of_entries(cpu_buffer); |
7a8e76a3 SR |
3288 | } |
3289 | ||
3290 | return entries; | |
3291 | } | |
c4f50183 | 3292 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
3293 | |
3294 | /** | |
67b394f7 | 3295 | * ring_buffer_overruns - get the number of overruns in buffer |
7a8e76a3 SR |
3296 | * @buffer: The ring buffer |
3297 | * | |
3298 | * Returns the total number of overruns in the ring buffer | |
3299 | * (all CPU entries) | |
3300 | */ | |
3301 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
3302 | { | |
3303 | struct ring_buffer_per_cpu *cpu_buffer; | |
3304 | unsigned long overruns = 0; | |
3305 | int cpu; | |
3306 | ||
3307 | /* if you care about this being correct, lock the buffer */ | |
3308 | for_each_buffer_cpu(buffer, cpu) { | |
3309 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3310 | overruns += local_read(&cpu_buffer->overrun); |
7a8e76a3 SR |
3311 | } |
3312 | ||
3313 | return overruns; | |
3314 | } | |
c4f50183 | 3315 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 3316 | |
642edba5 | 3317 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
3318 | { |
3319 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3320 | ||
d769041f SR |
3321 | /* Iterator usage is expected to have record disabled */ |
3322 | if (list_empty(&cpu_buffer->reader_page->list)) { | |
77ae365e SR |
3323 | iter->head_page = rb_set_head_page(cpu_buffer); |
3324 | if (unlikely(!iter->head_page)) | |
3325 | return; | |
3326 | iter->head = iter->head_page->read; | |
d769041f SR |
3327 | } else { |
3328 | iter->head_page = cpu_buffer->reader_page; | |
6f807acd | 3329 | iter->head = cpu_buffer->reader_page->read; |
d769041f SR |
3330 | } |
3331 | if (iter->head) | |
3332 | iter->read_stamp = cpu_buffer->read_stamp; | |
3333 | else | |
abc9b56d | 3334 | iter->read_stamp = iter->head_page->page->time_stamp; |
492a74f4 SR |
3335 | iter->cache_reader_page = cpu_buffer->reader_page; |
3336 | iter->cache_read = cpu_buffer->read; | |
642edba5 | 3337 | } |
f83c9d0f | 3338 | |
642edba5 SR |
3339 | /** |
3340 | * ring_buffer_iter_reset - reset an iterator | |
3341 | * @iter: The iterator to reset | |
3342 | * | |
3343 | * Resets the iterator, so that it will start from the beginning | |
3344 | * again. | |
3345 | */ | |
3346 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
3347 | { | |
554f786e | 3348 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
3349 | unsigned long flags; |
3350 | ||
554f786e SR |
3351 | if (!iter) |
3352 | return; | |
3353 | ||
3354 | cpu_buffer = iter->cpu_buffer; | |
3355 | ||
5389f6fa | 3356 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
642edba5 | 3357 | rb_iter_reset(iter); |
5389f6fa | 3358 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 3359 | } |
c4f50183 | 3360 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
3361 | |
3362 | /** | |
3363 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
3364 | * @iter: The iterator to check | |
3365 | */ | |
3366 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
3367 | { | |
3368 | struct ring_buffer_per_cpu *cpu_buffer; | |
3369 | ||
3370 | cpu_buffer = iter->cpu_buffer; | |
3371 | ||
bf41a158 SR |
3372 | return iter->head_page == cpu_buffer->commit_page && |
3373 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 3374 | } |
c4f50183 | 3375 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
3376 | |
3377 | static void | |
3378 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
3379 | struct ring_buffer_event *event) | |
3380 | { | |
3381 | u64 delta; | |
3382 | ||
334d4169 | 3383 | switch (event->type_len) { |
7a8e76a3 SR |
3384 | case RINGBUF_TYPE_PADDING: |
3385 | return; | |
3386 | ||
3387 | case RINGBUF_TYPE_TIME_EXTEND: | |
3388 | delta = event->array[0]; | |
3389 | delta <<= TS_SHIFT; | |
3390 | delta += event->time_delta; | |
3391 | cpu_buffer->read_stamp += delta; | |
3392 | return; | |
3393 | ||
3394 | case RINGBUF_TYPE_TIME_STAMP: | |
3395 | /* FIXME: not implemented */ | |
3396 | return; | |
3397 | ||
3398 | case RINGBUF_TYPE_DATA: | |
3399 | cpu_buffer->read_stamp += event->time_delta; | |
3400 | return; | |
3401 | ||
3402 | default: | |
3403 | BUG(); | |
3404 | } | |
3405 | return; | |
3406 | } | |
3407 | ||
3408 | static void | |
3409 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
3410 | struct ring_buffer_event *event) | |
3411 | { | |
3412 | u64 delta; | |
3413 | ||
334d4169 | 3414 | switch (event->type_len) { |
7a8e76a3 SR |
3415 | case RINGBUF_TYPE_PADDING: |
3416 | return; | |
3417 | ||
3418 | case RINGBUF_TYPE_TIME_EXTEND: | |
3419 | delta = event->array[0]; | |
3420 | delta <<= TS_SHIFT; | |
3421 | delta += event->time_delta; | |
3422 | iter->read_stamp += delta; | |
3423 | return; | |
3424 | ||
3425 | case RINGBUF_TYPE_TIME_STAMP: | |
3426 | /* FIXME: not implemented */ | |
3427 | return; | |
3428 | ||
3429 | case RINGBUF_TYPE_DATA: | |
3430 | iter->read_stamp += event->time_delta; | |
3431 | return; | |
3432 | ||
3433 | default: | |
3434 | BUG(); | |
3435 | } | |
3436 | return; | |
3437 | } | |
3438 | ||
d769041f SR |
3439 | static struct buffer_page * |
3440 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 3441 | { |
d769041f | 3442 | struct buffer_page *reader = NULL; |
66a8cb95 | 3443 | unsigned long overwrite; |
d769041f | 3444 | unsigned long flags; |
818e3dd3 | 3445 | int nr_loops = 0; |
77ae365e | 3446 | int ret; |
d769041f | 3447 | |
3e03fb7f | 3448 | local_irq_save(flags); |
0199c4e6 | 3449 | arch_spin_lock(&cpu_buffer->lock); |
d769041f SR |
3450 | |
3451 | again: | |
818e3dd3 SR |
3452 | /* |
3453 | * This should normally only loop twice. But because the | |
3454 | * start of the reader inserts an empty page, it causes | |
3455 | * a case where we will loop three times. There should be no | |
3456 | * reason to loop four times (that I know of). | |
3457 | */ | |
3e89c7bb | 3458 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
3459 | reader = NULL; |
3460 | goto out; | |
3461 | } | |
3462 | ||
d769041f SR |
3463 | reader = cpu_buffer->reader_page; |
3464 | ||
3465 | /* If there's more to read, return this page */ | |
bf41a158 | 3466 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
3467 | goto out; |
3468 | ||
3469 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
3470 | if (RB_WARN_ON(cpu_buffer, |
3471 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
3472 | goto out; | |
d769041f SR |
3473 | |
3474 | /* check if we caught up to the tail */ | |
3475 | reader = NULL; | |
bf41a158 | 3476 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 3477 | goto out; |
7a8e76a3 | 3478 | |
a5fb8331 SR |
3479 | /* Don't bother swapping if the ring buffer is empty */ |
3480 | if (rb_num_of_entries(cpu_buffer) == 0) | |
3481 | goto out; | |
3482 | ||
7a8e76a3 | 3483 | /* |
d769041f | 3484 | * Reset the reader page to size zero. |
7a8e76a3 | 3485 | */ |
77ae365e SR |
3486 | local_set(&cpu_buffer->reader_page->write, 0); |
3487 | local_set(&cpu_buffer->reader_page->entries, 0); | |
3488 | local_set(&cpu_buffer->reader_page->page->commit, 0); | |
ff0ff84a | 3489 | cpu_buffer->reader_page->real_end = 0; |
7a8e76a3 | 3490 | |
77ae365e SR |
3491 | spin: |
3492 | /* | |
3493 | * Splice the empty reader page into the list around the head. | |
3494 | */ | |
3495 | reader = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
3496 | if (!reader) |
3497 | goto out; | |
0e1ff5d7 | 3498 | cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next); |
d769041f | 3499 | cpu_buffer->reader_page->list.prev = reader->list.prev; |
bf41a158 | 3500 | |
3adc54fa SR |
3501 | /* |
3502 | * cpu_buffer->pages just needs to point to the buffer, it | |
3503 | * has no specific buffer page to point to. Lets move it out | |
25985edc | 3504 | * of our way so we don't accidentally swap it. |
3adc54fa SR |
3505 | */ |
3506 | cpu_buffer->pages = reader->list.prev; | |
3507 | ||
77ae365e SR |
3508 | /* The reader page will be pointing to the new head */ |
3509 | rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list); | |
7a8e76a3 | 3510 | |
66a8cb95 SR |
3511 | /* |
3512 | * We want to make sure we read the overruns after we set up our | |
3513 | * pointers to the next object. The writer side does a | |
3514 | * cmpxchg to cross pages which acts as the mb on the writer | |
3515 | * side. Note, the reader will constantly fail the swap | |
3516 | * while the writer is updating the pointers, so this | |
3517 | * guarantees that the overwrite recorded here is the one we | |
3518 | * want to compare with the last_overrun. | |
3519 | */ | |
3520 | smp_mb(); | |
3521 | overwrite = local_read(&(cpu_buffer->overrun)); | |
3522 | ||
77ae365e SR |
3523 | /* |
3524 | * Here's the tricky part. | |
3525 | * | |
3526 | * We need to move the pointer past the header page. | |
3527 | * But we can only do that if a writer is not currently | |
3528 | * moving it. The page before the header page has the | |
3529 | * flag bit '1' set if it is pointing to the page we want. | |
3530 | * but if the writer is in the process of moving it | |
3531 | * than it will be '2' or already moved '0'. | |
3532 | */ | |
3533 | ||
3534 | ret = rb_head_page_replace(reader, cpu_buffer->reader_page); | |
7a8e76a3 SR |
3535 | |
3536 | /* | |
77ae365e | 3537 | * If we did not convert it, then we must try again. |
7a8e76a3 | 3538 | */ |
77ae365e SR |
3539 | if (!ret) |
3540 | goto spin; | |
7a8e76a3 | 3541 | |
77ae365e SR |
3542 | /* |
3543 | * Yeah! We succeeded in replacing the page. | |
3544 | * | |
3545 | * Now make the new head point back to the reader page. | |
3546 | */ | |
5ded3dc6 | 3547 | rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list; |
77ae365e | 3548 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
d769041f SR |
3549 | |
3550 | /* Finally update the reader page to the new head */ | |
3551 | cpu_buffer->reader_page = reader; | |
3552 | rb_reset_reader_page(cpu_buffer); | |
3553 | ||
66a8cb95 SR |
3554 | if (overwrite != cpu_buffer->last_overrun) { |
3555 | cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun; | |
3556 | cpu_buffer->last_overrun = overwrite; | |
3557 | } | |
3558 | ||
d769041f SR |
3559 | goto again; |
3560 | ||
3561 | out: | |
0199c4e6 | 3562 | arch_spin_unlock(&cpu_buffer->lock); |
3e03fb7f | 3563 | local_irq_restore(flags); |
d769041f SR |
3564 | |
3565 | return reader; | |
3566 | } | |
3567 | ||
3568 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
3569 | { | |
3570 | struct ring_buffer_event *event; | |
3571 | struct buffer_page *reader; | |
3572 | unsigned length; | |
3573 | ||
3574 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 3575 | |
d769041f | 3576 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
3577 | if (RB_WARN_ON(cpu_buffer, !reader)) |
3578 | return; | |
7a8e76a3 | 3579 | |
d769041f SR |
3580 | event = rb_reader_event(cpu_buffer); |
3581 | ||
a1863c21 | 3582 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
e4906eff | 3583 | cpu_buffer->read++; |
d769041f SR |
3584 | |
3585 | rb_update_read_stamp(cpu_buffer, event); | |
3586 | ||
3587 | length = rb_event_length(event); | |
6f807acd | 3588 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
3589 | } |
3590 | ||
3591 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
3592 | { | |
7a8e76a3 SR |
3593 | struct ring_buffer_per_cpu *cpu_buffer; |
3594 | struct ring_buffer_event *event; | |
3595 | unsigned length; | |
3596 | ||
3597 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 SR |
3598 | |
3599 | /* | |
3600 | * Check if we are at the end of the buffer. | |
3601 | */ | |
bf41a158 | 3602 | if (iter->head >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
3603 | /* discarded commits can make the page empty */ |
3604 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 3605 | return; |
d769041f | 3606 | rb_inc_iter(iter); |
7a8e76a3 SR |
3607 | return; |
3608 | } | |
3609 | ||
3610 | event = rb_iter_head_event(iter); | |
3611 | ||
3612 | length = rb_event_length(event); | |
3613 | ||
3614 | /* | |
3615 | * This should not be called to advance the header if we are | |
3616 | * at the tail of the buffer. | |
3617 | */ | |
3e89c7bb | 3618 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 3619 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
3620 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
3621 | return; | |
7a8e76a3 SR |
3622 | |
3623 | rb_update_iter_read_stamp(iter, event); | |
3624 | ||
3625 | iter->head += length; | |
3626 | ||
3627 | /* check for end of page padding */ | |
bf41a158 SR |
3628 | if ((iter->head >= rb_page_size(iter->head_page)) && |
3629 | (iter->head_page != cpu_buffer->commit_page)) | |
771e0384 | 3630 | rb_inc_iter(iter); |
7a8e76a3 SR |
3631 | } |
3632 | ||
66a8cb95 SR |
3633 | static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer) |
3634 | { | |
3635 | return cpu_buffer->lost_events; | |
3636 | } | |
3637 | ||
f83c9d0f | 3638 | static struct ring_buffer_event * |
66a8cb95 SR |
3639 | rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts, |
3640 | unsigned long *lost_events) | |
7a8e76a3 | 3641 | { |
7a8e76a3 | 3642 | struct ring_buffer_event *event; |
d769041f | 3643 | struct buffer_page *reader; |
818e3dd3 | 3644 | int nr_loops = 0; |
7a8e76a3 | 3645 | |
7a8e76a3 | 3646 | again: |
818e3dd3 | 3647 | /* |
69d1b839 SR |
3648 | * We repeat when a time extend is encountered. |
3649 | * Since the time extend is always attached to a data event, | |
3650 | * we should never loop more than once. | |
3651 | * (We never hit the following condition more than twice). | |
818e3dd3 | 3652 | */ |
69d1b839 | 3653 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 3654 | return NULL; |
818e3dd3 | 3655 | |
d769041f SR |
3656 | reader = rb_get_reader_page(cpu_buffer); |
3657 | if (!reader) | |
7a8e76a3 SR |
3658 | return NULL; |
3659 | ||
d769041f | 3660 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 3661 | |
334d4169 | 3662 | switch (event->type_len) { |
7a8e76a3 | 3663 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3664 | if (rb_null_event(event)) |
3665 | RB_WARN_ON(cpu_buffer, 1); | |
3666 | /* | |
3667 | * Because the writer could be discarding every | |
3668 | * event it creates (which would probably be bad) | |
3669 | * if we were to go back to "again" then we may never | |
3670 | * catch up, and will trigger the warn on, or lock | |
3671 | * the box. Return the padding, and we will release | |
3672 | * the current locks, and try again. | |
3673 | */ | |
2d622719 | 3674 | return event; |
7a8e76a3 SR |
3675 | |
3676 | case RINGBUF_TYPE_TIME_EXTEND: | |
3677 | /* Internal data, OK to advance */ | |
d769041f | 3678 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3679 | goto again; |
3680 | ||
3681 | case RINGBUF_TYPE_TIME_STAMP: | |
3682 | /* FIXME: not implemented */ | |
d769041f | 3683 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3684 | goto again; |
3685 | ||
3686 | case RINGBUF_TYPE_DATA: | |
3687 | if (ts) { | |
3688 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
d8eeb2d3 | 3689 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, |
37886f6a | 3690 | cpu_buffer->cpu, ts); |
7a8e76a3 | 3691 | } |
66a8cb95 SR |
3692 | if (lost_events) |
3693 | *lost_events = rb_lost_events(cpu_buffer); | |
7a8e76a3 SR |
3694 | return event; |
3695 | ||
3696 | default: | |
3697 | BUG(); | |
3698 | } | |
3699 | ||
3700 | return NULL; | |
3701 | } | |
c4f50183 | 3702 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 3703 | |
f83c9d0f SR |
3704 | static struct ring_buffer_event * |
3705 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
3706 | { |
3707 | struct ring_buffer *buffer; | |
3708 | struct ring_buffer_per_cpu *cpu_buffer; | |
3709 | struct ring_buffer_event *event; | |
818e3dd3 | 3710 | int nr_loops = 0; |
7a8e76a3 | 3711 | |
7a8e76a3 SR |
3712 | cpu_buffer = iter->cpu_buffer; |
3713 | buffer = cpu_buffer->buffer; | |
3714 | ||
492a74f4 SR |
3715 | /* |
3716 | * Check if someone performed a consuming read to | |
3717 | * the buffer. A consuming read invalidates the iterator | |
3718 | * and we need to reset the iterator in this case. | |
3719 | */ | |
3720 | if (unlikely(iter->cache_read != cpu_buffer->read || | |
3721 | iter->cache_reader_page != cpu_buffer->reader_page)) | |
3722 | rb_iter_reset(iter); | |
3723 | ||
7a8e76a3 | 3724 | again: |
3c05d748 SR |
3725 | if (ring_buffer_iter_empty(iter)) |
3726 | return NULL; | |
3727 | ||
818e3dd3 | 3728 | /* |
69d1b839 SR |
3729 | * We repeat when a time extend is encountered. |
3730 | * Since the time extend is always attached to a data event, | |
3731 | * we should never loop more than once. | |
3732 | * (We never hit the following condition more than twice). | |
818e3dd3 | 3733 | */ |
69d1b839 | 3734 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 3735 | return NULL; |
818e3dd3 | 3736 | |
7a8e76a3 SR |
3737 | if (rb_per_cpu_empty(cpu_buffer)) |
3738 | return NULL; | |
3739 | ||
3c05d748 SR |
3740 | if (iter->head >= local_read(&iter->head_page->page->commit)) { |
3741 | rb_inc_iter(iter); | |
3742 | goto again; | |
3743 | } | |
3744 | ||
7a8e76a3 SR |
3745 | event = rb_iter_head_event(iter); |
3746 | ||
334d4169 | 3747 | switch (event->type_len) { |
7a8e76a3 | 3748 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3749 | if (rb_null_event(event)) { |
3750 | rb_inc_iter(iter); | |
3751 | goto again; | |
3752 | } | |
3753 | rb_advance_iter(iter); | |
3754 | return event; | |
7a8e76a3 SR |
3755 | |
3756 | case RINGBUF_TYPE_TIME_EXTEND: | |
3757 | /* Internal data, OK to advance */ | |
3758 | rb_advance_iter(iter); | |
3759 | goto again; | |
3760 | ||
3761 | case RINGBUF_TYPE_TIME_STAMP: | |
3762 | /* FIXME: not implemented */ | |
3763 | rb_advance_iter(iter); | |
3764 | goto again; | |
3765 | ||
3766 | case RINGBUF_TYPE_DATA: | |
3767 | if (ts) { | |
3768 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
3769 | ring_buffer_normalize_time_stamp(buffer, |
3770 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
3771 | } |
3772 | return event; | |
3773 | ||
3774 | default: | |
3775 | BUG(); | |
3776 | } | |
3777 | ||
3778 | return NULL; | |
3779 | } | |
c4f50183 | 3780 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 3781 | |
8d707e8e SR |
3782 | static inline int rb_ok_to_lock(void) |
3783 | { | |
3784 | /* | |
3785 | * If an NMI die dumps out the content of the ring buffer | |
3786 | * do not grab locks. We also permanently disable the ring | |
3787 | * buffer too. A one time deal is all you get from reading | |
3788 | * the ring buffer from an NMI. | |
3789 | */ | |
464e85eb | 3790 | if (likely(!in_nmi())) |
8d707e8e SR |
3791 | return 1; |
3792 | ||
3793 | tracing_off_permanent(); | |
3794 | return 0; | |
3795 | } | |
3796 | ||
f83c9d0f SR |
3797 | /** |
3798 | * ring_buffer_peek - peek at the next event to be read | |
3799 | * @buffer: The ring buffer to read | |
3800 | * @cpu: The cpu to peak at | |
3801 | * @ts: The timestamp counter of this event. | |
66a8cb95 | 3802 | * @lost_events: a variable to store if events were lost (may be NULL) |
f83c9d0f SR |
3803 | * |
3804 | * This will return the event that will be read next, but does | |
3805 | * not consume the data. | |
3806 | */ | |
3807 | struct ring_buffer_event * | |
66a8cb95 SR |
3808 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts, |
3809 | unsigned long *lost_events) | |
f83c9d0f SR |
3810 | { |
3811 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 3812 | struct ring_buffer_event *event; |
f83c9d0f | 3813 | unsigned long flags; |
8d707e8e | 3814 | int dolock; |
f83c9d0f | 3815 | |
554f786e | 3816 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3817 | return NULL; |
554f786e | 3818 | |
8d707e8e | 3819 | dolock = rb_ok_to_lock(); |
2d622719 | 3820 | again: |
8d707e8e SR |
3821 | local_irq_save(flags); |
3822 | if (dolock) | |
5389f6fa | 3823 | raw_spin_lock(&cpu_buffer->reader_lock); |
66a8cb95 | 3824 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
469535a5 RR |
3825 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
3826 | rb_advance_reader(cpu_buffer); | |
8d707e8e | 3827 | if (dolock) |
5389f6fa | 3828 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3829 | local_irq_restore(flags); |
f83c9d0f | 3830 | |
1b959e18 | 3831 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3832 | goto again; |
2d622719 | 3833 | |
f83c9d0f SR |
3834 | return event; |
3835 | } | |
3836 | ||
3837 | /** | |
3838 | * ring_buffer_iter_peek - peek at the next event to be read | |
3839 | * @iter: The ring buffer iterator | |
3840 | * @ts: The timestamp counter of this event. | |
3841 | * | |
3842 | * This will return the event that will be read next, but does | |
3843 | * not increment the iterator. | |
3844 | */ | |
3845 | struct ring_buffer_event * | |
3846 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
3847 | { | |
3848 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3849 | struct ring_buffer_event *event; | |
3850 | unsigned long flags; | |
3851 | ||
2d622719 | 3852 | again: |
5389f6fa | 3853 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3854 | event = rb_iter_peek(iter, ts); |
5389f6fa | 3855 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3856 | |
1b959e18 | 3857 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3858 | goto again; |
2d622719 | 3859 | |
f83c9d0f SR |
3860 | return event; |
3861 | } | |
3862 | ||
7a8e76a3 SR |
3863 | /** |
3864 | * ring_buffer_consume - return an event and consume it | |
3865 | * @buffer: The ring buffer to get the next event from | |
66a8cb95 SR |
3866 | * @cpu: the cpu to read the buffer from |
3867 | * @ts: a variable to store the timestamp (may be NULL) | |
3868 | * @lost_events: a variable to store if events were lost (may be NULL) | |
7a8e76a3 SR |
3869 | * |
3870 | * Returns the next event in the ring buffer, and that event is consumed. | |
3871 | * Meaning, that sequential reads will keep returning a different event, | |
3872 | * and eventually empty the ring buffer if the producer is slower. | |
3873 | */ | |
3874 | struct ring_buffer_event * | |
66a8cb95 SR |
3875 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts, |
3876 | unsigned long *lost_events) | |
7a8e76a3 | 3877 | { |
554f786e SR |
3878 | struct ring_buffer_per_cpu *cpu_buffer; |
3879 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 3880 | unsigned long flags; |
8d707e8e SR |
3881 | int dolock; |
3882 | ||
3883 | dolock = rb_ok_to_lock(); | |
7a8e76a3 | 3884 | |
2d622719 | 3885 | again: |
554f786e SR |
3886 | /* might be called in atomic */ |
3887 | preempt_disable(); | |
3888 | ||
9e01c1b7 | 3889 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 3890 | goto out; |
7a8e76a3 | 3891 | |
554f786e | 3892 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
3893 | local_irq_save(flags); |
3894 | if (dolock) | |
5389f6fa | 3895 | raw_spin_lock(&cpu_buffer->reader_lock); |
f83c9d0f | 3896 | |
66a8cb95 SR |
3897 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
3898 | if (event) { | |
3899 | cpu_buffer->lost_events = 0; | |
469535a5 | 3900 | rb_advance_reader(cpu_buffer); |
66a8cb95 | 3901 | } |
7a8e76a3 | 3902 | |
8d707e8e | 3903 | if (dolock) |
5389f6fa | 3904 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3905 | local_irq_restore(flags); |
f83c9d0f | 3906 | |
554f786e SR |
3907 | out: |
3908 | preempt_enable(); | |
3909 | ||
1b959e18 | 3910 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3911 | goto again; |
2d622719 | 3912 | |
7a8e76a3 SR |
3913 | return event; |
3914 | } | |
c4f50183 | 3915 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
3916 | |
3917 | /** | |
72c9ddfd | 3918 | * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer |
7a8e76a3 SR |
3919 | * @buffer: The ring buffer to read from |
3920 | * @cpu: The cpu buffer to iterate over | |
3921 | * | |
72c9ddfd DM |
3922 | * This performs the initial preparations necessary to iterate |
3923 | * through the buffer. Memory is allocated, buffer recording | |
3924 | * is disabled, and the iterator pointer is returned to the caller. | |
7a8e76a3 | 3925 | * |
72c9ddfd DM |
3926 | * Disabling buffer recordng prevents the reading from being |
3927 | * corrupted. This is not a consuming read, so a producer is not | |
3928 | * expected. | |
3929 | * | |
3930 | * After a sequence of ring_buffer_read_prepare calls, the user is | |
3931 | * expected to make at least one call to ring_buffer_prepare_sync. | |
3932 | * Afterwards, ring_buffer_read_start is invoked to get things going | |
3933 | * for real. | |
3934 | * | |
3935 | * This overall must be paired with ring_buffer_finish. | |
7a8e76a3 SR |
3936 | */ |
3937 | struct ring_buffer_iter * | |
72c9ddfd | 3938 | ring_buffer_read_prepare(struct ring_buffer *buffer, int cpu) |
7a8e76a3 SR |
3939 | { |
3940 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3941 | struct ring_buffer_iter *iter; |
7a8e76a3 | 3942 | |
9e01c1b7 | 3943 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3944 | return NULL; |
7a8e76a3 SR |
3945 | |
3946 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
3947 | if (!iter) | |
8aabee57 | 3948 | return NULL; |
7a8e76a3 SR |
3949 | |
3950 | cpu_buffer = buffer->buffers[cpu]; | |
3951 | ||
3952 | iter->cpu_buffer = cpu_buffer; | |
3953 | ||
83f40318 | 3954 | atomic_inc(&buffer->resize_disabled); |
7a8e76a3 | 3955 | atomic_inc(&cpu_buffer->record_disabled); |
72c9ddfd DM |
3956 | |
3957 | return iter; | |
3958 | } | |
3959 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare); | |
3960 | ||
3961 | /** | |
3962 | * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls | |
3963 | * | |
3964 | * All previously invoked ring_buffer_read_prepare calls to prepare | |
3965 | * iterators will be synchronized. Afterwards, read_buffer_read_start | |
3966 | * calls on those iterators are allowed. | |
3967 | */ | |
3968 | void | |
3969 | ring_buffer_read_prepare_sync(void) | |
3970 | { | |
7a8e76a3 | 3971 | synchronize_sched(); |
72c9ddfd DM |
3972 | } |
3973 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync); | |
3974 | ||
3975 | /** | |
3976 | * ring_buffer_read_start - start a non consuming read of the buffer | |
3977 | * @iter: The iterator returned by ring_buffer_read_prepare | |
3978 | * | |
3979 | * This finalizes the startup of an iteration through the buffer. | |
3980 | * The iterator comes from a call to ring_buffer_read_prepare and | |
3981 | * an intervening ring_buffer_read_prepare_sync must have been | |
3982 | * performed. | |
3983 | * | |
3984 | * Must be paired with ring_buffer_finish. | |
3985 | */ | |
3986 | void | |
3987 | ring_buffer_read_start(struct ring_buffer_iter *iter) | |
3988 | { | |
3989 | struct ring_buffer_per_cpu *cpu_buffer; | |
3990 | unsigned long flags; | |
3991 | ||
3992 | if (!iter) | |
3993 | return; | |
3994 | ||
3995 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 | 3996 | |
5389f6fa | 3997 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
0199c4e6 | 3998 | arch_spin_lock(&cpu_buffer->lock); |
642edba5 | 3999 | rb_iter_reset(iter); |
0199c4e6 | 4000 | arch_spin_unlock(&cpu_buffer->lock); |
5389f6fa | 4001 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 4002 | } |
c4f50183 | 4003 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
4004 | |
4005 | /** | |
4006 | * ring_buffer_finish - finish reading the iterator of the buffer | |
4007 | * @iter: The iterator retrieved by ring_buffer_start | |
4008 | * | |
4009 | * This re-enables the recording to the buffer, and frees the | |
4010 | * iterator. | |
4011 | */ | |
4012 | void | |
4013 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
4014 | { | |
4015 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
9366c1ba | 4016 | unsigned long flags; |
7a8e76a3 | 4017 | |
659f451f SR |
4018 | /* |
4019 | * Ring buffer is disabled from recording, here's a good place | |
9366c1ba SR |
4020 | * to check the integrity of the ring buffer. |
4021 | * Must prevent readers from trying to read, as the check | |
4022 | * clears the HEAD page and readers require it. | |
659f451f | 4023 | */ |
9366c1ba | 4024 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
659f451f | 4025 | rb_check_pages(cpu_buffer); |
9366c1ba | 4026 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
659f451f | 4027 | |
7a8e76a3 | 4028 | atomic_dec(&cpu_buffer->record_disabled); |
83f40318 | 4029 | atomic_dec(&cpu_buffer->buffer->resize_disabled); |
7a8e76a3 SR |
4030 | kfree(iter); |
4031 | } | |
c4f50183 | 4032 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
4033 | |
4034 | /** | |
4035 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
4036 | * @iter: The ring buffer iterator | |
4037 | * @ts: The time stamp of the event read. | |
4038 | * | |
4039 | * This reads the next event in the ring buffer and increments the iterator. | |
4040 | */ | |
4041 | struct ring_buffer_event * | |
4042 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
4043 | { | |
4044 | struct ring_buffer_event *event; | |
f83c9d0f SR |
4045 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
4046 | unsigned long flags; | |
7a8e76a3 | 4047 | |
5389f6fa | 4048 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
7e9391cf | 4049 | again: |
f83c9d0f | 4050 | event = rb_iter_peek(iter, ts); |
7a8e76a3 | 4051 | if (!event) |
f83c9d0f | 4052 | goto out; |
7a8e76a3 | 4053 | |
7e9391cf SR |
4054 | if (event->type_len == RINGBUF_TYPE_PADDING) |
4055 | goto again; | |
4056 | ||
7a8e76a3 | 4057 | rb_advance_iter(iter); |
f83c9d0f | 4058 | out: |
5389f6fa | 4059 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
4060 | |
4061 | return event; | |
4062 | } | |
c4f50183 | 4063 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
4064 | |
4065 | /** | |
4066 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
4067 | * @buffer: The ring buffer. | |
4068 | */ | |
438ced17 | 4069 | unsigned long ring_buffer_size(struct ring_buffer *buffer, int cpu) |
7a8e76a3 | 4070 | { |
438ced17 VN |
4071 | /* |
4072 | * Earlier, this method returned | |
4073 | * BUF_PAGE_SIZE * buffer->nr_pages | |
4074 | * Since the nr_pages field is now removed, we have converted this to | |
4075 | * return the per cpu buffer value. | |
4076 | */ | |
4077 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
4078 | return 0; | |
4079 | ||
4080 | return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages; | |
7a8e76a3 | 4081 | } |
c4f50183 | 4082 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
4083 | |
4084 | static void | |
4085 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
4086 | { | |
77ae365e SR |
4087 | rb_head_page_deactivate(cpu_buffer); |
4088 | ||
7a8e76a3 | 4089 | cpu_buffer->head_page |
3adc54fa | 4090 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 4091 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 4092 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 4093 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 4094 | |
6f807acd | 4095 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
4096 | |
4097 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
4098 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
4099 | ||
4100 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
5040b4b7 | 4101 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
bf41a158 | 4102 | local_set(&cpu_buffer->reader_page->write, 0); |
778c55d4 | 4103 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 4104 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 4105 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 4106 | |
c64e148a | 4107 | local_set(&cpu_buffer->entries_bytes, 0); |
77ae365e | 4108 | local_set(&cpu_buffer->overrun, 0); |
884bfe89 SP |
4109 | local_set(&cpu_buffer->commit_overrun, 0); |
4110 | local_set(&cpu_buffer->dropped_events, 0); | |
e4906eff | 4111 | local_set(&cpu_buffer->entries, 0); |
fa743953 SR |
4112 | local_set(&cpu_buffer->committing, 0); |
4113 | local_set(&cpu_buffer->commits, 0); | |
77ae365e | 4114 | cpu_buffer->read = 0; |
c64e148a | 4115 | cpu_buffer->read_bytes = 0; |
69507c06 SR |
4116 | |
4117 | cpu_buffer->write_stamp = 0; | |
4118 | cpu_buffer->read_stamp = 0; | |
77ae365e | 4119 | |
66a8cb95 SR |
4120 | cpu_buffer->lost_events = 0; |
4121 | cpu_buffer->last_overrun = 0; | |
4122 | ||
77ae365e | 4123 | rb_head_page_activate(cpu_buffer); |
7a8e76a3 SR |
4124 | } |
4125 | ||
4126 | /** | |
4127 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
4128 | * @buffer: The ring buffer to reset a per cpu buffer of | |
4129 | * @cpu: The CPU buffer to be reset | |
4130 | */ | |
4131 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
4132 | { | |
4133 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
4134 | unsigned long flags; | |
4135 | ||
9e01c1b7 | 4136 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4137 | return; |
7a8e76a3 | 4138 | |
83f40318 | 4139 | atomic_inc(&buffer->resize_disabled); |
41ede23e SR |
4140 | atomic_inc(&cpu_buffer->record_disabled); |
4141 | ||
83f40318 VN |
4142 | /* Make sure all commits have finished */ |
4143 | synchronize_sched(); | |
4144 | ||
5389f6fa | 4145 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 4146 | |
41b6a95d SR |
4147 | if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing))) |
4148 | goto out; | |
4149 | ||
0199c4e6 | 4150 | arch_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
4151 | |
4152 | rb_reset_cpu(cpu_buffer); | |
4153 | ||
0199c4e6 | 4154 | arch_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 4155 | |
41b6a95d | 4156 | out: |
5389f6fa | 4157 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
41ede23e SR |
4158 | |
4159 | atomic_dec(&cpu_buffer->record_disabled); | |
83f40318 | 4160 | atomic_dec(&buffer->resize_disabled); |
7a8e76a3 | 4161 | } |
c4f50183 | 4162 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
4163 | |
4164 | /** | |
4165 | * ring_buffer_reset - reset a ring buffer | |
4166 | * @buffer: The ring buffer to reset all cpu buffers | |
4167 | */ | |
4168 | void ring_buffer_reset(struct ring_buffer *buffer) | |
4169 | { | |
7a8e76a3 SR |
4170 | int cpu; |
4171 | ||
7a8e76a3 | 4172 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 4173 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 4174 | } |
c4f50183 | 4175 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
4176 | |
4177 | /** | |
4178 | * rind_buffer_empty - is the ring buffer empty? | |
4179 | * @buffer: The ring buffer to test | |
4180 | */ | |
4181 | int ring_buffer_empty(struct ring_buffer *buffer) | |
4182 | { | |
4183 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 4184 | unsigned long flags; |
8d707e8e | 4185 | int dolock; |
7a8e76a3 | 4186 | int cpu; |
d4788207 | 4187 | int ret; |
7a8e76a3 | 4188 | |
8d707e8e | 4189 | dolock = rb_ok_to_lock(); |
7a8e76a3 SR |
4190 | |
4191 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
4192 | for_each_buffer_cpu(buffer, cpu) { | |
4193 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e SR |
4194 | local_irq_save(flags); |
4195 | if (dolock) | |
5389f6fa | 4196 | raw_spin_lock(&cpu_buffer->reader_lock); |
d4788207 | 4197 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e | 4198 | if (dolock) |
5389f6fa | 4199 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e SR |
4200 | local_irq_restore(flags); |
4201 | ||
d4788207 | 4202 | if (!ret) |
7a8e76a3 SR |
4203 | return 0; |
4204 | } | |
554f786e | 4205 | |
7a8e76a3 SR |
4206 | return 1; |
4207 | } | |
c4f50183 | 4208 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
4209 | |
4210 | /** | |
4211 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
4212 | * @buffer: The ring buffer | |
4213 | * @cpu: The CPU buffer to test | |
4214 | */ | |
4215 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
4216 | { | |
4217 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 4218 | unsigned long flags; |
8d707e8e | 4219 | int dolock; |
8aabee57 | 4220 | int ret; |
7a8e76a3 | 4221 | |
9e01c1b7 | 4222 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4223 | return 1; |
7a8e76a3 | 4224 | |
8d707e8e SR |
4225 | dolock = rb_ok_to_lock(); |
4226 | ||
7a8e76a3 | 4227 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
4228 | local_irq_save(flags); |
4229 | if (dolock) | |
5389f6fa | 4230 | raw_spin_lock(&cpu_buffer->reader_lock); |
554f786e | 4231 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e | 4232 | if (dolock) |
5389f6fa | 4233 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 4234 | local_irq_restore(flags); |
554f786e SR |
4235 | |
4236 | return ret; | |
7a8e76a3 | 4237 | } |
c4f50183 | 4238 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 | 4239 | |
85bac32c | 4240 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
7a8e76a3 SR |
4241 | /** |
4242 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
4243 | * @buffer_a: One buffer to swap with | |
4244 | * @buffer_b: The other buffer to swap with | |
4245 | * | |
4246 | * This function is useful for tracers that want to take a "snapshot" | |
4247 | * of a CPU buffer and has another back up buffer lying around. | |
4248 | * it is expected that the tracer handles the cpu buffer not being | |
4249 | * used at the moment. | |
4250 | */ | |
4251 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
4252 | struct ring_buffer *buffer_b, int cpu) | |
4253 | { | |
4254 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
4255 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
4256 | int ret = -EINVAL; |
4257 | ||
9e01c1b7 RR |
4258 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
4259 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 4260 | goto out; |
7a8e76a3 | 4261 | |
438ced17 VN |
4262 | cpu_buffer_a = buffer_a->buffers[cpu]; |
4263 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
4264 | ||
7a8e76a3 | 4265 | /* At least make sure the two buffers are somewhat the same */ |
438ced17 | 4266 | if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages) |
554f786e SR |
4267 | goto out; |
4268 | ||
4269 | ret = -EAGAIN; | |
7a8e76a3 | 4270 | |
97b17efe | 4271 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 4272 | goto out; |
97b17efe SR |
4273 | |
4274 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 4275 | goto out; |
97b17efe SR |
4276 | |
4277 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 4278 | goto out; |
97b17efe | 4279 | |
97b17efe | 4280 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 4281 | goto out; |
97b17efe SR |
4282 | |
4283 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 4284 | goto out; |
97b17efe | 4285 | |
7a8e76a3 SR |
4286 | /* |
4287 | * We can't do a synchronize_sched here because this | |
4288 | * function can be called in atomic context. | |
4289 | * Normally this will be called from the same CPU as cpu. | |
4290 | * If not it's up to the caller to protect this. | |
4291 | */ | |
4292 | atomic_inc(&cpu_buffer_a->record_disabled); | |
4293 | atomic_inc(&cpu_buffer_b->record_disabled); | |
4294 | ||
98277991 SR |
4295 | ret = -EBUSY; |
4296 | if (local_read(&cpu_buffer_a->committing)) | |
4297 | goto out_dec; | |
4298 | if (local_read(&cpu_buffer_b->committing)) | |
4299 | goto out_dec; | |
4300 | ||
7a8e76a3 SR |
4301 | buffer_a->buffers[cpu] = cpu_buffer_b; |
4302 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
4303 | ||
4304 | cpu_buffer_b->buffer = buffer_a; | |
4305 | cpu_buffer_a->buffer = buffer_b; | |
4306 | ||
98277991 SR |
4307 | ret = 0; |
4308 | ||
4309 | out_dec: | |
7a8e76a3 SR |
4310 | atomic_dec(&cpu_buffer_a->record_disabled); |
4311 | atomic_dec(&cpu_buffer_b->record_disabled); | |
554f786e | 4312 | out: |
554f786e | 4313 | return ret; |
7a8e76a3 | 4314 | } |
c4f50183 | 4315 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
85bac32c | 4316 | #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */ |
7a8e76a3 | 4317 | |
8789a9e7 SR |
4318 | /** |
4319 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
4320 | * @buffer: the buffer to allocate for. | |
4321 | * | |
4322 | * This function is used in conjunction with ring_buffer_read_page. | |
4323 | * When reading a full page from the ring buffer, these functions | |
4324 | * can be used to speed up the process. The calling function should | |
4325 | * allocate a few pages first with this function. Then when it | |
4326 | * needs to get pages from the ring buffer, it passes the result | |
4327 | * of this function into ring_buffer_read_page, which will swap | |
4328 | * the page that was allocated, with the read page of the buffer. | |
4329 | * | |
4330 | * Returns: | |
4331 | * The page allocated, or NULL on error. | |
4332 | */ | |
7ea59064 | 4333 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer, int cpu) |
8789a9e7 | 4334 | { |
044fa782 | 4335 | struct buffer_data_page *bpage; |
7ea59064 | 4336 | struct page *page; |
8789a9e7 | 4337 | |
d7ec4bfe VN |
4338 | page = alloc_pages_node(cpu_to_node(cpu), |
4339 | GFP_KERNEL | __GFP_NORETRY, 0); | |
7ea59064 | 4340 | if (!page) |
8789a9e7 SR |
4341 | return NULL; |
4342 | ||
7ea59064 | 4343 | bpage = page_address(page); |
8789a9e7 | 4344 | |
ef7a4a16 SR |
4345 | rb_init_page(bpage); |
4346 | ||
044fa782 | 4347 | return bpage; |
8789a9e7 | 4348 | } |
d6ce96da | 4349 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
4350 | |
4351 | /** | |
4352 | * ring_buffer_free_read_page - free an allocated read page | |
4353 | * @buffer: the buffer the page was allocate for | |
4354 | * @data: the page to free | |
4355 | * | |
4356 | * Free a page allocated from ring_buffer_alloc_read_page. | |
4357 | */ | |
4358 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
4359 | { | |
4360 | free_page((unsigned long)data); | |
4361 | } | |
d6ce96da | 4362 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
4363 | |
4364 | /** | |
4365 | * ring_buffer_read_page - extract a page from the ring buffer | |
4366 | * @buffer: buffer to extract from | |
4367 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 4368 | * @len: amount to extract |
8789a9e7 SR |
4369 | * @cpu: the cpu of the buffer to extract |
4370 | * @full: should the extraction only happen when the page is full. | |
4371 | * | |
4372 | * This function will pull out a page from the ring buffer and consume it. | |
4373 | * @data_page must be the address of the variable that was returned | |
4374 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
4375 | * to swap with a page in the ring buffer. | |
4376 | * | |
4377 | * for example: | |
b85fa01e | 4378 | * rpage = ring_buffer_alloc_read_page(buffer); |
8789a9e7 SR |
4379 | * if (!rpage) |
4380 | * return error; | |
ef7a4a16 | 4381 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
4382 | * if (ret >= 0) |
4383 | * process_page(rpage, ret); | |
8789a9e7 SR |
4384 | * |
4385 | * When @full is set, the function will not return true unless | |
4386 | * the writer is off the reader page. | |
4387 | * | |
4388 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
4389 | * The ring buffer can be used anywhere in the kernel and can not | |
4390 | * blindly call wake_up. The layer that uses the ring buffer must be | |
4391 | * responsible for that. | |
4392 | * | |
4393 | * Returns: | |
667d2412 LJ |
4394 | * >=0 if data has been transferred, returns the offset of consumed data. |
4395 | * <0 if no data has been transferred. | |
8789a9e7 SR |
4396 | */ |
4397 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 4398 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
4399 | { |
4400 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
4401 | struct ring_buffer_event *event; | |
044fa782 | 4402 | struct buffer_data_page *bpage; |
ef7a4a16 | 4403 | struct buffer_page *reader; |
ff0ff84a | 4404 | unsigned long missed_events; |
8789a9e7 | 4405 | unsigned long flags; |
ef7a4a16 | 4406 | unsigned int commit; |
667d2412 | 4407 | unsigned int read; |
4f3640f8 | 4408 | u64 save_timestamp; |
667d2412 | 4409 | int ret = -1; |
8789a9e7 | 4410 | |
554f786e SR |
4411 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
4412 | goto out; | |
4413 | ||
474d32b6 SR |
4414 | /* |
4415 | * If len is not big enough to hold the page header, then | |
4416 | * we can not copy anything. | |
4417 | */ | |
4418 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 4419 | goto out; |
474d32b6 SR |
4420 | |
4421 | len -= BUF_PAGE_HDR_SIZE; | |
4422 | ||
8789a9e7 | 4423 | if (!data_page) |
554f786e | 4424 | goto out; |
8789a9e7 | 4425 | |
044fa782 SR |
4426 | bpage = *data_page; |
4427 | if (!bpage) | |
554f786e | 4428 | goto out; |
8789a9e7 | 4429 | |
5389f6fa | 4430 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4431 | |
ef7a4a16 SR |
4432 | reader = rb_get_reader_page(cpu_buffer); |
4433 | if (!reader) | |
554f786e | 4434 | goto out_unlock; |
8789a9e7 | 4435 | |
ef7a4a16 SR |
4436 | event = rb_reader_event(cpu_buffer); |
4437 | ||
4438 | read = reader->read; | |
4439 | commit = rb_page_commit(reader); | |
667d2412 | 4440 | |
66a8cb95 | 4441 | /* Check if any events were dropped */ |
ff0ff84a | 4442 | missed_events = cpu_buffer->lost_events; |
66a8cb95 | 4443 | |
8789a9e7 | 4444 | /* |
474d32b6 SR |
4445 | * If this page has been partially read or |
4446 | * if len is not big enough to read the rest of the page or | |
4447 | * a writer is still on the page, then | |
4448 | * we must copy the data from the page to the buffer. | |
4449 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 4450 | */ |
474d32b6 | 4451 | if (read || (len < (commit - read)) || |
ef7a4a16 | 4452 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 4453 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
4454 | unsigned int rpos = read; |
4455 | unsigned int pos = 0; | |
ef7a4a16 | 4456 | unsigned int size; |
8789a9e7 SR |
4457 | |
4458 | if (full) | |
554f786e | 4459 | goto out_unlock; |
8789a9e7 | 4460 | |
ef7a4a16 SR |
4461 | if (len > (commit - read)) |
4462 | len = (commit - read); | |
4463 | ||
69d1b839 SR |
4464 | /* Always keep the time extend and data together */ |
4465 | size = rb_event_ts_length(event); | |
ef7a4a16 SR |
4466 | |
4467 | if (len < size) | |
554f786e | 4468 | goto out_unlock; |
ef7a4a16 | 4469 | |
4f3640f8 SR |
4470 | /* save the current timestamp, since the user will need it */ |
4471 | save_timestamp = cpu_buffer->read_stamp; | |
4472 | ||
ef7a4a16 SR |
4473 | /* Need to copy one event at a time */ |
4474 | do { | |
e1e35927 DS |
4475 | /* We need the size of one event, because |
4476 | * rb_advance_reader only advances by one event, | |
4477 | * whereas rb_event_ts_length may include the size of | |
4478 | * one or two events. | |
4479 | * We have already ensured there's enough space if this | |
4480 | * is a time extend. */ | |
4481 | size = rb_event_length(event); | |
474d32b6 | 4482 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
4483 | |
4484 | len -= size; | |
4485 | ||
4486 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
4487 | rpos = reader->read; |
4488 | pos += size; | |
ef7a4a16 | 4489 | |
18fab912 HY |
4490 | if (rpos >= commit) |
4491 | break; | |
4492 | ||
ef7a4a16 | 4493 | event = rb_reader_event(cpu_buffer); |
69d1b839 SR |
4494 | /* Always keep the time extend and data together */ |
4495 | size = rb_event_ts_length(event); | |
e1e35927 | 4496 | } while (len >= size); |
667d2412 LJ |
4497 | |
4498 | /* update bpage */ | |
ef7a4a16 | 4499 | local_set(&bpage->commit, pos); |
4f3640f8 | 4500 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 4501 | |
474d32b6 SR |
4502 | /* we copied everything to the beginning */ |
4503 | read = 0; | |
8789a9e7 | 4504 | } else { |
afbab76a | 4505 | /* update the entry counter */ |
77ae365e | 4506 | cpu_buffer->read += rb_page_entries(reader); |
c64e148a | 4507 | cpu_buffer->read_bytes += BUF_PAGE_SIZE; |
afbab76a | 4508 | |
8789a9e7 | 4509 | /* swap the pages */ |
044fa782 | 4510 | rb_init_page(bpage); |
ef7a4a16 SR |
4511 | bpage = reader->page; |
4512 | reader->page = *data_page; | |
4513 | local_set(&reader->write, 0); | |
778c55d4 | 4514 | local_set(&reader->entries, 0); |
ef7a4a16 | 4515 | reader->read = 0; |
044fa782 | 4516 | *data_page = bpage; |
ff0ff84a SR |
4517 | |
4518 | /* | |
4519 | * Use the real_end for the data size, | |
4520 | * This gives us a chance to store the lost events | |
4521 | * on the page. | |
4522 | */ | |
4523 | if (reader->real_end) | |
4524 | local_set(&bpage->commit, reader->real_end); | |
8789a9e7 | 4525 | } |
667d2412 | 4526 | ret = read; |
8789a9e7 | 4527 | |
66a8cb95 | 4528 | cpu_buffer->lost_events = 0; |
2711ca23 SR |
4529 | |
4530 | commit = local_read(&bpage->commit); | |
66a8cb95 SR |
4531 | /* |
4532 | * Set a flag in the commit field if we lost events | |
4533 | */ | |
ff0ff84a | 4534 | if (missed_events) { |
ff0ff84a SR |
4535 | /* If there is room at the end of the page to save the |
4536 | * missed events, then record it there. | |
4537 | */ | |
4538 | if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) { | |
4539 | memcpy(&bpage->data[commit], &missed_events, | |
4540 | sizeof(missed_events)); | |
4541 | local_add(RB_MISSED_STORED, &bpage->commit); | |
2711ca23 | 4542 | commit += sizeof(missed_events); |
ff0ff84a | 4543 | } |
66a8cb95 | 4544 | local_add(RB_MISSED_EVENTS, &bpage->commit); |
ff0ff84a | 4545 | } |
66a8cb95 | 4546 | |
2711ca23 SR |
4547 | /* |
4548 | * This page may be off to user land. Zero it out here. | |
4549 | */ | |
4550 | if (commit < BUF_PAGE_SIZE) | |
4551 | memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit); | |
4552 | ||
554f786e | 4553 | out_unlock: |
5389f6fa | 4554 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4555 | |
554f786e | 4556 | out: |
8789a9e7 SR |
4557 | return ret; |
4558 | } | |
d6ce96da | 4559 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 4560 | |
59222efe | 4561 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
4562 | static int rb_cpu_notify(struct notifier_block *self, |
4563 | unsigned long action, void *hcpu) | |
554f786e SR |
4564 | { |
4565 | struct ring_buffer *buffer = | |
4566 | container_of(self, struct ring_buffer, cpu_notify); | |
4567 | long cpu = (long)hcpu; | |
438ced17 VN |
4568 | int cpu_i, nr_pages_same; |
4569 | unsigned int nr_pages; | |
554f786e SR |
4570 | |
4571 | switch (action) { | |
4572 | case CPU_UP_PREPARE: | |
4573 | case CPU_UP_PREPARE_FROZEN: | |
3f237a79 | 4574 | if (cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e SR |
4575 | return NOTIFY_OK; |
4576 | ||
438ced17 VN |
4577 | nr_pages = 0; |
4578 | nr_pages_same = 1; | |
4579 | /* check if all cpu sizes are same */ | |
4580 | for_each_buffer_cpu(buffer, cpu_i) { | |
4581 | /* fill in the size from first enabled cpu */ | |
4582 | if (nr_pages == 0) | |
4583 | nr_pages = buffer->buffers[cpu_i]->nr_pages; | |
4584 | if (nr_pages != buffer->buffers[cpu_i]->nr_pages) { | |
4585 | nr_pages_same = 0; | |
4586 | break; | |
4587 | } | |
4588 | } | |
4589 | /* allocate minimum pages, user can later expand it */ | |
4590 | if (!nr_pages_same) | |
4591 | nr_pages = 2; | |
554f786e | 4592 | buffer->buffers[cpu] = |
438ced17 | 4593 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
554f786e SR |
4594 | if (!buffer->buffers[cpu]) { |
4595 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
4596 | cpu); | |
4597 | return NOTIFY_OK; | |
4598 | } | |
4599 | smp_wmb(); | |
3f237a79 | 4600 | cpumask_set_cpu(cpu, buffer->cpumask); |
554f786e SR |
4601 | break; |
4602 | case CPU_DOWN_PREPARE: | |
4603 | case CPU_DOWN_PREPARE_FROZEN: | |
4604 | /* | |
4605 | * Do nothing. | |
4606 | * If we were to free the buffer, then the user would | |
4607 | * lose any trace that was in the buffer. | |
4608 | */ | |
4609 | break; | |
4610 | default: | |
4611 | break; | |
4612 | } | |
4613 | return NOTIFY_OK; | |
4614 | } | |
4615 | #endif |