Commit | Line | Data |
---|---|---|
bcea3f96 | 1 | // SPDX-License-Identifier: GPL-2.0 |
7a8e76a3 SR |
2 | /* |
3 | * Generic ring buffer | |
4 | * | |
5 | * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com> | |
6 | */ | |
af658dca | 7 | #include <linux/trace_events.h> |
7a8e76a3 | 8 | #include <linux/ring_buffer.h> |
14131f2f | 9 | #include <linux/trace_clock.h> |
e6017571 | 10 | #include <linux/sched/clock.h> |
0b07436d | 11 | #include <linux/trace_seq.h> |
7a8e76a3 | 12 | #include <linux/spinlock.h> |
15693458 | 13 | #include <linux/irq_work.h> |
7a8e76a3 | 14 | #include <linux/uaccess.h> |
a81bd80a | 15 | #include <linux/hardirq.h> |
6c43e554 | 16 | #include <linux/kthread.h> /* for self test */ |
7a8e76a3 SR |
17 | #include <linux/module.h> |
18 | #include <linux/percpu.h> | |
19 | #include <linux/mutex.h> | |
6c43e554 | 20 | #include <linux/delay.h> |
5a0e3ad6 | 21 | #include <linux/slab.h> |
7a8e76a3 SR |
22 | #include <linux/init.h> |
23 | #include <linux/hash.h> | |
24 | #include <linux/list.h> | |
554f786e | 25 | #include <linux/cpu.h> |
927e56db | 26 | #include <linux/oom.h> |
7a8e76a3 | 27 | |
79615760 | 28 | #include <asm/local.h> |
182e9f5f | 29 | |
83f40318 VN |
30 | static void update_pages_handler(struct work_struct *work); |
31 | ||
d1b182a8 SR |
32 | /* |
33 | * The ring buffer header is special. We must manually up keep it. | |
34 | */ | |
35 | int ring_buffer_print_entry_header(struct trace_seq *s) | |
36 | { | |
c0cd93aa SRRH |
37 | trace_seq_puts(s, "# compressed entry header\n"); |
38 | trace_seq_puts(s, "\ttype_len : 5 bits\n"); | |
39 | trace_seq_puts(s, "\ttime_delta : 27 bits\n"); | |
40 | trace_seq_puts(s, "\tarray : 32 bits\n"); | |
41 | trace_seq_putc(s, '\n'); | |
42 | trace_seq_printf(s, "\tpadding : type == %d\n", | |
43 | RINGBUF_TYPE_PADDING); | |
44 | trace_seq_printf(s, "\ttime_extend : type == %d\n", | |
45 | RINGBUF_TYPE_TIME_EXTEND); | |
dc4e2801 TZ |
46 | trace_seq_printf(s, "\ttime_stamp : type == %d\n", |
47 | RINGBUF_TYPE_TIME_STAMP); | |
c0cd93aa SRRH |
48 | trace_seq_printf(s, "\tdata max type_len == %d\n", |
49 | RINGBUF_TYPE_DATA_TYPE_LEN_MAX); | |
50 | ||
51 | return !trace_seq_has_overflowed(s); | |
d1b182a8 SR |
52 | } |
53 | ||
5cc98548 SR |
54 | /* |
55 | * The ring buffer is made up of a list of pages. A separate list of pages is | |
56 | * allocated for each CPU. A writer may only write to a buffer that is | |
57 | * associated with the CPU it is currently executing on. A reader may read | |
58 | * from any per cpu buffer. | |
59 | * | |
60 | * The reader is special. For each per cpu buffer, the reader has its own | |
61 | * reader page. When a reader has read the entire reader page, this reader | |
62 | * page is swapped with another page in the ring buffer. | |
63 | * | |
64 | * Now, as long as the writer is off the reader page, the reader can do what | |
65 | * ever it wants with that page. The writer will never write to that page | |
66 | * again (as long as it is out of the ring buffer). | |
67 | * | |
68 | * Here's some silly ASCII art. | |
69 | * | |
70 | * +------+ | |
71 | * |reader| RING BUFFER | |
72 | * |page | | |
73 | * +------+ +---+ +---+ +---+ | |
74 | * | |-->| |-->| | | |
75 | * +---+ +---+ +---+ | |
76 | * ^ | | |
77 | * | | | |
78 | * +---------------+ | |
79 | * | |
80 | * | |
81 | * +------+ | |
82 | * |reader| RING BUFFER | |
83 | * |page |------------------v | |
84 | * +------+ +---+ +---+ +---+ | |
85 | * | |-->| |-->| | | |
86 | * +---+ +---+ +---+ | |
87 | * ^ | | |
88 | * | | | |
89 | * +---------------+ | |
90 | * | |
91 | * | |
92 | * +------+ | |
93 | * |reader| RING BUFFER | |
94 | * |page |------------------v | |
95 | * +------+ +---+ +---+ +---+ | |
96 | * ^ | |-->| |-->| | | |
97 | * | +---+ +---+ +---+ | |
98 | * | | | |
99 | * | | | |
100 | * +------------------------------+ | |
101 | * | |
102 | * | |
103 | * +------+ | |
104 | * |buffer| RING BUFFER | |
105 | * |page |------------------v | |
106 | * +------+ +---+ +---+ +---+ | |
107 | * ^ | | | |-->| | | |
108 | * | New +---+ +---+ +---+ | |
109 | * | Reader------^ | | |
110 | * | page | | |
111 | * +------------------------------+ | |
112 | * | |
113 | * | |
114 | * After we make this swap, the reader can hand this page off to the splice | |
115 | * code and be done with it. It can even allocate a new page if it needs to | |
116 | * and swap that into the ring buffer. | |
117 | * | |
118 | * We will be using cmpxchg soon to make all this lockless. | |
119 | * | |
120 | */ | |
121 | ||
499e5470 SR |
122 | /* Used for individual buffers (after the counter) */ |
123 | #define RB_BUFFER_OFF (1 << 20) | |
a3583244 | 124 | |
499e5470 | 125 | #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data) |
033601a3 | 126 | |
e3d6bf0a | 127 | #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array)) |
67d34724 | 128 | #define RB_ALIGNMENT 4U |
334d4169 | 129 | #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
c7b09308 | 130 | #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */ |
334d4169 | 131 | |
649508f6 | 132 | #ifndef CONFIG_HAVE_64BIT_ALIGNED_ACCESS |
2271048d SR |
133 | # define RB_FORCE_8BYTE_ALIGNMENT 0 |
134 | # define RB_ARCH_ALIGNMENT RB_ALIGNMENT | |
135 | #else | |
136 | # define RB_FORCE_8BYTE_ALIGNMENT 1 | |
137 | # define RB_ARCH_ALIGNMENT 8U | |
138 | #endif | |
139 | ||
649508f6 JH |
140 | #define RB_ALIGN_DATA __aligned(RB_ARCH_ALIGNMENT) |
141 | ||
334d4169 LJ |
142 | /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */ |
143 | #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX | |
7a8e76a3 SR |
144 | |
145 | enum { | |
146 | RB_LEN_TIME_EXTEND = 8, | |
dc4e2801 | 147 | RB_LEN_TIME_STAMP = 8, |
7a8e76a3 SR |
148 | }; |
149 | ||
69d1b839 SR |
150 | #define skip_time_extend(event) \ |
151 | ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND)) | |
152 | ||
dc4e2801 TZ |
153 | #define extended_time(event) \ |
154 | (event->type_len >= RINGBUF_TYPE_TIME_EXTEND) | |
155 | ||
2d622719 TZ |
156 | static inline int rb_null_event(struct ring_buffer_event *event) |
157 | { | |
a1863c21 | 158 | return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta; |
2d622719 TZ |
159 | } |
160 | ||
161 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
162 | { | |
a1863c21 | 163 | /* padding has a NULL time_delta */ |
334d4169 | 164 | event->type_len = RINGBUF_TYPE_PADDING; |
2d622719 TZ |
165 | event->time_delta = 0; |
166 | } | |
167 | ||
34a148bf | 168 | static unsigned |
2d622719 | 169 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
170 | { |
171 | unsigned length; | |
172 | ||
334d4169 LJ |
173 | if (event->type_len) |
174 | length = event->type_len * RB_ALIGNMENT; | |
2d622719 TZ |
175 | else |
176 | length = event->array[0]; | |
177 | return length + RB_EVNT_HDR_SIZE; | |
178 | } | |
179 | ||
69d1b839 SR |
180 | /* |
181 | * Return the length of the given event. Will return | |
182 | * the length of the time extend if the event is a | |
183 | * time extend. | |
184 | */ | |
185 | static inline unsigned | |
2d622719 TZ |
186 | rb_event_length(struct ring_buffer_event *event) |
187 | { | |
334d4169 | 188 | switch (event->type_len) { |
7a8e76a3 | 189 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
190 | if (rb_null_event(event)) |
191 | /* undefined */ | |
192 | return -1; | |
334d4169 | 193 | return event->array[0] + RB_EVNT_HDR_SIZE; |
7a8e76a3 SR |
194 | |
195 | case RINGBUF_TYPE_TIME_EXTEND: | |
196 | return RB_LEN_TIME_EXTEND; | |
197 | ||
198 | case RINGBUF_TYPE_TIME_STAMP: | |
199 | return RB_LEN_TIME_STAMP; | |
200 | ||
201 | case RINGBUF_TYPE_DATA: | |
2d622719 | 202 | return rb_event_data_length(event); |
7a8e76a3 SR |
203 | default: |
204 | BUG(); | |
205 | } | |
206 | /* not hit */ | |
207 | return 0; | |
208 | } | |
209 | ||
69d1b839 SR |
210 | /* |
211 | * Return total length of time extend and data, | |
212 | * or just the event length for all other events. | |
213 | */ | |
214 | static inline unsigned | |
215 | rb_event_ts_length(struct ring_buffer_event *event) | |
216 | { | |
217 | unsigned len = 0; | |
218 | ||
dc4e2801 | 219 | if (extended_time(event)) { |
69d1b839 SR |
220 | /* time extends include the data event after it */ |
221 | len = RB_LEN_TIME_EXTEND; | |
222 | event = skip_time_extend(event); | |
223 | } | |
224 | return len + rb_event_length(event); | |
225 | } | |
226 | ||
7a8e76a3 SR |
227 | /** |
228 | * ring_buffer_event_length - return the length of the event | |
229 | * @event: the event to get the length of | |
69d1b839 SR |
230 | * |
231 | * Returns the size of the data load of a data event. | |
232 | * If the event is something other than a data event, it | |
233 | * returns the size of the event itself. With the exception | |
234 | * of a TIME EXTEND, where it still returns the size of the | |
235 | * data load of the data event after it. | |
7a8e76a3 SR |
236 | */ |
237 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
238 | { | |
69d1b839 SR |
239 | unsigned length; |
240 | ||
dc4e2801 | 241 | if (extended_time(event)) |
69d1b839 SR |
242 | event = skip_time_extend(event); |
243 | ||
244 | length = rb_event_length(event); | |
334d4169 | 245 | if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
465634ad RR |
246 | return length; |
247 | length -= RB_EVNT_HDR_SIZE; | |
248 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
249 | length -= sizeof(event->array[0]); | |
250 | return length; | |
7a8e76a3 | 251 | } |
c4f50183 | 252 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
253 | |
254 | /* inline for ring buffer fast paths */ | |
929ddbf3 | 255 | static __always_inline void * |
7a8e76a3 SR |
256 | rb_event_data(struct ring_buffer_event *event) |
257 | { | |
dc4e2801 | 258 | if (extended_time(event)) |
69d1b839 | 259 | event = skip_time_extend(event); |
334d4169 | 260 | BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX); |
7a8e76a3 | 261 | /* If length is in len field, then array[0] has the data */ |
334d4169 | 262 | if (event->type_len) |
7a8e76a3 SR |
263 | return (void *)&event->array[0]; |
264 | /* Otherwise length is in array[0] and array[1] has the data */ | |
265 | return (void *)&event->array[1]; | |
266 | } | |
267 | ||
268 | /** | |
269 | * ring_buffer_event_data - return the data of the event | |
270 | * @event: the event to get the data from | |
271 | */ | |
272 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
273 | { | |
274 | return rb_event_data(event); | |
275 | } | |
c4f50183 | 276 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
277 | |
278 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 279 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
280 | |
281 | #define TS_SHIFT 27 | |
282 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
283 | #define TS_DELTA_TEST (~TS_MASK) | |
284 | ||
dc4e2801 TZ |
285 | /** |
286 | * ring_buffer_event_time_stamp - return the event's extended timestamp | |
287 | * @event: the event to get the timestamp of | |
288 | * | |
289 | * Returns the extended timestamp associated with a data event. | |
290 | * An extended time_stamp is a 64-bit timestamp represented | |
291 | * internally in a special way that makes the best use of space | |
292 | * contained within a ring buffer event. This function decodes | |
293 | * it and maps it to a straight u64 value. | |
294 | */ | |
295 | u64 ring_buffer_event_time_stamp(struct ring_buffer_event *event) | |
296 | { | |
297 | u64 ts; | |
298 | ||
299 | ts = event->array[0]; | |
300 | ts <<= TS_SHIFT; | |
301 | ts += event->time_delta; | |
302 | ||
303 | return ts; | |
304 | } | |
305 | ||
66a8cb95 SR |
306 | /* Flag when events were overwritten */ |
307 | #define RB_MISSED_EVENTS (1 << 31) | |
ff0ff84a SR |
308 | /* Missed count stored at end */ |
309 | #define RB_MISSED_STORED (1 << 30) | |
66a8cb95 | 310 | |
45d8b80c SRV |
311 | #define RB_MISSED_FLAGS (RB_MISSED_EVENTS|RB_MISSED_STORED) |
312 | ||
abc9b56d | 313 | struct buffer_data_page { |
e4c2ce82 | 314 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 315 | local_t commit; /* write committed index */ |
649508f6 | 316 | unsigned char data[] RB_ALIGN_DATA; /* data of buffer page */ |
abc9b56d SR |
317 | }; |
318 | ||
77ae365e SR |
319 | /* |
320 | * Note, the buffer_page list must be first. The buffer pages | |
321 | * are allocated in cache lines, which means that each buffer | |
322 | * page will be at the beginning of a cache line, and thus | |
323 | * the least significant bits will be zero. We use this to | |
324 | * add flags in the list struct pointers, to make the ring buffer | |
325 | * lockless. | |
326 | */ | |
abc9b56d | 327 | struct buffer_page { |
778c55d4 | 328 | struct list_head list; /* list of buffer pages */ |
abc9b56d | 329 | local_t write; /* index for next write */ |
6f807acd | 330 | unsigned read; /* index for next read */ |
778c55d4 | 331 | local_t entries; /* entries on this page */ |
ff0ff84a | 332 | unsigned long real_end; /* real end of data */ |
abc9b56d | 333 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
334 | }; |
335 | ||
77ae365e SR |
336 | /* |
337 | * The buffer page counters, write and entries, must be reset | |
338 | * atomically when crossing page boundaries. To synchronize this | |
339 | * update, two counters are inserted into the number. One is | |
340 | * the actual counter for the write position or count on the page. | |
341 | * | |
342 | * The other is a counter of updaters. Before an update happens | |
343 | * the update partition of the counter is incremented. This will | |
344 | * allow the updater to update the counter atomically. | |
345 | * | |
346 | * The counter is 20 bits, and the state data is 12. | |
347 | */ | |
348 | #define RB_WRITE_MASK 0xfffff | |
349 | #define RB_WRITE_INTCNT (1 << 20) | |
350 | ||
044fa782 | 351 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 352 | { |
044fa782 | 353 | local_set(&bpage->commit, 0); |
abc9b56d SR |
354 | } |
355 | ||
474d32b6 SR |
356 | /** |
357 | * ring_buffer_page_len - the size of data on the page. | |
358 | * @page: The page to read | |
359 | * | |
360 | * Returns the amount of data on the page, including buffer page header. | |
361 | */ | |
ef7a4a16 SR |
362 | size_t ring_buffer_page_len(void *page) |
363 | { | |
45d8b80c SRV |
364 | struct buffer_data_page *bpage = page; |
365 | ||
366 | return (local_read(&bpage->commit) & ~RB_MISSED_FLAGS) | |
474d32b6 | 367 | + BUF_PAGE_HDR_SIZE; |
ef7a4a16 SR |
368 | } |
369 | ||
ed56829c SR |
370 | /* |
371 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
372 | * this issue out. | |
373 | */ | |
34a148bf | 374 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 375 | { |
34a148bf | 376 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 377 | kfree(bpage); |
ed56829c SR |
378 | } |
379 | ||
7a8e76a3 SR |
380 | /* |
381 | * We need to fit the time_stamp delta into 27 bits. | |
382 | */ | |
383 | static inline int test_time_stamp(u64 delta) | |
384 | { | |
385 | if (delta & TS_DELTA_TEST) | |
386 | return 1; | |
387 | return 0; | |
388 | } | |
389 | ||
474d32b6 | 390 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 | 391 | |
be957c44 SR |
392 | /* Max payload is BUF_PAGE_SIZE - header (8bytes) */ |
393 | #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) | |
394 | ||
d1b182a8 SR |
395 | int ring_buffer_print_page_header(struct trace_seq *s) |
396 | { | |
397 | struct buffer_data_page field; | |
c0cd93aa SRRH |
398 | |
399 | trace_seq_printf(s, "\tfield: u64 timestamp;\t" | |
400 | "offset:0;\tsize:%u;\tsigned:%u;\n", | |
401 | (unsigned int)sizeof(field.time_stamp), | |
402 | (unsigned int)is_signed_type(u64)); | |
403 | ||
404 | trace_seq_printf(s, "\tfield: local_t commit;\t" | |
405 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
406 | (unsigned int)offsetof(typeof(field), commit), | |
407 | (unsigned int)sizeof(field.commit), | |
408 | (unsigned int)is_signed_type(long)); | |
409 | ||
410 | trace_seq_printf(s, "\tfield: int overwrite;\t" | |
411 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
412 | (unsigned int)offsetof(typeof(field), commit), | |
413 | 1, | |
414 | (unsigned int)is_signed_type(long)); | |
415 | ||
416 | trace_seq_printf(s, "\tfield: char data;\t" | |
417 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
418 | (unsigned int)offsetof(typeof(field), data), | |
419 | (unsigned int)BUF_PAGE_SIZE, | |
420 | (unsigned int)is_signed_type(char)); | |
421 | ||
422 | return !trace_seq_has_overflowed(s); | |
d1b182a8 SR |
423 | } |
424 | ||
15693458 SRRH |
425 | struct rb_irq_work { |
426 | struct irq_work work; | |
427 | wait_queue_head_t waiters; | |
1e0d6714 | 428 | wait_queue_head_t full_waiters; |
15693458 | 429 | bool waiters_pending; |
1e0d6714 SRRH |
430 | bool full_waiters_pending; |
431 | bool wakeup_full; | |
15693458 SRRH |
432 | }; |
433 | ||
fcc742ea SRRH |
434 | /* |
435 | * Structure to hold event state and handle nested events. | |
436 | */ | |
437 | struct rb_event_info { | |
438 | u64 ts; | |
439 | u64 delta; | |
440 | unsigned long length; | |
441 | struct buffer_page *tail_page; | |
442 | int add_timestamp; | |
443 | }; | |
444 | ||
a497adb4 SRRH |
445 | /* |
446 | * Used for which event context the event is in. | |
447 | * NMI = 0 | |
448 | * IRQ = 1 | |
449 | * SOFTIRQ = 2 | |
450 | * NORMAL = 3 | |
451 | * | |
452 | * See trace_recursive_lock() comment below for more details. | |
453 | */ | |
454 | enum { | |
455 | RB_CTX_NMI, | |
456 | RB_CTX_IRQ, | |
457 | RB_CTX_SOFTIRQ, | |
458 | RB_CTX_NORMAL, | |
459 | RB_CTX_MAX | |
460 | }; | |
461 | ||
7a8e76a3 SR |
462 | /* |
463 | * head_page == tail_page && head == tail then buffer is empty. | |
464 | */ | |
465 | struct ring_buffer_per_cpu { | |
466 | int cpu; | |
985023de | 467 | atomic_t record_disabled; |
7a8e76a3 | 468 | struct ring_buffer *buffer; |
5389f6fa | 469 | raw_spinlock_t reader_lock; /* serialize readers */ |
445c8951 | 470 | arch_spinlock_t lock; |
7a8e76a3 | 471 | struct lock_class_key lock_key; |
73a757e6 | 472 | struct buffer_data_page *free_page; |
9b94a8fb | 473 | unsigned long nr_pages; |
58a09ec6 | 474 | unsigned int current_context; |
3adc54fa | 475 | struct list_head *pages; |
6f807acd SR |
476 | struct buffer_page *head_page; /* read from head */ |
477 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 478 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 479 | struct buffer_page *reader_page; |
66a8cb95 SR |
480 | unsigned long lost_events; |
481 | unsigned long last_overrun; | |
8e012066 | 482 | unsigned long nest; |
c64e148a | 483 | local_t entries_bytes; |
e4906eff | 484 | local_t entries; |
884bfe89 SP |
485 | local_t overrun; |
486 | local_t commit_overrun; | |
487 | local_t dropped_events; | |
fa743953 SR |
488 | local_t committing; |
489 | local_t commits; | |
77ae365e | 490 | unsigned long read; |
c64e148a | 491 | unsigned long read_bytes; |
7a8e76a3 SR |
492 | u64 write_stamp; |
493 | u64 read_stamp; | |
438ced17 | 494 | /* ring buffer pages to update, > 0 to add, < 0 to remove */ |
9b94a8fb | 495 | long nr_pages_to_update; |
438ced17 | 496 | struct list_head new_pages; /* new pages to add */ |
83f40318 | 497 | struct work_struct update_pages_work; |
05fdd70d | 498 | struct completion update_done; |
15693458 SRRH |
499 | |
500 | struct rb_irq_work irq_work; | |
7a8e76a3 SR |
501 | }; |
502 | ||
503 | struct ring_buffer { | |
7a8e76a3 SR |
504 | unsigned flags; |
505 | int cpus; | |
7a8e76a3 | 506 | atomic_t record_disabled; |
83f40318 | 507 | atomic_t resize_disabled; |
00f62f61 | 508 | cpumask_var_t cpumask; |
7a8e76a3 | 509 | |
1f8a6a10 PZ |
510 | struct lock_class_key *reader_lock_key; |
511 | ||
7a8e76a3 SR |
512 | struct mutex mutex; |
513 | ||
514 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 515 | |
b32614c0 | 516 | struct hlist_node node; |
37886f6a | 517 | u64 (*clock)(void); |
15693458 SRRH |
518 | |
519 | struct rb_irq_work irq_work; | |
00b41452 | 520 | bool time_stamp_abs; |
7a8e76a3 SR |
521 | }; |
522 | ||
523 | struct ring_buffer_iter { | |
524 | struct ring_buffer_per_cpu *cpu_buffer; | |
525 | unsigned long head; | |
526 | struct buffer_page *head_page; | |
492a74f4 SR |
527 | struct buffer_page *cache_reader_page; |
528 | unsigned long cache_read; | |
7a8e76a3 SR |
529 | u64 read_stamp; |
530 | }; | |
531 | ||
15693458 SRRH |
532 | /* |
533 | * rb_wake_up_waiters - wake up tasks waiting for ring buffer input | |
534 | * | |
535 | * Schedules a delayed work to wake up any task that is blocked on the | |
536 | * ring buffer waiters queue. | |
537 | */ | |
538 | static void rb_wake_up_waiters(struct irq_work *work) | |
539 | { | |
540 | struct rb_irq_work *rbwork = container_of(work, struct rb_irq_work, work); | |
541 | ||
542 | wake_up_all(&rbwork->waiters); | |
1e0d6714 SRRH |
543 | if (rbwork->wakeup_full) { |
544 | rbwork->wakeup_full = false; | |
545 | wake_up_all(&rbwork->full_waiters); | |
546 | } | |
15693458 SRRH |
547 | } |
548 | ||
549 | /** | |
550 | * ring_buffer_wait - wait for input to the ring buffer | |
551 | * @buffer: buffer to wait on | |
552 | * @cpu: the cpu buffer to wait on | |
e30f53aa | 553 | * @full: wait until a full page is available, if @cpu != RING_BUFFER_ALL_CPUS |
15693458 SRRH |
554 | * |
555 | * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon | |
556 | * as data is added to any of the @buffer's cpu buffers. Otherwise | |
557 | * it will wait for data to be added to a specific cpu buffer. | |
558 | */ | |
e30f53aa | 559 | int ring_buffer_wait(struct ring_buffer *buffer, int cpu, bool full) |
15693458 | 560 | { |
e30f53aa | 561 | struct ring_buffer_per_cpu *uninitialized_var(cpu_buffer); |
15693458 SRRH |
562 | DEFINE_WAIT(wait); |
563 | struct rb_irq_work *work; | |
e30f53aa | 564 | int ret = 0; |
15693458 SRRH |
565 | |
566 | /* | |
567 | * Depending on what the caller is waiting for, either any | |
568 | * data in any cpu buffer, or a specific buffer, put the | |
569 | * caller on the appropriate wait queue. | |
570 | */ | |
1e0d6714 | 571 | if (cpu == RING_BUFFER_ALL_CPUS) { |
15693458 | 572 | work = &buffer->irq_work; |
1e0d6714 SRRH |
573 | /* Full only makes sense on per cpu reads */ |
574 | full = false; | |
575 | } else { | |
8b8b3683 SRRH |
576 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
577 | return -ENODEV; | |
15693458 SRRH |
578 | cpu_buffer = buffer->buffers[cpu]; |
579 | work = &cpu_buffer->irq_work; | |
580 | } | |
581 | ||
582 | ||
e30f53aa | 583 | while (true) { |
1e0d6714 SRRH |
584 | if (full) |
585 | prepare_to_wait(&work->full_waiters, &wait, TASK_INTERRUPTIBLE); | |
586 | else | |
587 | prepare_to_wait(&work->waiters, &wait, TASK_INTERRUPTIBLE); | |
e30f53aa RV |
588 | |
589 | /* | |
590 | * The events can happen in critical sections where | |
591 | * checking a work queue can cause deadlocks. | |
592 | * After adding a task to the queue, this flag is set | |
593 | * only to notify events to try to wake up the queue | |
594 | * using irq_work. | |
595 | * | |
596 | * We don't clear it even if the buffer is no longer | |
597 | * empty. The flag only causes the next event to run | |
598 | * irq_work to do the work queue wake up. The worse | |
599 | * that can happen if we race with !trace_empty() is that | |
600 | * an event will cause an irq_work to try to wake up | |
601 | * an empty queue. | |
602 | * | |
603 | * There's no reason to protect this flag either, as | |
604 | * the work queue and irq_work logic will do the necessary | |
605 | * synchronization for the wake ups. The only thing | |
606 | * that is necessary is that the wake up happens after | |
607 | * a task has been queued. It's OK for spurious wake ups. | |
608 | */ | |
1e0d6714 SRRH |
609 | if (full) |
610 | work->full_waiters_pending = true; | |
611 | else | |
612 | work->waiters_pending = true; | |
e30f53aa RV |
613 | |
614 | if (signal_pending(current)) { | |
615 | ret = -EINTR; | |
616 | break; | |
617 | } | |
618 | ||
619 | if (cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) | |
620 | break; | |
621 | ||
622 | if (cpu != RING_BUFFER_ALL_CPUS && | |
623 | !ring_buffer_empty_cpu(buffer, cpu)) { | |
624 | unsigned long flags; | |
625 | bool pagebusy; | |
626 | ||
627 | if (!full) | |
628 | break; | |
629 | ||
630 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
631 | pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page; | |
632 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
633 | ||
634 | if (!pagebusy) | |
635 | break; | |
636 | } | |
15693458 | 637 | |
15693458 | 638 | schedule(); |
e30f53aa | 639 | } |
15693458 | 640 | |
1e0d6714 SRRH |
641 | if (full) |
642 | finish_wait(&work->full_waiters, &wait); | |
643 | else | |
644 | finish_wait(&work->waiters, &wait); | |
e30f53aa RV |
645 | |
646 | return ret; | |
15693458 SRRH |
647 | } |
648 | ||
649 | /** | |
650 | * ring_buffer_poll_wait - poll on buffer input | |
651 | * @buffer: buffer to wait on | |
652 | * @cpu: the cpu buffer to wait on | |
653 | * @filp: the file descriptor | |
654 | * @poll_table: The poll descriptor | |
655 | * | |
656 | * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon | |
657 | * as data is added to any of the @buffer's cpu buffers. Otherwise | |
658 | * it will wait for data to be added to a specific cpu buffer. | |
659 | * | |
a9a08845 | 660 | * Returns EPOLLIN | EPOLLRDNORM if data exists in the buffers, |
15693458 SRRH |
661 | * zero otherwise. |
662 | */ | |
ecf92700 | 663 | __poll_t ring_buffer_poll_wait(struct ring_buffer *buffer, int cpu, |
15693458 SRRH |
664 | struct file *filp, poll_table *poll_table) |
665 | { | |
666 | struct ring_buffer_per_cpu *cpu_buffer; | |
667 | struct rb_irq_work *work; | |
668 | ||
15693458 SRRH |
669 | if (cpu == RING_BUFFER_ALL_CPUS) |
670 | work = &buffer->irq_work; | |
671 | else { | |
6721cb60 SRRH |
672 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
673 | return -EINVAL; | |
674 | ||
15693458 SRRH |
675 | cpu_buffer = buffer->buffers[cpu]; |
676 | work = &cpu_buffer->irq_work; | |
677 | } | |
678 | ||
15693458 | 679 | poll_wait(filp, &work->waiters, poll_table); |
4ce97dbf JB |
680 | work->waiters_pending = true; |
681 | /* | |
682 | * There's a tight race between setting the waiters_pending and | |
683 | * checking if the ring buffer is empty. Once the waiters_pending bit | |
684 | * is set, the next event will wake the task up, but we can get stuck | |
685 | * if there's only a single event in. | |
686 | * | |
687 | * FIXME: Ideally, we need a memory barrier on the writer side as well, | |
688 | * but adding a memory barrier to all events will cause too much of a | |
689 | * performance hit in the fast path. We only need a memory barrier when | |
690 | * the buffer goes from empty to having content. But as this race is | |
691 | * extremely small, and it's not a problem if another event comes in, we | |
692 | * will fix it later. | |
693 | */ | |
694 | smp_mb(); | |
15693458 SRRH |
695 | |
696 | if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) || | |
697 | (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu))) | |
a9a08845 | 698 | return EPOLLIN | EPOLLRDNORM; |
15693458 SRRH |
699 | return 0; |
700 | } | |
701 | ||
f536aafc | 702 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
077c5407 SR |
703 | #define RB_WARN_ON(b, cond) \ |
704 | ({ \ | |
705 | int _____ret = unlikely(cond); \ | |
706 | if (_____ret) { \ | |
707 | if (__same_type(*(b), struct ring_buffer_per_cpu)) { \ | |
708 | struct ring_buffer_per_cpu *__b = \ | |
709 | (void *)b; \ | |
710 | atomic_inc(&__b->buffer->record_disabled); \ | |
711 | } else \ | |
712 | atomic_inc(&b->record_disabled); \ | |
713 | WARN_ON(1); \ | |
714 | } \ | |
715 | _____ret; \ | |
3e89c7bb | 716 | }) |
f536aafc | 717 | |
37886f6a SR |
718 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
719 | #define DEBUG_SHIFT 0 | |
720 | ||
6d3f1e12 | 721 | static inline u64 rb_time_stamp(struct ring_buffer *buffer) |
88eb0125 SR |
722 | { |
723 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
724 | return buffer->clock() << DEBUG_SHIFT; | |
725 | } | |
726 | ||
37886f6a SR |
727 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) |
728 | { | |
729 | u64 time; | |
730 | ||
731 | preempt_disable_notrace(); | |
6d3f1e12 | 732 | time = rb_time_stamp(buffer); |
37886f6a SR |
733 | preempt_enable_no_resched_notrace(); |
734 | ||
735 | return time; | |
736 | } | |
737 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
738 | ||
739 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
740 | int cpu, u64 *ts) | |
741 | { | |
742 | /* Just stupid testing the normalize function and deltas */ | |
743 | *ts >>= DEBUG_SHIFT; | |
744 | } | |
745 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
746 | ||
77ae365e SR |
747 | /* |
748 | * Making the ring buffer lockless makes things tricky. | |
749 | * Although writes only happen on the CPU that they are on, | |
750 | * and they only need to worry about interrupts. Reads can | |
751 | * happen on any CPU. | |
752 | * | |
753 | * The reader page is always off the ring buffer, but when the | |
754 | * reader finishes with a page, it needs to swap its page with | |
755 | * a new one from the buffer. The reader needs to take from | |
756 | * the head (writes go to the tail). But if a writer is in overwrite | |
757 | * mode and wraps, it must push the head page forward. | |
758 | * | |
759 | * Here lies the problem. | |
760 | * | |
761 | * The reader must be careful to replace only the head page, and | |
762 | * not another one. As described at the top of the file in the | |
763 | * ASCII art, the reader sets its old page to point to the next | |
764 | * page after head. It then sets the page after head to point to | |
765 | * the old reader page. But if the writer moves the head page | |
766 | * during this operation, the reader could end up with the tail. | |
767 | * | |
768 | * We use cmpxchg to help prevent this race. We also do something | |
769 | * special with the page before head. We set the LSB to 1. | |
770 | * | |
771 | * When the writer must push the page forward, it will clear the | |
772 | * bit that points to the head page, move the head, and then set | |
773 | * the bit that points to the new head page. | |
774 | * | |
775 | * We also don't want an interrupt coming in and moving the head | |
776 | * page on another writer. Thus we use the second LSB to catch | |
777 | * that too. Thus: | |
778 | * | |
779 | * head->list->prev->next bit 1 bit 0 | |
780 | * ------- ------- | |
781 | * Normal page 0 0 | |
782 | * Points to head page 0 1 | |
783 | * New head page 1 0 | |
784 | * | |
785 | * Note we can not trust the prev pointer of the head page, because: | |
786 | * | |
787 | * +----+ +-----+ +-----+ | |
788 | * | |------>| T |---X--->| N | | |
789 | * | |<------| | | | | |
790 | * +----+ +-----+ +-----+ | |
791 | * ^ ^ | | |
792 | * | +-----+ | | | |
793 | * +----------| R |----------+ | | |
794 | * | |<-----------+ | |
795 | * +-----+ | |
796 | * | |
797 | * Key: ---X--> HEAD flag set in pointer | |
798 | * T Tail page | |
799 | * R Reader page | |
800 | * N Next page | |
801 | * | |
802 | * (see __rb_reserve_next() to see where this happens) | |
803 | * | |
804 | * What the above shows is that the reader just swapped out | |
805 | * the reader page with a page in the buffer, but before it | |
806 | * could make the new header point back to the new page added | |
807 | * it was preempted by a writer. The writer moved forward onto | |
808 | * the new page added by the reader and is about to move forward | |
809 | * again. | |
810 | * | |
811 | * You can see, it is legitimate for the previous pointer of | |
812 | * the head (or any page) not to point back to itself. But only | |
6167c205 | 813 | * temporarily. |
77ae365e SR |
814 | */ |
815 | ||
816 | #define RB_PAGE_NORMAL 0UL | |
817 | #define RB_PAGE_HEAD 1UL | |
818 | #define RB_PAGE_UPDATE 2UL | |
819 | ||
820 | ||
821 | #define RB_FLAG_MASK 3UL | |
822 | ||
823 | /* PAGE_MOVED is not part of the mask */ | |
824 | #define RB_PAGE_MOVED 4UL | |
825 | ||
826 | /* | |
827 | * rb_list_head - remove any bit | |
828 | */ | |
829 | static struct list_head *rb_list_head(struct list_head *list) | |
830 | { | |
831 | unsigned long val = (unsigned long)list; | |
832 | ||
833 | return (struct list_head *)(val & ~RB_FLAG_MASK); | |
834 | } | |
835 | ||
836 | /* | |
6d3f1e12 | 837 | * rb_is_head_page - test if the given page is the head page |
77ae365e SR |
838 | * |
839 | * Because the reader may move the head_page pointer, we can | |
840 | * not trust what the head page is (it may be pointing to | |
841 | * the reader page). But if the next page is a header page, | |
842 | * its flags will be non zero. | |
843 | */ | |
42b16b3f | 844 | static inline int |
77ae365e SR |
845 | rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer, |
846 | struct buffer_page *page, struct list_head *list) | |
847 | { | |
848 | unsigned long val; | |
849 | ||
850 | val = (unsigned long)list->next; | |
851 | ||
852 | if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list) | |
853 | return RB_PAGE_MOVED; | |
854 | ||
855 | return val & RB_FLAG_MASK; | |
856 | } | |
857 | ||
858 | /* | |
859 | * rb_is_reader_page | |
860 | * | |
861 | * The unique thing about the reader page, is that, if the | |
862 | * writer is ever on it, the previous pointer never points | |
863 | * back to the reader page. | |
864 | */ | |
06ca3209 | 865 | static bool rb_is_reader_page(struct buffer_page *page) |
77ae365e SR |
866 | { |
867 | struct list_head *list = page->list.prev; | |
868 | ||
869 | return rb_list_head(list->next) != &page->list; | |
870 | } | |
871 | ||
872 | /* | |
873 | * rb_set_list_to_head - set a list_head to be pointing to head. | |
874 | */ | |
875 | static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer, | |
876 | struct list_head *list) | |
877 | { | |
878 | unsigned long *ptr; | |
879 | ||
880 | ptr = (unsigned long *)&list->next; | |
881 | *ptr |= RB_PAGE_HEAD; | |
882 | *ptr &= ~RB_PAGE_UPDATE; | |
883 | } | |
884 | ||
885 | /* | |
886 | * rb_head_page_activate - sets up head page | |
887 | */ | |
888 | static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer) | |
889 | { | |
890 | struct buffer_page *head; | |
891 | ||
892 | head = cpu_buffer->head_page; | |
893 | if (!head) | |
894 | return; | |
895 | ||
896 | /* | |
897 | * Set the previous list pointer to have the HEAD flag. | |
898 | */ | |
899 | rb_set_list_to_head(cpu_buffer, head->list.prev); | |
900 | } | |
901 | ||
902 | static void rb_list_head_clear(struct list_head *list) | |
903 | { | |
904 | unsigned long *ptr = (unsigned long *)&list->next; | |
905 | ||
906 | *ptr &= ~RB_FLAG_MASK; | |
907 | } | |
908 | ||
909 | /* | |
6167c205 | 910 | * rb_head_page_deactivate - clears head page ptr (for free list) |
77ae365e SR |
911 | */ |
912 | static void | |
913 | rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer) | |
914 | { | |
915 | struct list_head *hd; | |
916 | ||
917 | /* Go through the whole list and clear any pointers found. */ | |
918 | rb_list_head_clear(cpu_buffer->pages); | |
919 | ||
920 | list_for_each(hd, cpu_buffer->pages) | |
921 | rb_list_head_clear(hd); | |
922 | } | |
923 | ||
924 | static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer, | |
925 | struct buffer_page *head, | |
926 | struct buffer_page *prev, | |
927 | int old_flag, int new_flag) | |
928 | { | |
929 | struct list_head *list; | |
930 | unsigned long val = (unsigned long)&head->list; | |
931 | unsigned long ret; | |
932 | ||
933 | list = &prev->list; | |
934 | ||
935 | val &= ~RB_FLAG_MASK; | |
936 | ||
08a40816 SR |
937 | ret = cmpxchg((unsigned long *)&list->next, |
938 | val | old_flag, val | new_flag); | |
77ae365e SR |
939 | |
940 | /* check if the reader took the page */ | |
941 | if ((ret & ~RB_FLAG_MASK) != val) | |
942 | return RB_PAGE_MOVED; | |
943 | ||
944 | return ret & RB_FLAG_MASK; | |
945 | } | |
946 | ||
947 | static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer, | |
948 | struct buffer_page *head, | |
949 | struct buffer_page *prev, | |
950 | int old_flag) | |
951 | { | |
952 | return rb_head_page_set(cpu_buffer, head, prev, | |
953 | old_flag, RB_PAGE_UPDATE); | |
954 | } | |
955 | ||
956 | static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer, | |
957 | struct buffer_page *head, | |
958 | struct buffer_page *prev, | |
959 | int old_flag) | |
960 | { | |
961 | return rb_head_page_set(cpu_buffer, head, prev, | |
962 | old_flag, RB_PAGE_HEAD); | |
963 | } | |
964 | ||
965 | static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer, | |
966 | struct buffer_page *head, | |
967 | struct buffer_page *prev, | |
968 | int old_flag) | |
969 | { | |
970 | return rb_head_page_set(cpu_buffer, head, prev, | |
971 | old_flag, RB_PAGE_NORMAL); | |
972 | } | |
973 | ||
974 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
975 | struct buffer_page **bpage) | |
976 | { | |
977 | struct list_head *p = rb_list_head((*bpage)->list.next); | |
978 | ||
979 | *bpage = list_entry(p, struct buffer_page, list); | |
980 | } | |
981 | ||
982 | static struct buffer_page * | |
983 | rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer) | |
984 | { | |
985 | struct buffer_page *head; | |
986 | struct buffer_page *page; | |
987 | struct list_head *list; | |
988 | int i; | |
989 | ||
990 | if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page)) | |
991 | return NULL; | |
992 | ||
993 | /* sanity check */ | |
994 | list = cpu_buffer->pages; | |
995 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list)) | |
996 | return NULL; | |
997 | ||
998 | page = head = cpu_buffer->head_page; | |
999 | /* | |
1000 | * It is possible that the writer moves the header behind | |
1001 | * where we started, and we miss in one loop. | |
1002 | * A second loop should grab the header, but we'll do | |
1003 | * three loops just because I'm paranoid. | |
1004 | */ | |
1005 | for (i = 0; i < 3; i++) { | |
1006 | do { | |
1007 | if (rb_is_head_page(cpu_buffer, page, page->list.prev)) { | |
1008 | cpu_buffer->head_page = page; | |
1009 | return page; | |
1010 | } | |
1011 | rb_inc_page(cpu_buffer, &page); | |
1012 | } while (page != head); | |
1013 | } | |
1014 | ||
1015 | RB_WARN_ON(cpu_buffer, 1); | |
1016 | ||
1017 | return NULL; | |
1018 | } | |
1019 | ||
1020 | static int rb_head_page_replace(struct buffer_page *old, | |
1021 | struct buffer_page *new) | |
1022 | { | |
1023 | unsigned long *ptr = (unsigned long *)&old->list.prev->next; | |
1024 | unsigned long val; | |
1025 | unsigned long ret; | |
1026 | ||
1027 | val = *ptr & ~RB_FLAG_MASK; | |
1028 | val |= RB_PAGE_HEAD; | |
1029 | ||
08a40816 | 1030 | ret = cmpxchg(ptr, val, (unsigned long)&new->list); |
77ae365e SR |
1031 | |
1032 | return ret == val; | |
1033 | } | |
1034 | ||
1035 | /* | |
1036 | * rb_tail_page_update - move the tail page forward | |
77ae365e | 1037 | */ |
70004986 | 1038 | static void rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer, |
77ae365e SR |
1039 | struct buffer_page *tail_page, |
1040 | struct buffer_page *next_page) | |
1041 | { | |
77ae365e SR |
1042 | unsigned long old_entries; |
1043 | unsigned long old_write; | |
77ae365e SR |
1044 | |
1045 | /* | |
1046 | * The tail page now needs to be moved forward. | |
1047 | * | |
1048 | * We need to reset the tail page, but without messing | |
1049 | * with possible erasing of data brought in by interrupts | |
1050 | * that have moved the tail page and are currently on it. | |
1051 | * | |
1052 | * We add a counter to the write field to denote this. | |
1053 | */ | |
1054 | old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write); | |
1055 | old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries); | |
1056 | ||
1057 | /* | |
1058 | * Just make sure we have seen our old_write and synchronize | |
1059 | * with any interrupts that come in. | |
1060 | */ | |
1061 | barrier(); | |
1062 | ||
1063 | /* | |
1064 | * If the tail page is still the same as what we think | |
1065 | * it is, then it is up to us to update the tail | |
1066 | * pointer. | |
1067 | */ | |
8573636e | 1068 | if (tail_page == READ_ONCE(cpu_buffer->tail_page)) { |
77ae365e SR |
1069 | /* Zero the write counter */ |
1070 | unsigned long val = old_write & ~RB_WRITE_MASK; | |
1071 | unsigned long eval = old_entries & ~RB_WRITE_MASK; | |
1072 | ||
1073 | /* | |
1074 | * This will only succeed if an interrupt did | |
1075 | * not come in and change it. In which case, we | |
1076 | * do not want to modify it. | |
da706d8b LJ |
1077 | * |
1078 | * We add (void) to let the compiler know that we do not care | |
1079 | * about the return value of these functions. We use the | |
1080 | * cmpxchg to only update if an interrupt did not already | |
1081 | * do it for us. If the cmpxchg fails, we don't care. | |
77ae365e | 1082 | */ |
da706d8b LJ |
1083 | (void)local_cmpxchg(&next_page->write, old_write, val); |
1084 | (void)local_cmpxchg(&next_page->entries, old_entries, eval); | |
77ae365e SR |
1085 | |
1086 | /* | |
1087 | * No need to worry about races with clearing out the commit. | |
1088 | * it only can increment when a commit takes place. But that | |
1089 | * only happens in the outer most nested commit. | |
1090 | */ | |
1091 | local_set(&next_page->page->commit, 0); | |
1092 | ||
70004986 SRRH |
1093 | /* Again, either we update tail_page or an interrupt does */ |
1094 | (void)cmpxchg(&cpu_buffer->tail_page, tail_page, next_page); | |
77ae365e | 1095 | } |
77ae365e SR |
1096 | } |
1097 | ||
1098 | static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer, | |
1099 | struct buffer_page *bpage) | |
1100 | { | |
1101 | unsigned long val = (unsigned long)bpage; | |
1102 | ||
1103 | if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK)) | |
1104 | return 1; | |
1105 | ||
1106 | return 0; | |
1107 | } | |
1108 | ||
1109 | /** | |
1110 | * rb_check_list - make sure a pointer to a list has the last bits zero | |
1111 | */ | |
1112 | static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer, | |
1113 | struct list_head *list) | |
1114 | { | |
1115 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev)) | |
1116 | return 1; | |
1117 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next)) | |
1118 | return 1; | |
1119 | return 0; | |
1120 | } | |
1121 | ||
7a8e76a3 | 1122 | /** |
d611851b | 1123 | * rb_check_pages - integrity check of buffer pages |
7a8e76a3 SR |
1124 | * @cpu_buffer: CPU buffer with pages to test |
1125 | * | |
c3706f00 | 1126 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
1127 | * been corrupted. |
1128 | */ | |
1129 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
1130 | { | |
3adc54fa | 1131 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1132 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1133 | |
308f7eeb SR |
1134 | /* Reset the head page if it exists */ |
1135 | if (cpu_buffer->head_page) | |
1136 | rb_set_head_page(cpu_buffer); | |
1137 | ||
77ae365e SR |
1138 | rb_head_page_deactivate(cpu_buffer); |
1139 | ||
3e89c7bb SR |
1140 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
1141 | return -1; | |
1142 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
1143 | return -1; | |
7a8e76a3 | 1144 | |
77ae365e SR |
1145 | if (rb_check_list(cpu_buffer, head)) |
1146 | return -1; | |
1147 | ||
044fa782 | 1148 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 1149 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 1150 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
1151 | return -1; |
1152 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 1153 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 1154 | return -1; |
77ae365e SR |
1155 | if (rb_check_list(cpu_buffer, &bpage->list)) |
1156 | return -1; | |
7a8e76a3 SR |
1157 | } |
1158 | ||
77ae365e SR |
1159 | rb_head_page_activate(cpu_buffer); |
1160 | ||
7a8e76a3 SR |
1161 | return 0; |
1162 | } | |
1163 | ||
9b94a8fb | 1164 | static int __rb_allocate_pages(long nr_pages, struct list_head *pages, int cpu) |
7a8e76a3 | 1165 | { |
044fa782 | 1166 | struct buffer_page *bpage, *tmp; |
927e56db SRV |
1167 | bool user_thread = current->mm != NULL; |
1168 | gfp_t mflags; | |
9b94a8fb | 1169 | long i; |
3adc54fa | 1170 | |
927e56db SRV |
1171 | /* |
1172 | * Check if the available memory is there first. | |
1173 | * Note, si_mem_available() only gives us a rough estimate of available | |
1174 | * memory. It may not be accurate. But we don't care, we just want | |
1175 | * to prevent doing any allocation when it is obvious that it is | |
1176 | * not going to succeed. | |
1177 | */ | |
2a872fa4 SRV |
1178 | i = si_mem_available(); |
1179 | if (i < nr_pages) | |
1180 | return -ENOMEM; | |
1181 | ||
927e56db SRV |
1182 | /* |
1183 | * __GFP_RETRY_MAYFAIL flag makes sure that the allocation fails | |
1184 | * gracefully without invoking oom-killer and the system is not | |
1185 | * destabilized. | |
1186 | */ | |
1187 | mflags = GFP_KERNEL | __GFP_RETRY_MAYFAIL; | |
1188 | ||
1189 | /* | |
1190 | * If a user thread allocates too much, and si_mem_available() | |
1191 | * reports there's enough memory, even though there is not. | |
1192 | * Make sure the OOM killer kills this thread. This can happen | |
1193 | * even with RETRY_MAYFAIL because another task may be doing | |
1194 | * an allocation after this task has taken all memory. | |
1195 | * This is the task the OOM killer needs to take out during this | |
1196 | * loop, even if it was triggered by an allocation somewhere else. | |
1197 | */ | |
1198 | if (user_thread) | |
1199 | set_current_oom_origin(); | |
7a8e76a3 | 1200 | for (i = 0; i < nr_pages; i++) { |
7ea59064 | 1201 | struct page *page; |
927e56db | 1202 | |
044fa782 | 1203 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
927e56db | 1204 | mflags, cpu_to_node(cpu)); |
044fa782 | 1205 | if (!bpage) |
e4c2ce82 | 1206 | goto free_pages; |
77ae365e | 1207 | |
438ced17 | 1208 | list_add(&bpage->list, pages); |
77ae365e | 1209 | |
927e56db | 1210 | page = alloc_pages_node(cpu_to_node(cpu), mflags, 0); |
7ea59064 | 1211 | if (!page) |
7a8e76a3 | 1212 | goto free_pages; |
7ea59064 | 1213 | bpage->page = page_address(page); |
044fa782 | 1214 | rb_init_page(bpage->page); |
927e56db SRV |
1215 | |
1216 | if (user_thread && fatal_signal_pending(current)) | |
1217 | goto free_pages; | |
7a8e76a3 | 1218 | } |
927e56db SRV |
1219 | if (user_thread) |
1220 | clear_current_oom_origin(); | |
7a8e76a3 | 1221 | |
438ced17 VN |
1222 | return 0; |
1223 | ||
1224 | free_pages: | |
1225 | list_for_each_entry_safe(bpage, tmp, pages, list) { | |
1226 | list_del_init(&bpage->list); | |
1227 | free_buffer_page(bpage); | |
1228 | } | |
927e56db SRV |
1229 | if (user_thread) |
1230 | clear_current_oom_origin(); | |
438ced17 VN |
1231 | |
1232 | return -ENOMEM; | |
1233 | } | |
1234 | ||
1235 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
9b94a8fb | 1236 | unsigned long nr_pages) |
438ced17 VN |
1237 | { |
1238 | LIST_HEAD(pages); | |
1239 | ||
1240 | WARN_ON(!nr_pages); | |
1241 | ||
1242 | if (__rb_allocate_pages(nr_pages, &pages, cpu_buffer->cpu)) | |
1243 | return -ENOMEM; | |
1244 | ||
3adc54fa SR |
1245 | /* |
1246 | * The ring buffer page list is a circular list that does not | |
1247 | * start and end with a list head. All page list items point to | |
1248 | * other pages. | |
1249 | */ | |
1250 | cpu_buffer->pages = pages.next; | |
1251 | list_del(&pages); | |
7a8e76a3 | 1252 | |
438ced17 VN |
1253 | cpu_buffer->nr_pages = nr_pages; |
1254 | ||
7a8e76a3 SR |
1255 | rb_check_pages(cpu_buffer); |
1256 | ||
1257 | return 0; | |
7a8e76a3 SR |
1258 | } |
1259 | ||
1260 | static struct ring_buffer_per_cpu * | |
9b94a8fb | 1261 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, long nr_pages, int cpu) |
7a8e76a3 SR |
1262 | { |
1263 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 1264 | struct buffer_page *bpage; |
7ea59064 | 1265 | struct page *page; |
7a8e76a3 SR |
1266 | int ret; |
1267 | ||
1268 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
1269 | GFP_KERNEL, cpu_to_node(cpu)); | |
1270 | if (!cpu_buffer) | |
1271 | return NULL; | |
1272 | ||
1273 | cpu_buffer->cpu = cpu; | |
1274 | cpu_buffer->buffer = buffer; | |
5389f6fa | 1275 | raw_spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 1276 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
edc35bd7 | 1277 | cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
83f40318 | 1278 | INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler); |
05fdd70d | 1279 | init_completion(&cpu_buffer->update_done); |
15693458 | 1280 | init_irq_work(&cpu_buffer->irq_work.work, rb_wake_up_waiters); |
f1dc6725 | 1281 | init_waitqueue_head(&cpu_buffer->irq_work.waiters); |
1e0d6714 | 1282 | init_waitqueue_head(&cpu_buffer->irq_work.full_waiters); |
7a8e76a3 | 1283 | |
044fa782 | 1284 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 1285 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 1286 | if (!bpage) |
e4c2ce82 SR |
1287 | goto fail_free_buffer; |
1288 | ||
77ae365e SR |
1289 | rb_check_bpage(cpu_buffer, bpage); |
1290 | ||
044fa782 | 1291 | cpu_buffer->reader_page = bpage; |
7ea59064 VN |
1292 | page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0); |
1293 | if (!page) | |
e4c2ce82 | 1294 | goto fail_free_reader; |
7ea59064 | 1295 | bpage->page = page_address(page); |
044fa782 | 1296 | rb_init_page(bpage->page); |
e4c2ce82 | 1297 | |
d769041f | 1298 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
44b99462 | 1299 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
d769041f | 1300 | |
438ced17 | 1301 | ret = rb_allocate_pages(cpu_buffer, nr_pages); |
7a8e76a3 | 1302 | if (ret < 0) |
d769041f | 1303 | goto fail_free_reader; |
7a8e76a3 SR |
1304 | |
1305 | cpu_buffer->head_page | |
3adc54fa | 1306 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 1307 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 | 1308 | |
77ae365e SR |
1309 | rb_head_page_activate(cpu_buffer); |
1310 | ||
7a8e76a3 SR |
1311 | return cpu_buffer; |
1312 | ||
d769041f SR |
1313 | fail_free_reader: |
1314 | free_buffer_page(cpu_buffer->reader_page); | |
1315 | ||
7a8e76a3 SR |
1316 | fail_free_buffer: |
1317 | kfree(cpu_buffer); | |
1318 | return NULL; | |
1319 | } | |
1320 | ||
1321 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
1322 | { | |
3adc54fa | 1323 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1324 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1325 | |
d769041f SR |
1326 | free_buffer_page(cpu_buffer->reader_page); |
1327 | ||
77ae365e SR |
1328 | rb_head_page_deactivate(cpu_buffer); |
1329 | ||
3adc54fa SR |
1330 | if (head) { |
1331 | list_for_each_entry_safe(bpage, tmp, head, list) { | |
1332 | list_del_init(&bpage->list); | |
1333 | free_buffer_page(bpage); | |
1334 | } | |
1335 | bpage = list_entry(head, struct buffer_page, list); | |
044fa782 | 1336 | free_buffer_page(bpage); |
7a8e76a3 | 1337 | } |
3adc54fa | 1338 | |
7a8e76a3 SR |
1339 | kfree(cpu_buffer); |
1340 | } | |
1341 | ||
1342 | /** | |
d611851b | 1343 | * __ring_buffer_alloc - allocate a new ring_buffer |
68814b58 | 1344 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
1345 | * @flags: attributes to set for the ring buffer. |
1346 | * | |
1347 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
1348 | * flag. This flag means that the buffer will overwrite old data | |
1349 | * when the buffer wraps. If this flag is not set, the buffer will | |
1350 | * drop data when the tail hits the head. | |
1351 | */ | |
1f8a6a10 PZ |
1352 | struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
1353 | struct lock_class_key *key) | |
7a8e76a3 SR |
1354 | { |
1355 | struct ring_buffer *buffer; | |
9b94a8fb | 1356 | long nr_pages; |
7a8e76a3 | 1357 | int bsize; |
9b94a8fb | 1358 | int cpu; |
b32614c0 | 1359 | int ret; |
7a8e76a3 SR |
1360 | |
1361 | /* keep it in its own cache line */ | |
1362 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
1363 | GFP_KERNEL); | |
1364 | if (!buffer) | |
1365 | return NULL; | |
1366 | ||
b18cc3de | 1367 | if (!zalloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
9e01c1b7 RR |
1368 | goto fail_free_buffer; |
1369 | ||
438ced17 | 1370 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
7a8e76a3 | 1371 | buffer->flags = flags; |
37886f6a | 1372 | buffer->clock = trace_clock_local; |
1f8a6a10 | 1373 | buffer->reader_lock_key = key; |
7a8e76a3 | 1374 | |
15693458 | 1375 | init_irq_work(&buffer->irq_work.work, rb_wake_up_waiters); |
f1dc6725 | 1376 | init_waitqueue_head(&buffer->irq_work.waiters); |
15693458 | 1377 | |
7a8e76a3 | 1378 | /* need at least two pages */ |
438ced17 VN |
1379 | if (nr_pages < 2) |
1380 | nr_pages = 2; | |
7a8e76a3 | 1381 | |
7a8e76a3 SR |
1382 | buffer->cpus = nr_cpu_ids; |
1383 | ||
1384 | bsize = sizeof(void *) * nr_cpu_ids; | |
1385 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
1386 | GFP_KERNEL); | |
1387 | if (!buffer->buffers) | |
9e01c1b7 | 1388 | goto fail_free_cpumask; |
7a8e76a3 | 1389 | |
b32614c0 SAS |
1390 | cpu = raw_smp_processor_id(); |
1391 | cpumask_set_cpu(cpu, buffer->cpumask); | |
1392 | buffer->buffers[cpu] = rb_allocate_cpu_buffer(buffer, nr_pages, cpu); | |
1393 | if (!buffer->buffers[cpu]) | |
1394 | goto fail_free_buffers; | |
7a8e76a3 | 1395 | |
b32614c0 SAS |
1396 | ret = cpuhp_state_add_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node); |
1397 | if (ret < 0) | |
1398 | goto fail_free_buffers; | |
554f786e | 1399 | |
7a8e76a3 SR |
1400 | mutex_init(&buffer->mutex); |
1401 | ||
1402 | return buffer; | |
1403 | ||
1404 | fail_free_buffers: | |
1405 | for_each_buffer_cpu(buffer, cpu) { | |
1406 | if (buffer->buffers[cpu]) | |
1407 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1408 | } | |
1409 | kfree(buffer->buffers); | |
1410 | ||
9e01c1b7 RR |
1411 | fail_free_cpumask: |
1412 | free_cpumask_var(buffer->cpumask); | |
1413 | ||
7a8e76a3 SR |
1414 | fail_free_buffer: |
1415 | kfree(buffer); | |
1416 | return NULL; | |
1417 | } | |
1f8a6a10 | 1418 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
1419 | |
1420 | /** | |
1421 | * ring_buffer_free - free a ring buffer. | |
1422 | * @buffer: the buffer to free. | |
1423 | */ | |
1424 | void | |
1425 | ring_buffer_free(struct ring_buffer *buffer) | |
1426 | { | |
1427 | int cpu; | |
1428 | ||
b32614c0 | 1429 | cpuhp_state_remove_instance(CPUHP_TRACE_RB_PREPARE, &buffer->node); |
554f786e | 1430 | |
7a8e76a3 SR |
1431 | for_each_buffer_cpu(buffer, cpu) |
1432 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1433 | ||
bd3f0221 | 1434 | kfree(buffer->buffers); |
9e01c1b7 RR |
1435 | free_cpumask_var(buffer->cpumask); |
1436 | ||
7a8e76a3 SR |
1437 | kfree(buffer); |
1438 | } | |
c4f50183 | 1439 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 1440 | |
37886f6a SR |
1441 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
1442 | u64 (*clock)(void)) | |
1443 | { | |
1444 | buffer->clock = clock; | |
1445 | } | |
1446 | ||
00b41452 TZ |
1447 | void ring_buffer_set_time_stamp_abs(struct ring_buffer *buffer, bool abs) |
1448 | { | |
1449 | buffer->time_stamp_abs = abs; | |
1450 | } | |
1451 | ||
1452 | bool ring_buffer_time_stamp_abs(struct ring_buffer *buffer) | |
1453 | { | |
1454 | return buffer->time_stamp_abs; | |
1455 | } | |
1456 | ||
7a8e76a3 SR |
1457 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
1458 | ||
83f40318 VN |
1459 | static inline unsigned long rb_page_entries(struct buffer_page *bpage) |
1460 | { | |
1461 | return local_read(&bpage->entries) & RB_WRITE_MASK; | |
1462 | } | |
1463 | ||
1464 | static inline unsigned long rb_page_write(struct buffer_page *bpage) | |
1465 | { | |
1466 | return local_read(&bpage->write) & RB_WRITE_MASK; | |
1467 | } | |
1468 | ||
5040b4b7 | 1469 | static int |
9b94a8fb | 1470 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned long nr_pages) |
7a8e76a3 | 1471 | { |
83f40318 VN |
1472 | struct list_head *tail_page, *to_remove, *next_page; |
1473 | struct buffer_page *to_remove_page, *tmp_iter_page; | |
1474 | struct buffer_page *last_page, *first_page; | |
9b94a8fb | 1475 | unsigned long nr_removed; |
83f40318 VN |
1476 | unsigned long head_bit; |
1477 | int page_entries; | |
1478 | ||
1479 | head_bit = 0; | |
7a8e76a3 | 1480 | |
5389f6fa | 1481 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1482 | atomic_inc(&cpu_buffer->record_disabled); |
1483 | /* | |
1484 | * We don't race with the readers since we have acquired the reader | |
1485 | * lock. We also don't race with writers after disabling recording. | |
1486 | * This makes it easy to figure out the first and the last page to be | |
1487 | * removed from the list. We unlink all the pages in between including | |
1488 | * the first and last pages. This is done in a busy loop so that we | |
1489 | * lose the least number of traces. | |
1490 | * The pages are freed after we restart recording and unlock readers. | |
1491 | */ | |
1492 | tail_page = &cpu_buffer->tail_page->list; | |
77ae365e | 1493 | |
83f40318 VN |
1494 | /* |
1495 | * tail page might be on reader page, we remove the next page | |
1496 | * from the ring buffer | |
1497 | */ | |
1498 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
1499 | tail_page = rb_list_head(tail_page->next); | |
1500 | to_remove = tail_page; | |
1501 | ||
1502 | /* start of pages to remove */ | |
1503 | first_page = list_entry(rb_list_head(to_remove->next), | |
1504 | struct buffer_page, list); | |
1505 | ||
1506 | for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) { | |
1507 | to_remove = rb_list_head(to_remove)->next; | |
1508 | head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD; | |
7a8e76a3 | 1509 | } |
7a8e76a3 | 1510 | |
83f40318 | 1511 | next_page = rb_list_head(to_remove)->next; |
7a8e76a3 | 1512 | |
83f40318 VN |
1513 | /* |
1514 | * Now we remove all pages between tail_page and next_page. | |
1515 | * Make sure that we have head_bit value preserved for the | |
1516 | * next page | |
1517 | */ | |
1518 | tail_page->next = (struct list_head *)((unsigned long)next_page | | |
1519 | head_bit); | |
1520 | next_page = rb_list_head(next_page); | |
1521 | next_page->prev = tail_page; | |
1522 | ||
1523 | /* make sure pages points to a valid page in the ring buffer */ | |
1524 | cpu_buffer->pages = next_page; | |
1525 | ||
1526 | /* update head page */ | |
1527 | if (head_bit) | |
1528 | cpu_buffer->head_page = list_entry(next_page, | |
1529 | struct buffer_page, list); | |
1530 | ||
1531 | /* | |
1532 | * change read pointer to make sure any read iterators reset | |
1533 | * themselves | |
1534 | */ | |
1535 | cpu_buffer->read = 0; | |
1536 | ||
1537 | /* pages are removed, resume tracing and then free the pages */ | |
1538 | atomic_dec(&cpu_buffer->record_disabled); | |
5389f6fa | 1539 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1540 | |
1541 | RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)); | |
1542 | ||
1543 | /* last buffer page to remove */ | |
1544 | last_page = list_entry(rb_list_head(to_remove), struct buffer_page, | |
1545 | list); | |
1546 | tmp_iter_page = first_page; | |
1547 | ||
1548 | do { | |
83f36555 VN |
1549 | cond_resched(); |
1550 | ||
83f40318 VN |
1551 | to_remove_page = tmp_iter_page; |
1552 | rb_inc_page(cpu_buffer, &tmp_iter_page); | |
1553 | ||
1554 | /* update the counters */ | |
1555 | page_entries = rb_page_entries(to_remove_page); | |
1556 | if (page_entries) { | |
1557 | /* | |
1558 | * If something was added to this page, it was full | |
1559 | * since it is not the tail page. So we deduct the | |
1560 | * bytes consumed in ring buffer from here. | |
48fdc72f | 1561 | * Increment overrun to account for the lost events. |
83f40318 | 1562 | */ |
48fdc72f | 1563 | local_add(page_entries, &cpu_buffer->overrun); |
83f40318 VN |
1564 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
1565 | } | |
1566 | ||
1567 | /* | |
1568 | * We have already removed references to this list item, just | |
1569 | * free up the buffer_page and its page | |
1570 | */ | |
1571 | free_buffer_page(to_remove_page); | |
1572 | nr_removed--; | |
1573 | ||
1574 | } while (to_remove_page != last_page); | |
1575 | ||
1576 | RB_WARN_ON(cpu_buffer, nr_removed); | |
5040b4b7 VN |
1577 | |
1578 | return nr_removed == 0; | |
7a8e76a3 SR |
1579 | } |
1580 | ||
5040b4b7 VN |
1581 | static int |
1582 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 1583 | { |
5040b4b7 VN |
1584 | struct list_head *pages = &cpu_buffer->new_pages; |
1585 | int retries, success; | |
7a8e76a3 | 1586 | |
5389f6fa | 1587 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1588 | /* |
1589 | * We are holding the reader lock, so the reader page won't be swapped | |
1590 | * in the ring buffer. Now we are racing with the writer trying to | |
1591 | * move head page and the tail page. | |
1592 | * We are going to adapt the reader page update process where: | |
1593 | * 1. We first splice the start and end of list of new pages between | |
1594 | * the head page and its previous page. | |
1595 | * 2. We cmpxchg the prev_page->next to point from head page to the | |
1596 | * start of new pages list. | |
1597 | * 3. Finally, we update the head->prev to the end of new list. | |
1598 | * | |
1599 | * We will try this process 10 times, to make sure that we don't keep | |
1600 | * spinning. | |
1601 | */ | |
1602 | retries = 10; | |
1603 | success = 0; | |
1604 | while (retries--) { | |
1605 | struct list_head *head_page, *prev_page, *r; | |
1606 | struct list_head *last_page, *first_page; | |
1607 | struct list_head *head_page_with_bit; | |
77ae365e | 1608 | |
5040b4b7 | 1609 | head_page = &rb_set_head_page(cpu_buffer)->list; |
54f7be5b SR |
1610 | if (!head_page) |
1611 | break; | |
5040b4b7 VN |
1612 | prev_page = head_page->prev; |
1613 | ||
1614 | first_page = pages->next; | |
1615 | last_page = pages->prev; | |
1616 | ||
1617 | head_page_with_bit = (struct list_head *) | |
1618 | ((unsigned long)head_page | RB_PAGE_HEAD); | |
1619 | ||
1620 | last_page->next = head_page_with_bit; | |
1621 | first_page->prev = prev_page; | |
1622 | ||
1623 | r = cmpxchg(&prev_page->next, head_page_with_bit, first_page); | |
1624 | ||
1625 | if (r == head_page_with_bit) { | |
1626 | /* | |
1627 | * yay, we replaced the page pointer to our new list, | |
1628 | * now, we just have to update to head page's prev | |
1629 | * pointer to point to end of list | |
1630 | */ | |
1631 | head_page->prev = last_page; | |
1632 | success = 1; | |
1633 | break; | |
1634 | } | |
7a8e76a3 | 1635 | } |
7a8e76a3 | 1636 | |
5040b4b7 VN |
1637 | if (success) |
1638 | INIT_LIST_HEAD(pages); | |
1639 | /* | |
1640 | * If we weren't successful in adding in new pages, warn and stop | |
1641 | * tracing | |
1642 | */ | |
1643 | RB_WARN_ON(cpu_buffer, !success); | |
5389f6fa | 1644 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1645 | |
1646 | /* free pages if they weren't inserted */ | |
1647 | if (!success) { | |
1648 | struct buffer_page *bpage, *tmp; | |
1649 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, | |
1650 | list) { | |
1651 | list_del_init(&bpage->list); | |
1652 | free_buffer_page(bpage); | |
1653 | } | |
1654 | } | |
1655 | return success; | |
7a8e76a3 SR |
1656 | } |
1657 | ||
83f40318 | 1658 | static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer) |
438ced17 | 1659 | { |
5040b4b7 VN |
1660 | int success; |
1661 | ||
438ced17 | 1662 | if (cpu_buffer->nr_pages_to_update > 0) |
5040b4b7 | 1663 | success = rb_insert_pages(cpu_buffer); |
438ced17 | 1664 | else |
5040b4b7 VN |
1665 | success = rb_remove_pages(cpu_buffer, |
1666 | -cpu_buffer->nr_pages_to_update); | |
83f40318 | 1667 | |
5040b4b7 VN |
1668 | if (success) |
1669 | cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update; | |
83f40318 VN |
1670 | } |
1671 | ||
1672 | static void update_pages_handler(struct work_struct *work) | |
1673 | { | |
1674 | struct ring_buffer_per_cpu *cpu_buffer = container_of(work, | |
1675 | struct ring_buffer_per_cpu, update_pages_work); | |
1676 | rb_update_pages(cpu_buffer); | |
05fdd70d | 1677 | complete(&cpu_buffer->update_done); |
438ced17 VN |
1678 | } |
1679 | ||
7a8e76a3 SR |
1680 | /** |
1681 | * ring_buffer_resize - resize the ring buffer | |
1682 | * @buffer: the buffer to resize. | |
1683 | * @size: the new size. | |
d611851b | 1684 | * @cpu_id: the cpu buffer to resize |
7a8e76a3 | 1685 | * |
7a8e76a3 SR |
1686 | * Minimum size is 2 * BUF_PAGE_SIZE. |
1687 | * | |
83f40318 | 1688 | * Returns 0 on success and < 0 on failure. |
7a8e76a3 | 1689 | */ |
438ced17 VN |
1690 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size, |
1691 | int cpu_id) | |
7a8e76a3 SR |
1692 | { |
1693 | struct ring_buffer_per_cpu *cpu_buffer; | |
9b94a8fb | 1694 | unsigned long nr_pages; |
83f40318 | 1695 | int cpu, err = 0; |
7a8e76a3 | 1696 | |
ee51a1de IM |
1697 | /* |
1698 | * Always succeed at resizing a non-existent buffer: | |
1699 | */ | |
1700 | if (!buffer) | |
1701 | return size; | |
1702 | ||
6a31e1f1 SR |
1703 | /* Make sure the requested buffer exists */ |
1704 | if (cpu_id != RING_BUFFER_ALL_CPUS && | |
1705 | !cpumask_test_cpu(cpu_id, buffer->cpumask)) | |
1706 | return size; | |
1707 | ||
59643d15 | 1708 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
7a8e76a3 SR |
1709 | |
1710 | /* we need a minimum of two pages */ | |
59643d15 SRRH |
1711 | if (nr_pages < 2) |
1712 | nr_pages = 2; | |
7a8e76a3 | 1713 | |
59643d15 | 1714 | size = nr_pages * BUF_PAGE_SIZE; |
18421015 | 1715 | |
83f40318 VN |
1716 | /* |
1717 | * Don't succeed if resizing is disabled, as a reader might be | |
1718 | * manipulating the ring buffer and is expecting a sane state while | |
1719 | * this is true. | |
1720 | */ | |
1721 | if (atomic_read(&buffer->resize_disabled)) | |
1722 | return -EBUSY; | |
18421015 | 1723 | |
83f40318 | 1724 | /* prevent another thread from changing buffer sizes */ |
7a8e76a3 | 1725 | mutex_lock(&buffer->mutex); |
7a8e76a3 | 1726 | |
438ced17 VN |
1727 | if (cpu_id == RING_BUFFER_ALL_CPUS) { |
1728 | /* calculate the pages to update */ | |
7a8e76a3 SR |
1729 | for_each_buffer_cpu(buffer, cpu) { |
1730 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 1731 | |
438ced17 VN |
1732 | cpu_buffer->nr_pages_to_update = nr_pages - |
1733 | cpu_buffer->nr_pages; | |
438ced17 VN |
1734 | /* |
1735 | * nothing more to do for removing pages or no update | |
1736 | */ | |
1737 | if (cpu_buffer->nr_pages_to_update <= 0) | |
1738 | continue; | |
d7ec4bfe | 1739 | /* |
438ced17 VN |
1740 | * to add pages, make sure all new pages can be |
1741 | * allocated without receiving ENOMEM | |
d7ec4bfe | 1742 | */ |
438ced17 VN |
1743 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
1744 | if (__rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 | 1745 | &cpu_buffer->new_pages, cpu)) { |
438ced17 | 1746 | /* not enough memory for new pages */ |
83f40318 VN |
1747 | err = -ENOMEM; |
1748 | goto out_err; | |
1749 | } | |
1750 | } | |
1751 | ||
1752 | get_online_cpus(); | |
1753 | /* | |
1754 | * Fire off all the required work handlers | |
05fdd70d | 1755 | * We can't schedule on offline CPUs, but it's not necessary |
83f40318 VN |
1756 | * since we can change their buffer sizes without any race. |
1757 | */ | |
1758 | for_each_buffer_cpu(buffer, cpu) { | |
1759 | cpu_buffer = buffer->buffers[cpu]; | |
05fdd70d | 1760 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
1761 | continue; |
1762 | ||
021c5b34 CM |
1763 | /* Can't run something on an offline CPU. */ |
1764 | if (!cpu_online(cpu)) { | |
f5eb5588 SRRH |
1765 | rb_update_pages(cpu_buffer); |
1766 | cpu_buffer->nr_pages_to_update = 0; | |
1767 | } else { | |
05fdd70d VN |
1768 | schedule_work_on(cpu, |
1769 | &cpu_buffer->update_pages_work); | |
f5eb5588 | 1770 | } |
7a8e76a3 | 1771 | } |
7a8e76a3 | 1772 | |
438ced17 VN |
1773 | /* wait for all the updates to complete */ |
1774 | for_each_buffer_cpu(buffer, cpu) { | |
1775 | cpu_buffer = buffer->buffers[cpu]; | |
05fdd70d | 1776 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
1777 | continue; |
1778 | ||
05fdd70d VN |
1779 | if (cpu_online(cpu)) |
1780 | wait_for_completion(&cpu_buffer->update_done); | |
83f40318 | 1781 | cpu_buffer->nr_pages_to_update = 0; |
438ced17 | 1782 | } |
83f40318 VN |
1783 | |
1784 | put_online_cpus(); | |
438ced17 | 1785 | } else { |
6167c205 | 1786 | /* Make sure this CPU has been initialized */ |
8e49f418 VN |
1787 | if (!cpumask_test_cpu(cpu_id, buffer->cpumask)) |
1788 | goto out; | |
1789 | ||
438ced17 | 1790 | cpu_buffer = buffer->buffers[cpu_id]; |
83f40318 | 1791 | |
438ced17 VN |
1792 | if (nr_pages == cpu_buffer->nr_pages) |
1793 | goto out; | |
7a8e76a3 | 1794 | |
438ced17 VN |
1795 | cpu_buffer->nr_pages_to_update = nr_pages - |
1796 | cpu_buffer->nr_pages; | |
1797 | ||
1798 | INIT_LIST_HEAD(&cpu_buffer->new_pages); | |
1799 | if (cpu_buffer->nr_pages_to_update > 0 && | |
1800 | __rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 VN |
1801 | &cpu_buffer->new_pages, cpu_id)) { |
1802 | err = -ENOMEM; | |
1803 | goto out_err; | |
1804 | } | |
438ced17 | 1805 | |
83f40318 VN |
1806 | get_online_cpus(); |
1807 | ||
021c5b34 CM |
1808 | /* Can't run something on an offline CPU. */ |
1809 | if (!cpu_online(cpu_id)) | |
f5eb5588 SRRH |
1810 | rb_update_pages(cpu_buffer); |
1811 | else { | |
83f40318 VN |
1812 | schedule_work_on(cpu_id, |
1813 | &cpu_buffer->update_pages_work); | |
05fdd70d | 1814 | wait_for_completion(&cpu_buffer->update_done); |
f5eb5588 | 1815 | } |
83f40318 | 1816 | |
83f40318 | 1817 | cpu_buffer->nr_pages_to_update = 0; |
05fdd70d | 1818 | put_online_cpus(); |
438ced17 | 1819 | } |
7a8e76a3 SR |
1820 | |
1821 | out: | |
659f451f SR |
1822 | /* |
1823 | * The ring buffer resize can happen with the ring buffer | |
1824 | * enabled, so that the update disturbs the tracing as little | |
1825 | * as possible. But if the buffer is disabled, we do not need | |
1826 | * to worry about that, and we can take the time to verify | |
1827 | * that the buffer is not corrupt. | |
1828 | */ | |
1829 | if (atomic_read(&buffer->record_disabled)) { | |
1830 | atomic_inc(&buffer->record_disabled); | |
1831 | /* | |
1832 | * Even though the buffer was disabled, we must make sure | |
1833 | * that it is truly disabled before calling rb_check_pages. | |
1834 | * There could have been a race between checking | |
1835 | * record_disable and incrementing it. | |
1836 | */ | |
1837 | synchronize_sched(); | |
1838 | for_each_buffer_cpu(buffer, cpu) { | |
1839 | cpu_buffer = buffer->buffers[cpu]; | |
1840 | rb_check_pages(cpu_buffer); | |
1841 | } | |
1842 | atomic_dec(&buffer->record_disabled); | |
1843 | } | |
1844 | ||
7a8e76a3 | 1845 | mutex_unlock(&buffer->mutex); |
7a8e76a3 SR |
1846 | return size; |
1847 | ||
83f40318 | 1848 | out_err: |
438ced17 VN |
1849 | for_each_buffer_cpu(buffer, cpu) { |
1850 | struct buffer_page *bpage, *tmp; | |
83f40318 | 1851 | |
438ced17 | 1852 | cpu_buffer = buffer->buffers[cpu]; |
438ced17 | 1853 | cpu_buffer->nr_pages_to_update = 0; |
83f40318 | 1854 | |
438ced17 VN |
1855 | if (list_empty(&cpu_buffer->new_pages)) |
1856 | continue; | |
83f40318 | 1857 | |
438ced17 VN |
1858 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, |
1859 | list) { | |
1860 | list_del_init(&bpage->list); | |
1861 | free_buffer_page(bpage); | |
1862 | } | |
7a8e76a3 | 1863 | } |
641d2f63 | 1864 | mutex_unlock(&buffer->mutex); |
83f40318 | 1865 | return err; |
7a8e76a3 | 1866 | } |
c4f50183 | 1867 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 1868 | |
750912fa DS |
1869 | void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val) |
1870 | { | |
1871 | mutex_lock(&buffer->mutex); | |
1872 | if (val) | |
1873 | buffer->flags |= RB_FL_OVERWRITE; | |
1874 | else | |
1875 | buffer->flags &= ~RB_FL_OVERWRITE; | |
1876 | mutex_unlock(&buffer->mutex); | |
1877 | } | |
1878 | EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite); | |
1879 | ||
2289d567 | 1880 | static __always_inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 1881 | { |
044fa782 | 1882 | return bpage->page->data + index; |
7a8e76a3 SR |
1883 | } |
1884 | ||
2289d567 | 1885 | static __always_inline struct ring_buffer_event * |
d769041f | 1886 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1887 | { |
6f807acd SR |
1888 | return __rb_page_index(cpu_buffer->reader_page, |
1889 | cpu_buffer->reader_page->read); | |
1890 | } | |
1891 | ||
2289d567 | 1892 | static __always_inline struct ring_buffer_event * |
7a8e76a3 SR |
1893 | rb_iter_head_event(struct ring_buffer_iter *iter) |
1894 | { | |
6f807acd | 1895 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
1896 | } |
1897 | ||
2289d567 | 1898 | static __always_inline unsigned rb_page_commit(struct buffer_page *bpage) |
bf41a158 | 1899 | { |
abc9b56d | 1900 | return local_read(&bpage->page->commit); |
bf41a158 SR |
1901 | } |
1902 | ||
25985edc | 1903 | /* Size is determined by what has been committed */ |
2289d567 | 1904 | static __always_inline unsigned rb_page_size(struct buffer_page *bpage) |
bf41a158 SR |
1905 | { |
1906 | return rb_page_commit(bpage); | |
1907 | } | |
1908 | ||
2289d567 | 1909 | static __always_inline unsigned |
bf41a158 SR |
1910 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) |
1911 | { | |
1912 | return rb_page_commit(cpu_buffer->commit_page); | |
1913 | } | |
1914 | ||
2289d567 | 1915 | static __always_inline unsigned |
bf41a158 SR |
1916 | rb_event_index(struct ring_buffer_event *event) |
1917 | { | |
1918 | unsigned long addr = (unsigned long)event; | |
1919 | ||
22f470f8 | 1920 | return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE; |
bf41a158 SR |
1921 | } |
1922 | ||
34a148bf | 1923 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1924 | { |
1925 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1926 | ||
1927 | /* | |
1928 | * The iterator could be on the reader page (it starts there). | |
1929 | * But the head could have moved, since the reader was | |
1930 | * found. Check for this case and assign the iterator | |
1931 | * to the head page instead of next. | |
1932 | */ | |
1933 | if (iter->head_page == cpu_buffer->reader_page) | |
77ae365e | 1934 | iter->head_page = rb_set_head_page(cpu_buffer); |
d769041f SR |
1935 | else |
1936 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1937 | ||
abc9b56d | 1938 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1939 | iter->head = 0; |
1940 | } | |
1941 | ||
77ae365e SR |
1942 | /* |
1943 | * rb_handle_head_page - writer hit the head page | |
1944 | * | |
1945 | * Returns: +1 to retry page | |
1946 | * 0 to continue | |
1947 | * -1 on error | |
1948 | */ | |
1949 | static int | |
1950 | rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
1951 | struct buffer_page *tail_page, | |
1952 | struct buffer_page *next_page) | |
1953 | { | |
1954 | struct buffer_page *new_head; | |
1955 | int entries; | |
1956 | int type; | |
1957 | int ret; | |
1958 | ||
1959 | entries = rb_page_entries(next_page); | |
1960 | ||
1961 | /* | |
1962 | * The hard part is here. We need to move the head | |
1963 | * forward, and protect against both readers on | |
1964 | * other CPUs and writers coming in via interrupts. | |
1965 | */ | |
1966 | type = rb_head_page_set_update(cpu_buffer, next_page, tail_page, | |
1967 | RB_PAGE_HEAD); | |
1968 | ||
1969 | /* | |
1970 | * type can be one of four: | |
1971 | * NORMAL - an interrupt already moved it for us | |
1972 | * HEAD - we are the first to get here. | |
1973 | * UPDATE - we are the interrupt interrupting | |
1974 | * a current move. | |
1975 | * MOVED - a reader on another CPU moved the next | |
1976 | * pointer to its reader page. Give up | |
1977 | * and try again. | |
1978 | */ | |
1979 | ||
1980 | switch (type) { | |
1981 | case RB_PAGE_HEAD: | |
1982 | /* | |
1983 | * We changed the head to UPDATE, thus | |
1984 | * it is our responsibility to update | |
1985 | * the counters. | |
1986 | */ | |
1987 | local_add(entries, &cpu_buffer->overrun); | |
c64e148a | 1988 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
77ae365e SR |
1989 | |
1990 | /* | |
1991 | * The entries will be zeroed out when we move the | |
1992 | * tail page. | |
1993 | */ | |
1994 | ||
1995 | /* still more to do */ | |
1996 | break; | |
1997 | ||
1998 | case RB_PAGE_UPDATE: | |
1999 | /* | |
2000 | * This is an interrupt that interrupt the | |
2001 | * previous update. Still more to do. | |
2002 | */ | |
2003 | break; | |
2004 | case RB_PAGE_NORMAL: | |
2005 | /* | |
2006 | * An interrupt came in before the update | |
2007 | * and processed this for us. | |
2008 | * Nothing left to do. | |
2009 | */ | |
2010 | return 1; | |
2011 | case RB_PAGE_MOVED: | |
2012 | /* | |
2013 | * The reader is on another CPU and just did | |
2014 | * a swap with our next_page. | |
2015 | * Try again. | |
2016 | */ | |
2017 | return 1; | |
2018 | default: | |
2019 | RB_WARN_ON(cpu_buffer, 1); /* WTF??? */ | |
2020 | return -1; | |
2021 | } | |
2022 | ||
2023 | /* | |
2024 | * Now that we are here, the old head pointer is | |
2025 | * set to UPDATE. This will keep the reader from | |
2026 | * swapping the head page with the reader page. | |
2027 | * The reader (on another CPU) will spin till | |
2028 | * we are finished. | |
2029 | * | |
2030 | * We just need to protect against interrupts | |
2031 | * doing the job. We will set the next pointer | |
2032 | * to HEAD. After that, we set the old pointer | |
2033 | * to NORMAL, but only if it was HEAD before. | |
2034 | * otherwise we are an interrupt, and only | |
2035 | * want the outer most commit to reset it. | |
2036 | */ | |
2037 | new_head = next_page; | |
2038 | rb_inc_page(cpu_buffer, &new_head); | |
2039 | ||
2040 | ret = rb_head_page_set_head(cpu_buffer, new_head, next_page, | |
2041 | RB_PAGE_NORMAL); | |
2042 | ||
2043 | /* | |
2044 | * Valid returns are: | |
2045 | * HEAD - an interrupt came in and already set it. | |
2046 | * NORMAL - One of two things: | |
2047 | * 1) We really set it. | |
2048 | * 2) A bunch of interrupts came in and moved | |
2049 | * the page forward again. | |
2050 | */ | |
2051 | switch (ret) { | |
2052 | case RB_PAGE_HEAD: | |
2053 | case RB_PAGE_NORMAL: | |
2054 | /* OK */ | |
2055 | break; | |
2056 | default: | |
2057 | RB_WARN_ON(cpu_buffer, 1); | |
2058 | return -1; | |
2059 | } | |
2060 | ||
2061 | /* | |
2062 | * It is possible that an interrupt came in, | |
2063 | * set the head up, then more interrupts came in | |
2064 | * and moved it again. When we get back here, | |
2065 | * the page would have been set to NORMAL but we | |
2066 | * just set it back to HEAD. | |
2067 | * | |
2068 | * How do you detect this? Well, if that happened | |
2069 | * the tail page would have moved. | |
2070 | */ | |
2071 | if (ret == RB_PAGE_NORMAL) { | |
8573636e SRRH |
2072 | struct buffer_page *buffer_tail_page; |
2073 | ||
2074 | buffer_tail_page = READ_ONCE(cpu_buffer->tail_page); | |
77ae365e SR |
2075 | /* |
2076 | * If the tail had moved passed next, then we need | |
2077 | * to reset the pointer. | |
2078 | */ | |
8573636e SRRH |
2079 | if (buffer_tail_page != tail_page && |
2080 | buffer_tail_page != next_page) | |
77ae365e SR |
2081 | rb_head_page_set_normal(cpu_buffer, new_head, |
2082 | next_page, | |
2083 | RB_PAGE_HEAD); | |
2084 | } | |
2085 | ||
2086 | /* | |
2087 | * If this was the outer most commit (the one that | |
2088 | * changed the original pointer from HEAD to UPDATE), | |
2089 | * then it is up to us to reset it to NORMAL. | |
2090 | */ | |
2091 | if (type == RB_PAGE_HEAD) { | |
2092 | ret = rb_head_page_set_normal(cpu_buffer, next_page, | |
2093 | tail_page, | |
2094 | RB_PAGE_UPDATE); | |
2095 | if (RB_WARN_ON(cpu_buffer, | |
2096 | ret != RB_PAGE_UPDATE)) | |
2097 | return -1; | |
2098 | } | |
2099 | ||
2100 | return 0; | |
2101 | } | |
2102 | ||
c7b09308 SR |
2103 | static inline void |
2104 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
fcc742ea | 2105 | unsigned long tail, struct rb_event_info *info) |
c7b09308 | 2106 | { |
fcc742ea | 2107 | struct buffer_page *tail_page = info->tail_page; |
c7b09308 | 2108 | struct ring_buffer_event *event; |
fcc742ea | 2109 | unsigned long length = info->length; |
c7b09308 SR |
2110 | |
2111 | /* | |
2112 | * Only the event that crossed the page boundary | |
2113 | * must fill the old tail_page with padding. | |
2114 | */ | |
2115 | if (tail >= BUF_PAGE_SIZE) { | |
b3230c8b SR |
2116 | /* |
2117 | * If the page was filled, then we still need | |
2118 | * to update the real_end. Reset it to zero | |
2119 | * and the reader will ignore it. | |
2120 | */ | |
2121 | if (tail == BUF_PAGE_SIZE) | |
2122 | tail_page->real_end = 0; | |
2123 | ||
c7b09308 SR |
2124 | local_sub(length, &tail_page->write); |
2125 | return; | |
2126 | } | |
2127 | ||
2128 | event = __rb_page_index(tail_page, tail); | |
2129 | ||
c64e148a VN |
2130 | /* account for padding bytes */ |
2131 | local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes); | |
2132 | ||
ff0ff84a SR |
2133 | /* |
2134 | * Save the original length to the meta data. | |
2135 | * This will be used by the reader to add lost event | |
2136 | * counter. | |
2137 | */ | |
2138 | tail_page->real_end = tail; | |
2139 | ||
c7b09308 SR |
2140 | /* |
2141 | * If this event is bigger than the minimum size, then | |
2142 | * we need to be careful that we don't subtract the | |
2143 | * write counter enough to allow another writer to slip | |
2144 | * in on this page. | |
2145 | * We put in a discarded commit instead, to make sure | |
2146 | * that this space is not used again. | |
2147 | * | |
2148 | * If we are less than the minimum size, we don't need to | |
2149 | * worry about it. | |
2150 | */ | |
2151 | if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) { | |
2152 | /* No room for any events */ | |
2153 | ||
2154 | /* Mark the rest of the page with padding */ | |
2155 | rb_event_set_padding(event); | |
2156 | ||
2157 | /* Set the write back to the previous setting */ | |
2158 | local_sub(length, &tail_page->write); | |
2159 | return; | |
2160 | } | |
2161 | ||
2162 | /* Put in a discarded event */ | |
2163 | event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE; | |
2164 | event->type_len = RINGBUF_TYPE_PADDING; | |
2165 | /* time delta must be non zero */ | |
2166 | event->time_delta = 1; | |
c7b09308 SR |
2167 | |
2168 | /* Set write to end of buffer */ | |
2169 | length = (tail + length) - BUF_PAGE_SIZE; | |
2170 | local_sub(length, &tail_page->write); | |
2171 | } | |
6634ff26 | 2172 | |
4239c38f SRRH |
2173 | static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer); |
2174 | ||
747e94ae SR |
2175 | /* |
2176 | * This is the slow path, force gcc not to inline it. | |
2177 | */ | |
2178 | static noinline struct ring_buffer_event * | |
6634ff26 | 2179 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
fcc742ea | 2180 | unsigned long tail, struct rb_event_info *info) |
7a8e76a3 | 2181 | { |
fcc742ea | 2182 | struct buffer_page *tail_page = info->tail_page; |
5a50e33c | 2183 | struct buffer_page *commit_page = cpu_buffer->commit_page; |
7a8e76a3 | 2184 | struct ring_buffer *buffer = cpu_buffer->buffer; |
77ae365e SR |
2185 | struct buffer_page *next_page; |
2186 | int ret; | |
aa20ae84 SR |
2187 | |
2188 | next_page = tail_page; | |
2189 | ||
aa20ae84 SR |
2190 | rb_inc_page(cpu_buffer, &next_page); |
2191 | ||
aa20ae84 SR |
2192 | /* |
2193 | * If for some reason, we had an interrupt storm that made | |
2194 | * it all the way around the buffer, bail, and warn | |
2195 | * about it. | |
2196 | */ | |
2197 | if (unlikely(next_page == commit_page)) { | |
77ae365e | 2198 | local_inc(&cpu_buffer->commit_overrun); |
aa20ae84 SR |
2199 | goto out_reset; |
2200 | } | |
2201 | ||
77ae365e SR |
2202 | /* |
2203 | * This is where the fun begins! | |
2204 | * | |
2205 | * We are fighting against races between a reader that | |
2206 | * could be on another CPU trying to swap its reader | |
2207 | * page with the buffer head. | |
2208 | * | |
2209 | * We are also fighting against interrupts coming in and | |
2210 | * moving the head or tail on us as well. | |
2211 | * | |
2212 | * If the next page is the head page then we have filled | |
2213 | * the buffer, unless the commit page is still on the | |
2214 | * reader page. | |
2215 | */ | |
2216 | if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) { | |
aa20ae84 | 2217 | |
77ae365e SR |
2218 | /* |
2219 | * If the commit is not on the reader page, then | |
2220 | * move the header page. | |
2221 | */ | |
2222 | if (!rb_is_reader_page(cpu_buffer->commit_page)) { | |
2223 | /* | |
2224 | * If we are not in overwrite mode, | |
2225 | * this is easy, just stop here. | |
2226 | */ | |
884bfe89 SP |
2227 | if (!(buffer->flags & RB_FL_OVERWRITE)) { |
2228 | local_inc(&cpu_buffer->dropped_events); | |
77ae365e | 2229 | goto out_reset; |
884bfe89 | 2230 | } |
77ae365e SR |
2231 | |
2232 | ret = rb_handle_head_page(cpu_buffer, | |
2233 | tail_page, | |
2234 | next_page); | |
2235 | if (ret < 0) | |
2236 | goto out_reset; | |
2237 | if (ret) | |
2238 | goto out_again; | |
2239 | } else { | |
2240 | /* | |
2241 | * We need to be careful here too. The | |
2242 | * commit page could still be on the reader | |
2243 | * page. We could have a small buffer, and | |
2244 | * have filled up the buffer with events | |
2245 | * from interrupts and such, and wrapped. | |
2246 | * | |
2247 | * Note, if the tail page is also the on the | |
2248 | * reader_page, we let it move out. | |
2249 | */ | |
2250 | if (unlikely((cpu_buffer->commit_page != | |
2251 | cpu_buffer->tail_page) && | |
2252 | (cpu_buffer->commit_page == | |
2253 | cpu_buffer->reader_page))) { | |
2254 | local_inc(&cpu_buffer->commit_overrun); | |
2255 | goto out_reset; | |
2256 | } | |
aa20ae84 SR |
2257 | } |
2258 | } | |
2259 | ||
70004986 | 2260 | rb_tail_page_update(cpu_buffer, tail_page, next_page); |
aa20ae84 | 2261 | |
77ae365e | 2262 | out_again: |
aa20ae84 | 2263 | |
fcc742ea | 2264 | rb_reset_tail(cpu_buffer, tail, info); |
aa20ae84 | 2265 | |
4239c38f SRRH |
2266 | /* Commit what we have for now. */ |
2267 | rb_end_commit(cpu_buffer); | |
2268 | /* rb_end_commit() decs committing */ | |
2269 | local_inc(&cpu_buffer->committing); | |
2270 | ||
aa20ae84 SR |
2271 | /* fail and let the caller try again */ |
2272 | return ERR_PTR(-EAGAIN); | |
2273 | ||
45141d46 | 2274 | out_reset: |
6f3b3440 | 2275 | /* reset write */ |
fcc742ea | 2276 | rb_reset_tail(cpu_buffer, tail, info); |
6f3b3440 | 2277 | |
bf41a158 | 2278 | return NULL; |
7a8e76a3 SR |
2279 | } |
2280 | ||
d90fd774 SRRH |
2281 | /* Slow path, do not inline */ |
2282 | static noinline struct ring_buffer_event * | |
dc4e2801 | 2283 | rb_add_time_stamp(struct ring_buffer_event *event, u64 delta, bool abs) |
9826b273 | 2284 | { |
dc4e2801 TZ |
2285 | if (abs) |
2286 | event->type_len = RINGBUF_TYPE_TIME_STAMP; | |
2287 | else | |
2288 | event->type_len = RINGBUF_TYPE_TIME_EXTEND; | |
9826b273 | 2289 | |
dc4e2801 TZ |
2290 | /* Not the first event on the page, or not delta? */ |
2291 | if (abs || rb_event_index(event)) { | |
d90fd774 SRRH |
2292 | event->time_delta = delta & TS_MASK; |
2293 | event->array[0] = delta >> TS_SHIFT; | |
2294 | } else { | |
2295 | /* nope, just zero it */ | |
2296 | event->time_delta = 0; | |
2297 | event->array[0] = 0; | |
2298 | } | |
a4543a2f | 2299 | |
d90fd774 SRRH |
2300 | return skip_time_extend(event); |
2301 | } | |
a4543a2f | 2302 | |
cdb2a0a9 | 2303 | static inline bool rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
b7dc42fd SRRH |
2304 | struct ring_buffer_event *event); |
2305 | ||
d90fd774 SRRH |
2306 | /** |
2307 | * rb_update_event - update event type and data | |
2308 | * @event: the event to update | |
2309 | * @type: the type of event | |
2310 | * @length: the size of the event field in the ring buffer | |
2311 | * | |
2312 | * Update the type and data fields of the event. The length | |
2313 | * is the actual size that is written to the ring buffer, | |
2314 | * and with this, we can determine what to place into the | |
2315 | * data field. | |
2316 | */ | |
b7dc42fd | 2317 | static void |
d90fd774 SRRH |
2318 | rb_update_event(struct ring_buffer_per_cpu *cpu_buffer, |
2319 | struct ring_buffer_event *event, | |
2320 | struct rb_event_info *info) | |
2321 | { | |
2322 | unsigned length = info->length; | |
2323 | u64 delta = info->delta; | |
a4543a2f | 2324 | |
b7dc42fd SRRH |
2325 | /* Only a commit updates the timestamp */ |
2326 | if (unlikely(!rb_event_is_commit(cpu_buffer, event))) | |
2327 | delta = 0; | |
2328 | ||
a4543a2f | 2329 | /* |
d90fd774 | 2330 | * If we need to add a timestamp, then we |
6167c205 | 2331 | * add it to the start of the reserved space. |
a4543a2f | 2332 | */ |
d90fd774 | 2333 | if (unlikely(info->add_timestamp)) { |
dc4e2801 TZ |
2334 | bool abs = ring_buffer_time_stamp_abs(cpu_buffer->buffer); |
2335 | ||
2336 | event = rb_add_time_stamp(event, info->delta, abs); | |
d90fd774 SRRH |
2337 | length -= RB_LEN_TIME_EXTEND; |
2338 | delta = 0; | |
a4543a2f SRRH |
2339 | } |
2340 | ||
d90fd774 SRRH |
2341 | event->time_delta = delta; |
2342 | length -= RB_EVNT_HDR_SIZE; | |
2343 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) { | |
2344 | event->type_len = 0; | |
2345 | event->array[0] = length; | |
2346 | } else | |
2347 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
2348 | } | |
2349 | ||
2350 | static unsigned rb_calculate_event_length(unsigned length) | |
2351 | { | |
2352 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
2353 | ||
2354 | /* zero length can cause confusions */ | |
2355 | if (!length) | |
2356 | length++; | |
2357 | ||
2358 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) | |
2359 | length += sizeof(event.array[0]); | |
2360 | ||
2361 | length += RB_EVNT_HDR_SIZE; | |
2362 | length = ALIGN(length, RB_ARCH_ALIGNMENT); | |
2363 | ||
2364 | /* | |
2365 | * In case the time delta is larger than the 27 bits for it | |
2366 | * in the header, we need to add a timestamp. If another | |
2367 | * event comes in when trying to discard this one to increase | |
2368 | * the length, then the timestamp will be added in the allocated | |
2369 | * space of this event. If length is bigger than the size needed | |
2370 | * for the TIME_EXTEND, then padding has to be used. The events | |
2371 | * length must be either RB_LEN_TIME_EXTEND, or greater than or equal | |
2372 | * to RB_LEN_TIME_EXTEND + 8, as 8 is the minimum size for padding. | |
2373 | * As length is a multiple of 4, we only need to worry if it | |
2374 | * is 12 (RB_LEN_TIME_EXTEND + 4). | |
2375 | */ | |
2376 | if (length == RB_LEN_TIME_EXTEND + RB_ALIGNMENT) | |
2377 | length += RB_ALIGNMENT; | |
2378 | ||
2379 | return length; | |
2380 | } | |
2381 | ||
2382 | #ifndef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK | |
2383 | static inline bool sched_clock_stable(void) | |
2384 | { | |
2385 | return true; | |
2386 | } | |
2387 | #endif | |
2388 | ||
2389 | static inline int | |
2390 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
2391 | struct ring_buffer_event *event) | |
2392 | { | |
2393 | unsigned long new_index, old_index; | |
2394 | struct buffer_page *bpage; | |
2395 | unsigned long index; | |
2396 | unsigned long addr; | |
2397 | ||
2398 | new_index = rb_event_index(event); | |
2399 | old_index = new_index + rb_event_ts_length(event); | |
2400 | addr = (unsigned long)event; | |
2401 | addr &= PAGE_MASK; | |
2402 | ||
8573636e | 2403 | bpage = READ_ONCE(cpu_buffer->tail_page); |
d90fd774 SRRH |
2404 | |
2405 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { | |
2406 | unsigned long write_mask = | |
2407 | local_read(&bpage->write) & ~RB_WRITE_MASK; | |
2408 | unsigned long event_length = rb_event_length(event); | |
2409 | /* | |
2410 | * This is on the tail page. It is possible that | |
2411 | * a write could come in and move the tail page | |
2412 | * and write to the next page. That is fine | |
2413 | * because we just shorten what is on this page. | |
2414 | */ | |
2415 | old_index += write_mask; | |
2416 | new_index += write_mask; | |
2417 | index = local_cmpxchg(&bpage->write, old_index, new_index); | |
2418 | if (index == old_index) { | |
2419 | /* update counters */ | |
2420 | local_sub(event_length, &cpu_buffer->entries_bytes); | |
2421 | return 1; | |
2422 | } | |
2423 | } | |
2424 | ||
2425 | /* could not discard */ | |
2426 | return 0; | |
2427 | } | |
2428 | ||
2429 | static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer) | |
2430 | { | |
2431 | local_inc(&cpu_buffer->committing); | |
2432 | local_inc(&cpu_buffer->commits); | |
2433 | } | |
2434 | ||
38e11df1 | 2435 | static __always_inline void |
d90fd774 SRRH |
2436 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
2437 | { | |
2438 | unsigned long max_count; | |
2439 | ||
2440 | /* | |
2441 | * We only race with interrupts and NMIs on this CPU. | |
2442 | * If we own the commit event, then we can commit | |
2443 | * all others that interrupted us, since the interruptions | |
2444 | * are in stack format (they finish before they come | |
2445 | * back to us). This allows us to do a simple loop to | |
2446 | * assign the commit to the tail. | |
2447 | */ | |
2448 | again: | |
2449 | max_count = cpu_buffer->nr_pages * 100; | |
2450 | ||
8573636e | 2451 | while (cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page)) { |
d90fd774 SRRH |
2452 | if (RB_WARN_ON(cpu_buffer, !(--max_count))) |
2453 | return; | |
2454 | if (RB_WARN_ON(cpu_buffer, | |
2455 | rb_is_reader_page(cpu_buffer->tail_page))) | |
2456 | return; | |
2457 | local_set(&cpu_buffer->commit_page->page->commit, | |
2458 | rb_page_write(cpu_buffer->commit_page)); | |
2459 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); | |
70004986 SRRH |
2460 | /* Only update the write stamp if the page has an event */ |
2461 | if (rb_page_write(cpu_buffer->commit_page)) | |
2462 | cpu_buffer->write_stamp = | |
2463 | cpu_buffer->commit_page->page->time_stamp; | |
d90fd774 SRRH |
2464 | /* add barrier to keep gcc from optimizing too much */ |
2465 | barrier(); | |
2466 | } | |
2467 | while (rb_commit_index(cpu_buffer) != | |
2468 | rb_page_write(cpu_buffer->commit_page)) { | |
2469 | ||
2470 | local_set(&cpu_buffer->commit_page->page->commit, | |
2471 | rb_page_write(cpu_buffer->commit_page)); | |
2472 | RB_WARN_ON(cpu_buffer, | |
2473 | local_read(&cpu_buffer->commit_page->page->commit) & | |
2474 | ~RB_WRITE_MASK); | |
2475 | barrier(); | |
2476 | } | |
2477 | ||
2478 | /* again, keep gcc from optimizing */ | |
2479 | barrier(); | |
2480 | ||
2481 | /* | |
2482 | * If an interrupt came in just after the first while loop | |
2483 | * and pushed the tail page forward, we will be left with | |
2484 | * a dangling commit that will never go forward. | |
2485 | */ | |
8573636e | 2486 | if (unlikely(cpu_buffer->commit_page != READ_ONCE(cpu_buffer->tail_page))) |
d90fd774 SRRH |
2487 | goto again; |
2488 | } | |
2489 | ||
38e11df1 | 2490 | static __always_inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer) |
d90fd774 SRRH |
2491 | { |
2492 | unsigned long commits; | |
2493 | ||
2494 | if (RB_WARN_ON(cpu_buffer, | |
2495 | !local_read(&cpu_buffer->committing))) | |
2496 | return; | |
2497 | ||
2498 | again: | |
2499 | commits = local_read(&cpu_buffer->commits); | |
2500 | /* synchronize with interrupts */ | |
2501 | barrier(); | |
2502 | if (local_read(&cpu_buffer->committing) == 1) | |
2503 | rb_set_commit_to_write(cpu_buffer); | |
2504 | ||
2505 | local_dec(&cpu_buffer->committing); | |
2506 | ||
2507 | /* synchronize with interrupts */ | |
2508 | barrier(); | |
2509 | ||
2510 | /* | |
2511 | * Need to account for interrupts coming in between the | |
2512 | * updating of the commit page and the clearing of the | |
2513 | * committing counter. | |
2514 | */ | |
2515 | if (unlikely(local_read(&cpu_buffer->commits) != commits) && | |
2516 | !local_read(&cpu_buffer->committing)) { | |
2517 | local_inc(&cpu_buffer->committing); | |
2518 | goto again; | |
2519 | } | |
2520 | } | |
2521 | ||
2522 | static inline void rb_event_discard(struct ring_buffer_event *event) | |
2523 | { | |
dc4e2801 | 2524 | if (extended_time(event)) |
d90fd774 SRRH |
2525 | event = skip_time_extend(event); |
2526 | ||
2527 | /* array[0] holds the actual length for the discarded event */ | |
2528 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
2529 | event->type_len = RINGBUF_TYPE_PADDING; | |
2530 | /* time delta must be non zero */ | |
2531 | if (!event->time_delta) | |
2532 | event->time_delta = 1; | |
2533 | } | |
2534 | ||
babe3fce | 2535 | static __always_inline bool |
d90fd774 SRRH |
2536 | rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
2537 | struct ring_buffer_event *event) | |
2538 | { | |
2539 | unsigned long addr = (unsigned long)event; | |
2540 | unsigned long index; | |
2541 | ||
2542 | index = rb_event_index(event); | |
2543 | addr &= PAGE_MASK; | |
2544 | ||
2545 | return cpu_buffer->commit_page->page == (void *)addr && | |
2546 | rb_commit_index(cpu_buffer) == index; | |
2547 | } | |
2548 | ||
babe3fce | 2549 | static __always_inline void |
d90fd774 SRRH |
2550 | rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer, |
2551 | struct ring_buffer_event *event) | |
2552 | { | |
2553 | u64 delta; | |
2554 | ||
2555 | /* | |
2556 | * The event first in the commit queue updates the | |
2557 | * time stamp. | |
2558 | */ | |
2559 | if (rb_event_is_commit(cpu_buffer, event)) { | |
2560 | /* | |
2561 | * A commit event that is first on a page | |
2562 | * updates the write timestamp with the page stamp | |
2563 | */ | |
2564 | if (!rb_event_index(event)) | |
2565 | cpu_buffer->write_stamp = | |
2566 | cpu_buffer->commit_page->page->time_stamp; | |
2567 | else if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
dc4e2801 | 2568 | delta = ring_buffer_event_time_stamp(event); |
d90fd774 | 2569 | cpu_buffer->write_stamp += delta; |
dc4e2801 TZ |
2570 | } else if (event->type_len == RINGBUF_TYPE_TIME_STAMP) { |
2571 | delta = ring_buffer_event_time_stamp(event); | |
2572 | cpu_buffer->write_stamp = delta; | |
d90fd774 SRRH |
2573 | } else |
2574 | cpu_buffer->write_stamp += event->time_delta; | |
2575 | } | |
2576 | } | |
2577 | ||
2578 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, | |
2579 | struct ring_buffer_event *event) | |
2580 | { | |
2581 | local_inc(&cpu_buffer->entries); | |
2582 | rb_update_write_stamp(cpu_buffer, event); | |
2583 | rb_end_commit(cpu_buffer); | |
2584 | } | |
2585 | ||
2586 | static __always_inline void | |
2587 | rb_wakeups(struct ring_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer) | |
2588 | { | |
2589 | bool pagebusy; | |
2590 | ||
2591 | if (buffer->irq_work.waiters_pending) { | |
2592 | buffer->irq_work.waiters_pending = false; | |
2593 | /* irq_work_queue() supplies it's own memory barriers */ | |
2594 | irq_work_queue(&buffer->irq_work.work); | |
2595 | } | |
2596 | ||
2597 | if (cpu_buffer->irq_work.waiters_pending) { | |
2598 | cpu_buffer->irq_work.waiters_pending = false; | |
2599 | /* irq_work_queue() supplies it's own memory barriers */ | |
2600 | irq_work_queue(&cpu_buffer->irq_work.work); | |
2601 | } | |
2602 | ||
2603 | pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page; | |
2604 | ||
2605 | if (!pagebusy && cpu_buffer->irq_work.full_waiters_pending) { | |
2606 | cpu_buffer->irq_work.wakeup_full = true; | |
2607 | cpu_buffer->irq_work.full_waiters_pending = false; | |
2608 | /* irq_work_queue() supplies it's own memory barriers */ | |
2609 | irq_work_queue(&cpu_buffer->irq_work.work); | |
2610 | } | |
2611 | } | |
2612 | ||
2613 | /* | |
2614 | * The lock and unlock are done within a preempt disable section. | |
2615 | * The current_context per_cpu variable can only be modified | |
2616 | * by the current task between lock and unlock. But it can | |
a0e3a18f SRV |
2617 | * be modified more than once via an interrupt. To pass this |
2618 | * information from the lock to the unlock without having to | |
2619 | * access the 'in_interrupt()' functions again (which do show | |
2620 | * a bit of overhead in something as critical as function tracing, | |
2621 | * we use a bitmask trick. | |
d90fd774 | 2622 | * |
a0e3a18f SRV |
2623 | * bit 0 = NMI context |
2624 | * bit 1 = IRQ context | |
2625 | * bit 2 = SoftIRQ context | |
2626 | * bit 3 = normal context. | |
d90fd774 | 2627 | * |
a0e3a18f SRV |
2628 | * This works because this is the order of contexts that can |
2629 | * preempt other contexts. A SoftIRQ never preempts an IRQ | |
2630 | * context. | |
2631 | * | |
2632 | * When the context is determined, the corresponding bit is | |
2633 | * checked and set (if it was set, then a recursion of that context | |
2634 | * happened). | |
2635 | * | |
2636 | * On unlock, we need to clear this bit. To do so, just subtract | |
2637 | * 1 from the current_context and AND it to itself. | |
2638 | * | |
2639 | * (binary) | |
2640 | * 101 - 1 = 100 | |
2641 | * 101 & 100 = 100 (clearing bit zero) | |
2642 | * | |
2643 | * 1010 - 1 = 1001 | |
2644 | * 1010 & 1001 = 1000 (clearing bit 1) | |
2645 | * | |
2646 | * The least significant bit can be cleared this way, and it | |
2647 | * just so happens that it is the same bit corresponding to | |
2648 | * the current context. | |
d90fd774 SRRH |
2649 | */ |
2650 | ||
2651 | static __always_inline int | |
2652 | trace_recursive_lock(struct ring_buffer_per_cpu *cpu_buffer) | |
2653 | { | |
a0e3a18f SRV |
2654 | unsigned int val = cpu_buffer->current_context; |
2655 | unsigned long pc = preempt_count(); | |
2656 | int bit; | |
2657 | ||
2658 | if (!(pc & (NMI_MASK | HARDIRQ_MASK | SOFTIRQ_OFFSET))) | |
2659 | bit = RB_CTX_NORMAL; | |
2660 | else | |
2661 | bit = pc & NMI_MASK ? RB_CTX_NMI : | |
0164e0d7 | 2662 | pc & HARDIRQ_MASK ? RB_CTX_IRQ : RB_CTX_SOFTIRQ; |
a0e3a18f | 2663 | |
8e012066 | 2664 | if (unlikely(val & (1 << (bit + cpu_buffer->nest)))) |
d90fd774 SRRH |
2665 | return 1; |
2666 | ||
8e012066 | 2667 | val |= (1 << (bit + cpu_buffer->nest)); |
a0e3a18f | 2668 | cpu_buffer->current_context = val; |
d90fd774 SRRH |
2669 | |
2670 | return 0; | |
2671 | } | |
2672 | ||
2673 | static __always_inline void | |
2674 | trace_recursive_unlock(struct ring_buffer_per_cpu *cpu_buffer) | |
2675 | { | |
8e012066 SRV |
2676 | cpu_buffer->current_context &= |
2677 | cpu_buffer->current_context - (1 << cpu_buffer->nest); | |
2678 | } | |
2679 | ||
2680 | /* The recursive locking above uses 4 bits */ | |
2681 | #define NESTED_BITS 4 | |
2682 | ||
2683 | /** | |
2684 | * ring_buffer_nest_start - Allow to trace while nested | |
2685 | * @buffer: The ring buffer to modify | |
2686 | * | |
6167c205 | 2687 | * The ring buffer has a safety mechanism to prevent recursion. |
8e012066 SRV |
2688 | * But there may be a case where a trace needs to be done while |
2689 | * tracing something else. In this case, calling this function | |
2690 | * will allow this function to nest within a currently active | |
2691 | * ring_buffer_lock_reserve(). | |
2692 | * | |
2693 | * Call this function before calling another ring_buffer_lock_reserve() and | |
2694 | * call ring_buffer_nest_end() after the nested ring_buffer_unlock_commit(). | |
2695 | */ | |
2696 | void ring_buffer_nest_start(struct ring_buffer *buffer) | |
2697 | { | |
2698 | struct ring_buffer_per_cpu *cpu_buffer; | |
2699 | int cpu; | |
2700 | ||
2701 | /* Enabled by ring_buffer_nest_end() */ | |
2702 | preempt_disable_notrace(); | |
2703 | cpu = raw_smp_processor_id(); | |
2704 | cpu_buffer = buffer->buffers[cpu]; | |
6167c205 | 2705 | /* This is the shift value for the above recursive locking */ |
8e012066 SRV |
2706 | cpu_buffer->nest += NESTED_BITS; |
2707 | } | |
2708 | ||
2709 | /** | |
2710 | * ring_buffer_nest_end - Allow to trace while nested | |
2711 | * @buffer: The ring buffer to modify | |
2712 | * | |
2713 | * Must be called after ring_buffer_nest_start() and after the | |
2714 | * ring_buffer_unlock_commit(). | |
2715 | */ | |
2716 | void ring_buffer_nest_end(struct ring_buffer *buffer) | |
2717 | { | |
2718 | struct ring_buffer_per_cpu *cpu_buffer; | |
2719 | int cpu; | |
2720 | ||
2721 | /* disabled by ring_buffer_nest_start() */ | |
2722 | cpu = raw_smp_processor_id(); | |
2723 | cpu_buffer = buffer->buffers[cpu]; | |
6167c205 | 2724 | /* This is the shift value for the above recursive locking */ |
8e012066 SRV |
2725 | cpu_buffer->nest -= NESTED_BITS; |
2726 | preempt_enable_notrace(); | |
d90fd774 SRRH |
2727 | } |
2728 | ||
2729 | /** | |
2730 | * ring_buffer_unlock_commit - commit a reserved | |
2731 | * @buffer: The buffer to commit to | |
2732 | * @event: The event pointer to commit. | |
2733 | * | |
2734 | * This commits the data to the ring buffer, and releases any locks held. | |
2735 | * | |
2736 | * Must be paired with ring_buffer_lock_reserve. | |
2737 | */ | |
2738 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
2739 | struct ring_buffer_event *event) | |
2740 | { | |
2741 | struct ring_buffer_per_cpu *cpu_buffer; | |
2742 | int cpu = raw_smp_processor_id(); | |
2743 | ||
2744 | cpu_buffer = buffer->buffers[cpu]; | |
2745 | ||
2746 | rb_commit(cpu_buffer, event); | |
2747 | ||
2748 | rb_wakeups(buffer, cpu_buffer); | |
2749 | ||
2750 | trace_recursive_unlock(cpu_buffer); | |
2751 | ||
2752 | preempt_enable_notrace(); | |
2753 | ||
2754 | return 0; | |
2755 | } | |
2756 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); | |
2757 | ||
2758 | static noinline void | |
2759 | rb_handle_timestamp(struct ring_buffer_per_cpu *cpu_buffer, | |
d90fd774 SRRH |
2760 | struct rb_event_info *info) |
2761 | { | |
d90fd774 SRRH |
2762 | WARN_ONCE(info->delta > (1ULL << 59), |
2763 | KERN_WARNING "Delta way too big! %llu ts=%llu write stamp = %llu\n%s", | |
2764 | (unsigned long long)info->delta, | |
2765 | (unsigned long long)info->ts, | |
2766 | (unsigned long long)cpu_buffer->write_stamp, | |
2767 | sched_clock_stable() ? "" : | |
2768 | "If you just came from a suspend/resume,\n" | |
2769 | "please switch to the trace global clock:\n" | |
913ea4d0 CW |
2770 | " echo global > /sys/kernel/debug/tracing/trace_clock\n" |
2771 | "or add trace_clock=global to the kernel command line\n"); | |
b7dc42fd | 2772 | info->add_timestamp = 1; |
9826b273 SRRH |
2773 | } |
2774 | ||
6634ff26 SR |
2775 | static struct ring_buffer_event * |
2776 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
fcc742ea | 2777 | struct rb_event_info *info) |
6634ff26 | 2778 | { |
6634ff26 | 2779 | struct ring_buffer_event *event; |
fcc742ea | 2780 | struct buffer_page *tail_page; |
6634ff26 | 2781 | unsigned long tail, write; |
b7dc42fd SRRH |
2782 | |
2783 | /* | |
2784 | * If the time delta since the last event is too big to | |
2785 | * hold in the time field of the event, then we append a | |
2786 | * TIME EXTEND event ahead of the data event. | |
2787 | */ | |
2788 | if (unlikely(info->add_timestamp)) | |
2789 | info->length += RB_LEN_TIME_EXTEND; | |
69d1b839 | 2790 | |
8573636e SRRH |
2791 | /* Don't let the compiler play games with cpu_buffer->tail_page */ |
2792 | tail_page = info->tail_page = READ_ONCE(cpu_buffer->tail_page); | |
fcc742ea | 2793 | write = local_add_return(info->length, &tail_page->write); |
77ae365e SR |
2794 | |
2795 | /* set write to only the index of the write */ | |
2796 | write &= RB_WRITE_MASK; | |
fcc742ea | 2797 | tail = write - info->length; |
6634ff26 | 2798 | |
6634ff26 | 2799 | /* |
a4543a2f | 2800 | * If this is the first commit on the page, then it has the same |
b7dc42fd | 2801 | * timestamp as the page itself. |
6634ff26 | 2802 | */ |
dc4e2801 | 2803 | if (!tail && !ring_buffer_time_stamp_abs(cpu_buffer->buffer)) |
a4543a2f SRRH |
2804 | info->delta = 0; |
2805 | ||
b7dc42fd SRRH |
2806 | /* See if we shot pass the end of this buffer page */ |
2807 | if (unlikely(write > BUF_PAGE_SIZE)) | |
2808 | return rb_move_tail(cpu_buffer, tail, info); | |
a4543a2f | 2809 | |
b7dc42fd SRRH |
2810 | /* We reserved something on the buffer */ |
2811 | ||
2812 | event = __rb_page_index(tail_page, tail); | |
a4543a2f SRRH |
2813 | rb_update_event(cpu_buffer, event, info); |
2814 | ||
2815 | local_inc(&tail_page->entries); | |
6634ff26 | 2816 | |
b7dc42fd SRRH |
2817 | /* |
2818 | * If this is the first commit on the page, then update | |
2819 | * its timestamp. | |
2820 | */ | |
2821 | if (!tail) | |
2822 | tail_page->page->time_stamp = info->ts; | |
2823 | ||
c64e148a | 2824 | /* account for these added bytes */ |
fcc742ea | 2825 | local_add(info->length, &cpu_buffer->entries_bytes); |
c64e148a | 2826 | |
6634ff26 SR |
2827 | return event; |
2828 | } | |
2829 | ||
fa7ffb39 | 2830 | static __always_inline struct ring_buffer_event * |
62f0b3eb SR |
2831 | rb_reserve_next_event(struct ring_buffer *buffer, |
2832 | struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 2833 | unsigned long length) |
7a8e76a3 SR |
2834 | { |
2835 | struct ring_buffer_event *event; | |
fcc742ea | 2836 | struct rb_event_info info; |
818e3dd3 | 2837 | int nr_loops = 0; |
b7dc42fd | 2838 | u64 diff; |
7a8e76a3 | 2839 | |
fa743953 SR |
2840 | rb_start_commit(cpu_buffer); |
2841 | ||
85bac32c | 2842 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
62f0b3eb SR |
2843 | /* |
2844 | * Due to the ability to swap a cpu buffer from a buffer | |
2845 | * it is possible it was swapped before we committed. | |
2846 | * (committing stops a swap). We check for it here and | |
2847 | * if it happened, we have to fail the write. | |
2848 | */ | |
2849 | barrier(); | |
6aa7de05 | 2850 | if (unlikely(READ_ONCE(cpu_buffer->buffer) != buffer)) { |
62f0b3eb SR |
2851 | local_dec(&cpu_buffer->committing); |
2852 | local_dec(&cpu_buffer->commits); | |
2853 | return NULL; | |
2854 | } | |
85bac32c | 2855 | #endif |
b7dc42fd | 2856 | |
fcc742ea | 2857 | info.length = rb_calculate_event_length(length); |
a4543a2f | 2858 | again: |
b7dc42fd SRRH |
2859 | info.add_timestamp = 0; |
2860 | info.delta = 0; | |
2861 | ||
818e3dd3 SR |
2862 | /* |
2863 | * We allow for interrupts to reenter here and do a trace. | |
2864 | * If one does, it will cause this original code to loop | |
2865 | * back here. Even with heavy interrupts happening, this | |
2866 | * should only happen a few times in a row. If this happens | |
2867 | * 1000 times in a row, there must be either an interrupt | |
2868 | * storm or we have something buggy. | |
2869 | * Bail! | |
2870 | */ | |
3e89c7bb | 2871 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
fa743953 | 2872 | goto out_fail; |
818e3dd3 | 2873 | |
b7dc42fd SRRH |
2874 | info.ts = rb_time_stamp(cpu_buffer->buffer); |
2875 | diff = info.ts - cpu_buffer->write_stamp; | |
2876 | ||
2877 | /* make sure this diff is calculated here */ | |
2878 | barrier(); | |
2879 | ||
dc4e2801 TZ |
2880 | if (ring_buffer_time_stamp_abs(buffer)) { |
2881 | info.delta = info.ts; | |
2882 | rb_handle_timestamp(cpu_buffer, &info); | |
2883 | } else /* Did the write stamp get updated already? */ | |
2884 | if (likely(info.ts >= cpu_buffer->write_stamp)) { | |
b7dc42fd SRRH |
2885 | info.delta = diff; |
2886 | if (unlikely(test_time_stamp(info.delta))) | |
2887 | rb_handle_timestamp(cpu_buffer, &info); | |
2888 | } | |
2889 | ||
fcc742ea SRRH |
2890 | event = __rb_reserve_next(cpu_buffer, &info); |
2891 | ||
bd1b7cd3 SRRH |
2892 | if (unlikely(PTR_ERR(event) == -EAGAIN)) { |
2893 | if (info.add_timestamp) | |
2894 | info.length -= RB_LEN_TIME_EXTEND; | |
bf41a158 | 2895 | goto again; |
bd1b7cd3 | 2896 | } |
bf41a158 | 2897 | |
fa743953 SR |
2898 | if (!event) |
2899 | goto out_fail; | |
7a8e76a3 | 2900 | |
7a8e76a3 | 2901 | return event; |
fa743953 SR |
2902 | |
2903 | out_fail: | |
2904 | rb_end_commit(cpu_buffer); | |
2905 | return NULL; | |
7a8e76a3 SR |
2906 | } |
2907 | ||
2908 | /** | |
2909 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
2910 | * @buffer: the ring buffer to reserve from | |
2911 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 | 2912 | * |
6167c205 | 2913 | * Returns a reserved event on the ring buffer to copy directly to. |
7a8e76a3 SR |
2914 | * The user of this interface will need to get the body to write into |
2915 | * and can use the ring_buffer_event_data() interface. | |
2916 | * | |
2917 | * The length is the length of the data needed, not the event length | |
2918 | * which also includes the event header. | |
2919 | * | |
2920 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
2921 | * If NULL is returned, then nothing has been allocated or locked. | |
2922 | */ | |
2923 | struct ring_buffer_event * | |
0a987751 | 2924 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
2925 | { |
2926 | struct ring_buffer_per_cpu *cpu_buffer; | |
2927 | struct ring_buffer_event *event; | |
5168ae50 | 2928 | int cpu; |
7a8e76a3 | 2929 | |
bf41a158 | 2930 | /* If we are tracing schedule, we don't want to recurse */ |
5168ae50 | 2931 | preempt_disable_notrace(); |
bf41a158 | 2932 | |
3205f806 | 2933 | if (unlikely(atomic_read(&buffer->record_disabled))) |
58a09ec6 | 2934 | goto out; |
261842b7 | 2935 | |
7a8e76a3 SR |
2936 | cpu = raw_smp_processor_id(); |
2937 | ||
3205f806 | 2938 | if (unlikely(!cpumask_test_cpu(cpu, buffer->cpumask))) |
d769041f | 2939 | goto out; |
7a8e76a3 SR |
2940 | |
2941 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 2942 | |
3205f806 | 2943 | if (unlikely(atomic_read(&cpu_buffer->record_disabled))) |
d769041f | 2944 | goto out; |
7a8e76a3 | 2945 | |
3205f806 | 2946 | if (unlikely(length > BUF_MAX_DATA_SIZE)) |
bf41a158 | 2947 | goto out; |
7a8e76a3 | 2948 | |
58a09ec6 SRRH |
2949 | if (unlikely(trace_recursive_lock(cpu_buffer))) |
2950 | goto out; | |
2951 | ||
62f0b3eb | 2952 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 2953 | if (!event) |
58a09ec6 | 2954 | goto out_unlock; |
7a8e76a3 SR |
2955 | |
2956 | return event; | |
2957 | ||
58a09ec6 SRRH |
2958 | out_unlock: |
2959 | trace_recursive_unlock(cpu_buffer); | |
d769041f | 2960 | out: |
5168ae50 | 2961 | preempt_enable_notrace(); |
7a8e76a3 SR |
2962 | return NULL; |
2963 | } | |
c4f50183 | 2964 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 | 2965 | |
a1863c21 SR |
2966 | /* |
2967 | * Decrement the entries to the page that an event is on. | |
2968 | * The event does not even need to exist, only the pointer | |
2969 | * to the page it is on. This may only be called before the commit | |
2970 | * takes place. | |
2971 | */ | |
2972 | static inline void | |
2973 | rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer, | |
2974 | struct ring_buffer_event *event) | |
2975 | { | |
2976 | unsigned long addr = (unsigned long)event; | |
2977 | struct buffer_page *bpage = cpu_buffer->commit_page; | |
2978 | struct buffer_page *start; | |
2979 | ||
2980 | addr &= PAGE_MASK; | |
2981 | ||
2982 | /* Do the likely case first */ | |
2983 | if (likely(bpage->page == (void *)addr)) { | |
2984 | local_dec(&bpage->entries); | |
2985 | return; | |
2986 | } | |
2987 | ||
2988 | /* | |
2989 | * Because the commit page may be on the reader page we | |
2990 | * start with the next page and check the end loop there. | |
2991 | */ | |
2992 | rb_inc_page(cpu_buffer, &bpage); | |
2993 | start = bpage; | |
2994 | do { | |
2995 | if (bpage->page == (void *)addr) { | |
2996 | local_dec(&bpage->entries); | |
2997 | return; | |
2998 | } | |
2999 | rb_inc_page(cpu_buffer, &bpage); | |
3000 | } while (bpage != start); | |
3001 | ||
3002 | /* commit not part of this buffer?? */ | |
3003 | RB_WARN_ON(cpu_buffer, 1); | |
3004 | } | |
3005 | ||
fa1b47dd SR |
3006 | /** |
3007 | * ring_buffer_commit_discard - discard an event that has not been committed | |
3008 | * @buffer: the ring buffer | |
3009 | * @event: non committed event to discard | |
3010 | * | |
dc892f73 SR |
3011 | * Sometimes an event that is in the ring buffer needs to be ignored. |
3012 | * This function lets the user discard an event in the ring buffer | |
3013 | * and then that event will not be read later. | |
3014 | * | |
6167c205 | 3015 | * This function only works if it is called before the item has been |
dc892f73 | 3016 | * committed. It will try to free the event from the ring buffer |
fa1b47dd SR |
3017 | * if another event has not been added behind it. |
3018 | * | |
3019 | * If another event has been added behind it, it will set the event | |
3020 | * up as discarded, and perform the commit. | |
3021 | * | |
3022 | * If this function is called, do not call ring_buffer_unlock_commit on | |
3023 | * the event. | |
3024 | */ | |
3025 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
3026 | struct ring_buffer_event *event) | |
3027 | { | |
3028 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
3029 | int cpu; |
3030 | ||
3031 | /* The event is discarded regardless */ | |
f3b9aae1 | 3032 | rb_event_discard(event); |
fa1b47dd | 3033 | |
fa743953 SR |
3034 | cpu = smp_processor_id(); |
3035 | cpu_buffer = buffer->buffers[cpu]; | |
3036 | ||
fa1b47dd SR |
3037 | /* |
3038 | * This must only be called if the event has not been | |
3039 | * committed yet. Thus we can assume that preemption | |
3040 | * is still disabled. | |
3041 | */ | |
fa743953 | 3042 | RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing)); |
fa1b47dd | 3043 | |
a1863c21 | 3044 | rb_decrement_entry(cpu_buffer, event); |
0f2541d2 | 3045 | if (rb_try_to_discard(cpu_buffer, event)) |
edd813bf | 3046 | goto out; |
fa1b47dd SR |
3047 | |
3048 | /* | |
3049 | * The commit is still visible by the reader, so we | |
a1863c21 | 3050 | * must still update the timestamp. |
fa1b47dd | 3051 | */ |
a1863c21 | 3052 | rb_update_write_stamp(cpu_buffer, event); |
fa1b47dd | 3053 | out: |
fa743953 | 3054 | rb_end_commit(cpu_buffer); |
fa1b47dd | 3055 | |
58a09ec6 | 3056 | trace_recursive_unlock(cpu_buffer); |
f3b9aae1 | 3057 | |
5168ae50 | 3058 | preempt_enable_notrace(); |
fa1b47dd SR |
3059 | |
3060 | } | |
3061 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
3062 | ||
7a8e76a3 SR |
3063 | /** |
3064 | * ring_buffer_write - write data to the buffer without reserving | |
3065 | * @buffer: The ring buffer to write to. | |
3066 | * @length: The length of the data being written (excluding the event header) | |
3067 | * @data: The data to write to the buffer. | |
3068 | * | |
3069 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
3070 | * one function. If you already have the data to write to the buffer, it | |
3071 | * may be easier to simply call this function. | |
3072 | * | |
3073 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
3074 | * and not the length of the event which would hold the header. | |
3075 | */ | |
3076 | int ring_buffer_write(struct ring_buffer *buffer, | |
01e3e710 DS |
3077 | unsigned long length, |
3078 | void *data) | |
7a8e76a3 SR |
3079 | { |
3080 | struct ring_buffer_per_cpu *cpu_buffer; | |
3081 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
3082 | void *body; |
3083 | int ret = -EBUSY; | |
5168ae50 | 3084 | int cpu; |
7a8e76a3 | 3085 | |
5168ae50 | 3086 | preempt_disable_notrace(); |
bf41a158 | 3087 | |
52fbe9cd LJ |
3088 | if (atomic_read(&buffer->record_disabled)) |
3089 | goto out; | |
3090 | ||
7a8e76a3 SR |
3091 | cpu = raw_smp_processor_id(); |
3092 | ||
9e01c1b7 | 3093 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 3094 | goto out; |
7a8e76a3 SR |
3095 | |
3096 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
3097 | |
3098 | if (atomic_read(&cpu_buffer->record_disabled)) | |
3099 | goto out; | |
3100 | ||
be957c44 SR |
3101 | if (length > BUF_MAX_DATA_SIZE) |
3102 | goto out; | |
3103 | ||
985e871b SRRH |
3104 | if (unlikely(trace_recursive_lock(cpu_buffer))) |
3105 | goto out; | |
3106 | ||
62f0b3eb | 3107 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 3108 | if (!event) |
985e871b | 3109 | goto out_unlock; |
7a8e76a3 SR |
3110 | |
3111 | body = rb_event_data(event); | |
3112 | ||
3113 | memcpy(body, data, length); | |
3114 | ||
3115 | rb_commit(cpu_buffer, event); | |
3116 | ||
15693458 SRRH |
3117 | rb_wakeups(buffer, cpu_buffer); |
3118 | ||
7a8e76a3 | 3119 | ret = 0; |
985e871b SRRH |
3120 | |
3121 | out_unlock: | |
3122 | trace_recursive_unlock(cpu_buffer); | |
3123 | ||
7a8e76a3 | 3124 | out: |
5168ae50 | 3125 | preempt_enable_notrace(); |
7a8e76a3 SR |
3126 | |
3127 | return ret; | |
3128 | } | |
c4f50183 | 3129 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 3130 | |
da58834c | 3131 | static bool rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
3132 | { |
3133 | struct buffer_page *reader = cpu_buffer->reader_page; | |
77ae365e | 3134 | struct buffer_page *head = rb_set_head_page(cpu_buffer); |
bf41a158 SR |
3135 | struct buffer_page *commit = cpu_buffer->commit_page; |
3136 | ||
77ae365e SR |
3137 | /* In case of error, head will be NULL */ |
3138 | if (unlikely(!head)) | |
da58834c | 3139 | return true; |
77ae365e | 3140 | |
bf41a158 SR |
3141 | return reader->read == rb_page_commit(reader) && |
3142 | (commit == reader || | |
3143 | (commit == head && | |
3144 | head->read == rb_page_commit(commit))); | |
3145 | } | |
3146 | ||
7a8e76a3 SR |
3147 | /** |
3148 | * ring_buffer_record_disable - stop all writes into the buffer | |
3149 | * @buffer: The ring buffer to stop writes to. | |
3150 | * | |
3151 | * This prevents all writes to the buffer. Any attempt to write | |
3152 | * to the buffer after this will fail and return NULL. | |
3153 | * | |
3154 | * The caller should call synchronize_sched() after this. | |
3155 | */ | |
3156 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
3157 | { | |
3158 | atomic_inc(&buffer->record_disabled); | |
3159 | } | |
c4f50183 | 3160 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
3161 | |
3162 | /** | |
3163 | * ring_buffer_record_enable - enable writes to the buffer | |
3164 | * @buffer: The ring buffer to enable writes | |
3165 | * | |
3166 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 3167 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
3168 | */ |
3169 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
3170 | { | |
3171 | atomic_dec(&buffer->record_disabled); | |
3172 | } | |
c4f50183 | 3173 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 | 3174 | |
499e5470 SR |
3175 | /** |
3176 | * ring_buffer_record_off - stop all writes into the buffer | |
3177 | * @buffer: The ring buffer to stop writes to. | |
3178 | * | |
3179 | * This prevents all writes to the buffer. Any attempt to write | |
3180 | * to the buffer after this will fail and return NULL. | |
3181 | * | |
3182 | * This is different than ring_buffer_record_disable() as | |
87abb3b1 | 3183 | * it works like an on/off switch, where as the disable() version |
499e5470 SR |
3184 | * must be paired with a enable(). |
3185 | */ | |
3186 | void ring_buffer_record_off(struct ring_buffer *buffer) | |
3187 | { | |
3188 | unsigned int rd; | |
3189 | unsigned int new_rd; | |
3190 | ||
3191 | do { | |
3192 | rd = atomic_read(&buffer->record_disabled); | |
3193 | new_rd = rd | RB_BUFFER_OFF; | |
3194 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
3195 | } | |
3196 | EXPORT_SYMBOL_GPL(ring_buffer_record_off); | |
3197 | ||
3198 | /** | |
3199 | * ring_buffer_record_on - restart writes into the buffer | |
3200 | * @buffer: The ring buffer to start writes to. | |
3201 | * | |
3202 | * This enables all writes to the buffer that was disabled by | |
3203 | * ring_buffer_record_off(). | |
3204 | * | |
3205 | * This is different than ring_buffer_record_enable() as | |
87abb3b1 | 3206 | * it works like an on/off switch, where as the enable() version |
499e5470 SR |
3207 | * must be paired with a disable(). |
3208 | */ | |
3209 | void ring_buffer_record_on(struct ring_buffer *buffer) | |
3210 | { | |
3211 | unsigned int rd; | |
3212 | unsigned int new_rd; | |
3213 | ||
3214 | do { | |
3215 | rd = atomic_read(&buffer->record_disabled); | |
3216 | new_rd = rd & ~RB_BUFFER_OFF; | |
3217 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
3218 | } | |
3219 | EXPORT_SYMBOL_GPL(ring_buffer_record_on); | |
3220 | ||
3221 | /** | |
3222 | * ring_buffer_record_is_on - return true if the ring buffer can write | |
3223 | * @buffer: The ring buffer to see if write is enabled | |
3224 | * | |
3225 | * Returns true if the ring buffer is in a state that it accepts writes. | |
3226 | */ | |
3ebea280 | 3227 | bool ring_buffer_record_is_on(struct ring_buffer *buffer) |
499e5470 SR |
3228 | { |
3229 | return !atomic_read(&buffer->record_disabled); | |
3230 | } | |
3231 | ||
73c8d894 MH |
3232 | /** |
3233 | * ring_buffer_record_is_set_on - return true if the ring buffer is set writable | |
3234 | * @buffer: The ring buffer to see if write is set enabled | |
3235 | * | |
3236 | * Returns true if the ring buffer is set writable by ring_buffer_record_on(). | |
3237 | * Note that this does NOT mean it is in a writable state. | |
3238 | * | |
3239 | * It may return true when the ring buffer has been disabled by | |
3240 | * ring_buffer_record_disable(), as that is a temporary disabling of | |
3241 | * the ring buffer. | |
3242 | */ | |
d7224c0e | 3243 | bool ring_buffer_record_is_set_on(struct ring_buffer *buffer) |
73c8d894 MH |
3244 | { |
3245 | return !(atomic_read(&buffer->record_disabled) & RB_BUFFER_OFF); | |
3246 | } | |
3247 | ||
7a8e76a3 SR |
3248 | /** |
3249 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
3250 | * @buffer: The ring buffer to stop writes to. | |
3251 | * @cpu: The CPU buffer to stop | |
3252 | * | |
3253 | * This prevents all writes to the buffer. Any attempt to write | |
3254 | * to the buffer after this will fail and return NULL. | |
3255 | * | |
3256 | * The caller should call synchronize_sched() after this. | |
3257 | */ | |
3258 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
3259 | { | |
3260 | struct ring_buffer_per_cpu *cpu_buffer; | |
3261 | ||
9e01c1b7 | 3262 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3263 | return; |
7a8e76a3 SR |
3264 | |
3265 | cpu_buffer = buffer->buffers[cpu]; | |
3266 | atomic_inc(&cpu_buffer->record_disabled); | |
3267 | } | |
c4f50183 | 3268 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
3269 | |
3270 | /** | |
3271 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
3272 | * @buffer: The ring buffer to enable writes | |
3273 | * @cpu: The CPU to enable. | |
3274 | * | |
3275 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 3276 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
3277 | */ |
3278 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
3279 | { | |
3280 | struct ring_buffer_per_cpu *cpu_buffer; | |
3281 | ||
9e01c1b7 | 3282 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3283 | return; |
7a8e76a3 SR |
3284 | |
3285 | cpu_buffer = buffer->buffers[cpu]; | |
3286 | atomic_dec(&cpu_buffer->record_disabled); | |
3287 | } | |
c4f50183 | 3288 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 | 3289 | |
f6195aa0 SR |
3290 | /* |
3291 | * The total entries in the ring buffer is the running counter | |
3292 | * of entries entered into the ring buffer, minus the sum of | |
3293 | * the entries read from the ring buffer and the number of | |
3294 | * entries that were overwritten. | |
3295 | */ | |
3296 | static inline unsigned long | |
3297 | rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer) | |
3298 | { | |
3299 | return local_read(&cpu_buffer->entries) - | |
3300 | (local_read(&cpu_buffer->overrun) + cpu_buffer->read); | |
3301 | } | |
3302 | ||
c64e148a VN |
3303 | /** |
3304 | * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer | |
3305 | * @buffer: The ring buffer | |
3306 | * @cpu: The per CPU buffer to read from. | |
3307 | */ | |
50ecf2c3 | 3308 | u64 ring_buffer_oldest_event_ts(struct ring_buffer *buffer, int cpu) |
c64e148a VN |
3309 | { |
3310 | unsigned long flags; | |
3311 | struct ring_buffer_per_cpu *cpu_buffer; | |
3312 | struct buffer_page *bpage; | |
da830e58 | 3313 | u64 ret = 0; |
c64e148a VN |
3314 | |
3315 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3316 | return 0; | |
3317 | ||
3318 | cpu_buffer = buffer->buffers[cpu]; | |
7115e3fc | 3319 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
3320 | /* |
3321 | * if the tail is on reader_page, oldest time stamp is on the reader | |
3322 | * page | |
3323 | */ | |
3324 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
3325 | bpage = cpu_buffer->reader_page; | |
3326 | else | |
3327 | bpage = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
3328 | if (bpage) |
3329 | ret = bpage->page->time_stamp; | |
7115e3fc | 3330 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
3331 | |
3332 | return ret; | |
3333 | } | |
3334 | EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts); | |
3335 | ||
3336 | /** | |
3337 | * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer | |
3338 | * @buffer: The ring buffer | |
3339 | * @cpu: The per CPU buffer to read from. | |
3340 | */ | |
3341 | unsigned long ring_buffer_bytes_cpu(struct ring_buffer *buffer, int cpu) | |
3342 | { | |
3343 | struct ring_buffer_per_cpu *cpu_buffer; | |
3344 | unsigned long ret; | |
3345 | ||
3346 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3347 | return 0; | |
3348 | ||
3349 | cpu_buffer = buffer->buffers[cpu]; | |
3350 | ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes; | |
3351 | ||
3352 | return ret; | |
3353 | } | |
3354 | EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu); | |
3355 | ||
7a8e76a3 SR |
3356 | /** |
3357 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
3358 | * @buffer: The ring buffer | |
3359 | * @cpu: The per CPU buffer to get the entries from. | |
3360 | */ | |
3361 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
3362 | { | |
3363 | struct ring_buffer_per_cpu *cpu_buffer; | |
3364 | ||
9e01c1b7 | 3365 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3366 | return 0; |
7a8e76a3 SR |
3367 | |
3368 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 3369 | |
f6195aa0 | 3370 | return rb_num_of_entries(cpu_buffer); |
7a8e76a3 | 3371 | } |
c4f50183 | 3372 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
3373 | |
3374 | /** | |
884bfe89 SP |
3375 | * ring_buffer_overrun_cpu - get the number of overruns caused by the ring |
3376 | * buffer wrapping around (only if RB_FL_OVERWRITE is on). | |
7a8e76a3 SR |
3377 | * @buffer: The ring buffer |
3378 | * @cpu: The per CPU buffer to get the number of overruns from | |
3379 | */ | |
3380 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3381 | { | |
3382 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3383 | unsigned long ret; |
7a8e76a3 | 3384 | |
9e01c1b7 | 3385 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3386 | return 0; |
7a8e76a3 SR |
3387 | |
3388 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3389 | ret = local_read(&cpu_buffer->overrun); |
554f786e SR |
3390 | |
3391 | return ret; | |
7a8e76a3 | 3392 | } |
c4f50183 | 3393 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 3394 | |
f0d2c681 | 3395 | /** |
884bfe89 SP |
3396 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by |
3397 | * commits failing due to the buffer wrapping around while there are uncommitted | |
3398 | * events, such as during an interrupt storm. | |
f0d2c681 SR |
3399 | * @buffer: The ring buffer |
3400 | * @cpu: The per CPU buffer to get the number of overruns from | |
3401 | */ | |
3402 | unsigned long | |
3403 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3404 | { | |
3405 | struct ring_buffer_per_cpu *cpu_buffer; | |
3406 | unsigned long ret; | |
3407 | ||
3408 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3409 | return 0; | |
3410 | ||
3411 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3412 | ret = local_read(&cpu_buffer->commit_overrun); |
f0d2c681 SR |
3413 | |
3414 | return ret; | |
3415 | } | |
3416 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
3417 | ||
884bfe89 SP |
3418 | /** |
3419 | * ring_buffer_dropped_events_cpu - get the number of dropped events caused by | |
3420 | * the ring buffer filling up (only if RB_FL_OVERWRITE is off). | |
3421 | * @buffer: The ring buffer | |
3422 | * @cpu: The per CPU buffer to get the number of overruns from | |
3423 | */ | |
3424 | unsigned long | |
3425 | ring_buffer_dropped_events_cpu(struct ring_buffer *buffer, int cpu) | |
3426 | { | |
3427 | struct ring_buffer_per_cpu *cpu_buffer; | |
3428 | unsigned long ret; | |
3429 | ||
3430 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3431 | return 0; | |
3432 | ||
3433 | cpu_buffer = buffer->buffers[cpu]; | |
3434 | ret = local_read(&cpu_buffer->dropped_events); | |
3435 | ||
3436 | return ret; | |
3437 | } | |
3438 | EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu); | |
3439 | ||
ad964704 SRRH |
3440 | /** |
3441 | * ring_buffer_read_events_cpu - get the number of events successfully read | |
3442 | * @buffer: The ring buffer | |
3443 | * @cpu: The per CPU buffer to get the number of events read | |
3444 | */ | |
3445 | unsigned long | |
3446 | ring_buffer_read_events_cpu(struct ring_buffer *buffer, int cpu) | |
3447 | { | |
3448 | struct ring_buffer_per_cpu *cpu_buffer; | |
3449 | ||
3450 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3451 | return 0; | |
3452 | ||
3453 | cpu_buffer = buffer->buffers[cpu]; | |
3454 | return cpu_buffer->read; | |
3455 | } | |
3456 | EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu); | |
3457 | ||
7a8e76a3 SR |
3458 | /** |
3459 | * ring_buffer_entries - get the number of entries in a buffer | |
3460 | * @buffer: The ring buffer | |
3461 | * | |
3462 | * Returns the total number of entries in the ring buffer | |
3463 | * (all CPU entries) | |
3464 | */ | |
3465 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
3466 | { | |
3467 | struct ring_buffer_per_cpu *cpu_buffer; | |
3468 | unsigned long entries = 0; | |
3469 | int cpu; | |
3470 | ||
3471 | /* if you care about this being correct, lock the buffer */ | |
3472 | for_each_buffer_cpu(buffer, cpu) { | |
3473 | cpu_buffer = buffer->buffers[cpu]; | |
f6195aa0 | 3474 | entries += rb_num_of_entries(cpu_buffer); |
7a8e76a3 SR |
3475 | } |
3476 | ||
3477 | return entries; | |
3478 | } | |
c4f50183 | 3479 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
3480 | |
3481 | /** | |
67b394f7 | 3482 | * ring_buffer_overruns - get the number of overruns in buffer |
7a8e76a3 SR |
3483 | * @buffer: The ring buffer |
3484 | * | |
3485 | * Returns the total number of overruns in the ring buffer | |
3486 | * (all CPU entries) | |
3487 | */ | |
3488 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
3489 | { | |
3490 | struct ring_buffer_per_cpu *cpu_buffer; | |
3491 | unsigned long overruns = 0; | |
3492 | int cpu; | |
3493 | ||
3494 | /* if you care about this being correct, lock the buffer */ | |
3495 | for_each_buffer_cpu(buffer, cpu) { | |
3496 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3497 | overruns += local_read(&cpu_buffer->overrun); |
7a8e76a3 SR |
3498 | } |
3499 | ||
3500 | return overruns; | |
3501 | } | |
c4f50183 | 3502 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 3503 | |
642edba5 | 3504 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
3505 | { |
3506 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3507 | ||
d769041f | 3508 | /* Iterator usage is expected to have record disabled */ |
651e22f2 SRRH |
3509 | iter->head_page = cpu_buffer->reader_page; |
3510 | iter->head = cpu_buffer->reader_page->read; | |
3511 | ||
3512 | iter->cache_reader_page = iter->head_page; | |
24607f11 | 3513 | iter->cache_read = cpu_buffer->read; |
651e22f2 | 3514 | |
d769041f SR |
3515 | if (iter->head) |
3516 | iter->read_stamp = cpu_buffer->read_stamp; | |
3517 | else | |
abc9b56d | 3518 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 3519 | } |
f83c9d0f | 3520 | |
642edba5 SR |
3521 | /** |
3522 | * ring_buffer_iter_reset - reset an iterator | |
3523 | * @iter: The iterator to reset | |
3524 | * | |
3525 | * Resets the iterator, so that it will start from the beginning | |
3526 | * again. | |
3527 | */ | |
3528 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
3529 | { | |
554f786e | 3530 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
3531 | unsigned long flags; |
3532 | ||
554f786e SR |
3533 | if (!iter) |
3534 | return; | |
3535 | ||
3536 | cpu_buffer = iter->cpu_buffer; | |
3537 | ||
5389f6fa | 3538 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
642edba5 | 3539 | rb_iter_reset(iter); |
5389f6fa | 3540 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 3541 | } |
c4f50183 | 3542 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
3543 | |
3544 | /** | |
3545 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
3546 | * @iter: The iterator to check | |
3547 | */ | |
3548 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
3549 | { | |
3550 | struct ring_buffer_per_cpu *cpu_buffer; | |
78f7a45d SRV |
3551 | struct buffer_page *reader; |
3552 | struct buffer_page *head_page; | |
3553 | struct buffer_page *commit_page; | |
3554 | unsigned commit; | |
7a8e76a3 SR |
3555 | |
3556 | cpu_buffer = iter->cpu_buffer; | |
3557 | ||
78f7a45d SRV |
3558 | /* Remember, trace recording is off when iterator is in use */ |
3559 | reader = cpu_buffer->reader_page; | |
3560 | head_page = cpu_buffer->head_page; | |
3561 | commit_page = cpu_buffer->commit_page; | |
3562 | commit = rb_page_commit(commit_page); | |
3563 | ||
3564 | return ((iter->head_page == commit_page && iter->head == commit) || | |
3565 | (iter->head_page == reader && commit_page == head_page && | |
3566 | head_page->read == commit && | |
3567 | iter->head == rb_page_commit(cpu_buffer->reader_page))); | |
7a8e76a3 | 3568 | } |
c4f50183 | 3569 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
3570 | |
3571 | static void | |
3572 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
3573 | struct ring_buffer_event *event) | |
3574 | { | |
3575 | u64 delta; | |
3576 | ||
334d4169 | 3577 | switch (event->type_len) { |
7a8e76a3 SR |
3578 | case RINGBUF_TYPE_PADDING: |
3579 | return; | |
3580 | ||
3581 | case RINGBUF_TYPE_TIME_EXTEND: | |
dc4e2801 | 3582 | delta = ring_buffer_event_time_stamp(event); |
7a8e76a3 SR |
3583 | cpu_buffer->read_stamp += delta; |
3584 | return; | |
3585 | ||
3586 | case RINGBUF_TYPE_TIME_STAMP: | |
dc4e2801 TZ |
3587 | delta = ring_buffer_event_time_stamp(event); |
3588 | cpu_buffer->read_stamp = delta; | |
7a8e76a3 SR |
3589 | return; |
3590 | ||
3591 | case RINGBUF_TYPE_DATA: | |
3592 | cpu_buffer->read_stamp += event->time_delta; | |
3593 | return; | |
3594 | ||
3595 | default: | |
3596 | BUG(); | |
3597 | } | |
3598 | return; | |
3599 | } | |
3600 | ||
3601 | static void | |
3602 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
3603 | struct ring_buffer_event *event) | |
3604 | { | |
3605 | u64 delta; | |
3606 | ||
334d4169 | 3607 | switch (event->type_len) { |
7a8e76a3 SR |
3608 | case RINGBUF_TYPE_PADDING: |
3609 | return; | |
3610 | ||
3611 | case RINGBUF_TYPE_TIME_EXTEND: | |
dc4e2801 | 3612 | delta = ring_buffer_event_time_stamp(event); |
7a8e76a3 SR |
3613 | iter->read_stamp += delta; |
3614 | return; | |
3615 | ||
3616 | case RINGBUF_TYPE_TIME_STAMP: | |
dc4e2801 TZ |
3617 | delta = ring_buffer_event_time_stamp(event); |
3618 | iter->read_stamp = delta; | |
7a8e76a3 SR |
3619 | return; |
3620 | ||
3621 | case RINGBUF_TYPE_DATA: | |
3622 | iter->read_stamp += event->time_delta; | |
3623 | return; | |
3624 | ||
3625 | default: | |
3626 | BUG(); | |
3627 | } | |
3628 | return; | |
3629 | } | |
3630 | ||
d769041f SR |
3631 | static struct buffer_page * |
3632 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 3633 | { |
d769041f | 3634 | struct buffer_page *reader = NULL; |
66a8cb95 | 3635 | unsigned long overwrite; |
d769041f | 3636 | unsigned long flags; |
818e3dd3 | 3637 | int nr_loops = 0; |
77ae365e | 3638 | int ret; |
d769041f | 3639 | |
3e03fb7f | 3640 | local_irq_save(flags); |
0199c4e6 | 3641 | arch_spin_lock(&cpu_buffer->lock); |
d769041f SR |
3642 | |
3643 | again: | |
818e3dd3 SR |
3644 | /* |
3645 | * This should normally only loop twice. But because the | |
3646 | * start of the reader inserts an empty page, it causes | |
3647 | * a case where we will loop three times. There should be no | |
3648 | * reason to loop four times (that I know of). | |
3649 | */ | |
3e89c7bb | 3650 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
3651 | reader = NULL; |
3652 | goto out; | |
3653 | } | |
3654 | ||
d769041f SR |
3655 | reader = cpu_buffer->reader_page; |
3656 | ||
3657 | /* If there's more to read, return this page */ | |
bf41a158 | 3658 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
3659 | goto out; |
3660 | ||
3661 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
3662 | if (RB_WARN_ON(cpu_buffer, |
3663 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
3664 | goto out; | |
d769041f SR |
3665 | |
3666 | /* check if we caught up to the tail */ | |
3667 | reader = NULL; | |
bf41a158 | 3668 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 3669 | goto out; |
7a8e76a3 | 3670 | |
a5fb8331 SR |
3671 | /* Don't bother swapping if the ring buffer is empty */ |
3672 | if (rb_num_of_entries(cpu_buffer) == 0) | |
3673 | goto out; | |
3674 | ||
7a8e76a3 | 3675 | /* |
d769041f | 3676 | * Reset the reader page to size zero. |
7a8e76a3 | 3677 | */ |
77ae365e SR |
3678 | local_set(&cpu_buffer->reader_page->write, 0); |
3679 | local_set(&cpu_buffer->reader_page->entries, 0); | |
3680 | local_set(&cpu_buffer->reader_page->page->commit, 0); | |
ff0ff84a | 3681 | cpu_buffer->reader_page->real_end = 0; |
7a8e76a3 | 3682 | |
77ae365e SR |
3683 | spin: |
3684 | /* | |
3685 | * Splice the empty reader page into the list around the head. | |
3686 | */ | |
3687 | reader = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
3688 | if (!reader) |
3689 | goto out; | |
0e1ff5d7 | 3690 | cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next); |
d769041f | 3691 | cpu_buffer->reader_page->list.prev = reader->list.prev; |
bf41a158 | 3692 | |
3adc54fa SR |
3693 | /* |
3694 | * cpu_buffer->pages just needs to point to the buffer, it | |
3695 | * has no specific buffer page to point to. Lets move it out | |
25985edc | 3696 | * of our way so we don't accidentally swap it. |
3adc54fa SR |
3697 | */ |
3698 | cpu_buffer->pages = reader->list.prev; | |
3699 | ||
77ae365e SR |
3700 | /* The reader page will be pointing to the new head */ |
3701 | rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list); | |
7a8e76a3 | 3702 | |
66a8cb95 SR |
3703 | /* |
3704 | * We want to make sure we read the overruns after we set up our | |
3705 | * pointers to the next object. The writer side does a | |
3706 | * cmpxchg to cross pages which acts as the mb on the writer | |
3707 | * side. Note, the reader will constantly fail the swap | |
3708 | * while the writer is updating the pointers, so this | |
3709 | * guarantees that the overwrite recorded here is the one we | |
3710 | * want to compare with the last_overrun. | |
3711 | */ | |
3712 | smp_mb(); | |
3713 | overwrite = local_read(&(cpu_buffer->overrun)); | |
3714 | ||
77ae365e SR |
3715 | /* |
3716 | * Here's the tricky part. | |
3717 | * | |
3718 | * We need to move the pointer past the header page. | |
3719 | * But we can only do that if a writer is not currently | |
3720 | * moving it. The page before the header page has the | |
3721 | * flag bit '1' set if it is pointing to the page we want. | |
3722 | * but if the writer is in the process of moving it | |
3723 | * than it will be '2' or already moved '0'. | |
3724 | */ | |
3725 | ||
3726 | ret = rb_head_page_replace(reader, cpu_buffer->reader_page); | |
7a8e76a3 SR |
3727 | |
3728 | /* | |
77ae365e | 3729 | * If we did not convert it, then we must try again. |
7a8e76a3 | 3730 | */ |
77ae365e SR |
3731 | if (!ret) |
3732 | goto spin; | |
7a8e76a3 | 3733 | |
77ae365e SR |
3734 | /* |
3735 | * Yeah! We succeeded in replacing the page. | |
3736 | * | |
3737 | * Now make the new head point back to the reader page. | |
3738 | */ | |
5ded3dc6 | 3739 | rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list; |
77ae365e | 3740 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
d769041f SR |
3741 | |
3742 | /* Finally update the reader page to the new head */ | |
3743 | cpu_buffer->reader_page = reader; | |
b81f472a | 3744 | cpu_buffer->reader_page->read = 0; |
d769041f | 3745 | |
66a8cb95 SR |
3746 | if (overwrite != cpu_buffer->last_overrun) { |
3747 | cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun; | |
3748 | cpu_buffer->last_overrun = overwrite; | |
3749 | } | |
3750 | ||
d769041f SR |
3751 | goto again; |
3752 | ||
3753 | out: | |
b81f472a SRRH |
3754 | /* Update the read_stamp on the first event */ |
3755 | if (reader && reader->read == 0) | |
3756 | cpu_buffer->read_stamp = reader->page->time_stamp; | |
3757 | ||
0199c4e6 | 3758 | arch_spin_unlock(&cpu_buffer->lock); |
3e03fb7f | 3759 | local_irq_restore(flags); |
d769041f SR |
3760 | |
3761 | return reader; | |
3762 | } | |
3763 | ||
3764 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
3765 | { | |
3766 | struct ring_buffer_event *event; | |
3767 | struct buffer_page *reader; | |
3768 | unsigned length; | |
3769 | ||
3770 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 3771 | |
d769041f | 3772 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
3773 | if (RB_WARN_ON(cpu_buffer, !reader)) |
3774 | return; | |
7a8e76a3 | 3775 | |
d769041f SR |
3776 | event = rb_reader_event(cpu_buffer); |
3777 | ||
a1863c21 | 3778 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
e4906eff | 3779 | cpu_buffer->read++; |
d769041f SR |
3780 | |
3781 | rb_update_read_stamp(cpu_buffer, event); | |
3782 | ||
3783 | length = rb_event_length(event); | |
6f807acd | 3784 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
3785 | } |
3786 | ||
3787 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
3788 | { | |
7a8e76a3 SR |
3789 | struct ring_buffer_per_cpu *cpu_buffer; |
3790 | struct ring_buffer_event *event; | |
3791 | unsigned length; | |
3792 | ||
3793 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 SR |
3794 | |
3795 | /* | |
3796 | * Check if we are at the end of the buffer. | |
3797 | */ | |
bf41a158 | 3798 | if (iter->head >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
3799 | /* discarded commits can make the page empty */ |
3800 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 3801 | return; |
d769041f | 3802 | rb_inc_iter(iter); |
7a8e76a3 SR |
3803 | return; |
3804 | } | |
3805 | ||
3806 | event = rb_iter_head_event(iter); | |
3807 | ||
3808 | length = rb_event_length(event); | |
3809 | ||
3810 | /* | |
3811 | * This should not be called to advance the header if we are | |
3812 | * at the tail of the buffer. | |
3813 | */ | |
3e89c7bb | 3814 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 3815 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
3816 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
3817 | return; | |
7a8e76a3 SR |
3818 | |
3819 | rb_update_iter_read_stamp(iter, event); | |
3820 | ||
3821 | iter->head += length; | |
3822 | ||
3823 | /* check for end of page padding */ | |
bf41a158 SR |
3824 | if ((iter->head >= rb_page_size(iter->head_page)) && |
3825 | (iter->head_page != cpu_buffer->commit_page)) | |
771e0384 | 3826 | rb_inc_iter(iter); |
7a8e76a3 SR |
3827 | } |
3828 | ||
66a8cb95 SR |
3829 | static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer) |
3830 | { | |
3831 | return cpu_buffer->lost_events; | |
3832 | } | |
3833 | ||
f83c9d0f | 3834 | static struct ring_buffer_event * |
66a8cb95 SR |
3835 | rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts, |
3836 | unsigned long *lost_events) | |
7a8e76a3 | 3837 | { |
7a8e76a3 | 3838 | struct ring_buffer_event *event; |
d769041f | 3839 | struct buffer_page *reader; |
818e3dd3 | 3840 | int nr_loops = 0; |
7a8e76a3 | 3841 | |
dc4e2801 TZ |
3842 | if (ts) |
3843 | *ts = 0; | |
7a8e76a3 | 3844 | again: |
818e3dd3 | 3845 | /* |
69d1b839 SR |
3846 | * We repeat when a time extend is encountered. |
3847 | * Since the time extend is always attached to a data event, | |
3848 | * we should never loop more than once. | |
3849 | * (We never hit the following condition more than twice). | |
818e3dd3 | 3850 | */ |
69d1b839 | 3851 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 3852 | return NULL; |
818e3dd3 | 3853 | |
d769041f SR |
3854 | reader = rb_get_reader_page(cpu_buffer); |
3855 | if (!reader) | |
7a8e76a3 SR |
3856 | return NULL; |
3857 | ||
d769041f | 3858 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 3859 | |
334d4169 | 3860 | switch (event->type_len) { |
7a8e76a3 | 3861 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3862 | if (rb_null_event(event)) |
3863 | RB_WARN_ON(cpu_buffer, 1); | |
3864 | /* | |
3865 | * Because the writer could be discarding every | |
3866 | * event it creates (which would probably be bad) | |
3867 | * if we were to go back to "again" then we may never | |
3868 | * catch up, and will trigger the warn on, or lock | |
3869 | * the box. Return the padding, and we will release | |
3870 | * the current locks, and try again. | |
3871 | */ | |
2d622719 | 3872 | return event; |
7a8e76a3 SR |
3873 | |
3874 | case RINGBUF_TYPE_TIME_EXTEND: | |
3875 | /* Internal data, OK to advance */ | |
d769041f | 3876 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3877 | goto again; |
3878 | ||
3879 | case RINGBUF_TYPE_TIME_STAMP: | |
dc4e2801 TZ |
3880 | if (ts) { |
3881 | *ts = ring_buffer_event_time_stamp(event); | |
3882 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, | |
3883 | cpu_buffer->cpu, ts); | |
3884 | } | |
3885 | /* Internal data, OK to advance */ | |
d769041f | 3886 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3887 | goto again; |
3888 | ||
3889 | case RINGBUF_TYPE_DATA: | |
dc4e2801 | 3890 | if (ts && !(*ts)) { |
7a8e76a3 | 3891 | *ts = cpu_buffer->read_stamp + event->time_delta; |
d8eeb2d3 | 3892 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, |
37886f6a | 3893 | cpu_buffer->cpu, ts); |
7a8e76a3 | 3894 | } |
66a8cb95 SR |
3895 | if (lost_events) |
3896 | *lost_events = rb_lost_events(cpu_buffer); | |
7a8e76a3 SR |
3897 | return event; |
3898 | ||
3899 | default: | |
3900 | BUG(); | |
3901 | } | |
3902 | ||
3903 | return NULL; | |
3904 | } | |
c4f50183 | 3905 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 3906 | |
f83c9d0f SR |
3907 | static struct ring_buffer_event * |
3908 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
3909 | { |
3910 | struct ring_buffer *buffer; | |
3911 | struct ring_buffer_per_cpu *cpu_buffer; | |
3912 | struct ring_buffer_event *event; | |
818e3dd3 | 3913 | int nr_loops = 0; |
7a8e76a3 | 3914 | |
dc4e2801 TZ |
3915 | if (ts) |
3916 | *ts = 0; | |
3917 | ||
7a8e76a3 SR |
3918 | cpu_buffer = iter->cpu_buffer; |
3919 | buffer = cpu_buffer->buffer; | |
3920 | ||
492a74f4 SR |
3921 | /* |
3922 | * Check if someone performed a consuming read to | |
3923 | * the buffer. A consuming read invalidates the iterator | |
3924 | * and we need to reset the iterator in this case. | |
3925 | */ | |
3926 | if (unlikely(iter->cache_read != cpu_buffer->read || | |
3927 | iter->cache_reader_page != cpu_buffer->reader_page)) | |
3928 | rb_iter_reset(iter); | |
3929 | ||
7a8e76a3 | 3930 | again: |
3c05d748 SR |
3931 | if (ring_buffer_iter_empty(iter)) |
3932 | return NULL; | |
3933 | ||
818e3dd3 | 3934 | /* |
021de3d9 SRRH |
3935 | * We repeat when a time extend is encountered or we hit |
3936 | * the end of the page. Since the time extend is always attached | |
3937 | * to a data event, we should never loop more than three times. | |
3938 | * Once for going to next page, once on time extend, and | |
3939 | * finally once to get the event. | |
3940 | * (We never hit the following condition more than thrice). | |
818e3dd3 | 3941 | */ |
021de3d9 | 3942 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) |
818e3dd3 | 3943 | return NULL; |
818e3dd3 | 3944 | |
7a8e76a3 SR |
3945 | if (rb_per_cpu_empty(cpu_buffer)) |
3946 | return NULL; | |
3947 | ||
10e83fd0 | 3948 | if (iter->head >= rb_page_size(iter->head_page)) { |
3c05d748 SR |
3949 | rb_inc_iter(iter); |
3950 | goto again; | |
3951 | } | |
3952 | ||
7a8e76a3 SR |
3953 | event = rb_iter_head_event(iter); |
3954 | ||
334d4169 | 3955 | switch (event->type_len) { |
7a8e76a3 | 3956 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3957 | if (rb_null_event(event)) { |
3958 | rb_inc_iter(iter); | |
3959 | goto again; | |
3960 | } | |
3961 | rb_advance_iter(iter); | |
3962 | return event; | |
7a8e76a3 SR |
3963 | |
3964 | case RINGBUF_TYPE_TIME_EXTEND: | |
3965 | /* Internal data, OK to advance */ | |
3966 | rb_advance_iter(iter); | |
3967 | goto again; | |
3968 | ||
3969 | case RINGBUF_TYPE_TIME_STAMP: | |
dc4e2801 TZ |
3970 | if (ts) { |
3971 | *ts = ring_buffer_event_time_stamp(event); | |
3972 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, | |
3973 | cpu_buffer->cpu, ts); | |
3974 | } | |
3975 | /* Internal data, OK to advance */ | |
7a8e76a3 SR |
3976 | rb_advance_iter(iter); |
3977 | goto again; | |
3978 | ||
3979 | case RINGBUF_TYPE_DATA: | |
dc4e2801 | 3980 | if (ts && !(*ts)) { |
7a8e76a3 | 3981 | *ts = iter->read_stamp + event->time_delta; |
37886f6a SR |
3982 | ring_buffer_normalize_time_stamp(buffer, |
3983 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
3984 | } |
3985 | return event; | |
3986 | ||
3987 | default: | |
3988 | BUG(); | |
3989 | } | |
3990 | ||
3991 | return NULL; | |
3992 | } | |
c4f50183 | 3993 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 3994 | |
289a5a25 | 3995 | static inline bool rb_reader_lock(struct ring_buffer_per_cpu *cpu_buffer) |
8d707e8e | 3996 | { |
289a5a25 SRRH |
3997 | if (likely(!in_nmi())) { |
3998 | raw_spin_lock(&cpu_buffer->reader_lock); | |
3999 | return true; | |
4000 | } | |
4001 | ||
8d707e8e SR |
4002 | /* |
4003 | * If an NMI die dumps out the content of the ring buffer | |
289a5a25 SRRH |
4004 | * trylock must be used to prevent a deadlock if the NMI |
4005 | * preempted a task that holds the ring buffer locks. If | |
4006 | * we get the lock then all is fine, if not, then continue | |
4007 | * to do the read, but this can corrupt the ring buffer, | |
4008 | * so it must be permanently disabled from future writes. | |
4009 | * Reading from NMI is a oneshot deal. | |
8d707e8e | 4010 | */ |
289a5a25 SRRH |
4011 | if (raw_spin_trylock(&cpu_buffer->reader_lock)) |
4012 | return true; | |
8d707e8e | 4013 | |
289a5a25 SRRH |
4014 | /* Continue without locking, but disable the ring buffer */ |
4015 | atomic_inc(&cpu_buffer->record_disabled); | |
4016 | return false; | |
4017 | } | |
4018 | ||
4019 | static inline void | |
4020 | rb_reader_unlock(struct ring_buffer_per_cpu *cpu_buffer, bool locked) | |
4021 | { | |
4022 | if (likely(locked)) | |
4023 | raw_spin_unlock(&cpu_buffer->reader_lock); | |
4024 | return; | |
8d707e8e SR |
4025 | } |
4026 | ||
f83c9d0f SR |
4027 | /** |
4028 | * ring_buffer_peek - peek at the next event to be read | |
4029 | * @buffer: The ring buffer to read | |
4030 | * @cpu: The cpu to peak at | |
4031 | * @ts: The timestamp counter of this event. | |
66a8cb95 | 4032 | * @lost_events: a variable to store if events were lost (may be NULL) |
f83c9d0f SR |
4033 | * |
4034 | * This will return the event that will be read next, but does | |
4035 | * not consume the data. | |
4036 | */ | |
4037 | struct ring_buffer_event * | |
66a8cb95 SR |
4038 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts, |
4039 | unsigned long *lost_events) | |
f83c9d0f SR |
4040 | { |
4041 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 4042 | struct ring_buffer_event *event; |
f83c9d0f | 4043 | unsigned long flags; |
289a5a25 | 4044 | bool dolock; |
f83c9d0f | 4045 | |
554f786e | 4046 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4047 | return NULL; |
554f786e | 4048 | |
2d622719 | 4049 | again: |
8d707e8e | 4050 | local_irq_save(flags); |
289a5a25 | 4051 | dolock = rb_reader_lock(cpu_buffer); |
66a8cb95 | 4052 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
469535a5 RR |
4053 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
4054 | rb_advance_reader(cpu_buffer); | |
289a5a25 | 4055 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e | 4056 | local_irq_restore(flags); |
f83c9d0f | 4057 | |
1b959e18 | 4058 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 4059 | goto again; |
2d622719 | 4060 | |
f83c9d0f SR |
4061 | return event; |
4062 | } | |
4063 | ||
4064 | /** | |
4065 | * ring_buffer_iter_peek - peek at the next event to be read | |
4066 | * @iter: The ring buffer iterator | |
4067 | * @ts: The timestamp counter of this event. | |
4068 | * | |
4069 | * This will return the event that will be read next, but does | |
4070 | * not increment the iterator. | |
4071 | */ | |
4072 | struct ring_buffer_event * | |
4073 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
4074 | { | |
4075 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
4076 | struct ring_buffer_event *event; | |
4077 | unsigned long flags; | |
4078 | ||
2d622719 | 4079 | again: |
5389f6fa | 4080 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 4081 | event = rb_iter_peek(iter, ts); |
5389f6fa | 4082 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 4083 | |
1b959e18 | 4084 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 4085 | goto again; |
2d622719 | 4086 | |
f83c9d0f SR |
4087 | return event; |
4088 | } | |
4089 | ||
7a8e76a3 SR |
4090 | /** |
4091 | * ring_buffer_consume - return an event and consume it | |
4092 | * @buffer: The ring buffer to get the next event from | |
66a8cb95 SR |
4093 | * @cpu: the cpu to read the buffer from |
4094 | * @ts: a variable to store the timestamp (may be NULL) | |
4095 | * @lost_events: a variable to store if events were lost (may be NULL) | |
7a8e76a3 SR |
4096 | * |
4097 | * Returns the next event in the ring buffer, and that event is consumed. | |
4098 | * Meaning, that sequential reads will keep returning a different event, | |
4099 | * and eventually empty the ring buffer if the producer is slower. | |
4100 | */ | |
4101 | struct ring_buffer_event * | |
66a8cb95 SR |
4102 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts, |
4103 | unsigned long *lost_events) | |
7a8e76a3 | 4104 | { |
554f786e SR |
4105 | struct ring_buffer_per_cpu *cpu_buffer; |
4106 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 4107 | unsigned long flags; |
289a5a25 | 4108 | bool dolock; |
7a8e76a3 | 4109 | |
2d622719 | 4110 | again: |
554f786e SR |
4111 | /* might be called in atomic */ |
4112 | preempt_disable(); | |
4113 | ||
9e01c1b7 | 4114 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 4115 | goto out; |
7a8e76a3 | 4116 | |
554f786e | 4117 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e | 4118 | local_irq_save(flags); |
289a5a25 | 4119 | dolock = rb_reader_lock(cpu_buffer); |
f83c9d0f | 4120 | |
66a8cb95 SR |
4121 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
4122 | if (event) { | |
4123 | cpu_buffer->lost_events = 0; | |
469535a5 | 4124 | rb_advance_reader(cpu_buffer); |
66a8cb95 | 4125 | } |
7a8e76a3 | 4126 | |
289a5a25 | 4127 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e | 4128 | local_irq_restore(flags); |
f83c9d0f | 4129 | |
554f786e SR |
4130 | out: |
4131 | preempt_enable(); | |
4132 | ||
1b959e18 | 4133 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 4134 | goto again; |
2d622719 | 4135 | |
7a8e76a3 SR |
4136 | return event; |
4137 | } | |
c4f50183 | 4138 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
4139 | |
4140 | /** | |
72c9ddfd | 4141 | * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer |
7a8e76a3 SR |
4142 | * @buffer: The ring buffer to read from |
4143 | * @cpu: The cpu buffer to iterate over | |
4144 | * | |
72c9ddfd DM |
4145 | * This performs the initial preparations necessary to iterate |
4146 | * through the buffer. Memory is allocated, buffer recording | |
4147 | * is disabled, and the iterator pointer is returned to the caller. | |
7a8e76a3 | 4148 | * |
6167c205 | 4149 | * Disabling buffer recording prevents the reading from being |
72c9ddfd DM |
4150 | * corrupted. This is not a consuming read, so a producer is not |
4151 | * expected. | |
4152 | * | |
4153 | * After a sequence of ring_buffer_read_prepare calls, the user is | |
d611851b | 4154 | * expected to make at least one call to ring_buffer_read_prepare_sync. |
72c9ddfd DM |
4155 | * Afterwards, ring_buffer_read_start is invoked to get things going |
4156 | * for real. | |
4157 | * | |
d611851b | 4158 | * This overall must be paired with ring_buffer_read_finish. |
7a8e76a3 SR |
4159 | */ |
4160 | struct ring_buffer_iter * | |
72c9ddfd | 4161 | ring_buffer_read_prepare(struct ring_buffer *buffer, int cpu) |
7a8e76a3 SR |
4162 | { |
4163 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 4164 | struct ring_buffer_iter *iter; |
7a8e76a3 | 4165 | |
9e01c1b7 | 4166 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4167 | return NULL; |
7a8e76a3 SR |
4168 | |
4169 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
4170 | if (!iter) | |
8aabee57 | 4171 | return NULL; |
7a8e76a3 SR |
4172 | |
4173 | cpu_buffer = buffer->buffers[cpu]; | |
4174 | ||
4175 | iter->cpu_buffer = cpu_buffer; | |
4176 | ||
83f40318 | 4177 | atomic_inc(&buffer->resize_disabled); |
7a8e76a3 | 4178 | atomic_inc(&cpu_buffer->record_disabled); |
72c9ddfd DM |
4179 | |
4180 | return iter; | |
4181 | } | |
4182 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare); | |
4183 | ||
4184 | /** | |
4185 | * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls | |
4186 | * | |
4187 | * All previously invoked ring_buffer_read_prepare calls to prepare | |
4188 | * iterators will be synchronized. Afterwards, read_buffer_read_start | |
4189 | * calls on those iterators are allowed. | |
4190 | */ | |
4191 | void | |
4192 | ring_buffer_read_prepare_sync(void) | |
4193 | { | |
7a8e76a3 | 4194 | synchronize_sched(); |
72c9ddfd DM |
4195 | } |
4196 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync); | |
4197 | ||
4198 | /** | |
4199 | * ring_buffer_read_start - start a non consuming read of the buffer | |
4200 | * @iter: The iterator returned by ring_buffer_read_prepare | |
4201 | * | |
4202 | * This finalizes the startup of an iteration through the buffer. | |
4203 | * The iterator comes from a call to ring_buffer_read_prepare and | |
4204 | * an intervening ring_buffer_read_prepare_sync must have been | |
4205 | * performed. | |
4206 | * | |
d611851b | 4207 | * Must be paired with ring_buffer_read_finish. |
72c9ddfd DM |
4208 | */ |
4209 | void | |
4210 | ring_buffer_read_start(struct ring_buffer_iter *iter) | |
4211 | { | |
4212 | struct ring_buffer_per_cpu *cpu_buffer; | |
4213 | unsigned long flags; | |
4214 | ||
4215 | if (!iter) | |
4216 | return; | |
4217 | ||
4218 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 | 4219 | |
5389f6fa | 4220 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
0199c4e6 | 4221 | arch_spin_lock(&cpu_buffer->lock); |
642edba5 | 4222 | rb_iter_reset(iter); |
0199c4e6 | 4223 | arch_spin_unlock(&cpu_buffer->lock); |
5389f6fa | 4224 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 4225 | } |
c4f50183 | 4226 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
4227 | |
4228 | /** | |
d611851b | 4229 | * ring_buffer_read_finish - finish reading the iterator of the buffer |
7a8e76a3 SR |
4230 | * @iter: The iterator retrieved by ring_buffer_start |
4231 | * | |
4232 | * This re-enables the recording to the buffer, and frees the | |
4233 | * iterator. | |
4234 | */ | |
4235 | void | |
4236 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
4237 | { | |
4238 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
9366c1ba | 4239 | unsigned long flags; |
7a8e76a3 | 4240 | |
659f451f SR |
4241 | /* |
4242 | * Ring buffer is disabled from recording, here's a good place | |
9366c1ba SR |
4243 | * to check the integrity of the ring buffer. |
4244 | * Must prevent readers from trying to read, as the check | |
4245 | * clears the HEAD page and readers require it. | |
659f451f | 4246 | */ |
9366c1ba | 4247 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
659f451f | 4248 | rb_check_pages(cpu_buffer); |
9366c1ba | 4249 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
659f451f | 4250 | |
7a8e76a3 | 4251 | atomic_dec(&cpu_buffer->record_disabled); |
83f40318 | 4252 | atomic_dec(&cpu_buffer->buffer->resize_disabled); |
7a8e76a3 SR |
4253 | kfree(iter); |
4254 | } | |
c4f50183 | 4255 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
4256 | |
4257 | /** | |
4258 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
4259 | * @iter: The ring buffer iterator | |
4260 | * @ts: The time stamp of the event read. | |
4261 | * | |
4262 | * This reads the next event in the ring buffer and increments the iterator. | |
4263 | */ | |
4264 | struct ring_buffer_event * | |
4265 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
4266 | { | |
4267 | struct ring_buffer_event *event; | |
f83c9d0f SR |
4268 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
4269 | unsigned long flags; | |
7a8e76a3 | 4270 | |
5389f6fa | 4271 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
7e9391cf | 4272 | again: |
f83c9d0f | 4273 | event = rb_iter_peek(iter, ts); |
7a8e76a3 | 4274 | if (!event) |
f83c9d0f | 4275 | goto out; |
7a8e76a3 | 4276 | |
7e9391cf SR |
4277 | if (event->type_len == RINGBUF_TYPE_PADDING) |
4278 | goto again; | |
4279 | ||
7a8e76a3 | 4280 | rb_advance_iter(iter); |
f83c9d0f | 4281 | out: |
5389f6fa | 4282 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
4283 | |
4284 | return event; | |
4285 | } | |
c4f50183 | 4286 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
4287 | |
4288 | /** | |
4289 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
4290 | * @buffer: The ring buffer. | |
4291 | */ | |
438ced17 | 4292 | unsigned long ring_buffer_size(struct ring_buffer *buffer, int cpu) |
7a8e76a3 | 4293 | { |
438ced17 VN |
4294 | /* |
4295 | * Earlier, this method returned | |
4296 | * BUF_PAGE_SIZE * buffer->nr_pages | |
4297 | * Since the nr_pages field is now removed, we have converted this to | |
4298 | * return the per cpu buffer value. | |
4299 | */ | |
4300 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
4301 | return 0; | |
4302 | ||
4303 | return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages; | |
7a8e76a3 | 4304 | } |
c4f50183 | 4305 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
4306 | |
4307 | static void | |
4308 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
4309 | { | |
77ae365e SR |
4310 | rb_head_page_deactivate(cpu_buffer); |
4311 | ||
7a8e76a3 | 4312 | cpu_buffer->head_page |
3adc54fa | 4313 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 4314 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 4315 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 4316 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 4317 | |
6f807acd | 4318 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
4319 | |
4320 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
4321 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
4322 | ||
4323 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
5040b4b7 | 4324 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
bf41a158 | 4325 | local_set(&cpu_buffer->reader_page->write, 0); |
778c55d4 | 4326 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 4327 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 4328 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 4329 | |
c64e148a | 4330 | local_set(&cpu_buffer->entries_bytes, 0); |
77ae365e | 4331 | local_set(&cpu_buffer->overrun, 0); |
884bfe89 SP |
4332 | local_set(&cpu_buffer->commit_overrun, 0); |
4333 | local_set(&cpu_buffer->dropped_events, 0); | |
e4906eff | 4334 | local_set(&cpu_buffer->entries, 0); |
fa743953 SR |
4335 | local_set(&cpu_buffer->committing, 0); |
4336 | local_set(&cpu_buffer->commits, 0); | |
77ae365e | 4337 | cpu_buffer->read = 0; |
c64e148a | 4338 | cpu_buffer->read_bytes = 0; |
69507c06 SR |
4339 | |
4340 | cpu_buffer->write_stamp = 0; | |
4341 | cpu_buffer->read_stamp = 0; | |
77ae365e | 4342 | |
66a8cb95 SR |
4343 | cpu_buffer->lost_events = 0; |
4344 | cpu_buffer->last_overrun = 0; | |
4345 | ||
77ae365e | 4346 | rb_head_page_activate(cpu_buffer); |
7a8e76a3 SR |
4347 | } |
4348 | ||
4349 | /** | |
4350 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
4351 | * @buffer: The ring buffer to reset a per cpu buffer of | |
4352 | * @cpu: The CPU buffer to be reset | |
4353 | */ | |
4354 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
4355 | { | |
4356 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
4357 | unsigned long flags; | |
4358 | ||
9e01c1b7 | 4359 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4360 | return; |
7a8e76a3 | 4361 | |
83f40318 | 4362 | atomic_inc(&buffer->resize_disabled); |
41ede23e SR |
4363 | atomic_inc(&cpu_buffer->record_disabled); |
4364 | ||
83f40318 VN |
4365 | /* Make sure all commits have finished */ |
4366 | synchronize_sched(); | |
4367 | ||
5389f6fa | 4368 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 4369 | |
41b6a95d SR |
4370 | if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing))) |
4371 | goto out; | |
4372 | ||
0199c4e6 | 4373 | arch_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
4374 | |
4375 | rb_reset_cpu(cpu_buffer); | |
4376 | ||
0199c4e6 | 4377 | arch_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 4378 | |
41b6a95d | 4379 | out: |
5389f6fa | 4380 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
41ede23e SR |
4381 | |
4382 | atomic_dec(&cpu_buffer->record_disabled); | |
83f40318 | 4383 | atomic_dec(&buffer->resize_disabled); |
7a8e76a3 | 4384 | } |
c4f50183 | 4385 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
4386 | |
4387 | /** | |
4388 | * ring_buffer_reset - reset a ring buffer | |
4389 | * @buffer: The ring buffer to reset all cpu buffers | |
4390 | */ | |
4391 | void ring_buffer_reset(struct ring_buffer *buffer) | |
4392 | { | |
7a8e76a3 SR |
4393 | int cpu; |
4394 | ||
7a8e76a3 | 4395 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 4396 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 4397 | } |
c4f50183 | 4398 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
4399 | |
4400 | /** | |
4401 | * rind_buffer_empty - is the ring buffer empty? | |
4402 | * @buffer: The ring buffer to test | |
4403 | */ | |
3d4e204d | 4404 | bool ring_buffer_empty(struct ring_buffer *buffer) |
7a8e76a3 SR |
4405 | { |
4406 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 4407 | unsigned long flags; |
289a5a25 | 4408 | bool dolock; |
7a8e76a3 | 4409 | int cpu; |
d4788207 | 4410 | int ret; |
7a8e76a3 SR |
4411 | |
4412 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
4413 | for_each_buffer_cpu(buffer, cpu) { | |
4414 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e | 4415 | local_irq_save(flags); |
289a5a25 | 4416 | dolock = rb_reader_lock(cpu_buffer); |
d4788207 | 4417 | ret = rb_per_cpu_empty(cpu_buffer); |
289a5a25 | 4418 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e SR |
4419 | local_irq_restore(flags); |
4420 | ||
d4788207 | 4421 | if (!ret) |
3d4e204d | 4422 | return false; |
7a8e76a3 | 4423 | } |
554f786e | 4424 | |
3d4e204d | 4425 | return true; |
7a8e76a3 | 4426 | } |
c4f50183 | 4427 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
4428 | |
4429 | /** | |
4430 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
4431 | * @buffer: The ring buffer | |
4432 | * @cpu: The CPU buffer to test | |
4433 | */ | |
3d4e204d | 4434 | bool ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) |
7a8e76a3 SR |
4435 | { |
4436 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 4437 | unsigned long flags; |
289a5a25 | 4438 | bool dolock; |
8aabee57 | 4439 | int ret; |
7a8e76a3 | 4440 | |
9e01c1b7 | 4441 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
3d4e204d | 4442 | return true; |
7a8e76a3 SR |
4443 | |
4444 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e | 4445 | local_irq_save(flags); |
289a5a25 | 4446 | dolock = rb_reader_lock(cpu_buffer); |
554f786e | 4447 | ret = rb_per_cpu_empty(cpu_buffer); |
289a5a25 | 4448 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e | 4449 | local_irq_restore(flags); |
554f786e SR |
4450 | |
4451 | return ret; | |
7a8e76a3 | 4452 | } |
c4f50183 | 4453 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 | 4454 | |
85bac32c | 4455 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
7a8e76a3 SR |
4456 | /** |
4457 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
4458 | * @buffer_a: One buffer to swap with | |
4459 | * @buffer_b: The other buffer to swap with | |
4460 | * | |
4461 | * This function is useful for tracers that want to take a "snapshot" | |
4462 | * of a CPU buffer and has another back up buffer lying around. | |
4463 | * it is expected that the tracer handles the cpu buffer not being | |
4464 | * used at the moment. | |
4465 | */ | |
4466 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
4467 | struct ring_buffer *buffer_b, int cpu) | |
4468 | { | |
4469 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
4470 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
4471 | int ret = -EINVAL; |
4472 | ||
9e01c1b7 RR |
4473 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
4474 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 4475 | goto out; |
7a8e76a3 | 4476 | |
438ced17 VN |
4477 | cpu_buffer_a = buffer_a->buffers[cpu]; |
4478 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
4479 | ||
7a8e76a3 | 4480 | /* At least make sure the two buffers are somewhat the same */ |
438ced17 | 4481 | if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages) |
554f786e SR |
4482 | goto out; |
4483 | ||
4484 | ret = -EAGAIN; | |
7a8e76a3 | 4485 | |
97b17efe | 4486 | if (atomic_read(&buffer_a->record_disabled)) |
554f786e | 4487 | goto out; |
97b17efe SR |
4488 | |
4489 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 4490 | goto out; |
97b17efe | 4491 | |
97b17efe | 4492 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 4493 | goto out; |
97b17efe SR |
4494 | |
4495 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 4496 | goto out; |
97b17efe | 4497 | |
7a8e76a3 SR |
4498 | /* |
4499 | * We can't do a synchronize_sched here because this | |
4500 | * function can be called in atomic context. | |
4501 | * Normally this will be called from the same CPU as cpu. | |
4502 | * If not it's up to the caller to protect this. | |
4503 | */ | |
4504 | atomic_inc(&cpu_buffer_a->record_disabled); | |
4505 | atomic_inc(&cpu_buffer_b->record_disabled); | |
4506 | ||
98277991 SR |
4507 | ret = -EBUSY; |
4508 | if (local_read(&cpu_buffer_a->committing)) | |
4509 | goto out_dec; | |
4510 | if (local_read(&cpu_buffer_b->committing)) | |
4511 | goto out_dec; | |
4512 | ||
7a8e76a3 SR |
4513 | buffer_a->buffers[cpu] = cpu_buffer_b; |
4514 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
4515 | ||
4516 | cpu_buffer_b->buffer = buffer_a; | |
4517 | cpu_buffer_a->buffer = buffer_b; | |
4518 | ||
98277991 SR |
4519 | ret = 0; |
4520 | ||
4521 | out_dec: | |
7a8e76a3 SR |
4522 | atomic_dec(&cpu_buffer_a->record_disabled); |
4523 | atomic_dec(&cpu_buffer_b->record_disabled); | |
554f786e | 4524 | out: |
554f786e | 4525 | return ret; |
7a8e76a3 | 4526 | } |
c4f50183 | 4527 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
85bac32c | 4528 | #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */ |
7a8e76a3 | 4529 | |
8789a9e7 SR |
4530 | /** |
4531 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
4532 | * @buffer: the buffer to allocate for. | |
d611851b | 4533 | * @cpu: the cpu buffer to allocate. |
8789a9e7 SR |
4534 | * |
4535 | * This function is used in conjunction with ring_buffer_read_page. | |
4536 | * When reading a full page from the ring buffer, these functions | |
4537 | * can be used to speed up the process. The calling function should | |
4538 | * allocate a few pages first with this function. Then when it | |
4539 | * needs to get pages from the ring buffer, it passes the result | |
4540 | * of this function into ring_buffer_read_page, which will swap | |
4541 | * the page that was allocated, with the read page of the buffer. | |
4542 | * | |
4543 | * Returns: | |
a7e52ad7 | 4544 | * The page allocated, or ERR_PTR |
8789a9e7 | 4545 | */ |
7ea59064 | 4546 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer, int cpu) |
8789a9e7 | 4547 | { |
a7e52ad7 | 4548 | struct ring_buffer_per_cpu *cpu_buffer; |
73a757e6 SRV |
4549 | struct buffer_data_page *bpage = NULL; |
4550 | unsigned long flags; | |
7ea59064 | 4551 | struct page *page; |
8789a9e7 | 4552 | |
a7e52ad7 SRV |
4553 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
4554 | return ERR_PTR(-ENODEV); | |
4555 | ||
4556 | cpu_buffer = buffer->buffers[cpu]; | |
73a757e6 SRV |
4557 | local_irq_save(flags); |
4558 | arch_spin_lock(&cpu_buffer->lock); | |
4559 | ||
4560 | if (cpu_buffer->free_page) { | |
4561 | bpage = cpu_buffer->free_page; | |
4562 | cpu_buffer->free_page = NULL; | |
4563 | } | |
4564 | ||
4565 | arch_spin_unlock(&cpu_buffer->lock); | |
4566 | local_irq_restore(flags); | |
4567 | ||
4568 | if (bpage) | |
4569 | goto out; | |
4570 | ||
d7ec4bfe VN |
4571 | page = alloc_pages_node(cpu_to_node(cpu), |
4572 | GFP_KERNEL | __GFP_NORETRY, 0); | |
7ea59064 | 4573 | if (!page) |
a7e52ad7 | 4574 | return ERR_PTR(-ENOMEM); |
8789a9e7 | 4575 | |
7ea59064 | 4576 | bpage = page_address(page); |
8789a9e7 | 4577 | |
73a757e6 | 4578 | out: |
ef7a4a16 SR |
4579 | rb_init_page(bpage); |
4580 | ||
044fa782 | 4581 | return bpage; |
8789a9e7 | 4582 | } |
d6ce96da | 4583 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
4584 | |
4585 | /** | |
4586 | * ring_buffer_free_read_page - free an allocated read page | |
4587 | * @buffer: the buffer the page was allocate for | |
73a757e6 | 4588 | * @cpu: the cpu buffer the page came from |
8789a9e7 SR |
4589 | * @data: the page to free |
4590 | * | |
4591 | * Free a page allocated from ring_buffer_alloc_read_page. | |
4592 | */ | |
73a757e6 | 4593 | void ring_buffer_free_read_page(struct ring_buffer *buffer, int cpu, void *data) |
8789a9e7 | 4594 | { |
73a757e6 SRV |
4595 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; |
4596 | struct buffer_data_page *bpage = data; | |
ae415fa4 | 4597 | struct page *page = virt_to_page(bpage); |
73a757e6 SRV |
4598 | unsigned long flags; |
4599 | ||
ae415fa4 SRV |
4600 | /* If the page is still in use someplace else, we can't reuse it */ |
4601 | if (page_ref_count(page) > 1) | |
4602 | goto out; | |
4603 | ||
73a757e6 SRV |
4604 | local_irq_save(flags); |
4605 | arch_spin_lock(&cpu_buffer->lock); | |
4606 | ||
4607 | if (!cpu_buffer->free_page) { | |
4608 | cpu_buffer->free_page = bpage; | |
4609 | bpage = NULL; | |
4610 | } | |
4611 | ||
4612 | arch_spin_unlock(&cpu_buffer->lock); | |
4613 | local_irq_restore(flags); | |
4614 | ||
ae415fa4 | 4615 | out: |
73a757e6 | 4616 | free_page((unsigned long)bpage); |
8789a9e7 | 4617 | } |
d6ce96da | 4618 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
4619 | |
4620 | /** | |
4621 | * ring_buffer_read_page - extract a page from the ring buffer | |
4622 | * @buffer: buffer to extract from | |
4623 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 4624 | * @len: amount to extract |
8789a9e7 SR |
4625 | * @cpu: the cpu of the buffer to extract |
4626 | * @full: should the extraction only happen when the page is full. | |
4627 | * | |
4628 | * This function will pull out a page from the ring buffer and consume it. | |
4629 | * @data_page must be the address of the variable that was returned | |
4630 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
4631 | * to swap with a page in the ring buffer. | |
4632 | * | |
4633 | * for example: | |
d611851b | 4634 | * rpage = ring_buffer_alloc_read_page(buffer, cpu); |
a7e52ad7 SRV |
4635 | * if (IS_ERR(rpage)) |
4636 | * return PTR_ERR(rpage); | |
ef7a4a16 | 4637 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
4638 | * if (ret >= 0) |
4639 | * process_page(rpage, ret); | |
8789a9e7 SR |
4640 | * |
4641 | * When @full is set, the function will not return true unless | |
4642 | * the writer is off the reader page. | |
4643 | * | |
4644 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
4645 | * The ring buffer can be used anywhere in the kernel and can not | |
4646 | * blindly call wake_up. The layer that uses the ring buffer must be | |
4647 | * responsible for that. | |
4648 | * | |
4649 | * Returns: | |
667d2412 LJ |
4650 | * >=0 if data has been transferred, returns the offset of consumed data. |
4651 | * <0 if no data has been transferred. | |
8789a9e7 SR |
4652 | */ |
4653 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 4654 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
4655 | { |
4656 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
4657 | struct ring_buffer_event *event; | |
044fa782 | 4658 | struct buffer_data_page *bpage; |
ef7a4a16 | 4659 | struct buffer_page *reader; |
ff0ff84a | 4660 | unsigned long missed_events; |
8789a9e7 | 4661 | unsigned long flags; |
ef7a4a16 | 4662 | unsigned int commit; |
667d2412 | 4663 | unsigned int read; |
4f3640f8 | 4664 | u64 save_timestamp; |
667d2412 | 4665 | int ret = -1; |
8789a9e7 | 4666 | |
554f786e SR |
4667 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
4668 | goto out; | |
4669 | ||
474d32b6 SR |
4670 | /* |
4671 | * If len is not big enough to hold the page header, then | |
4672 | * we can not copy anything. | |
4673 | */ | |
4674 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 4675 | goto out; |
474d32b6 SR |
4676 | |
4677 | len -= BUF_PAGE_HDR_SIZE; | |
4678 | ||
8789a9e7 | 4679 | if (!data_page) |
554f786e | 4680 | goto out; |
8789a9e7 | 4681 | |
044fa782 SR |
4682 | bpage = *data_page; |
4683 | if (!bpage) | |
554f786e | 4684 | goto out; |
8789a9e7 | 4685 | |
5389f6fa | 4686 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4687 | |
ef7a4a16 SR |
4688 | reader = rb_get_reader_page(cpu_buffer); |
4689 | if (!reader) | |
554f786e | 4690 | goto out_unlock; |
8789a9e7 | 4691 | |
ef7a4a16 SR |
4692 | event = rb_reader_event(cpu_buffer); |
4693 | ||
4694 | read = reader->read; | |
4695 | commit = rb_page_commit(reader); | |
667d2412 | 4696 | |
66a8cb95 | 4697 | /* Check if any events were dropped */ |
ff0ff84a | 4698 | missed_events = cpu_buffer->lost_events; |
66a8cb95 | 4699 | |
8789a9e7 | 4700 | /* |
474d32b6 SR |
4701 | * If this page has been partially read or |
4702 | * if len is not big enough to read the rest of the page or | |
4703 | * a writer is still on the page, then | |
4704 | * we must copy the data from the page to the buffer. | |
4705 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 4706 | */ |
474d32b6 | 4707 | if (read || (len < (commit - read)) || |
ef7a4a16 | 4708 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 4709 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
4710 | unsigned int rpos = read; |
4711 | unsigned int pos = 0; | |
ef7a4a16 | 4712 | unsigned int size; |
8789a9e7 SR |
4713 | |
4714 | if (full) | |
554f786e | 4715 | goto out_unlock; |
8789a9e7 | 4716 | |
ef7a4a16 SR |
4717 | if (len > (commit - read)) |
4718 | len = (commit - read); | |
4719 | ||
69d1b839 SR |
4720 | /* Always keep the time extend and data together */ |
4721 | size = rb_event_ts_length(event); | |
ef7a4a16 SR |
4722 | |
4723 | if (len < size) | |
554f786e | 4724 | goto out_unlock; |
ef7a4a16 | 4725 | |
4f3640f8 SR |
4726 | /* save the current timestamp, since the user will need it */ |
4727 | save_timestamp = cpu_buffer->read_stamp; | |
4728 | ||
ef7a4a16 SR |
4729 | /* Need to copy one event at a time */ |
4730 | do { | |
e1e35927 DS |
4731 | /* We need the size of one event, because |
4732 | * rb_advance_reader only advances by one event, | |
4733 | * whereas rb_event_ts_length may include the size of | |
4734 | * one or two events. | |
4735 | * We have already ensured there's enough space if this | |
4736 | * is a time extend. */ | |
4737 | size = rb_event_length(event); | |
474d32b6 | 4738 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
4739 | |
4740 | len -= size; | |
4741 | ||
4742 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
4743 | rpos = reader->read; |
4744 | pos += size; | |
ef7a4a16 | 4745 | |
18fab912 HY |
4746 | if (rpos >= commit) |
4747 | break; | |
4748 | ||
ef7a4a16 | 4749 | event = rb_reader_event(cpu_buffer); |
69d1b839 SR |
4750 | /* Always keep the time extend and data together */ |
4751 | size = rb_event_ts_length(event); | |
e1e35927 | 4752 | } while (len >= size); |
667d2412 LJ |
4753 | |
4754 | /* update bpage */ | |
ef7a4a16 | 4755 | local_set(&bpage->commit, pos); |
4f3640f8 | 4756 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 4757 | |
474d32b6 SR |
4758 | /* we copied everything to the beginning */ |
4759 | read = 0; | |
8789a9e7 | 4760 | } else { |
afbab76a | 4761 | /* update the entry counter */ |
77ae365e | 4762 | cpu_buffer->read += rb_page_entries(reader); |
c64e148a | 4763 | cpu_buffer->read_bytes += BUF_PAGE_SIZE; |
afbab76a | 4764 | |
8789a9e7 | 4765 | /* swap the pages */ |
044fa782 | 4766 | rb_init_page(bpage); |
ef7a4a16 SR |
4767 | bpage = reader->page; |
4768 | reader->page = *data_page; | |
4769 | local_set(&reader->write, 0); | |
778c55d4 | 4770 | local_set(&reader->entries, 0); |
ef7a4a16 | 4771 | reader->read = 0; |
044fa782 | 4772 | *data_page = bpage; |
ff0ff84a SR |
4773 | |
4774 | /* | |
4775 | * Use the real_end for the data size, | |
4776 | * This gives us a chance to store the lost events | |
4777 | * on the page. | |
4778 | */ | |
4779 | if (reader->real_end) | |
4780 | local_set(&bpage->commit, reader->real_end); | |
8789a9e7 | 4781 | } |
667d2412 | 4782 | ret = read; |
8789a9e7 | 4783 | |
66a8cb95 | 4784 | cpu_buffer->lost_events = 0; |
2711ca23 SR |
4785 | |
4786 | commit = local_read(&bpage->commit); | |
66a8cb95 SR |
4787 | /* |
4788 | * Set a flag in the commit field if we lost events | |
4789 | */ | |
ff0ff84a | 4790 | if (missed_events) { |
ff0ff84a SR |
4791 | /* If there is room at the end of the page to save the |
4792 | * missed events, then record it there. | |
4793 | */ | |
4794 | if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) { | |
4795 | memcpy(&bpage->data[commit], &missed_events, | |
4796 | sizeof(missed_events)); | |
4797 | local_add(RB_MISSED_STORED, &bpage->commit); | |
2711ca23 | 4798 | commit += sizeof(missed_events); |
ff0ff84a | 4799 | } |
66a8cb95 | 4800 | local_add(RB_MISSED_EVENTS, &bpage->commit); |
ff0ff84a | 4801 | } |
66a8cb95 | 4802 | |
2711ca23 SR |
4803 | /* |
4804 | * This page may be off to user land. Zero it out here. | |
4805 | */ | |
4806 | if (commit < BUF_PAGE_SIZE) | |
4807 | memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit); | |
4808 | ||
554f786e | 4809 | out_unlock: |
5389f6fa | 4810 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4811 | |
554f786e | 4812 | out: |
8789a9e7 SR |
4813 | return ret; |
4814 | } | |
d6ce96da | 4815 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 4816 | |
b32614c0 SAS |
4817 | /* |
4818 | * We only allocate new buffers, never free them if the CPU goes down. | |
4819 | * If we were to free the buffer, then the user would lose any trace that was in | |
4820 | * the buffer. | |
4821 | */ | |
4822 | int trace_rb_cpu_prepare(unsigned int cpu, struct hlist_node *node) | |
554f786e | 4823 | { |
b32614c0 | 4824 | struct ring_buffer *buffer; |
9b94a8fb SRRH |
4825 | long nr_pages_same; |
4826 | int cpu_i; | |
4827 | unsigned long nr_pages; | |
554f786e | 4828 | |
b32614c0 SAS |
4829 | buffer = container_of(node, struct ring_buffer, node); |
4830 | if (cpumask_test_cpu(cpu, buffer->cpumask)) | |
4831 | return 0; | |
4832 | ||
4833 | nr_pages = 0; | |
4834 | nr_pages_same = 1; | |
4835 | /* check if all cpu sizes are same */ | |
4836 | for_each_buffer_cpu(buffer, cpu_i) { | |
4837 | /* fill in the size from first enabled cpu */ | |
4838 | if (nr_pages == 0) | |
4839 | nr_pages = buffer->buffers[cpu_i]->nr_pages; | |
4840 | if (nr_pages != buffer->buffers[cpu_i]->nr_pages) { | |
4841 | nr_pages_same = 0; | |
4842 | break; | |
554f786e | 4843 | } |
554f786e | 4844 | } |
b32614c0 SAS |
4845 | /* allocate minimum pages, user can later expand it */ |
4846 | if (!nr_pages_same) | |
4847 | nr_pages = 2; | |
4848 | buffer->buffers[cpu] = | |
4849 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); | |
4850 | if (!buffer->buffers[cpu]) { | |
4851 | WARN(1, "failed to allocate ring buffer on CPU %u\n", | |
4852 | cpu); | |
4853 | return -ENOMEM; | |
4854 | } | |
4855 | smp_wmb(); | |
4856 | cpumask_set_cpu(cpu, buffer->cpumask); | |
4857 | return 0; | |
554f786e | 4858 | } |
6c43e554 SRRH |
4859 | |
4860 | #ifdef CONFIG_RING_BUFFER_STARTUP_TEST | |
4861 | /* | |
4862 | * This is a basic integrity check of the ring buffer. | |
4863 | * Late in the boot cycle this test will run when configured in. | |
4864 | * It will kick off a thread per CPU that will go into a loop | |
4865 | * writing to the per cpu ring buffer various sizes of data. | |
4866 | * Some of the data will be large items, some small. | |
4867 | * | |
4868 | * Another thread is created that goes into a spin, sending out | |
4869 | * IPIs to the other CPUs to also write into the ring buffer. | |
4870 | * this is to test the nesting ability of the buffer. | |
4871 | * | |
4872 | * Basic stats are recorded and reported. If something in the | |
4873 | * ring buffer should happen that's not expected, a big warning | |
4874 | * is displayed and all ring buffers are disabled. | |
4875 | */ | |
4876 | static struct task_struct *rb_threads[NR_CPUS] __initdata; | |
4877 | ||
4878 | struct rb_test_data { | |
4879 | struct ring_buffer *buffer; | |
4880 | unsigned long events; | |
4881 | unsigned long bytes_written; | |
4882 | unsigned long bytes_alloc; | |
4883 | unsigned long bytes_dropped; | |
4884 | unsigned long events_nested; | |
4885 | unsigned long bytes_written_nested; | |
4886 | unsigned long bytes_alloc_nested; | |
4887 | unsigned long bytes_dropped_nested; | |
4888 | int min_size_nested; | |
4889 | int max_size_nested; | |
4890 | int max_size; | |
4891 | int min_size; | |
4892 | int cpu; | |
4893 | int cnt; | |
4894 | }; | |
4895 | ||
4896 | static struct rb_test_data rb_data[NR_CPUS] __initdata; | |
4897 | ||
4898 | /* 1 meg per cpu */ | |
4899 | #define RB_TEST_BUFFER_SIZE 1048576 | |
4900 | ||
4901 | static char rb_string[] __initdata = | |
4902 | "abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\" | |
4903 | "?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890" | |
4904 | "!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv"; | |
4905 | ||
4906 | static bool rb_test_started __initdata; | |
4907 | ||
4908 | struct rb_item { | |
4909 | int size; | |
4910 | char str[]; | |
4911 | }; | |
4912 | ||
4913 | static __init int rb_write_something(struct rb_test_data *data, bool nested) | |
4914 | { | |
4915 | struct ring_buffer_event *event; | |
4916 | struct rb_item *item; | |
4917 | bool started; | |
4918 | int event_len; | |
4919 | int size; | |
4920 | int len; | |
4921 | int cnt; | |
4922 | ||
4923 | /* Have nested writes different that what is written */ | |
4924 | cnt = data->cnt + (nested ? 27 : 0); | |
4925 | ||
4926 | /* Multiply cnt by ~e, to make some unique increment */ | |
4927 | size = (data->cnt * 68 / 25) % (sizeof(rb_string) - 1); | |
4928 | ||
4929 | len = size + sizeof(struct rb_item); | |
4930 | ||
4931 | started = rb_test_started; | |
4932 | /* read rb_test_started before checking buffer enabled */ | |
4933 | smp_rmb(); | |
4934 | ||
4935 | event = ring_buffer_lock_reserve(data->buffer, len); | |
4936 | if (!event) { | |
4937 | /* Ignore dropped events before test starts. */ | |
4938 | if (started) { | |
4939 | if (nested) | |
4940 | data->bytes_dropped += len; | |
4941 | else | |
4942 | data->bytes_dropped_nested += len; | |
4943 | } | |
4944 | return len; | |
4945 | } | |
4946 | ||
4947 | event_len = ring_buffer_event_length(event); | |
4948 | ||
4949 | if (RB_WARN_ON(data->buffer, event_len < len)) | |
4950 | goto out; | |
4951 | ||
4952 | item = ring_buffer_event_data(event); | |
4953 | item->size = size; | |
4954 | memcpy(item->str, rb_string, size); | |
4955 | ||
4956 | if (nested) { | |
4957 | data->bytes_alloc_nested += event_len; | |
4958 | data->bytes_written_nested += len; | |
4959 | data->events_nested++; | |
4960 | if (!data->min_size_nested || len < data->min_size_nested) | |
4961 | data->min_size_nested = len; | |
4962 | if (len > data->max_size_nested) | |
4963 | data->max_size_nested = len; | |
4964 | } else { | |
4965 | data->bytes_alloc += event_len; | |
4966 | data->bytes_written += len; | |
4967 | data->events++; | |
4968 | if (!data->min_size || len < data->min_size) | |
4969 | data->max_size = len; | |
4970 | if (len > data->max_size) | |
4971 | data->max_size = len; | |
4972 | } | |
4973 | ||
4974 | out: | |
4975 | ring_buffer_unlock_commit(data->buffer, event); | |
4976 | ||
4977 | return 0; | |
4978 | } | |
4979 | ||
4980 | static __init int rb_test(void *arg) | |
4981 | { | |
4982 | struct rb_test_data *data = arg; | |
4983 | ||
4984 | while (!kthread_should_stop()) { | |
4985 | rb_write_something(data, false); | |
4986 | data->cnt++; | |
4987 | ||
4988 | set_current_state(TASK_INTERRUPTIBLE); | |
4989 | /* Now sleep between a min of 100-300us and a max of 1ms */ | |
4990 | usleep_range(((data->cnt % 3) + 1) * 100, 1000); | |
4991 | } | |
4992 | ||
4993 | return 0; | |
4994 | } | |
4995 | ||
4996 | static __init void rb_ipi(void *ignore) | |
4997 | { | |
4998 | struct rb_test_data *data; | |
4999 | int cpu = smp_processor_id(); | |
5000 | ||
5001 | data = &rb_data[cpu]; | |
5002 | rb_write_something(data, true); | |
5003 | } | |
5004 | ||
5005 | static __init int rb_hammer_test(void *arg) | |
5006 | { | |
5007 | while (!kthread_should_stop()) { | |
5008 | ||
5009 | /* Send an IPI to all cpus to write data! */ | |
5010 | smp_call_function(rb_ipi, NULL, 1); | |
5011 | /* No sleep, but for non preempt, let others run */ | |
5012 | schedule(); | |
5013 | } | |
5014 | ||
5015 | return 0; | |
5016 | } | |
5017 | ||
5018 | static __init int test_ringbuffer(void) | |
5019 | { | |
5020 | struct task_struct *rb_hammer; | |
5021 | struct ring_buffer *buffer; | |
5022 | int cpu; | |
5023 | int ret = 0; | |
5024 | ||
5025 | pr_info("Running ring buffer tests...\n"); | |
5026 | ||
5027 | buffer = ring_buffer_alloc(RB_TEST_BUFFER_SIZE, RB_FL_OVERWRITE); | |
5028 | if (WARN_ON(!buffer)) | |
5029 | return 0; | |
5030 | ||
5031 | /* Disable buffer so that threads can't write to it yet */ | |
5032 | ring_buffer_record_off(buffer); | |
5033 | ||
5034 | for_each_online_cpu(cpu) { | |
5035 | rb_data[cpu].buffer = buffer; | |
5036 | rb_data[cpu].cpu = cpu; | |
5037 | rb_data[cpu].cnt = cpu; | |
5038 | rb_threads[cpu] = kthread_create(rb_test, &rb_data[cpu], | |
5039 | "rbtester/%d", cpu); | |
62277de7 | 5040 | if (WARN_ON(IS_ERR(rb_threads[cpu]))) { |
6c43e554 | 5041 | pr_cont("FAILED\n"); |
62277de7 | 5042 | ret = PTR_ERR(rb_threads[cpu]); |
6c43e554 SRRH |
5043 | goto out_free; |
5044 | } | |
5045 | ||
5046 | kthread_bind(rb_threads[cpu], cpu); | |
5047 | wake_up_process(rb_threads[cpu]); | |
5048 | } | |
5049 | ||
5050 | /* Now create the rb hammer! */ | |
5051 | rb_hammer = kthread_run(rb_hammer_test, NULL, "rbhammer"); | |
62277de7 | 5052 | if (WARN_ON(IS_ERR(rb_hammer))) { |
6c43e554 | 5053 | pr_cont("FAILED\n"); |
62277de7 | 5054 | ret = PTR_ERR(rb_hammer); |
6c43e554 SRRH |
5055 | goto out_free; |
5056 | } | |
5057 | ||
5058 | ring_buffer_record_on(buffer); | |
5059 | /* | |
5060 | * Show buffer is enabled before setting rb_test_started. | |
5061 | * Yes there's a small race window where events could be | |
5062 | * dropped and the thread wont catch it. But when a ring | |
5063 | * buffer gets enabled, there will always be some kind of | |
5064 | * delay before other CPUs see it. Thus, we don't care about | |
5065 | * those dropped events. We care about events dropped after | |
5066 | * the threads see that the buffer is active. | |
5067 | */ | |
5068 | smp_wmb(); | |
5069 | rb_test_started = true; | |
5070 | ||
5071 | set_current_state(TASK_INTERRUPTIBLE); | |
5072 | /* Just run for 10 seconds */; | |
5073 | schedule_timeout(10 * HZ); | |
5074 | ||
5075 | kthread_stop(rb_hammer); | |
5076 | ||
5077 | out_free: | |
5078 | for_each_online_cpu(cpu) { | |
5079 | if (!rb_threads[cpu]) | |
5080 | break; | |
5081 | kthread_stop(rb_threads[cpu]); | |
5082 | } | |
5083 | if (ret) { | |
5084 | ring_buffer_free(buffer); | |
5085 | return ret; | |
5086 | } | |
5087 | ||
5088 | /* Report! */ | |
5089 | pr_info("finished\n"); | |
5090 | for_each_online_cpu(cpu) { | |
5091 | struct ring_buffer_event *event; | |
5092 | struct rb_test_data *data = &rb_data[cpu]; | |
5093 | struct rb_item *item; | |
5094 | unsigned long total_events; | |
5095 | unsigned long total_dropped; | |
5096 | unsigned long total_written; | |
5097 | unsigned long total_alloc; | |
5098 | unsigned long total_read = 0; | |
5099 | unsigned long total_size = 0; | |
5100 | unsigned long total_len = 0; | |
5101 | unsigned long total_lost = 0; | |
5102 | unsigned long lost; | |
5103 | int big_event_size; | |
5104 | int small_event_size; | |
5105 | ||
5106 | ret = -1; | |
5107 | ||
5108 | total_events = data->events + data->events_nested; | |
5109 | total_written = data->bytes_written + data->bytes_written_nested; | |
5110 | total_alloc = data->bytes_alloc + data->bytes_alloc_nested; | |
5111 | total_dropped = data->bytes_dropped + data->bytes_dropped_nested; | |
5112 | ||
5113 | big_event_size = data->max_size + data->max_size_nested; | |
5114 | small_event_size = data->min_size + data->min_size_nested; | |
5115 | ||
5116 | pr_info("CPU %d:\n", cpu); | |
5117 | pr_info(" events: %ld\n", total_events); | |
5118 | pr_info(" dropped bytes: %ld\n", total_dropped); | |
5119 | pr_info(" alloced bytes: %ld\n", total_alloc); | |
5120 | pr_info(" written bytes: %ld\n", total_written); | |
5121 | pr_info(" biggest event: %d\n", big_event_size); | |
5122 | pr_info(" smallest event: %d\n", small_event_size); | |
5123 | ||
5124 | if (RB_WARN_ON(buffer, total_dropped)) | |
5125 | break; | |
5126 | ||
5127 | ret = 0; | |
5128 | ||
5129 | while ((event = ring_buffer_consume(buffer, cpu, NULL, &lost))) { | |
5130 | total_lost += lost; | |
5131 | item = ring_buffer_event_data(event); | |
5132 | total_len += ring_buffer_event_length(event); | |
5133 | total_size += item->size + sizeof(struct rb_item); | |
5134 | if (memcmp(&item->str[0], rb_string, item->size) != 0) { | |
5135 | pr_info("FAILED!\n"); | |
5136 | pr_info("buffer had: %.*s\n", item->size, item->str); | |
5137 | pr_info("expected: %.*s\n", item->size, rb_string); | |
5138 | RB_WARN_ON(buffer, 1); | |
5139 | ret = -1; | |
5140 | break; | |
5141 | } | |
5142 | total_read++; | |
5143 | } | |
5144 | if (ret) | |
5145 | break; | |
5146 | ||
5147 | ret = -1; | |
5148 | ||
5149 | pr_info(" read events: %ld\n", total_read); | |
5150 | pr_info(" lost events: %ld\n", total_lost); | |
5151 | pr_info(" total events: %ld\n", total_lost + total_read); | |
5152 | pr_info(" recorded len bytes: %ld\n", total_len); | |
5153 | pr_info(" recorded size bytes: %ld\n", total_size); | |
5154 | if (total_lost) | |
5155 | pr_info(" With dropped events, record len and size may not match\n" | |
5156 | " alloced and written from above\n"); | |
5157 | if (!total_lost) { | |
5158 | if (RB_WARN_ON(buffer, total_len != total_alloc || | |
5159 | total_size != total_written)) | |
5160 | break; | |
5161 | } | |
5162 | if (RB_WARN_ON(buffer, total_lost + total_read != total_events)) | |
5163 | break; | |
5164 | ||
5165 | ret = 0; | |
5166 | } | |
5167 | if (!ret) | |
5168 | pr_info("Ring buffer PASSED!\n"); | |
5169 | ||
5170 | ring_buffer_free(buffer); | |
5171 | return 0; | |
5172 | } | |
5173 | ||
5174 | late_initcall(test_ringbuffer); | |
5175 | #endif /* CONFIG_RING_BUFFER_STARTUP_TEST */ |