Commit | Line | Data |
---|---|---|
7a8e76a3 SR |
1 | /* |
2 | * Generic ring buffer | |
3 | * | |
4 | * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com> | |
5 | */ | |
af658dca | 6 | #include <linux/trace_events.h> |
7a8e76a3 | 7 | #include <linux/ring_buffer.h> |
14131f2f | 8 | #include <linux/trace_clock.h> |
0b07436d | 9 | #include <linux/trace_seq.h> |
7a8e76a3 | 10 | #include <linux/spinlock.h> |
15693458 | 11 | #include <linux/irq_work.h> |
7a8e76a3 | 12 | #include <linux/uaccess.h> |
a81bd80a | 13 | #include <linux/hardirq.h> |
6c43e554 | 14 | #include <linux/kthread.h> /* for self test */ |
1744a21d | 15 | #include <linux/kmemcheck.h> |
7a8e76a3 SR |
16 | #include <linux/module.h> |
17 | #include <linux/percpu.h> | |
18 | #include <linux/mutex.h> | |
6c43e554 | 19 | #include <linux/delay.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
7a8e76a3 SR |
21 | #include <linux/init.h> |
22 | #include <linux/hash.h> | |
23 | #include <linux/list.h> | |
554f786e | 24 | #include <linux/cpu.h> |
7a8e76a3 | 25 | |
79615760 | 26 | #include <asm/local.h> |
182e9f5f | 27 | |
83f40318 VN |
28 | static void update_pages_handler(struct work_struct *work); |
29 | ||
d1b182a8 SR |
30 | /* |
31 | * The ring buffer header is special. We must manually up keep it. | |
32 | */ | |
33 | int ring_buffer_print_entry_header(struct trace_seq *s) | |
34 | { | |
c0cd93aa SRRH |
35 | trace_seq_puts(s, "# compressed entry header\n"); |
36 | trace_seq_puts(s, "\ttype_len : 5 bits\n"); | |
37 | trace_seq_puts(s, "\ttime_delta : 27 bits\n"); | |
38 | trace_seq_puts(s, "\tarray : 32 bits\n"); | |
39 | trace_seq_putc(s, '\n'); | |
40 | trace_seq_printf(s, "\tpadding : type == %d\n", | |
41 | RINGBUF_TYPE_PADDING); | |
42 | trace_seq_printf(s, "\ttime_extend : type == %d\n", | |
43 | RINGBUF_TYPE_TIME_EXTEND); | |
44 | trace_seq_printf(s, "\tdata max type_len == %d\n", | |
45 | RINGBUF_TYPE_DATA_TYPE_LEN_MAX); | |
46 | ||
47 | return !trace_seq_has_overflowed(s); | |
d1b182a8 SR |
48 | } |
49 | ||
5cc98548 SR |
50 | /* |
51 | * The ring buffer is made up of a list of pages. A separate list of pages is | |
52 | * allocated for each CPU. A writer may only write to a buffer that is | |
53 | * associated with the CPU it is currently executing on. A reader may read | |
54 | * from any per cpu buffer. | |
55 | * | |
56 | * The reader is special. For each per cpu buffer, the reader has its own | |
57 | * reader page. When a reader has read the entire reader page, this reader | |
58 | * page is swapped with another page in the ring buffer. | |
59 | * | |
60 | * Now, as long as the writer is off the reader page, the reader can do what | |
61 | * ever it wants with that page. The writer will never write to that page | |
62 | * again (as long as it is out of the ring buffer). | |
63 | * | |
64 | * Here's some silly ASCII art. | |
65 | * | |
66 | * +------+ | |
67 | * |reader| RING BUFFER | |
68 | * |page | | |
69 | * +------+ +---+ +---+ +---+ | |
70 | * | |-->| |-->| | | |
71 | * +---+ +---+ +---+ | |
72 | * ^ | | |
73 | * | | | |
74 | * +---------------+ | |
75 | * | |
76 | * | |
77 | * +------+ | |
78 | * |reader| RING BUFFER | |
79 | * |page |------------------v | |
80 | * +------+ +---+ +---+ +---+ | |
81 | * | |-->| |-->| | | |
82 | * +---+ +---+ +---+ | |
83 | * ^ | | |
84 | * | | | |
85 | * +---------------+ | |
86 | * | |
87 | * | |
88 | * +------+ | |
89 | * |reader| RING BUFFER | |
90 | * |page |------------------v | |
91 | * +------+ +---+ +---+ +---+ | |
92 | * ^ | |-->| |-->| | | |
93 | * | +---+ +---+ +---+ | |
94 | * | | | |
95 | * | | | |
96 | * +------------------------------+ | |
97 | * | |
98 | * | |
99 | * +------+ | |
100 | * |buffer| RING BUFFER | |
101 | * |page |------------------v | |
102 | * +------+ +---+ +---+ +---+ | |
103 | * ^ | | | |-->| | | |
104 | * | New +---+ +---+ +---+ | |
105 | * | Reader------^ | | |
106 | * | page | | |
107 | * +------------------------------+ | |
108 | * | |
109 | * | |
110 | * After we make this swap, the reader can hand this page off to the splice | |
111 | * code and be done with it. It can even allocate a new page if it needs to | |
112 | * and swap that into the ring buffer. | |
113 | * | |
114 | * We will be using cmpxchg soon to make all this lockless. | |
115 | * | |
116 | */ | |
117 | ||
499e5470 SR |
118 | /* Used for individual buffers (after the counter) */ |
119 | #define RB_BUFFER_OFF (1 << 20) | |
a3583244 | 120 | |
499e5470 | 121 | #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data) |
033601a3 | 122 | |
e3d6bf0a | 123 | #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array)) |
67d34724 | 124 | #define RB_ALIGNMENT 4U |
334d4169 | 125 | #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
c7b09308 | 126 | #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */ |
334d4169 | 127 | |
649508f6 | 128 | #ifndef CONFIG_HAVE_64BIT_ALIGNED_ACCESS |
2271048d SR |
129 | # define RB_FORCE_8BYTE_ALIGNMENT 0 |
130 | # define RB_ARCH_ALIGNMENT RB_ALIGNMENT | |
131 | #else | |
132 | # define RB_FORCE_8BYTE_ALIGNMENT 1 | |
133 | # define RB_ARCH_ALIGNMENT 8U | |
134 | #endif | |
135 | ||
649508f6 JH |
136 | #define RB_ALIGN_DATA __aligned(RB_ARCH_ALIGNMENT) |
137 | ||
334d4169 LJ |
138 | /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */ |
139 | #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX | |
7a8e76a3 SR |
140 | |
141 | enum { | |
142 | RB_LEN_TIME_EXTEND = 8, | |
143 | RB_LEN_TIME_STAMP = 16, | |
144 | }; | |
145 | ||
69d1b839 SR |
146 | #define skip_time_extend(event) \ |
147 | ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND)) | |
148 | ||
2d622719 TZ |
149 | static inline int rb_null_event(struct ring_buffer_event *event) |
150 | { | |
a1863c21 | 151 | return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta; |
2d622719 TZ |
152 | } |
153 | ||
154 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
155 | { | |
a1863c21 | 156 | /* padding has a NULL time_delta */ |
334d4169 | 157 | event->type_len = RINGBUF_TYPE_PADDING; |
2d622719 TZ |
158 | event->time_delta = 0; |
159 | } | |
160 | ||
34a148bf | 161 | static unsigned |
2d622719 | 162 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
163 | { |
164 | unsigned length; | |
165 | ||
334d4169 LJ |
166 | if (event->type_len) |
167 | length = event->type_len * RB_ALIGNMENT; | |
2d622719 TZ |
168 | else |
169 | length = event->array[0]; | |
170 | return length + RB_EVNT_HDR_SIZE; | |
171 | } | |
172 | ||
69d1b839 SR |
173 | /* |
174 | * Return the length of the given event. Will return | |
175 | * the length of the time extend if the event is a | |
176 | * time extend. | |
177 | */ | |
178 | static inline unsigned | |
2d622719 TZ |
179 | rb_event_length(struct ring_buffer_event *event) |
180 | { | |
334d4169 | 181 | switch (event->type_len) { |
7a8e76a3 | 182 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
183 | if (rb_null_event(event)) |
184 | /* undefined */ | |
185 | return -1; | |
334d4169 | 186 | return event->array[0] + RB_EVNT_HDR_SIZE; |
7a8e76a3 SR |
187 | |
188 | case RINGBUF_TYPE_TIME_EXTEND: | |
189 | return RB_LEN_TIME_EXTEND; | |
190 | ||
191 | case RINGBUF_TYPE_TIME_STAMP: | |
192 | return RB_LEN_TIME_STAMP; | |
193 | ||
194 | case RINGBUF_TYPE_DATA: | |
2d622719 | 195 | return rb_event_data_length(event); |
7a8e76a3 SR |
196 | default: |
197 | BUG(); | |
198 | } | |
199 | /* not hit */ | |
200 | return 0; | |
201 | } | |
202 | ||
69d1b839 SR |
203 | /* |
204 | * Return total length of time extend and data, | |
205 | * or just the event length for all other events. | |
206 | */ | |
207 | static inline unsigned | |
208 | rb_event_ts_length(struct ring_buffer_event *event) | |
209 | { | |
210 | unsigned len = 0; | |
211 | ||
212 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
213 | /* time extends include the data event after it */ | |
214 | len = RB_LEN_TIME_EXTEND; | |
215 | event = skip_time_extend(event); | |
216 | } | |
217 | return len + rb_event_length(event); | |
218 | } | |
219 | ||
7a8e76a3 SR |
220 | /** |
221 | * ring_buffer_event_length - return the length of the event | |
222 | * @event: the event to get the length of | |
69d1b839 SR |
223 | * |
224 | * Returns the size of the data load of a data event. | |
225 | * If the event is something other than a data event, it | |
226 | * returns the size of the event itself. With the exception | |
227 | * of a TIME EXTEND, where it still returns the size of the | |
228 | * data load of the data event after it. | |
7a8e76a3 SR |
229 | */ |
230 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
231 | { | |
69d1b839 SR |
232 | unsigned length; |
233 | ||
234 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) | |
235 | event = skip_time_extend(event); | |
236 | ||
237 | length = rb_event_length(event); | |
334d4169 | 238 | if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
465634ad RR |
239 | return length; |
240 | length -= RB_EVNT_HDR_SIZE; | |
241 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
242 | length -= sizeof(event->array[0]); | |
243 | return length; | |
7a8e76a3 | 244 | } |
c4f50183 | 245 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
246 | |
247 | /* inline for ring buffer fast paths */ | |
34a148bf | 248 | static void * |
7a8e76a3 SR |
249 | rb_event_data(struct ring_buffer_event *event) |
250 | { | |
69d1b839 SR |
251 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) |
252 | event = skip_time_extend(event); | |
334d4169 | 253 | BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX); |
7a8e76a3 | 254 | /* If length is in len field, then array[0] has the data */ |
334d4169 | 255 | if (event->type_len) |
7a8e76a3 SR |
256 | return (void *)&event->array[0]; |
257 | /* Otherwise length is in array[0] and array[1] has the data */ | |
258 | return (void *)&event->array[1]; | |
259 | } | |
260 | ||
261 | /** | |
262 | * ring_buffer_event_data - return the data of the event | |
263 | * @event: the event to get the data from | |
264 | */ | |
265 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
266 | { | |
267 | return rb_event_data(event); | |
268 | } | |
c4f50183 | 269 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
270 | |
271 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 272 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
273 | |
274 | #define TS_SHIFT 27 | |
275 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
276 | #define TS_DELTA_TEST (~TS_MASK) | |
277 | ||
66a8cb95 SR |
278 | /* Flag when events were overwritten */ |
279 | #define RB_MISSED_EVENTS (1 << 31) | |
ff0ff84a SR |
280 | /* Missed count stored at end */ |
281 | #define RB_MISSED_STORED (1 << 30) | |
66a8cb95 | 282 | |
abc9b56d | 283 | struct buffer_data_page { |
e4c2ce82 | 284 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 285 | local_t commit; /* write committed index */ |
649508f6 | 286 | unsigned char data[] RB_ALIGN_DATA; /* data of buffer page */ |
abc9b56d SR |
287 | }; |
288 | ||
77ae365e SR |
289 | /* |
290 | * Note, the buffer_page list must be first. The buffer pages | |
291 | * are allocated in cache lines, which means that each buffer | |
292 | * page will be at the beginning of a cache line, and thus | |
293 | * the least significant bits will be zero. We use this to | |
294 | * add flags in the list struct pointers, to make the ring buffer | |
295 | * lockless. | |
296 | */ | |
abc9b56d | 297 | struct buffer_page { |
778c55d4 | 298 | struct list_head list; /* list of buffer pages */ |
abc9b56d | 299 | local_t write; /* index for next write */ |
6f807acd | 300 | unsigned read; /* index for next read */ |
778c55d4 | 301 | local_t entries; /* entries on this page */ |
ff0ff84a | 302 | unsigned long real_end; /* real end of data */ |
abc9b56d | 303 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
304 | }; |
305 | ||
77ae365e SR |
306 | /* |
307 | * The buffer page counters, write and entries, must be reset | |
308 | * atomically when crossing page boundaries. To synchronize this | |
309 | * update, two counters are inserted into the number. One is | |
310 | * the actual counter for the write position or count on the page. | |
311 | * | |
312 | * The other is a counter of updaters. Before an update happens | |
313 | * the update partition of the counter is incremented. This will | |
314 | * allow the updater to update the counter atomically. | |
315 | * | |
316 | * The counter is 20 bits, and the state data is 12. | |
317 | */ | |
318 | #define RB_WRITE_MASK 0xfffff | |
319 | #define RB_WRITE_INTCNT (1 << 20) | |
320 | ||
044fa782 | 321 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 322 | { |
044fa782 | 323 | local_set(&bpage->commit, 0); |
abc9b56d SR |
324 | } |
325 | ||
474d32b6 SR |
326 | /** |
327 | * ring_buffer_page_len - the size of data on the page. | |
328 | * @page: The page to read | |
329 | * | |
330 | * Returns the amount of data on the page, including buffer page header. | |
331 | */ | |
ef7a4a16 SR |
332 | size_t ring_buffer_page_len(void *page) |
333 | { | |
474d32b6 SR |
334 | return local_read(&((struct buffer_data_page *)page)->commit) |
335 | + BUF_PAGE_HDR_SIZE; | |
ef7a4a16 SR |
336 | } |
337 | ||
ed56829c SR |
338 | /* |
339 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
340 | * this issue out. | |
341 | */ | |
34a148bf | 342 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 343 | { |
34a148bf | 344 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 345 | kfree(bpage); |
ed56829c SR |
346 | } |
347 | ||
7a8e76a3 SR |
348 | /* |
349 | * We need to fit the time_stamp delta into 27 bits. | |
350 | */ | |
351 | static inline int test_time_stamp(u64 delta) | |
352 | { | |
353 | if (delta & TS_DELTA_TEST) | |
354 | return 1; | |
355 | return 0; | |
356 | } | |
357 | ||
474d32b6 | 358 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 | 359 | |
be957c44 SR |
360 | /* Max payload is BUF_PAGE_SIZE - header (8bytes) */ |
361 | #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) | |
362 | ||
d1b182a8 SR |
363 | int ring_buffer_print_page_header(struct trace_seq *s) |
364 | { | |
365 | struct buffer_data_page field; | |
c0cd93aa SRRH |
366 | |
367 | trace_seq_printf(s, "\tfield: u64 timestamp;\t" | |
368 | "offset:0;\tsize:%u;\tsigned:%u;\n", | |
369 | (unsigned int)sizeof(field.time_stamp), | |
370 | (unsigned int)is_signed_type(u64)); | |
371 | ||
372 | trace_seq_printf(s, "\tfield: local_t commit;\t" | |
373 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
374 | (unsigned int)offsetof(typeof(field), commit), | |
375 | (unsigned int)sizeof(field.commit), | |
376 | (unsigned int)is_signed_type(long)); | |
377 | ||
378 | trace_seq_printf(s, "\tfield: int overwrite;\t" | |
379 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
380 | (unsigned int)offsetof(typeof(field), commit), | |
381 | 1, | |
382 | (unsigned int)is_signed_type(long)); | |
383 | ||
384 | trace_seq_printf(s, "\tfield: char data;\t" | |
385 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
386 | (unsigned int)offsetof(typeof(field), data), | |
387 | (unsigned int)BUF_PAGE_SIZE, | |
388 | (unsigned int)is_signed_type(char)); | |
389 | ||
390 | return !trace_seq_has_overflowed(s); | |
d1b182a8 SR |
391 | } |
392 | ||
15693458 SRRH |
393 | struct rb_irq_work { |
394 | struct irq_work work; | |
395 | wait_queue_head_t waiters; | |
1e0d6714 | 396 | wait_queue_head_t full_waiters; |
15693458 | 397 | bool waiters_pending; |
1e0d6714 SRRH |
398 | bool full_waiters_pending; |
399 | bool wakeup_full; | |
15693458 SRRH |
400 | }; |
401 | ||
fcc742ea SRRH |
402 | /* |
403 | * Structure to hold event state and handle nested events. | |
404 | */ | |
405 | struct rb_event_info { | |
406 | u64 ts; | |
407 | u64 delta; | |
408 | unsigned long length; | |
409 | struct buffer_page *tail_page; | |
410 | int add_timestamp; | |
411 | }; | |
412 | ||
a497adb4 SRRH |
413 | /* |
414 | * Used for which event context the event is in. | |
415 | * NMI = 0 | |
416 | * IRQ = 1 | |
417 | * SOFTIRQ = 2 | |
418 | * NORMAL = 3 | |
419 | * | |
420 | * See trace_recursive_lock() comment below for more details. | |
421 | */ | |
422 | enum { | |
423 | RB_CTX_NMI, | |
424 | RB_CTX_IRQ, | |
425 | RB_CTX_SOFTIRQ, | |
426 | RB_CTX_NORMAL, | |
427 | RB_CTX_MAX | |
428 | }; | |
429 | ||
7a8e76a3 SR |
430 | /* |
431 | * head_page == tail_page && head == tail then buffer is empty. | |
432 | */ | |
433 | struct ring_buffer_per_cpu { | |
434 | int cpu; | |
985023de | 435 | atomic_t record_disabled; |
7a8e76a3 | 436 | struct ring_buffer *buffer; |
5389f6fa | 437 | raw_spinlock_t reader_lock; /* serialize readers */ |
445c8951 | 438 | arch_spinlock_t lock; |
7a8e76a3 | 439 | struct lock_class_key lock_key; |
438ced17 | 440 | unsigned int nr_pages; |
58a09ec6 | 441 | unsigned int current_context; |
3adc54fa | 442 | struct list_head *pages; |
6f807acd SR |
443 | struct buffer_page *head_page; /* read from head */ |
444 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 445 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 446 | struct buffer_page *reader_page; |
66a8cb95 SR |
447 | unsigned long lost_events; |
448 | unsigned long last_overrun; | |
c64e148a | 449 | local_t entries_bytes; |
e4906eff | 450 | local_t entries; |
884bfe89 SP |
451 | local_t overrun; |
452 | local_t commit_overrun; | |
453 | local_t dropped_events; | |
fa743953 SR |
454 | local_t committing; |
455 | local_t commits; | |
77ae365e | 456 | unsigned long read; |
c64e148a | 457 | unsigned long read_bytes; |
7a8e76a3 SR |
458 | u64 write_stamp; |
459 | u64 read_stamp; | |
438ced17 VN |
460 | /* ring buffer pages to update, > 0 to add, < 0 to remove */ |
461 | int nr_pages_to_update; | |
462 | struct list_head new_pages; /* new pages to add */ | |
83f40318 | 463 | struct work_struct update_pages_work; |
05fdd70d | 464 | struct completion update_done; |
15693458 SRRH |
465 | |
466 | struct rb_irq_work irq_work; | |
7a8e76a3 SR |
467 | }; |
468 | ||
469 | struct ring_buffer { | |
7a8e76a3 SR |
470 | unsigned flags; |
471 | int cpus; | |
7a8e76a3 | 472 | atomic_t record_disabled; |
83f40318 | 473 | atomic_t resize_disabled; |
00f62f61 | 474 | cpumask_var_t cpumask; |
7a8e76a3 | 475 | |
1f8a6a10 PZ |
476 | struct lock_class_key *reader_lock_key; |
477 | ||
7a8e76a3 SR |
478 | struct mutex mutex; |
479 | ||
480 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 481 | |
59222efe | 482 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
483 | struct notifier_block cpu_notify; |
484 | #endif | |
37886f6a | 485 | u64 (*clock)(void); |
15693458 SRRH |
486 | |
487 | struct rb_irq_work irq_work; | |
7a8e76a3 SR |
488 | }; |
489 | ||
490 | struct ring_buffer_iter { | |
491 | struct ring_buffer_per_cpu *cpu_buffer; | |
492 | unsigned long head; | |
493 | struct buffer_page *head_page; | |
492a74f4 SR |
494 | struct buffer_page *cache_reader_page; |
495 | unsigned long cache_read; | |
7a8e76a3 SR |
496 | u64 read_stamp; |
497 | }; | |
498 | ||
15693458 SRRH |
499 | /* |
500 | * rb_wake_up_waiters - wake up tasks waiting for ring buffer input | |
501 | * | |
502 | * Schedules a delayed work to wake up any task that is blocked on the | |
503 | * ring buffer waiters queue. | |
504 | */ | |
505 | static void rb_wake_up_waiters(struct irq_work *work) | |
506 | { | |
507 | struct rb_irq_work *rbwork = container_of(work, struct rb_irq_work, work); | |
508 | ||
509 | wake_up_all(&rbwork->waiters); | |
1e0d6714 SRRH |
510 | if (rbwork->wakeup_full) { |
511 | rbwork->wakeup_full = false; | |
512 | wake_up_all(&rbwork->full_waiters); | |
513 | } | |
15693458 SRRH |
514 | } |
515 | ||
516 | /** | |
517 | * ring_buffer_wait - wait for input to the ring buffer | |
518 | * @buffer: buffer to wait on | |
519 | * @cpu: the cpu buffer to wait on | |
e30f53aa | 520 | * @full: wait until a full page is available, if @cpu != RING_BUFFER_ALL_CPUS |
15693458 SRRH |
521 | * |
522 | * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon | |
523 | * as data is added to any of the @buffer's cpu buffers. Otherwise | |
524 | * it will wait for data to be added to a specific cpu buffer. | |
525 | */ | |
e30f53aa | 526 | int ring_buffer_wait(struct ring_buffer *buffer, int cpu, bool full) |
15693458 | 527 | { |
e30f53aa | 528 | struct ring_buffer_per_cpu *uninitialized_var(cpu_buffer); |
15693458 SRRH |
529 | DEFINE_WAIT(wait); |
530 | struct rb_irq_work *work; | |
e30f53aa | 531 | int ret = 0; |
15693458 SRRH |
532 | |
533 | /* | |
534 | * Depending on what the caller is waiting for, either any | |
535 | * data in any cpu buffer, or a specific buffer, put the | |
536 | * caller on the appropriate wait queue. | |
537 | */ | |
1e0d6714 | 538 | if (cpu == RING_BUFFER_ALL_CPUS) { |
15693458 | 539 | work = &buffer->irq_work; |
1e0d6714 SRRH |
540 | /* Full only makes sense on per cpu reads */ |
541 | full = false; | |
542 | } else { | |
8b8b3683 SRRH |
543 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
544 | return -ENODEV; | |
15693458 SRRH |
545 | cpu_buffer = buffer->buffers[cpu]; |
546 | work = &cpu_buffer->irq_work; | |
547 | } | |
548 | ||
549 | ||
e30f53aa | 550 | while (true) { |
1e0d6714 SRRH |
551 | if (full) |
552 | prepare_to_wait(&work->full_waiters, &wait, TASK_INTERRUPTIBLE); | |
553 | else | |
554 | prepare_to_wait(&work->waiters, &wait, TASK_INTERRUPTIBLE); | |
e30f53aa RV |
555 | |
556 | /* | |
557 | * The events can happen in critical sections where | |
558 | * checking a work queue can cause deadlocks. | |
559 | * After adding a task to the queue, this flag is set | |
560 | * only to notify events to try to wake up the queue | |
561 | * using irq_work. | |
562 | * | |
563 | * We don't clear it even if the buffer is no longer | |
564 | * empty. The flag only causes the next event to run | |
565 | * irq_work to do the work queue wake up. The worse | |
566 | * that can happen if we race with !trace_empty() is that | |
567 | * an event will cause an irq_work to try to wake up | |
568 | * an empty queue. | |
569 | * | |
570 | * There's no reason to protect this flag either, as | |
571 | * the work queue and irq_work logic will do the necessary | |
572 | * synchronization for the wake ups. The only thing | |
573 | * that is necessary is that the wake up happens after | |
574 | * a task has been queued. It's OK for spurious wake ups. | |
575 | */ | |
1e0d6714 SRRH |
576 | if (full) |
577 | work->full_waiters_pending = true; | |
578 | else | |
579 | work->waiters_pending = true; | |
e30f53aa RV |
580 | |
581 | if (signal_pending(current)) { | |
582 | ret = -EINTR; | |
583 | break; | |
584 | } | |
585 | ||
586 | if (cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) | |
587 | break; | |
588 | ||
589 | if (cpu != RING_BUFFER_ALL_CPUS && | |
590 | !ring_buffer_empty_cpu(buffer, cpu)) { | |
591 | unsigned long flags; | |
592 | bool pagebusy; | |
593 | ||
594 | if (!full) | |
595 | break; | |
596 | ||
597 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
598 | pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page; | |
599 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
600 | ||
601 | if (!pagebusy) | |
602 | break; | |
603 | } | |
15693458 | 604 | |
15693458 | 605 | schedule(); |
e30f53aa | 606 | } |
15693458 | 607 | |
1e0d6714 SRRH |
608 | if (full) |
609 | finish_wait(&work->full_waiters, &wait); | |
610 | else | |
611 | finish_wait(&work->waiters, &wait); | |
e30f53aa RV |
612 | |
613 | return ret; | |
15693458 SRRH |
614 | } |
615 | ||
616 | /** | |
617 | * ring_buffer_poll_wait - poll on buffer input | |
618 | * @buffer: buffer to wait on | |
619 | * @cpu: the cpu buffer to wait on | |
620 | * @filp: the file descriptor | |
621 | * @poll_table: The poll descriptor | |
622 | * | |
623 | * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon | |
624 | * as data is added to any of the @buffer's cpu buffers. Otherwise | |
625 | * it will wait for data to be added to a specific cpu buffer. | |
626 | * | |
627 | * Returns POLLIN | POLLRDNORM if data exists in the buffers, | |
628 | * zero otherwise. | |
629 | */ | |
630 | int ring_buffer_poll_wait(struct ring_buffer *buffer, int cpu, | |
631 | struct file *filp, poll_table *poll_table) | |
632 | { | |
633 | struct ring_buffer_per_cpu *cpu_buffer; | |
634 | struct rb_irq_work *work; | |
635 | ||
15693458 SRRH |
636 | if (cpu == RING_BUFFER_ALL_CPUS) |
637 | work = &buffer->irq_work; | |
638 | else { | |
6721cb60 SRRH |
639 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
640 | return -EINVAL; | |
641 | ||
15693458 SRRH |
642 | cpu_buffer = buffer->buffers[cpu]; |
643 | work = &cpu_buffer->irq_work; | |
644 | } | |
645 | ||
15693458 | 646 | poll_wait(filp, &work->waiters, poll_table); |
4ce97dbf JB |
647 | work->waiters_pending = true; |
648 | /* | |
649 | * There's a tight race between setting the waiters_pending and | |
650 | * checking if the ring buffer is empty. Once the waiters_pending bit | |
651 | * is set, the next event will wake the task up, but we can get stuck | |
652 | * if there's only a single event in. | |
653 | * | |
654 | * FIXME: Ideally, we need a memory barrier on the writer side as well, | |
655 | * but adding a memory barrier to all events will cause too much of a | |
656 | * performance hit in the fast path. We only need a memory barrier when | |
657 | * the buffer goes from empty to having content. But as this race is | |
658 | * extremely small, and it's not a problem if another event comes in, we | |
659 | * will fix it later. | |
660 | */ | |
661 | smp_mb(); | |
15693458 SRRH |
662 | |
663 | if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) || | |
664 | (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu))) | |
665 | return POLLIN | POLLRDNORM; | |
666 | return 0; | |
667 | } | |
668 | ||
f536aafc | 669 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
077c5407 SR |
670 | #define RB_WARN_ON(b, cond) \ |
671 | ({ \ | |
672 | int _____ret = unlikely(cond); \ | |
673 | if (_____ret) { \ | |
674 | if (__same_type(*(b), struct ring_buffer_per_cpu)) { \ | |
675 | struct ring_buffer_per_cpu *__b = \ | |
676 | (void *)b; \ | |
677 | atomic_inc(&__b->buffer->record_disabled); \ | |
678 | } else \ | |
679 | atomic_inc(&b->record_disabled); \ | |
680 | WARN_ON(1); \ | |
681 | } \ | |
682 | _____ret; \ | |
3e89c7bb | 683 | }) |
f536aafc | 684 | |
37886f6a SR |
685 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
686 | #define DEBUG_SHIFT 0 | |
687 | ||
6d3f1e12 | 688 | static inline u64 rb_time_stamp(struct ring_buffer *buffer) |
88eb0125 SR |
689 | { |
690 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
691 | return buffer->clock() << DEBUG_SHIFT; | |
692 | } | |
693 | ||
37886f6a SR |
694 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) |
695 | { | |
696 | u64 time; | |
697 | ||
698 | preempt_disable_notrace(); | |
6d3f1e12 | 699 | time = rb_time_stamp(buffer); |
37886f6a SR |
700 | preempt_enable_no_resched_notrace(); |
701 | ||
702 | return time; | |
703 | } | |
704 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
705 | ||
706 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
707 | int cpu, u64 *ts) | |
708 | { | |
709 | /* Just stupid testing the normalize function and deltas */ | |
710 | *ts >>= DEBUG_SHIFT; | |
711 | } | |
712 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
713 | ||
77ae365e SR |
714 | /* |
715 | * Making the ring buffer lockless makes things tricky. | |
716 | * Although writes only happen on the CPU that they are on, | |
717 | * and they only need to worry about interrupts. Reads can | |
718 | * happen on any CPU. | |
719 | * | |
720 | * The reader page is always off the ring buffer, but when the | |
721 | * reader finishes with a page, it needs to swap its page with | |
722 | * a new one from the buffer. The reader needs to take from | |
723 | * the head (writes go to the tail). But if a writer is in overwrite | |
724 | * mode and wraps, it must push the head page forward. | |
725 | * | |
726 | * Here lies the problem. | |
727 | * | |
728 | * The reader must be careful to replace only the head page, and | |
729 | * not another one. As described at the top of the file in the | |
730 | * ASCII art, the reader sets its old page to point to the next | |
731 | * page after head. It then sets the page after head to point to | |
732 | * the old reader page. But if the writer moves the head page | |
733 | * during this operation, the reader could end up with the tail. | |
734 | * | |
735 | * We use cmpxchg to help prevent this race. We also do something | |
736 | * special with the page before head. We set the LSB to 1. | |
737 | * | |
738 | * When the writer must push the page forward, it will clear the | |
739 | * bit that points to the head page, move the head, and then set | |
740 | * the bit that points to the new head page. | |
741 | * | |
742 | * We also don't want an interrupt coming in and moving the head | |
743 | * page on another writer. Thus we use the second LSB to catch | |
744 | * that too. Thus: | |
745 | * | |
746 | * head->list->prev->next bit 1 bit 0 | |
747 | * ------- ------- | |
748 | * Normal page 0 0 | |
749 | * Points to head page 0 1 | |
750 | * New head page 1 0 | |
751 | * | |
752 | * Note we can not trust the prev pointer of the head page, because: | |
753 | * | |
754 | * +----+ +-----+ +-----+ | |
755 | * | |------>| T |---X--->| N | | |
756 | * | |<------| | | | | |
757 | * +----+ +-----+ +-----+ | |
758 | * ^ ^ | | |
759 | * | +-----+ | | | |
760 | * +----------| R |----------+ | | |
761 | * | |<-----------+ | |
762 | * +-----+ | |
763 | * | |
764 | * Key: ---X--> HEAD flag set in pointer | |
765 | * T Tail page | |
766 | * R Reader page | |
767 | * N Next page | |
768 | * | |
769 | * (see __rb_reserve_next() to see where this happens) | |
770 | * | |
771 | * What the above shows is that the reader just swapped out | |
772 | * the reader page with a page in the buffer, but before it | |
773 | * could make the new header point back to the new page added | |
774 | * it was preempted by a writer. The writer moved forward onto | |
775 | * the new page added by the reader and is about to move forward | |
776 | * again. | |
777 | * | |
778 | * You can see, it is legitimate for the previous pointer of | |
779 | * the head (or any page) not to point back to itself. But only | |
780 | * temporarially. | |
781 | */ | |
782 | ||
783 | #define RB_PAGE_NORMAL 0UL | |
784 | #define RB_PAGE_HEAD 1UL | |
785 | #define RB_PAGE_UPDATE 2UL | |
786 | ||
787 | ||
788 | #define RB_FLAG_MASK 3UL | |
789 | ||
790 | /* PAGE_MOVED is not part of the mask */ | |
791 | #define RB_PAGE_MOVED 4UL | |
792 | ||
793 | /* | |
794 | * rb_list_head - remove any bit | |
795 | */ | |
796 | static struct list_head *rb_list_head(struct list_head *list) | |
797 | { | |
798 | unsigned long val = (unsigned long)list; | |
799 | ||
800 | return (struct list_head *)(val & ~RB_FLAG_MASK); | |
801 | } | |
802 | ||
803 | /* | |
6d3f1e12 | 804 | * rb_is_head_page - test if the given page is the head page |
77ae365e SR |
805 | * |
806 | * Because the reader may move the head_page pointer, we can | |
807 | * not trust what the head page is (it may be pointing to | |
808 | * the reader page). But if the next page is a header page, | |
809 | * its flags will be non zero. | |
810 | */ | |
42b16b3f | 811 | static inline int |
77ae365e SR |
812 | rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer, |
813 | struct buffer_page *page, struct list_head *list) | |
814 | { | |
815 | unsigned long val; | |
816 | ||
817 | val = (unsigned long)list->next; | |
818 | ||
819 | if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list) | |
820 | return RB_PAGE_MOVED; | |
821 | ||
822 | return val & RB_FLAG_MASK; | |
823 | } | |
824 | ||
825 | /* | |
826 | * rb_is_reader_page | |
827 | * | |
828 | * The unique thing about the reader page, is that, if the | |
829 | * writer is ever on it, the previous pointer never points | |
830 | * back to the reader page. | |
831 | */ | |
832 | static int rb_is_reader_page(struct buffer_page *page) | |
833 | { | |
834 | struct list_head *list = page->list.prev; | |
835 | ||
836 | return rb_list_head(list->next) != &page->list; | |
837 | } | |
838 | ||
839 | /* | |
840 | * rb_set_list_to_head - set a list_head to be pointing to head. | |
841 | */ | |
842 | static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer, | |
843 | struct list_head *list) | |
844 | { | |
845 | unsigned long *ptr; | |
846 | ||
847 | ptr = (unsigned long *)&list->next; | |
848 | *ptr |= RB_PAGE_HEAD; | |
849 | *ptr &= ~RB_PAGE_UPDATE; | |
850 | } | |
851 | ||
852 | /* | |
853 | * rb_head_page_activate - sets up head page | |
854 | */ | |
855 | static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer) | |
856 | { | |
857 | struct buffer_page *head; | |
858 | ||
859 | head = cpu_buffer->head_page; | |
860 | if (!head) | |
861 | return; | |
862 | ||
863 | /* | |
864 | * Set the previous list pointer to have the HEAD flag. | |
865 | */ | |
866 | rb_set_list_to_head(cpu_buffer, head->list.prev); | |
867 | } | |
868 | ||
869 | static void rb_list_head_clear(struct list_head *list) | |
870 | { | |
871 | unsigned long *ptr = (unsigned long *)&list->next; | |
872 | ||
873 | *ptr &= ~RB_FLAG_MASK; | |
874 | } | |
875 | ||
876 | /* | |
877 | * rb_head_page_dactivate - clears head page ptr (for free list) | |
878 | */ | |
879 | static void | |
880 | rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer) | |
881 | { | |
882 | struct list_head *hd; | |
883 | ||
884 | /* Go through the whole list and clear any pointers found. */ | |
885 | rb_list_head_clear(cpu_buffer->pages); | |
886 | ||
887 | list_for_each(hd, cpu_buffer->pages) | |
888 | rb_list_head_clear(hd); | |
889 | } | |
890 | ||
891 | static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer, | |
892 | struct buffer_page *head, | |
893 | struct buffer_page *prev, | |
894 | int old_flag, int new_flag) | |
895 | { | |
896 | struct list_head *list; | |
897 | unsigned long val = (unsigned long)&head->list; | |
898 | unsigned long ret; | |
899 | ||
900 | list = &prev->list; | |
901 | ||
902 | val &= ~RB_FLAG_MASK; | |
903 | ||
08a40816 SR |
904 | ret = cmpxchg((unsigned long *)&list->next, |
905 | val | old_flag, val | new_flag); | |
77ae365e SR |
906 | |
907 | /* check if the reader took the page */ | |
908 | if ((ret & ~RB_FLAG_MASK) != val) | |
909 | return RB_PAGE_MOVED; | |
910 | ||
911 | return ret & RB_FLAG_MASK; | |
912 | } | |
913 | ||
914 | static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer, | |
915 | struct buffer_page *head, | |
916 | struct buffer_page *prev, | |
917 | int old_flag) | |
918 | { | |
919 | return rb_head_page_set(cpu_buffer, head, prev, | |
920 | old_flag, RB_PAGE_UPDATE); | |
921 | } | |
922 | ||
923 | static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer, | |
924 | struct buffer_page *head, | |
925 | struct buffer_page *prev, | |
926 | int old_flag) | |
927 | { | |
928 | return rb_head_page_set(cpu_buffer, head, prev, | |
929 | old_flag, RB_PAGE_HEAD); | |
930 | } | |
931 | ||
932 | static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer, | |
933 | struct buffer_page *head, | |
934 | struct buffer_page *prev, | |
935 | int old_flag) | |
936 | { | |
937 | return rb_head_page_set(cpu_buffer, head, prev, | |
938 | old_flag, RB_PAGE_NORMAL); | |
939 | } | |
940 | ||
941 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
942 | struct buffer_page **bpage) | |
943 | { | |
944 | struct list_head *p = rb_list_head((*bpage)->list.next); | |
945 | ||
946 | *bpage = list_entry(p, struct buffer_page, list); | |
947 | } | |
948 | ||
949 | static struct buffer_page * | |
950 | rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer) | |
951 | { | |
952 | struct buffer_page *head; | |
953 | struct buffer_page *page; | |
954 | struct list_head *list; | |
955 | int i; | |
956 | ||
957 | if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page)) | |
958 | return NULL; | |
959 | ||
960 | /* sanity check */ | |
961 | list = cpu_buffer->pages; | |
962 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list)) | |
963 | return NULL; | |
964 | ||
965 | page = head = cpu_buffer->head_page; | |
966 | /* | |
967 | * It is possible that the writer moves the header behind | |
968 | * where we started, and we miss in one loop. | |
969 | * A second loop should grab the header, but we'll do | |
970 | * three loops just because I'm paranoid. | |
971 | */ | |
972 | for (i = 0; i < 3; i++) { | |
973 | do { | |
974 | if (rb_is_head_page(cpu_buffer, page, page->list.prev)) { | |
975 | cpu_buffer->head_page = page; | |
976 | return page; | |
977 | } | |
978 | rb_inc_page(cpu_buffer, &page); | |
979 | } while (page != head); | |
980 | } | |
981 | ||
982 | RB_WARN_ON(cpu_buffer, 1); | |
983 | ||
984 | return NULL; | |
985 | } | |
986 | ||
987 | static int rb_head_page_replace(struct buffer_page *old, | |
988 | struct buffer_page *new) | |
989 | { | |
990 | unsigned long *ptr = (unsigned long *)&old->list.prev->next; | |
991 | unsigned long val; | |
992 | unsigned long ret; | |
993 | ||
994 | val = *ptr & ~RB_FLAG_MASK; | |
995 | val |= RB_PAGE_HEAD; | |
996 | ||
08a40816 | 997 | ret = cmpxchg(ptr, val, (unsigned long)&new->list); |
77ae365e SR |
998 | |
999 | return ret == val; | |
1000 | } | |
1001 | ||
1002 | /* | |
1003 | * rb_tail_page_update - move the tail page forward | |
1004 | * | |
1005 | * Returns 1 if moved tail page, 0 if someone else did. | |
1006 | */ | |
1007 | static int rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer, | |
1008 | struct buffer_page *tail_page, | |
1009 | struct buffer_page *next_page) | |
1010 | { | |
1011 | struct buffer_page *old_tail; | |
1012 | unsigned long old_entries; | |
1013 | unsigned long old_write; | |
1014 | int ret = 0; | |
1015 | ||
1016 | /* | |
1017 | * The tail page now needs to be moved forward. | |
1018 | * | |
1019 | * We need to reset the tail page, but without messing | |
1020 | * with possible erasing of data brought in by interrupts | |
1021 | * that have moved the tail page and are currently on it. | |
1022 | * | |
1023 | * We add a counter to the write field to denote this. | |
1024 | */ | |
1025 | old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write); | |
1026 | old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries); | |
1027 | ||
1028 | /* | |
1029 | * Just make sure we have seen our old_write and synchronize | |
1030 | * with any interrupts that come in. | |
1031 | */ | |
1032 | barrier(); | |
1033 | ||
1034 | /* | |
1035 | * If the tail page is still the same as what we think | |
1036 | * it is, then it is up to us to update the tail | |
1037 | * pointer. | |
1038 | */ | |
1039 | if (tail_page == cpu_buffer->tail_page) { | |
1040 | /* Zero the write counter */ | |
1041 | unsigned long val = old_write & ~RB_WRITE_MASK; | |
1042 | unsigned long eval = old_entries & ~RB_WRITE_MASK; | |
1043 | ||
1044 | /* | |
1045 | * This will only succeed if an interrupt did | |
1046 | * not come in and change it. In which case, we | |
1047 | * do not want to modify it. | |
da706d8b LJ |
1048 | * |
1049 | * We add (void) to let the compiler know that we do not care | |
1050 | * about the return value of these functions. We use the | |
1051 | * cmpxchg to only update if an interrupt did not already | |
1052 | * do it for us. If the cmpxchg fails, we don't care. | |
77ae365e | 1053 | */ |
da706d8b LJ |
1054 | (void)local_cmpxchg(&next_page->write, old_write, val); |
1055 | (void)local_cmpxchg(&next_page->entries, old_entries, eval); | |
77ae365e SR |
1056 | |
1057 | /* | |
1058 | * No need to worry about races with clearing out the commit. | |
1059 | * it only can increment when a commit takes place. But that | |
1060 | * only happens in the outer most nested commit. | |
1061 | */ | |
1062 | local_set(&next_page->page->commit, 0); | |
1063 | ||
1064 | old_tail = cmpxchg(&cpu_buffer->tail_page, | |
1065 | tail_page, next_page); | |
1066 | ||
1067 | if (old_tail == tail_page) | |
1068 | ret = 1; | |
1069 | } | |
1070 | ||
1071 | return ret; | |
1072 | } | |
1073 | ||
1074 | static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer, | |
1075 | struct buffer_page *bpage) | |
1076 | { | |
1077 | unsigned long val = (unsigned long)bpage; | |
1078 | ||
1079 | if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK)) | |
1080 | return 1; | |
1081 | ||
1082 | return 0; | |
1083 | } | |
1084 | ||
1085 | /** | |
1086 | * rb_check_list - make sure a pointer to a list has the last bits zero | |
1087 | */ | |
1088 | static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer, | |
1089 | struct list_head *list) | |
1090 | { | |
1091 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev)) | |
1092 | return 1; | |
1093 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next)) | |
1094 | return 1; | |
1095 | return 0; | |
1096 | } | |
1097 | ||
7a8e76a3 | 1098 | /** |
d611851b | 1099 | * rb_check_pages - integrity check of buffer pages |
7a8e76a3 SR |
1100 | * @cpu_buffer: CPU buffer with pages to test |
1101 | * | |
c3706f00 | 1102 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
1103 | * been corrupted. |
1104 | */ | |
1105 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
1106 | { | |
3adc54fa | 1107 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1108 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1109 | |
308f7eeb SR |
1110 | /* Reset the head page if it exists */ |
1111 | if (cpu_buffer->head_page) | |
1112 | rb_set_head_page(cpu_buffer); | |
1113 | ||
77ae365e SR |
1114 | rb_head_page_deactivate(cpu_buffer); |
1115 | ||
3e89c7bb SR |
1116 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
1117 | return -1; | |
1118 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
1119 | return -1; | |
7a8e76a3 | 1120 | |
77ae365e SR |
1121 | if (rb_check_list(cpu_buffer, head)) |
1122 | return -1; | |
1123 | ||
044fa782 | 1124 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 1125 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 1126 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
1127 | return -1; |
1128 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 1129 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 1130 | return -1; |
77ae365e SR |
1131 | if (rb_check_list(cpu_buffer, &bpage->list)) |
1132 | return -1; | |
7a8e76a3 SR |
1133 | } |
1134 | ||
77ae365e SR |
1135 | rb_head_page_activate(cpu_buffer); |
1136 | ||
7a8e76a3 SR |
1137 | return 0; |
1138 | } | |
1139 | ||
438ced17 | 1140 | static int __rb_allocate_pages(int nr_pages, struct list_head *pages, int cpu) |
7a8e76a3 | 1141 | { |
438ced17 | 1142 | int i; |
044fa782 | 1143 | struct buffer_page *bpage, *tmp; |
3adc54fa | 1144 | |
7a8e76a3 | 1145 | for (i = 0; i < nr_pages; i++) { |
7ea59064 | 1146 | struct page *page; |
d7ec4bfe VN |
1147 | /* |
1148 | * __GFP_NORETRY flag makes sure that the allocation fails | |
1149 | * gracefully without invoking oom-killer and the system is | |
1150 | * not destabilized. | |
1151 | */ | |
044fa782 | 1152 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
d7ec4bfe | 1153 | GFP_KERNEL | __GFP_NORETRY, |
438ced17 | 1154 | cpu_to_node(cpu)); |
044fa782 | 1155 | if (!bpage) |
e4c2ce82 | 1156 | goto free_pages; |
77ae365e | 1157 | |
438ced17 | 1158 | list_add(&bpage->list, pages); |
77ae365e | 1159 | |
438ced17 | 1160 | page = alloc_pages_node(cpu_to_node(cpu), |
d7ec4bfe | 1161 | GFP_KERNEL | __GFP_NORETRY, 0); |
7ea59064 | 1162 | if (!page) |
7a8e76a3 | 1163 | goto free_pages; |
7ea59064 | 1164 | bpage->page = page_address(page); |
044fa782 | 1165 | rb_init_page(bpage->page); |
7a8e76a3 SR |
1166 | } |
1167 | ||
438ced17 VN |
1168 | return 0; |
1169 | ||
1170 | free_pages: | |
1171 | list_for_each_entry_safe(bpage, tmp, pages, list) { | |
1172 | list_del_init(&bpage->list); | |
1173 | free_buffer_page(bpage); | |
1174 | } | |
1175 | ||
1176 | return -ENOMEM; | |
1177 | } | |
1178 | ||
1179 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
1180 | unsigned nr_pages) | |
1181 | { | |
1182 | LIST_HEAD(pages); | |
1183 | ||
1184 | WARN_ON(!nr_pages); | |
1185 | ||
1186 | if (__rb_allocate_pages(nr_pages, &pages, cpu_buffer->cpu)) | |
1187 | return -ENOMEM; | |
1188 | ||
3adc54fa SR |
1189 | /* |
1190 | * The ring buffer page list is a circular list that does not | |
1191 | * start and end with a list head. All page list items point to | |
1192 | * other pages. | |
1193 | */ | |
1194 | cpu_buffer->pages = pages.next; | |
1195 | list_del(&pages); | |
7a8e76a3 | 1196 | |
438ced17 VN |
1197 | cpu_buffer->nr_pages = nr_pages; |
1198 | ||
7a8e76a3 SR |
1199 | rb_check_pages(cpu_buffer); |
1200 | ||
1201 | return 0; | |
7a8e76a3 SR |
1202 | } |
1203 | ||
1204 | static struct ring_buffer_per_cpu * | |
438ced17 | 1205 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int nr_pages, int cpu) |
7a8e76a3 SR |
1206 | { |
1207 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 1208 | struct buffer_page *bpage; |
7ea59064 | 1209 | struct page *page; |
7a8e76a3 SR |
1210 | int ret; |
1211 | ||
1212 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
1213 | GFP_KERNEL, cpu_to_node(cpu)); | |
1214 | if (!cpu_buffer) | |
1215 | return NULL; | |
1216 | ||
1217 | cpu_buffer->cpu = cpu; | |
1218 | cpu_buffer->buffer = buffer; | |
5389f6fa | 1219 | raw_spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 1220 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
edc35bd7 | 1221 | cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
83f40318 | 1222 | INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler); |
05fdd70d | 1223 | init_completion(&cpu_buffer->update_done); |
15693458 | 1224 | init_irq_work(&cpu_buffer->irq_work.work, rb_wake_up_waiters); |
f1dc6725 | 1225 | init_waitqueue_head(&cpu_buffer->irq_work.waiters); |
1e0d6714 | 1226 | init_waitqueue_head(&cpu_buffer->irq_work.full_waiters); |
7a8e76a3 | 1227 | |
044fa782 | 1228 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 1229 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 1230 | if (!bpage) |
e4c2ce82 SR |
1231 | goto fail_free_buffer; |
1232 | ||
77ae365e SR |
1233 | rb_check_bpage(cpu_buffer, bpage); |
1234 | ||
044fa782 | 1235 | cpu_buffer->reader_page = bpage; |
7ea59064 VN |
1236 | page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0); |
1237 | if (!page) | |
e4c2ce82 | 1238 | goto fail_free_reader; |
7ea59064 | 1239 | bpage->page = page_address(page); |
044fa782 | 1240 | rb_init_page(bpage->page); |
e4c2ce82 | 1241 | |
d769041f | 1242 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
44b99462 | 1243 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
d769041f | 1244 | |
438ced17 | 1245 | ret = rb_allocate_pages(cpu_buffer, nr_pages); |
7a8e76a3 | 1246 | if (ret < 0) |
d769041f | 1247 | goto fail_free_reader; |
7a8e76a3 SR |
1248 | |
1249 | cpu_buffer->head_page | |
3adc54fa | 1250 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 1251 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 | 1252 | |
77ae365e SR |
1253 | rb_head_page_activate(cpu_buffer); |
1254 | ||
7a8e76a3 SR |
1255 | return cpu_buffer; |
1256 | ||
d769041f SR |
1257 | fail_free_reader: |
1258 | free_buffer_page(cpu_buffer->reader_page); | |
1259 | ||
7a8e76a3 SR |
1260 | fail_free_buffer: |
1261 | kfree(cpu_buffer); | |
1262 | return NULL; | |
1263 | } | |
1264 | ||
1265 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
1266 | { | |
3adc54fa | 1267 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1268 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1269 | |
d769041f SR |
1270 | free_buffer_page(cpu_buffer->reader_page); |
1271 | ||
77ae365e SR |
1272 | rb_head_page_deactivate(cpu_buffer); |
1273 | ||
3adc54fa SR |
1274 | if (head) { |
1275 | list_for_each_entry_safe(bpage, tmp, head, list) { | |
1276 | list_del_init(&bpage->list); | |
1277 | free_buffer_page(bpage); | |
1278 | } | |
1279 | bpage = list_entry(head, struct buffer_page, list); | |
044fa782 | 1280 | free_buffer_page(bpage); |
7a8e76a3 | 1281 | } |
3adc54fa | 1282 | |
7a8e76a3 SR |
1283 | kfree(cpu_buffer); |
1284 | } | |
1285 | ||
59222efe | 1286 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
1287 | static int rb_cpu_notify(struct notifier_block *self, |
1288 | unsigned long action, void *hcpu); | |
554f786e SR |
1289 | #endif |
1290 | ||
7a8e76a3 | 1291 | /** |
d611851b | 1292 | * __ring_buffer_alloc - allocate a new ring_buffer |
68814b58 | 1293 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
1294 | * @flags: attributes to set for the ring buffer. |
1295 | * | |
1296 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
1297 | * flag. This flag means that the buffer will overwrite old data | |
1298 | * when the buffer wraps. If this flag is not set, the buffer will | |
1299 | * drop data when the tail hits the head. | |
1300 | */ | |
1f8a6a10 PZ |
1301 | struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
1302 | struct lock_class_key *key) | |
7a8e76a3 SR |
1303 | { |
1304 | struct ring_buffer *buffer; | |
1305 | int bsize; | |
438ced17 | 1306 | int cpu, nr_pages; |
7a8e76a3 SR |
1307 | |
1308 | /* keep it in its own cache line */ | |
1309 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
1310 | GFP_KERNEL); | |
1311 | if (!buffer) | |
1312 | return NULL; | |
1313 | ||
9e01c1b7 RR |
1314 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
1315 | goto fail_free_buffer; | |
1316 | ||
438ced17 | 1317 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
7a8e76a3 | 1318 | buffer->flags = flags; |
37886f6a | 1319 | buffer->clock = trace_clock_local; |
1f8a6a10 | 1320 | buffer->reader_lock_key = key; |
7a8e76a3 | 1321 | |
15693458 | 1322 | init_irq_work(&buffer->irq_work.work, rb_wake_up_waiters); |
f1dc6725 | 1323 | init_waitqueue_head(&buffer->irq_work.waiters); |
15693458 | 1324 | |
7a8e76a3 | 1325 | /* need at least two pages */ |
438ced17 VN |
1326 | if (nr_pages < 2) |
1327 | nr_pages = 2; | |
7a8e76a3 | 1328 | |
3bf832ce FW |
1329 | /* |
1330 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
1331 | * in early initcall, it will not be notified of secondary cpus. | |
1332 | * In that off case, we need to allocate for all possible cpus. | |
1333 | */ | |
1334 | #ifdef CONFIG_HOTPLUG_CPU | |
d39ad278 | 1335 | cpu_notifier_register_begin(); |
554f786e | 1336 | cpumask_copy(buffer->cpumask, cpu_online_mask); |
3bf832ce FW |
1337 | #else |
1338 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
1339 | #endif | |
7a8e76a3 SR |
1340 | buffer->cpus = nr_cpu_ids; |
1341 | ||
1342 | bsize = sizeof(void *) * nr_cpu_ids; | |
1343 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
1344 | GFP_KERNEL); | |
1345 | if (!buffer->buffers) | |
9e01c1b7 | 1346 | goto fail_free_cpumask; |
7a8e76a3 SR |
1347 | |
1348 | for_each_buffer_cpu(buffer, cpu) { | |
1349 | buffer->buffers[cpu] = | |
438ced17 | 1350 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
7a8e76a3 SR |
1351 | if (!buffer->buffers[cpu]) |
1352 | goto fail_free_buffers; | |
1353 | } | |
1354 | ||
59222efe | 1355 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1356 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
1357 | buffer->cpu_notify.priority = 0; | |
d39ad278 SB |
1358 | __register_cpu_notifier(&buffer->cpu_notify); |
1359 | cpu_notifier_register_done(); | |
554f786e SR |
1360 | #endif |
1361 | ||
7a8e76a3 SR |
1362 | mutex_init(&buffer->mutex); |
1363 | ||
1364 | return buffer; | |
1365 | ||
1366 | fail_free_buffers: | |
1367 | for_each_buffer_cpu(buffer, cpu) { | |
1368 | if (buffer->buffers[cpu]) | |
1369 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1370 | } | |
1371 | kfree(buffer->buffers); | |
1372 | ||
9e01c1b7 RR |
1373 | fail_free_cpumask: |
1374 | free_cpumask_var(buffer->cpumask); | |
d39ad278 SB |
1375 | #ifdef CONFIG_HOTPLUG_CPU |
1376 | cpu_notifier_register_done(); | |
1377 | #endif | |
9e01c1b7 | 1378 | |
7a8e76a3 SR |
1379 | fail_free_buffer: |
1380 | kfree(buffer); | |
1381 | return NULL; | |
1382 | } | |
1f8a6a10 | 1383 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
1384 | |
1385 | /** | |
1386 | * ring_buffer_free - free a ring buffer. | |
1387 | * @buffer: the buffer to free. | |
1388 | */ | |
1389 | void | |
1390 | ring_buffer_free(struct ring_buffer *buffer) | |
1391 | { | |
1392 | int cpu; | |
1393 | ||
59222efe | 1394 | #ifdef CONFIG_HOTPLUG_CPU |
d39ad278 SB |
1395 | cpu_notifier_register_begin(); |
1396 | __unregister_cpu_notifier(&buffer->cpu_notify); | |
554f786e SR |
1397 | #endif |
1398 | ||
7a8e76a3 SR |
1399 | for_each_buffer_cpu(buffer, cpu) |
1400 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1401 | ||
d39ad278 SB |
1402 | #ifdef CONFIG_HOTPLUG_CPU |
1403 | cpu_notifier_register_done(); | |
1404 | #endif | |
554f786e | 1405 | |
bd3f0221 | 1406 | kfree(buffer->buffers); |
9e01c1b7 RR |
1407 | free_cpumask_var(buffer->cpumask); |
1408 | ||
7a8e76a3 SR |
1409 | kfree(buffer); |
1410 | } | |
c4f50183 | 1411 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 1412 | |
37886f6a SR |
1413 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
1414 | u64 (*clock)(void)) | |
1415 | { | |
1416 | buffer->clock = clock; | |
1417 | } | |
1418 | ||
7a8e76a3 SR |
1419 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
1420 | ||
83f40318 VN |
1421 | static inline unsigned long rb_page_entries(struct buffer_page *bpage) |
1422 | { | |
1423 | return local_read(&bpage->entries) & RB_WRITE_MASK; | |
1424 | } | |
1425 | ||
1426 | static inline unsigned long rb_page_write(struct buffer_page *bpage) | |
1427 | { | |
1428 | return local_read(&bpage->write) & RB_WRITE_MASK; | |
1429 | } | |
1430 | ||
5040b4b7 | 1431 | static int |
83f40318 | 1432 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned int nr_pages) |
7a8e76a3 | 1433 | { |
83f40318 VN |
1434 | struct list_head *tail_page, *to_remove, *next_page; |
1435 | struct buffer_page *to_remove_page, *tmp_iter_page; | |
1436 | struct buffer_page *last_page, *first_page; | |
1437 | unsigned int nr_removed; | |
1438 | unsigned long head_bit; | |
1439 | int page_entries; | |
1440 | ||
1441 | head_bit = 0; | |
7a8e76a3 | 1442 | |
5389f6fa | 1443 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1444 | atomic_inc(&cpu_buffer->record_disabled); |
1445 | /* | |
1446 | * We don't race with the readers since we have acquired the reader | |
1447 | * lock. We also don't race with writers after disabling recording. | |
1448 | * This makes it easy to figure out the first and the last page to be | |
1449 | * removed from the list. We unlink all the pages in between including | |
1450 | * the first and last pages. This is done in a busy loop so that we | |
1451 | * lose the least number of traces. | |
1452 | * The pages are freed after we restart recording and unlock readers. | |
1453 | */ | |
1454 | tail_page = &cpu_buffer->tail_page->list; | |
77ae365e | 1455 | |
83f40318 VN |
1456 | /* |
1457 | * tail page might be on reader page, we remove the next page | |
1458 | * from the ring buffer | |
1459 | */ | |
1460 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
1461 | tail_page = rb_list_head(tail_page->next); | |
1462 | to_remove = tail_page; | |
1463 | ||
1464 | /* start of pages to remove */ | |
1465 | first_page = list_entry(rb_list_head(to_remove->next), | |
1466 | struct buffer_page, list); | |
1467 | ||
1468 | for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) { | |
1469 | to_remove = rb_list_head(to_remove)->next; | |
1470 | head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD; | |
7a8e76a3 | 1471 | } |
7a8e76a3 | 1472 | |
83f40318 | 1473 | next_page = rb_list_head(to_remove)->next; |
7a8e76a3 | 1474 | |
83f40318 VN |
1475 | /* |
1476 | * Now we remove all pages between tail_page and next_page. | |
1477 | * Make sure that we have head_bit value preserved for the | |
1478 | * next page | |
1479 | */ | |
1480 | tail_page->next = (struct list_head *)((unsigned long)next_page | | |
1481 | head_bit); | |
1482 | next_page = rb_list_head(next_page); | |
1483 | next_page->prev = tail_page; | |
1484 | ||
1485 | /* make sure pages points to a valid page in the ring buffer */ | |
1486 | cpu_buffer->pages = next_page; | |
1487 | ||
1488 | /* update head page */ | |
1489 | if (head_bit) | |
1490 | cpu_buffer->head_page = list_entry(next_page, | |
1491 | struct buffer_page, list); | |
1492 | ||
1493 | /* | |
1494 | * change read pointer to make sure any read iterators reset | |
1495 | * themselves | |
1496 | */ | |
1497 | cpu_buffer->read = 0; | |
1498 | ||
1499 | /* pages are removed, resume tracing and then free the pages */ | |
1500 | atomic_dec(&cpu_buffer->record_disabled); | |
5389f6fa | 1501 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1502 | |
1503 | RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)); | |
1504 | ||
1505 | /* last buffer page to remove */ | |
1506 | last_page = list_entry(rb_list_head(to_remove), struct buffer_page, | |
1507 | list); | |
1508 | tmp_iter_page = first_page; | |
1509 | ||
1510 | do { | |
1511 | to_remove_page = tmp_iter_page; | |
1512 | rb_inc_page(cpu_buffer, &tmp_iter_page); | |
1513 | ||
1514 | /* update the counters */ | |
1515 | page_entries = rb_page_entries(to_remove_page); | |
1516 | if (page_entries) { | |
1517 | /* | |
1518 | * If something was added to this page, it was full | |
1519 | * since it is not the tail page. So we deduct the | |
1520 | * bytes consumed in ring buffer from here. | |
48fdc72f | 1521 | * Increment overrun to account for the lost events. |
83f40318 | 1522 | */ |
48fdc72f | 1523 | local_add(page_entries, &cpu_buffer->overrun); |
83f40318 VN |
1524 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
1525 | } | |
1526 | ||
1527 | /* | |
1528 | * We have already removed references to this list item, just | |
1529 | * free up the buffer_page and its page | |
1530 | */ | |
1531 | free_buffer_page(to_remove_page); | |
1532 | nr_removed--; | |
1533 | ||
1534 | } while (to_remove_page != last_page); | |
1535 | ||
1536 | RB_WARN_ON(cpu_buffer, nr_removed); | |
5040b4b7 VN |
1537 | |
1538 | return nr_removed == 0; | |
7a8e76a3 SR |
1539 | } |
1540 | ||
5040b4b7 VN |
1541 | static int |
1542 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 1543 | { |
5040b4b7 VN |
1544 | struct list_head *pages = &cpu_buffer->new_pages; |
1545 | int retries, success; | |
7a8e76a3 | 1546 | |
5389f6fa | 1547 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1548 | /* |
1549 | * We are holding the reader lock, so the reader page won't be swapped | |
1550 | * in the ring buffer. Now we are racing with the writer trying to | |
1551 | * move head page and the tail page. | |
1552 | * We are going to adapt the reader page update process where: | |
1553 | * 1. We first splice the start and end of list of new pages between | |
1554 | * the head page and its previous page. | |
1555 | * 2. We cmpxchg the prev_page->next to point from head page to the | |
1556 | * start of new pages list. | |
1557 | * 3. Finally, we update the head->prev to the end of new list. | |
1558 | * | |
1559 | * We will try this process 10 times, to make sure that we don't keep | |
1560 | * spinning. | |
1561 | */ | |
1562 | retries = 10; | |
1563 | success = 0; | |
1564 | while (retries--) { | |
1565 | struct list_head *head_page, *prev_page, *r; | |
1566 | struct list_head *last_page, *first_page; | |
1567 | struct list_head *head_page_with_bit; | |
77ae365e | 1568 | |
5040b4b7 | 1569 | head_page = &rb_set_head_page(cpu_buffer)->list; |
54f7be5b SR |
1570 | if (!head_page) |
1571 | break; | |
5040b4b7 VN |
1572 | prev_page = head_page->prev; |
1573 | ||
1574 | first_page = pages->next; | |
1575 | last_page = pages->prev; | |
1576 | ||
1577 | head_page_with_bit = (struct list_head *) | |
1578 | ((unsigned long)head_page | RB_PAGE_HEAD); | |
1579 | ||
1580 | last_page->next = head_page_with_bit; | |
1581 | first_page->prev = prev_page; | |
1582 | ||
1583 | r = cmpxchg(&prev_page->next, head_page_with_bit, first_page); | |
1584 | ||
1585 | if (r == head_page_with_bit) { | |
1586 | /* | |
1587 | * yay, we replaced the page pointer to our new list, | |
1588 | * now, we just have to update to head page's prev | |
1589 | * pointer to point to end of list | |
1590 | */ | |
1591 | head_page->prev = last_page; | |
1592 | success = 1; | |
1593 | break; | |
1594 | } | |
7a8e76a3 | 1595 | } |
7a8e76a3 | 1596 | |
5040b4b7 VN |
1597 | if (success) |
1598 | INIT_LIST_HEAD(pages); | |
1599 | /* | |
1600 | * If we weren't successful in adding in new pages, warn and stop | |
1601 | * tracing | |
1602 | */ | |
1603 | RB_WARN_ON(cpu_buffer, !success); | |
5389f6fa | 1604 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1605 | |
1606 | /* free pages if they weren't inserted */ | |
1607 | if (!success) { | |
1608 | struct buffer_page *bpage, *tmp; | |
1609 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, | |
1610 | list) { | |
1611 | list_del_init(&bpage->list); | |
1612 | free_buffer_page(bpage); | |
1613 | } | |
1614 | } | |
1615 | return success; | |
7a8e76a3 SR |
1616 | } |
1617 | ||
83f40318 | 1618 | static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer) |
438ced17 | 1619 | { |
5040b4b7 VN |
1620 | int success; |
1621 | ||
438ced17 | 1622 | if (cpu_buffer->nr_pages_to_update > 0) |
5040b4b7 | 1623 | success = rb_insert_pages(cpu_buffer); |
438ced17 | 1624 | else |
5040b4b7 VN |
1625 | success = rb_remove_pages(cpu_buffer, |
1626 | -cpu_buffer->nr_pages_to_update); | |
83f40318 | 1627 | |
5040b4b7 VN |
1628 | if (success) |
1629 | cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update; | |
83f40318 VN |
1630 | } |
1631 | ||
1632 | static void update_pages_handler(struct work_struct *work) | |
1633 | { | |
1634 | struct ring_buffer_per_cpu *cpu_buffer = container_of(work, | |
1635 | struct ring_buffer_per_cpu, update_pages_work); | |
1636 | rb_update_pages(cpu_buffer); | |
05fdd70d | 1637 | complete(&cpu_buffer->update_done); |
438ced17 VN |
1638 | } |
1639 | ||
7a8e76a3 SR |
1640 | /** |
1641 | * ring_buffer_resize - resize the ring buffer | |
1642 | * @buffer: the buffer to resize. | |
1643 | * @size: the new size. | |
d611851b | 1644 | * @cpu_id: the cpu buffer to resize |
7a8e76a3 | 1645 | * |
7a8e76a3 SR |
1646 | * Minimum size is 2 * BUF_PAGE_SIZE. |
1647 | * | |
83f40318 | 1648 | * Returns 0 on success and < 0 on failure. |
7a8e76a3 | 1649 | */ |
438ced17 VN |
1650 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size, |
1651 | int cpu_id) | |
7a8e76a3 SR |
1652 | { |
1653 | struct ring_buffer_per_cpu *cpu_buffer; | |
438ced17 | 1654 | unsigned nr_pages; |
83f40318 | 1655 | int cpu, err = 0; |
7a8e76a3 | 1656 | |
ee51a1de IM |
1657 | /* |
1658 | * Always succeed at resizing a non-existent buffer: | |
1659 | */ | |
1660 | if (!buffer) | |
1661 | return size; | |
1662 | ||
6a31e1f1 SR |
1663 | /* Make sure the requested buffer exists */ |
1664 | if (cpu_id != RING_BUFFER_ALL_CPUS && | |
1665 | !cpumask_test_cpu(cpu_id, buffer->cpumask)) | |
1666 | return size; | |
1667 | ||
7a8e76a3 SR |
1668 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
1669 | size *= BUF_PAGE_SIZE; | |
7a8e76a3 SR |
1670 | |
1671 | /* we need a minimum of two pages */ | |
1672 | if (size < BUF_PAGE_SIZE * 2) | |
1673 | size = BUF_PAGE_SIZE * 2; | |
1674 | ||
83f40318 | 1675 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
18421015 | 1676 | |
83f40318 VN |
1677 | /* |
1678 | * Don't succeed if resizing is disabled, as a reader might be | |
1679 | * manipulating the ring buffer and is expecting a sane state while | |
1680 | * this is true. | |
1681 | */ | |
1682 | if (atomic_read(&buffer->resize_disabled)) | |
1683 | return -EBUSY; | |
18421015 | 1684 | |
83f40318 | 1685 | /* prevent another thread from changing buffer sizes */ |
7a8e76a3 | 1686 | mutex_lock(&buffer->mutex); |
7a8e76a3 | 1687 | |
438ced17 VN |
1688 | if (cpu_id == RING_BUFFER_ALL_CPUS) { |
1689 | /* calculate the pages to update */ | |
7a8e76a3 SR |
1690 | for_each_buffer_cpu(buffer, cpu) { |
1691 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 1692 | |
438ced17 VN |
1693 | cpu_buffer->nr_pages_to_update = nr_pages - |
1694 | cpu_buffer->nr_pages; | |
438ced17 VN |
1695 | /* |
1696 | * nothing more to do for removing pages or no update | |
1697 | */ | |
1698 | if (cpu_buffer->nr_pages_to_update <= 0) | |
1699 | continue; | |
d7ec4bfe | 1700 | /* |
438ced17 VN |
1701 | * to add pages, make sure all new pages can be |
1702 | * allocated without receiving ENOMEM | |
d7ec4bfe | 1703 | */ |
438ced17 VN |
1704 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
1705 | if (__rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 | 1706 | &cpu_buffer->new_pages, cpu)) { |
438ced17 | 1707 | /* not enough memory for new pages */ |
83f40318 VN |
1708 | err = -ENOMEM; |
1709 | goto out_err; | |
1710 | } | |
1711 | } | |
1712 | ||
1713 | get_online_cpus(); | |
1714 | /* | |
1715 | * Fire off all the required work handlers | |
05fdd70d | 1716 | * We can't schedule on offline CPUs, but it's not necessary |
83f40318 VN |
1717 | * since we can change their buffer sizes without any race. |
1718 | */ | |
1719 | for_each_buffer_cpu(buffer, cpu) { | |
1720 | cpu_buffer = buffer->buffers[cpu]; | |
05fdd70d | 1721 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
1722 | continue; |
1723 | ||
021c5b34 CM |
1724 | /* Can't run something on an offline CPU. */ |
1725 | if (!cpu_online(cpu)) { | |
f5eb5588 SRRH |
1726 | rb_update_pages(cpu_buffer); |
1727 | cpu_buffer->nr_pages_to_update = 0; | |
1728 | } else { | |
05fdd70d VN |
1729 | schedule_work_on(cpu, |
1730 | &cpu_buffer->update_pages_work); | |
f5eb5588 | 1731 | } |
7a8e76a3 | 1732 | } |
7a8e76a3 | 1733 | |
438ced17 VN |
1734 | /* wait for all the updates to complete */ |
1735 | for_each_buffer_cpu(buffer, cpu) { | |
1736 | cpu_buffer = buffer->buffers[cpu]; | |
05fdd70d | 1737 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
1738 | continue; |
1739 | ||
05fdd70d VN |
1740 | if (cpu_online(cpu)) |
1741 | wait_for_completion(&cpu_buffer->update_done); | |
83f40318 | 1742 | cpu_buffer->nr_pages_to_update = 0; |
438ced17 | 1743 | } |
83f40318 VN |
1744 | |
1745 | put_online_cpus(); | |
438ced17 | 1746 | } else { |
8e49f418 VN |
1747 | /* Make sure this CPU has been intitialized */ |
1748 | if (!cpumask_test_cpu(cpu_id, buffer->cpumask)) | |
1749 | goto out; | |
1750 | ||
438ced17 | 1751 | cpu_buffer = buffer->buffers[cpu_id]; |
83f40318 | 1752 | |
438ced17 VN |
1753 | if (nr_pages == cpu_buffer->nr_pages) |
1754 | goto out; | |
7a8e76a3 | 1755 | |
438ced17 VN |
1756 | cpu_buffer->nr_pages_to_update = nr_pages - |
1757 | cpu_buffer->nr_pages; | |
1758 | ||
1759 | INIT_LIST_HEAD(&cpu_buffer->new_pages); | |
1760 | if (cpu_buffer->nr_pages_to_update > 0 && | |
1761 | __rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 VN |
1762 | &cpu_buffer->new_pages, cpu_id)) { |
1763 | err = -ENOMEM; | |
1764 | goto out_err; | |
1765 | } | |
438ced17 | 1766 | |
83f40318 VN |
1767 | get_online_cpus(); |
1768 | ||
021c5b34 CM |
1769 | /* Can't run something on an offline CPU. */ |
1770 | if (!cpu_online(cpu_id)) | |
f5eb5588 SRRH |
1771 | rb_update_pages(cpu_buffer); |
1772 | else { | |
83f40318 VN |
1773 | schedule_work_on(cpu_id, |
1774 | &cpu_buffer->update_pages_work); | |
05fdd70d | 1775 | wait_for_completion(&cpu_buffer->update_done); |
f5eb5588 | 1776 | } |
83f40318 | 1777 | |
83f40318 | 1778 | cpu_buffer->nr_pages_to_update = 0; |
05fdd70d | 1779 | put_online_cpus(); |
438ced17 | 1780 | } |
7a8e76a3 SR |
1781 | |
1782 | out: | |
659f451f SR |
1783 | /* |
1784 | * The ring buffer resize can happen with the ring buffer | |
1785 | * enabled, so that the update disturbs the tracing as little | |
1786 | * as possible. But if the buffer is disabled, we do not need | |
1787 | * to worry about that, and we can take the time to verify | |
1788 | * that the buffer is not corrupt. | |
1789 | */ | |
1790 | if (atomic_read(&buffer->record_disabled)) { | |
1791 | atomic_inc(&buffer->record_disabled); | |
1792 | /* | |
1793 | * Even though the buffer was disabled, we must make sure | |
1794 | * that it is truly disabled before calling rb_check_pages. | |
1795 | * There could have been a race between checking | |
1796 | * record_disable and incrementing it. | |
1797 | */ | |
1798 | synchronize_sched(); | |
1799 | for_each_buffer_cpu(buffer, cpu) { | |
1800 | cpu_buffer = buffer->buffers[cpu]; | |
1801 | rb_check_pages(cpu_buffer); | |
1802 | } | |
1803 | atomic_dec(&buffer->record_disabled); | |
1804 | } | |
1805 | ||
7a8e76a3 | 1806 | mutex_unlock(&buffer->mutex); |
7a8e76a3 SR |
1807 | return size; |
1808 | ||
83f40318 | 1809 | out_err: |
438ced17 VN |
1810 | for_each_buffer_cpu(buffer, cpu) { |
1811 | struct buffer_page *bpage, *tmp; | |
83f40318 | 1812 | |
438ced17 | 1813 | cpu_buffer = buffer->buffers[cpu]; |
438ced17 | 1814 | cpu_buffer->nr_pages_to_update = 0; |
83f40318 | 1815 | |
438ced17 VN |
1816 | if (list_empty(&cpu_buffer->new_pages)) |
1817 | continue; | |
83f40318 | 1818 | |
438ced17 VN |
1819 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, |
1820 | list) { | |
1821 | list_del_init(&bpage->list); | |
1822 | free_buffer_page(bpage); | |
1823 | } | |
7a8e76a3 | 1824 | } |
641d2f63 | 1825 | mutex_unlock(&buffer->mutex); |
83f40318 | 1826 | return err; |
7a8e76a3 | 1827 | } |
c4f50183 | 1828 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 1829 | |
750912fa DS |
1830 | void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val) |
1831 | { | |
1832 | mutex_lock(&buffer->mutex); | |
1833 | if (val) | |
1834 | buffer->flags |= RB_FL_OVERWRITE; | |
1835 | else | |
1836 | buffer->flags &= ~RB_FL_OVERWRITE; | |
1837 | mutex_unlock(&buffer->mutex); | |
1838 | } | |
1839 | EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite); | |
1840 | ||
8789a9e7 | 1841 | static inline void * |
044fa782 | 1842 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 1843 | { |
044fa782 | 1844 | return bpage->data + index; |
8789a9e7 SR |
1845 | } |
1846 | ||
044fa782 | 1847 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 1848 | { |
044fa782 | 1849 | return bpage->page->data + index; |
7a8e76a3 SR |
1850 | } |
1851 | ||
1852 | static inline struct ring_buffer_event * | |
d769041f | 1853 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1854 | { |
6f807acd SR |
1855 | return __rb_page_index(cpu_buffer->reader_page, |
1856 | cpu_buffer->reader_page->read); | |
1857 | } | |
1858 | ||
7a8e76a3 SR |
1859 | static inline struct ring_buffer_event * |
1860 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
1861 | { | |
6f807acd | 1862 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
1863 | } |
1864 | ||
bf41a158 SR |
1865 | static inline unsigned rb_page_commit(struct buffer_page *bpage) |
1866 | { | |
abc9b56d | 1867 | return local_read(&bpage->page->commit); |
bf41a158 SR |
1868 | } |
1869 | ||
25985edc | 1870 | /* Size is determined by what has been committed */ |
bf41a158 SR |
1871 | static inline unsigned rb_page_size(struct buffer_page *bpage) |
1872 | { | |
1873 | return rb_page_commit(bpage); | |
1874 | } | |
1875 | ||
1876 | static inline unsigned | |
1877 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
1878 | { | |
1879 | return rb_page_commit(cpu_buffer->commit_page); | |
1880 | } | |
1881 | ||
bf41a158 SR |
1882 | static inline unsigned |
1883 | rb_event_index(struct ring_buffer_event *event) | |
1884 | { | |
1885 | unsigned long addr = (unsigned long)event; | |
1886 | ||
22f470f8 | 1887 | return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE; |
bf41a158 SR |
1888 | } |
1889 | ||
0f0c85fc | 1890 | static inline int |
fa743953 SR |
1891 | rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1892 | struct ring_buffer_event *event) | |
bf41a158 SR |
1893 | { |
1894 | unsigned long addr = (unsigned long)event; | |
1895 | unsigned long index; | |
1896 | ||
1897 | index = rb_event_index(event); | |
1898 | addr &= PAGE_MASK; | |
1899 | ||
1900 | return cpu_buffer->commit_page->page == (void *)addr && | |
1901 | rb_commit_index(cpu_buffer) == index; | |
1902 | } | |
1903 | ||
34a148bf | 1904 | static void |
bf41a158 | 1905 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1906 | { |
77ae365e SR |
1907 | unsigned long max_count; |
1908 | ||
bf41a158 SR |
1909 | /* |
1910 | * We only race with interrupts and NMIs on this CPU. | |
1911 | * If we own the commit event, then we can commit | |
1912 | * all others that interrupted us, since the interruptions | |
1913 | * are in stack format (they finish before they come | |
1914 | * back to us). This allows us to do a simple loop to | |
1915 | * assign the commit to the tail. | |
1916 | */ | |
a8ccf1d6 | 1917 | again: |
438ced17 | 1918 | max_count = cpu_buffer->nr_pages * 100; |
77ae365e | 1919 | |
bf41a158 | 1920 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
77ae365e SR |
1921 | if (RB_WARN_ON(cpu_buffer, !(--max_count))) |
1922 | return; | |
1923 | if (RB_WARN_ON(cpu_buffer, | |
1924 | rb_is_reader_page(cpu_buffer->tail_page))) | |
1925 | return; | |
1926 | local_set(&cpu_buffer->commit_page->page->commit, | |
1927 | rb_page_write(cpu_buffer->commit_page)); | |
bf41a158 | 1928 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); |
abc9b56d SR |
1929 | cpu_buffer->write_stamp = |
1930 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1931 | /* add barrier to keep gcc from optimizing too much */ |
1932 | barrier(); | |
1933 | } | |
1934 | while (rb_commit_index(cpu_buffer) != | |
1935 | rb_page_write(cpu_buffer->commit_page)) { | |
77ae365e SR |
1936 | |
1937 | local_set(&cpu_buffer->commit_page->page->commit, | |
1938 | rb_page_write(cpu_buffer->commit_page)); | |
1939 | RB_WARN_ON(cpu_buffer, | |
1940 | local_read(&cpu_buffer->commit_page->page->commit) & | |
1941 | ~RB_WRITE_MASK); | |
bf41a158 SR |
1942 | barrier(); |
1943 | } | |
a8ccf1d6 SR |
1944 | |
1945 | /* again, keep gcc from optimizing */ | |
1946 | barrier(); | |
1947 | ||
1948 | /* | |
1949 | * If an interrupt came in just after the first while loop | |
1950 | * and pushed the tail page forward, we will be left with | |
1951 | * a dangling commit that will never go forward. | |
1952 | */ | |
1953 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1954 | goto again; | |
7a8e76a3 SR |
1955 | } |
1956 | ||
d769041f | 1957 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1958 | { |
abc9b56d | 1959 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1960 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1961 | } |
1962 | ||
34a148bf | 1963 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1964 | { |
1965 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1966 | ||
1967 | /* | |
1968 | * The iterator could be on the reader page (it starts there). | |
1969 | * But the head could have moved, since the reader was | |
1970 | * found. Check for this case and assign the iterator | |
1971 | * to the head page instead of next. | |
1972 | */ | |
1973 | if (iter->head_page == cpu_buffer->reader_page) | |
77ae365e | 1974 | iter->head_page = rb_set_head_page(cpu_buffer); |
d769041f SR |
1975 | else |
1976 | rb_inc_page(cpu_buffer, &iter->head_page); | |
1977 | ||
abc9b56d | 1978 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
1979 | iter->head = 0; |
1980 | } | |
1981 | ||
69d1b839 SR |
1982 | /* Slow path, do not inline */ |
1983 | static noinline struct ring_buffer_event * | |
1984 | rb_add_time_stamp(struct ring_buffer_event *event, u64 delta) | |
1985 | { | |
1986 | event->type_len = RINGBUF_TYPE_TIME_EXTEND; | |
1987 | ||
1988 | /* Not the first event on the page? */ | |
1989 | if (rb_event_index(event)) { | |
1990 | event->time_delta = delta & TS_MASK; | |
1991 | event->array[0] = delta >> TS_SHIFT; | |
1992 | } else { | |
1993 | /* nope, just zero it */ | |
1994 | event->time_delta = 0; | |
1995 | event->array[0] = 0; | |
1996 | } | |
1997 | ||
1998 | return skip_time_extend(event); | |
1999 | } | |
2000 | ||
7a8e76a3 | 2001 | /** |
01e3e710 | 2002 | * rb_update_event - update event type and data |
021de3d9 | 2003 | * @event: the event to update |
7a8e76a3 SR |
2004 | * @type: the type of event |
2005 | * @length: the size of the event field in the ring buffer | |
2006 | * | |
2007 | * Update the type and data fields of the event. The length | |
2008 | * is the actual size that is written to the ring buffer, | |
2009 | * and with this, we can determine what to place into the | |
2010 | * data field. | |
2011 | */ | |
a4543a2f | 2012 | static void __always_inline |
69d1b839 | 2013 | rb_update_event(struct ring_buffer_per_cpu *cpu_buffer, |
fcc742ea SRRH |
2014 | struct ring_buffer_event *event, |
2015 | struct rb_event_info *info) | |
7a8e76a3 | 2016 | { |
fcc742ea SRRH |
2017 | unsigned length = info->length; |
2018 | u64 delta = info->delta; | |
2019 | ||
69d1b839 SR |
2020 | /* |
2021 | * If we need to add a timestamp, then we | |
2022 | * add it to the start of the resevered space. | |
2023 | */ | |
fcc742ea | 2024 | if (unlikely(info->add_timestamp)) { |
69d1b839 SR |
2025 | event = rb_add_time_stamp(event, delta); |
2026 | length -= RB_LEN_TIME_EXTEND; | |
2027 | delta = 0; | |
7a8e76a3 | 2028 | } |
69d1b839 SR |
2029 | |
2030 | event->time_delta = delta; | |
2031 | length -= RB_EVNT_HDR_SIZE; | |
2032 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) { | |
2033 | event->type_len = 0; | |
2034 | event->array[0] = length; | |
2035 | } else | |
2036 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
7a8e76a3 SR |
2037 | } |
2038 | ||
77ae365e SR |
2039 | /* |
2040 | * rb_handle_head_page - writer hit the head page | |
2041 | * | |
2042 | * Returns: +1 to retry page | |
2043 | * 0 to continue | |
2044 | * -1 on error | |
2045 | */ | |
2046 | static int | |
2047 | rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
2048 | struct buffer_page *tail_page, | |
2049 | struct buffer_page *next_page) | |
2050 | { | |
2051 | struct buffer_page *new_head; | |
2052 | int entries; | |
2053 | int type; | |
2054 | int ret; | |
2055 | ||
2056 | entries = rb_page_entries(next_page); | |
2057 | ||
2058 | /* | |
2059 | * The hard part is here. We need to move the head | |
2060 | * forward, and protect against both readers on | |
2061 | * other CPUs and writers coming in via interrupts. | |
2062 | */ | |
2063 | type = rb_head_page_set_update(cpu_buffer, next_page, tail_page, | |
2064 | RB_PAGE_HEAD); | |
2065 | ||
2066 | /* | |
2067 | * type can be one of four: | |
2068 | * NORMAL - an interrupt already moved it for us | |
2069 | * HEAD - we are the first to get here. | |
2070 | * UPDATE - we are the interrupt interrupting | |
2071 | * a current move. | |
2072 | * MOVED - a reader on another CPU moved the next | |
2073 | * pointer to its reader page. Give up | |
2074 | * and try again. | |
2075 | */ | |
2076 | ||
2077 | switch (type) { | |
2078 | case RB_PAGE_HEAD: | |
2079 | /* | |
2080 | * We changed the head to UPDATE, thus | |
2081 | * it is our responsibility to update | |
2082 | * the counters. | |
2083 | */ | |
2084 | local_add(entries, &cpu_buffer->overrun); | |
c64e148a | 2085 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
77ae365e SR |
2086 | |
2087 | /* | |
2088 | * The entries will be zeroed out when we move the | |
2089 | * tail page. | |
2090 | */ | |
2091 | ||
2092 | /* still more to do */ | |
2093 | break; | |
2094 | ||
2095 | case RB_PAGE_UPDATE: | |
2096 | /* | |
2097 | * This is an interrupt that interrupt the | |
2098 | * previous update. Still more to do. | |
2099 | */ | |
2100 | break; | |
2101 | case RB_PAGE_NORMAL: | |
2102 | /* | |
2103 | * An interrupt came in before the update | |
2104 | * and processed this for us. | |
2105 | * Nothing left to do. | |
2106 | */ | |
2107 | return 1; | |
2108 | case RB_PAGE_MOVED: | |
2109 | /* | |
2110 | * The reader is on another CPU and just did | |
2111 | * a swap with our next_page. | |
2112 | * Try again. | |
2113 | */ | |
2114 | return 1; | |
2115 | default: | |
2116 | RB_WARN_ON(cpu_buffer, 1); /* WTF??? */ | |
2117 | return -1; | |
2118 | } | |
2119 | ||
2120 | /* | |
2121 | * Now that we are here, the old head pointer is | |
2122 | * set to UPDATE. This will keep the reader from | |
2123 | * swapping the head page with the reader page. | |
2124 | * The reader (on another CPU) will spin till | |
2125 | * we are finished. | |
2126 | * | |
2127 | * We just need to protect against interrupts | |
2128 | * doing the job. We will set the next pointer | |
2129 | * to HEAD. After that, we set the old pointer | |
2130 | * to NORMAL, but only if it was HEAD before. | |
2131 | * otherwise we are an interrupt, and only | |
2132 | * want the outer most commit to reset it. | |
2133 | */ | |
2134 | new_head = next_page; | |
2135 | rb_inc_page(cpu_buffer, &new_head); | |
2136 | ||
2137 | ret = rb_head_page_set_head(cpu_buffer, new_head, next_page, | |
2138 | RB_PAGE_NORMAL); | |
2139 | ||
2140 | /* | |
2141 | * Valid returns are: | |
2142 | * HEAD - an interrupt came in and already set it. | |
2143 | * NORMAL - One of two things: | |
2144 | * 1) We really set it. | |
2145 | * 2) A bunch of interrupts came in and moved | |
2146 | * the page forward again. | |
2147 | */ | |
2148 | switch (ret) { | |
2149 | case RB_PAGE_HEAD: | |
2150 | case RB_PAGE_NORMAL: | |
2151 | /* OK */ | |
2152 | break; | |
2153 | default: | |
2154 | RB_WARN_ON(cpu_buffer, 1); | |
2155 | return -1; | |
2156 | } | |
2157 | ||
2158 | /* | |
2159 | * It is possible that an interrupt came in, | |
2160 | * set the head up, then more interrupts came in | |
2161 | * and moved it again. When we get back here, | |
2162 | * the page would have been set to NORMAL but we | |
2163 | * just set it back to HEAD. | |
2164 | * | |
2165 | * How do you detect this? Well, if that happened | |
2166 | * the tail page would have moved. | |
2167 | */ | |
2168 | if (ret == RB_PAGE_NORMAL) { | |
2169 | /* | |
2170 | * If the tail had moved passed next, then we need | |
2171 | * to reset the pointer. | |
2172 | */ | |
2173 | if (cpu_buffer->tail_page != tail_page && | |
2174 | cpu_buffer->tail_page != next_page) | |
2175 | rb_head_page_set_normal(cpu_buffer, new_head, | |
2176 | next_page, | |
2177 | RB_PAGE_HEAD); | |
2178 | } | |
2179 | ||
2180 | /* | |
2181 | * If this was the outer most commit (the one that | |
2182 | * changed the original pointer from HEAD to UPDATE), | |
2183 | * then it is up to us to reset it to NORMAL. | |
2184 | */ | |
2185 | if (type == RB_PAGE_HEAD) { | |
2186 | ret = rb_head_page_set_normal(cpu_buffer, next_page, | |
2187 | tail_page, | |
2188 | RB_PAGE_UPDATE); | |
2189 | if (RB_WARN_ON(cpu_buffer, | |
2190 | ret != RB_PAGE_UPDATE)) | |
2191 | return -1; | |
2192 | } | |
2193 | ||
2194 | return 0; | |
2195 | } | |
2196 | ||
34a148bf | 2197 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
2198 | { |
2199 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
2200 | ||
2201 | /* zero length can cause confusions */ | |
2202 | if (!length) | |
a497adb4 | 2203 | length++; |
7a8e76a3 | 2204 | |
2271048d | 2205 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) |
7a8e76a3 SR |
2206 | length += sizeof(event.array[0]); |
2207 | ||
2208 | length += RB_EVNT_HDR_SIZE; | |
2271048d | 2209 | length = ALIGN(length, RB_ARCH_ALIGNMENT); |
7a8e76a3 | 2210 | |
7d75e683 SRRH |
2211 | /* |
2212 | * In case the time delta is larger than the 27 bits for it | |
2213 | * in the header, we need to add a timestamp. If another | |
2214 | * event comes in when trying to discard this one to increase | |
2215 | * the length, then the timestamp will be added in the allocated | |
2216 | * space of this event. If length is bigger than the size needed | |
2217 | * for the TIME_EXTEND, then padding has to be used. The events | |
2218 | * length must be either RB_LEN_TIME_EXTEND, or greater than or equal | |
2219 | * to RB_LEN_TIME_EXTEND + 8, as 8 is the minimum size for padding. | |
2220 | * As length is a multiple of 4, we only need to worry if it | |
2221 | * is 12 (RB_LEN_TIME_EXTEND + 4). | |
2222 | */ | |
2223 | if (length == RB_LEN_TIME_EXTEND + RB_ALIGNMENT) | |
2224 | length += RB_ALIGNMENT; | |
2225 | ||
7a8e76a3 SR |
2226 | return length; |
2227 | } | |
2228 | ||
c7b09308 SR |
2229 | static inline void |
2230 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
fcc742ea | 2231 | unsigned long tail, struct rb_event_info *info) |
c7b09308 | 2232 | { |
fcc742ea | 2233 | struct buffer_page *tail_page = info->tail_page; |
c7b09308 | 2234 | struct ring_buffer_event *event; |
fcc742ea | 2235 | unsigned long length = info->length; |
c7b09308 SR |
2236 | |
2237 | /* | |
2238 | * Only the event that crossed the page boundary | |
2239 | * must fill the old tail_page with padding. | |
2240 | */ | |
2241 | if (tail >= BUF_PAGE_SIZE) { | |
b3230c8b SR |
2242 | /* |
2243 | * If the page was filled, then we still need | |
2244 | * to update the real_end. Reset it to zero | |
2245 | * and the reader will ignore it. | |
2246 | */ | |
2247 | if (tail == BUF_PAGE_SIZE) | |
2248 | tail_page->real_end = 0; | |
2249 | ||
c7b09308 SR |
2250 | local_sub(length, &tail_page->write); |
2251 | return; | |
2252 | } | |
2253 | ||
2254 | event = __rb_page_index(tail_page, tail); | |
b0b7065b | 2255 | kmemcheck_annotate_bitfield(event, bitfield); |
c7b09308 | 2256 | |
c64e148a VN |
2257 | /* account for padding bytes */ |
2258 | local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes); | |
2259 | ||
ff0ff84a SR |
2260 | /* |
2261 | * Save the original length to the meta data. | |
2262 | * This will be used by the reader to add lost event | |
2263 | * counter. | |
2264 | */ | |
2265 | tail_page->real_end = tail; | |
2266 | ||
c7b09308 SR |
2267 | /* |
2268 | * If this event is bigger than the minimum size, then | |
2269 | * we need to be careful that we don't subtract the | |
2270 | * write counter enough to allow another writer to slip | |
2271 | * in on this page. | |
2272 | * We put in a discarded commit instead, to make sure | |
2273 | * that this space is not used again. | |
2274 | * | |
2275 | * If we are less than the minimum size, we don't need to | |
2276 | * worry about it. | |
2277 | */ | |
2278 | if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) { | |
2279 | /* No room for any events */ | |
2280 | ||
2281 | /* Mark the rest of the page with padding */ | |
2282 | rb_event_set_padding(event); | |
2283 | ||
2284 | /* Set the write back to the previous setting */ | |
2285 | local_sub(length, &tail_page->write); | |
2286 | return; | |
2287 | } | |
2288 | ||
2289 | /* Put in a discarded event */ | |
2290 | event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE; | |
2291 | event->type_len = RINGBUF_TYPE_PADDING; | |
2292 | /* time delta must be non zero */ | |
2293 | event->time_delta = 1; | |
c7b09308 SR |
2294 | |
2295 | /* Set write to end of buffer */ | |
2296 | length = (tail + length) - BUF_PAGE_SIZE; | |
2297 | local_sub(length, &tail_page->write); | |
2298 | } | |
6634ff26 | 2299 | |
a4543a2f SRRH |
2300 | static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer); |
2301 | ||
747e94ae SR |
2302 | /* |
2303 | * This is the slow path, force gcc not to inline it. | |
2304 | */ | |
2305 | static noinline struct ring_buffer_event * | |
6634ff26 | 2306 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
fcc742ea | 2307 | unsigned long tail, struct rb_event_info *info) |
7a8e76a3 | 2308 | { |
fcc742ea | 2309 | struct buffer_page *tail_page = info->tail_page; |
5a50e33c | 2310 | struct buffer_page *commit_page = cpu_buffer->commit_page; |
7a8e76a3 | 2311 | struct ring_buffer *buffer = cpu_buffer->buffer; |
77ae365e SR |
2312 | struct buffer_page *next_page; |
2313 | int ret; | |
fcc742ea | 2314 | u64 ts; |
aa20ae84 | 2315 | |
a4543a2f SRRH |
2316 | /* |
2317 | * If the event had a timestamp attached to it, remove it. | |
2318 | * The first event on a page (nested or not) always uses | |
2319 | * the full timestamp of the new page. | |
2320 | */ | |
2321 | if (info->add_timestamp) { | |
2322 | info->add_timestamp = 0; | |
2323 | info->length -= RB_LEN_TIME_EXTEND; | |
2324 | } | |
2325 | ||
aa20ae84 SR |
2326 | next_page = tail_page; |
2327 | ||
aa20ae84 SR |
2328 | rb_inc_page(cpu_buffer, &next_page); |
2329 | ||
aa20ae84 SR |
2330 | /* |
2331 | * If for some reason, we had an interrupt storm that made | |
2332 | * it all the way around the buffer, bail, and warn | |
2333 | * about it. | |
2334 | */ | |
2335 | if (unlikely(next_page == commit_page)) { | |
77ae365e | 2336 | local_inc(&cpu_buffer->commit_overrun); |
aa20ae84 SR |
2337 | goto out_reset; |
2338 | } | |
2339 | ||
77ae365e SR |
2340 | /* |
2341 | * This is where the fun begins! | |
2342 | * | |
2343 | * We are fighting against races between a reader that | |
2344 | * could be on another CPU trying to swap its reader | |
2345 | * page with the buffer head. | |
2346 | * | |
2347 | * We are also fighting against interrupts coming in and | |
2348 | * moving the head or tail on us as well. | |
2349 | * | |
2350 | * If the next page is the head page then we have filled | |
2351 | * the buffer, unless the commit page is still on the | |
2352 | * reader page. | |
2353 | */ | |
2354 | if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) { | |
aa20ae84 | 2355 | |
77ae365e SR |
2356 | /* |
2357 | * If the commit is not on the reader page, then | |
2358 | * move the header page. | |
2359 | */ | |
2360 | if (!rb_is_reader_page(cpu_buffer->commit_page)) { | |
2361 | /* | |
2362 | * If we are not in overwrite mode, | |
2363 | * this is easy, just stop here. | |
2364 | */ | |
884bfe89 SP |
2365 | if (!(buffer->flags & RB_FL_OVERWRITE)) { |
2366 | local_inc(&cpu_buffer->dropped_events); | |
77ae365e | 2367 | goto out_reset; |
884bfe89 | 2368 | } |
77ae365e SR |
2369 | |
2370 | ret = rb_handle_head_page(cpu_buffer, | |
2371 | tail_page, | |
2372 | next_page); | |
2373 | if (ret < 0) | |
2374 | goto out_reset; | |
2375 | if (ret) | |
2376 | goto out_again; | |
2377 | } else { | |
2378 | /* | |
2379 | * We need to be careful here too. The | |
2380 | * commit page could still be on the reader | |
2381 | * page. We could have a small buffer, and | |
2382 | * have filled up the buffer with events | |
2383 | * from interrupts and such, and wrapped. | |
2384 | * | |
2385 | * Note, if the tail page is also the on the | |
2386 | * reader_page, we let it move out. | |
2387 | */ | |
2388 | if (unlikely((cpu_buffer->commit_page != | |
2389 | cpu_buffer->tail_page) && | |
2390 | (cpu_buffer->commit_page == | |
2391 | cpu_buffer->reader_page))) { | |
2392 | local_inc(&cpu_buffer->commit_overrun); | |
2393 | goto out_reset; | |
2394 | } | |
aa20ae84 SR |
2395 | } |
2396 | } | |
2397 | ||
77ae365e SR |
2398 | ret = rb_tail_page_update(cpu_buffer, tail_page, next_page); |
2399 | if (ret) { | |
2400 | /* | |
2401 | * Nested commits always have zero deltas, so | |
2402 | * just reread the time stamp | |
2403 | */ | |
e8bc43e8 SR |
2404 | ts = rb_time_stamp(buffer); |
2405 | next_page->page->time_stamp = ts; | |
aa20ae84 SR |
2406 | } |
2407 | ||
77ae365e | 2408 | out_again: |
aa20ae84 | 2409 | |
fcc742ea | 2410 | rb_reset_tail(cpu_buffer, tail, info); |
aa20ae84 | 2411 | |
a4543a2f SRRH |
2412 | /* Commit what we have for now to update timestamps */ |
2413 | rb_end_commit(cpu_buffer); | |
2414 | /* rb_end_commit() decs committing */ | |
2415 | local_inc(&cpu_buffer->committing); | |
2416 | ||
aa20ae84 SR |
2417 | /* fail and let the caller try again */ |
2418 | return ERR_PTR(-EAGAIN); | |
2419 | ||
45141d46 | 2420 | out_reset: |
6f3b3440 | 2421 | /* reset write */ |
fcc742ea | 2422 | rb_reset_tail(cpu_buffer, tail, info); |
6f3b3440 | 2423 | |
bf41a158 | 2424 | return NULL; |
7a8e76a3 SR |
2425 | } |
2426 | ||
9826b273 SRRH |
2427 | #ifndef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK |
2428 | static inline bool sched_clock_stable(void) | |
2429 | { | |
2430 | return true; | |
2431 | } | |
2432 | #endif | |
2433 | ||
a4543a2f SRRH |
2434 | static inline int |
2435 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
2436 | struct ring_buffer_event *event); | |
2437 | static inline void rb_event_discard(struct ring_buffer_event *event); | |
2438 | static void | |
2439 | rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
2440 | struct ring_buffer_event *event); | |
2441 | ||
9826b273 SRRH |
2442 | static noinline void |
2443 | rb_handle_timestamp(struct ring_buffer_per_cpu *cpu_buffer, | |
a4543a2f | 2444 | struct ring_buffer_event *event, |
9826b273 SRRH |
2445 | struct rb_event_info *info) |
2446 | { | |
a4543a2f SRRH |
2447 | struct ring_buffer_event *padding; |
2448 | int length; | |
2449 | int size; | |
2450 | ||
9826b273 SRRH |
2451 | WARN_ONCE(info->delta > (1ULL << 59), |
2452 | KERN_WARNING "Delta way too big! %llu ts=%llu write stamp = %llu\n%s", | |
2453 | (unsigned long long)info->delta, | |
2454 | (unsigned long long)info->ts, | |
2455 | (unsigned long long)cpu_buffer->write_stamp, | |
2456 | sched_clock_stable() ? "" : | |
2457 | "If you just came from a suspend/resume,\n" | |
2458 | "please switch to the trace global clock:\n" | |
2459 | " echo global > /sys/kernel/debug/tracing/trace_clock\n"); | |
a4543a2f SRRH |
2460 | |
2461 | /* | |
2462 | * Discarding this event to add a timestamp in front, but | |
2463 | * we still need to update the length of it to perform the discard. | |
2464 | */ | |
2465 | rb_update_event(cpu_buffer, event, info); | |
2466 | ||
2467 | if (rb_try_to_discard(cpu_buffer, event)) { | |
2468 | info->add_timestamp = 1; | |
2469 | /* | |
2470 | * The time delta since the last event is too big to | |
2471 | * hold in the time field of the event, then we append a | |
2472 | * TIME EXTEND event ahead of the data event. | |
2473 | */ | |
2474 | info->length += RB_LEN_TIME_EXTEND; | |
2475 | return; | |
2476 | } | |
2477 | ||
2478 | /* | |
2479 | * Humpf! An event came in after this one, and because it is not a | |
2480 | * commit, it will have a delta of zero, thus, it will take on | |
2481 | * the timestamp of the previous commit, which happened a long time | |
2482 | * ago (we need to add a timestamp, remember?). | |
2483 | * We need to add the timestamp here. A timestamp is a fixed size | |
2484 | * of 8 bytes. That means the rest of the event needs to be | |
2485 | * padding. | |
2486 | */ | |
2487 | size = info->length - RB_LEN_TIME_EXTEND; | |
2488 | ||
2489 | /* The padding will have a delta of 1 */ | |
2490 | if (size) | |
2491 | info->delta--; | |
2492 | ||
2493 | padding = rb_add_time_stamp(event, info->delta); | |
2494 | ||
2495 | if (size) { | |
2496 | length = info->length; | |
2497 | info->delta = 0; | |
2498 | info->length = size; | |
2499 | rb_update_event(cpu_buffer, padding, info); | |
2500 | ||
2501 | rb_event_discard(padding); | |
2502 | ||
2503 | /* Still visible, need to update write_stamp */ | |
2504 | rb_update_write_stamp(cpu_buffer, event); | |
2505 | ||
2506 | /* Still need to commit the padding. */ | |
2507 | rb_end_commit(cpu_buffer); | |
2508 | ||
2509 | /* rb_end_commit() decs committing */ | |
2510 | local_inc(&cpu_buffer->committing); | |
2511 | ||
2512 | /* The next iteration still uses the original length */ | |
2513 | info->length = length; | |
2514 | } | |
9826b273 SRRH |
2515 | } |
2516 | ||
6634ff26 SR |
2517 | static struct ring_buffer_event * |
2518 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
fcc742ea | 2519 | struct rb_event_info *info) |
6634ff26 | 2520 | { |
6634ff26 | 2521 | struct ring_buffer_event *event; |
fcc742ea | 2522 | struct buffer_page *tail_page; |
6634ff26 | 2523 | unsigned long tail, write; |
a4543a2f | 2524 | bool is_commit; |
69d1b839 | 2525 | |
fcc742ea SRRH |
2526 | tail_page = info->tail_page = cpu_buffer->tail_page; |
2527 | write = local_add_return(info->length, &tail_page->write); | |
77ae365e SR |
2528 | |
2529 | /* set write to only the index of the write */ | |
2530 | write &= RB_WRITE_MASK; | |
fcc742ea | 2531 | tail = write - info->length; |
6634ff26 SR |
2532 | |
2533 | /* See if we shot pass the end of this buffer page */ | |
747e94ae | 2534 | if (unlikely(write > BUF_PAGE_SIZE)) |
fcc742ea | 2535 | return rb_move_tail(cpu_buffer, tail, info); |
6634ff26 SR |
2536 | |
2537 | /* We reserved something on the buffer */ | |
6634ff26 | 2538 | event = __rb_page_index(tail_page, tail); |
6634ff26 SR |
2539 | |
2540 | /* | |
a4543a2f SRRH |
2541 | * If this is the first commit on the page, then it has the same |
2542 | * timestamp as the page itself, otherwise we need to figure out | |
2543 | * the delta. | |
6634ff26 | 2544 | */ |
a4543a2f SRRH |
2545 | info->ts = rb_time_stamp(cpu_buffer->buffer); |
2546 | is_commit = rb_event_is_commit(cpu_buffer, event); | |
2547 | ||
2548 | /* Commits are special (non nested events) */ | |
2549 | info->delta = is_commit ? info->ts - cpu_buffer->write_stamp : 0; | |
2550 | ||
2551 | if (!tail) { | |
2552 | /* | |
2553 | * If this is the first commit on the page, set the | |
2554 | * page to its timestamp. | |
2555 | */ | |
fcc742ea | 2556 | tail_page->page->time_stamp = info->ts; |
a4543a2f SRRH |
2557 | info->delta = 0; |
2558 | ||
2559 | } else if (unlikely(test_time_stamp(info->delta)) && | |
2560 | !info->add_timestamp) { | |
2561 | rb_handle_timestamp(cpu_buffer, event, info); | |
2562 | return ERR_PTR(-EAGAIN); | |
2563 | } | |
2564 | ||
2565 | kmemcheck_annotate_bitfield(event, bitfield); | |
2566 | rb_update_event(cpu_buffer, event, info); | |
2567 | ||
2568 | local_inc(&tail_page->entries); | |
6634ff26 | 2569 | |
c64e148a | 2570 | /* account for these added bytes */ |
fcc742ea | 2571 | local_add(info->length, &cpu_buffer->entries_bytes); |
c64e148a | 2572 | |
6634ff26 SR |
2573 | return event; |
2574 | } | |
2575 | ||
edd813bf SR |
2576 | static inline int |
2577 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
2578 | struct ring_buffer_event *event) | |
2579 | { | |
2580 | unsigned long new_index, old_index; | |
2581 | struct buffer_page *bpage; | |
2582 | unsigned long index; | |
2583 | unsigned long addr; | |
2584 | ||
2585 | new_index = rb_event_index(event); | |
69d1b839 | 2586 | old_index = new_index + rb_event_ts_length(event); |
edd813bf SR |
2587 | addr = (unsigned long)event; |
2588 | addr &= PAGE_MASK; | |
2589 | ||
2590 | bpage = cpu_buffer->tail_page; | |
2591 | ||
2592 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { | |
77ae365e SR |
2593 | unsigned long write_mask = |
2594 | local_read(&bpage->write) & ~RB_WRITE_MASK; | |
c64e148a | 2595 | unsigned long event_length = rb_event_length(event); |
edd813bf SR |
2596 | /* |
2597 | * This is on the tail page. It is possible that | |
2598 | * a write could come in and move the tail page | |
2599 | * and write to the next page. That is fine | |
2600 | * because we just shorten what is on this page. | |
2601 | */ | |
77ae365e SR |
2602 | old_index += write_mask; |
2603 | new_index += write_mask; | |
edd813bf | 2604 | index = local_cmpxchg(&bpage->write, old_index, new_index); |
c64e148a VN |
2605 | if (index == old_index) { |
2606 | /* update counters */ | |
2607 | local_sub(event_length, &cpu_buffer->entries_bytes); | |
edd813bf | 2608 | return 1; |
c64e148a | 2609 | } |
edd813bf SR |
2610 | } |
2611 | ||
2612 | /* could not discard */ | |
2613 | return 0; | |
2614 | } | |
2615 | ||
fa743953 SR |
2616 | static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer) |
2617 | { | |
2618 | local_inc(&cpu_buffer->committing); | |
2619 | local_inc(&cpu_buffer->commits); | |
2620 | } | |
2621 | ||
d9abde21 | 2622 | static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer) |
fa743953 SR |
2623 | { |
2624 | unsigned long commits; | |
2625 | ||
2626 | if (RB_WARN_ON(cpu_buffer, | |
2627 | !local_read(&cpu_buffer->committing))) | |
2628 | return; | |
2629 | ||
2630 | again: | |
2631 | commits = local_read(&cpu_buffer->commits); | |
2632 | /* synchronize with interrupts */ | |
2633 | barrier(); | |
2634 | if (local_read(&cpu_buffer->committing) == 1) | |
2635 | rb_set_commit_to_write(cpu_buffer); | |
2636 | ||
2637 | local_dec(&cpu_buffer->committing); | |
2638 | ||
2639 | /* synchronize with interrupts */ | |
2640 | barrier(); | |
2641 | ||
2642 | /* | |
2643 | * Need to account for interrupts coming in between the | |
2644 | * updating of the commit page and the clearing of the | |
2645 | * committing counter. | |
2646 | */ | |
2647 | if (unlikely(local_read(&cpu_buffer->commits) != commits) && | |
2648 | !local_read(&cpu_buffer->committing)) { | |
2649 | local_inc(&cpu_buffer->committing); | |
2650 | goto again; | |
2651 | } | |
2652 | } | |
2653 | ||
7a8e76a3 | 2654 | static struct ring_buffer_event * |
62f0b3eb SR |
2655 | rb_reserve_next_event(struct ring_buffer *buffer, |
2656 | struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 2657 | unsigned long length) |
7a8e76a3 SR |
2658 | { |
2659 | struct ring_buffer_event *event; | |
fcc742ea | 2660 | struct rb_event_info info; |
818e3dd3 | 2661 | int nr_loops = 0; |
7a8e76a3 | 2662 | |
fa743953 SR |
2663 | rb_start_commit(cpu_buffer); |
2664 | ||
85bac32c | 2665 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
62f0b3eb SR |
2666 | /* |
2667 | * Due to the ability to swap a cpu buffer from a buffer | |
2668 | * it is possible it was swapped before we committed. | |
2669 | * (committing stops a swap). We check for it here and | |
2670 | * if it happened, we have to fail the write. | |
2671 | */ | |
2672 | barrier(); | |
2673 | if (unlikely(ACCESS_ONCE(cpu_buffer->buffer) != buffer)) { | |
2674 | local_dec(&cpu_buffer->committing); | |
2675 | local_dec(&cpu_buffer->commits); | |
2676 | return NULL; | |
2677 | } | |
85bac32c | 2678 | #endif |
fcc742ea | 2679 | info.length = rb_calculate_event_length(length); |
fcc742ea | 2680 | info.add_timestamp = 0; |
a4543a2f | 2681 | again: |
818e3dd3 SR |
2682 | /* |
2683 | * We allow for interrupts to reenter here and do a trace. | |
2684 | * If one does, it will cause this original code to loop | |
2685 | * back here. Even with heavy interrupts happening, this | |
2686 | * should only happen a few times in a row. If this happens | |
2687 | * 1000 times in a row, there must be either an interrupt | |
2688 | * storm or we have something buggy. | |
2689 | * Bail! | |
2690 | */ | |
3e89c7bb | 2691 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
fa743953 | 2692 | goto out_fail; |
818e3dd3 | 2693 | |
fcc742ea SRRH |
2694 | event = __rb_reserve_next(cpu_buffer, &info); |
2695 | ||
168b6b1d | 2696 | if (unlikely(PTR_ERR(event) == -EAGAIN)) |
bf41a158 SR |
2697 | goto again; |
2698 | ||
fa743953 SR |
2699 | if (!event) |
2700 | goto out_fail; | |
7a8e76a3 | 2701 | |
7a8e76a3 | 2702 | return event; |
fa743953 SR |
2703 | |
2704 | out_fail: | |
2705 | rb_end_commit(cpu_buffer); | |
2706 | return NULL; | |
7a8e76a3 SR |
2707 | } |
2708 | ||
567cd4da SR |
2709 | /* |
2710 | * The lock and unlock are done within a preempt disable section. | |
2711 | * The current_context per_cpu variable can only be modified | |
2712 | * by the current task between lock and unlock. But it can | |
2713 | * be modified more than once via an interrupt. To pass this | |
2714 | * information from the lock to the unlock without having to | |
2715 | * access the 'in_interrupt()' functions again (which do show | |
2716 | * a bit of overhead in something as critical as function tracing, | |
2717 | * we use a bitmask trick. | |
2718 | * | |
2719 | * bit 0 = NMI context | |
2720 | * bit 1 = IRQ context | |
2721 | * bit 2 = SoftIRQ context | |
2722 | * bit 3 = normal context. | |
2723 | * | |
2724 | * This works because this is the order of contexts that can | |
2725 | * preempt other contexts. A SoftIRQ never preempts an IRQ | |
2726 | * context. | |
2727 | * | |
2728 | * When the context is determined, the corresponding bit is | |
2729 | * checked and set (if it was set, then a recursion of that context | |
2730 | * happened). | |
2731 | * | |
2732 | * On unlock, we need to clear this bit. To do so, just subtract | |
2733 | * 1 from the current_context and AND it to itself. | |
2734 | * | |
2735 | * (binary) | |
2736 | * 101 - 1 = 100 | |
2737 | * 101 & 100 = 100 (clearing bit zero) | |
2738 | * | |
2739 | * 1010 - 1 = 1001 | |
2740 | * 1010 & 1001 = 1000 (clearing bit 1) | |
2741 | * | |
2742 | * The least significant bit can be cleared this way, and it | |
2743 | * just so happens that it is the same bit corresponding to | |
2744 | * the current context. | |
2745 | */ | |
261842b7 | 2746 | |
58a09ec6 SRRH |
2747 | static __always_inline int |
2748 | trace_recursive_lock(struct ring_buffer_per_cpu *cpu_buffer) | |
261842b7 | 2749 | { |
58a09ec6 | 2750 | unsigned int val = cpu_buffer->current_context; |
567cd4da | 2751 | int bit; |
d9abde21 | 2752 | |
567cd4da SR |
2753 | if (in_interrupt()) { |
2754 | if (in_nmi()) | |
a497adb4 | 2755 | bit = RB_CTX_NMI; |
567cd4da | 2756 | else if (in_irq()) |
a497adb4 | 2757 | bit = RB_CTX_IRQ; |
567cd4da | 2758 | else |
a497adb4 | 2759 | bit = RB_CTX_SOFTIRQ; |
567cd4da | 2760 | } else |
a497adb4 | 2761 | bit = RB_CTX_NORMAL; |
d9abde21 | 2762 | |
567cd4da SR |
2763 | if (unlikely(val & (1 << bit))) |
2764 | return 1; | |
d9abde21 | 2765 | |
567cd4da | 2766 | val |= (1 << bit); |
58a09ec6 | 2767 | cpu_buffer->current_context = val; |
d9abde21 | 2768 | |
567cd4da | 2769 | return 0; |
261842b7 SR |
2770 | } |
2771 | ||
58a09ec6 SRRH |
2772 | static __always_inline void |
2773 | trace_recursive_unlock(struct ring_buffer_per_cpu *cpu_buffer) | |
261842b7 | 2774 | { |
58a09ec6 | 2775 | cpu_buffer->current_context &= cpu_buffer->current_context - 1; |
261842b7 SR |
2776 | } |
2777 | ||
7a8e76a3 SR |
2778 | /** |
2779 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
2780 | * @buffer: the ring buffer to reserve from | |
2781 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
2782 | * |
2783 | * Returns a reseverd event on the ring buffer to copy directly to. | |
2784 | * The user of this interface will need to get the body to write into | |
2785 | * and can use the ring_buffer_event_data() interface. | |
2786 | * | |
2787 | * The length is the length of the data needed, not the event length | |
2788 | * which also includes the event header. | |
2789 | * | |
2790 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
2791 | * If NULL is returned, then nothing has been allocated or locked. | |
2792 | */ | |
2793 | struct ring_buffer_event * | |
0a987751 | 2794 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
2795 | { |
2796 | struct ring_buffer_per_cpu *cpu_buffer; | |
2797 | struct ring_buffer_event *event; | |
5168ae50 | 2798 | int cpu; |
7a8e76a3 | 2799 | |
bf41a158 | 2800 | /* If we are tracing schedule, we don't want to recurse */ |
5168ae50 | 2801 | preempt_disable_notrace(); |
bf41a158 | 2802 | |
3205f806 | 2803 | if (unlikely(atomic_read(&buffer->record_disabled))) |
58a09ec6 | 2804 | goto out; |
261842b7 | 2805 | |
7a8e76a3 SR |
2806 | cpu = raw_smp_processor_id(); |
2807 | ||
3205f806 | 2808 | if (unlikely(!cpumask_test_cpu(cpu, buffer->cpumask))) |
d769041f | 2809 | goto out; |
7a8e76a3 SR |
2810 | |
2811 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 2812 | |
3205f806 | 2813 | if (unlikely(atomic_read(&cpu_buffer->record_disabled))) |
d769041f | 2814 | goto out; |
7a8e76a3 | 2815 | |
3205f806 | 2816 | if (unlikely(length > BUF_MAX_DATA_SIZE)) |
bf41a158 | 2817 | goto out; |
7a8e76a3 | 2818 | |
58a09ec6 SRRH |
2819 | if (unlikely(trace_recursive_lock(cpu_buffer))) |
2820 | goto out; | |
2821 | ||
62f0b3eb | 2822 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 2823 | if (!event) |
58a09ec6 | 2824 | goto out_unlock; |
7a8e76a3 SR |
2825 | |
2826 | return event; | |
2827 | ||
58a09ec6 SRRH |
2828 | out_unlock: |
2829 | trace_recursive_unlock(cpu_buffer); | |
d769041f | 2830 | out: |
5168ae50 | 2831 | preempt_enable_notrace(); |
7a8e76a3 SR |
2832 | return NULL; |
2833 | } | |
c4f50183 | 2834 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 | 2835 | |
a1863c21 SR |
2836 | static void |
2837 | rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
7a8e76a3 SR |
2838 | struct ring_buffer_event *event) |
2839 | { | |
69d1b839 SR |
2840 | u64 delta; |
2841 | ||
fa743953 SR |
2842 | /* |
2843 | * The event first in the commit queue updates the | |
2844 | * time stamp. | |
2845 | */ | |
69d1b839 SR |
2846 | if (rb_event_is_commit(cpu_buffer, event)) { |
2847 | /* | |
2848 | * A commit event that is first on a page | |
2849 | * updates the write timestamp with the page stamp | |
2850 | */ | |
2851 | if (!rb_event_index(event)) | |
2852 | cpu_buffer->write_stamp = | |
2853 | cpu_buffer->commit_page->page->time_stamp; | |
2854 | else if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
2855 | delta = event->array[0]; | |
2856 | delta <<= TS_SHIFT; | |
2857 | delta += event->time_delta; | |
2858 | cpu_buffer->write_stamp += delta; | |
2859 | } else | |
2860 | cpu_buffer->write_stamp += event->time_delta; | |
2861 | } | |
a1863c21 | 2862 | } |
bf41a158 | 2863 | |
a1863c21 SR |
2864 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, |
2865 | struct ring_buffer_event *event) | |
2866 | { | |
2867 | local_inc(&cpu_buffer->entries); | |
2868 | rb_update_write_stamp(cpu_buffer, event); | |
fa743953 | 2869 | rb_end_commit(cpu_buffer); |
7a8e76a3 SR |
2870 | } |
2871 | ||
15693458 SRRH |
2872 | static __always_inline void |
2873 | rb_wakeups(struct ring_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer) | |
2874 | { | |
1e0d6714 SRRH |
2875 | bool pagebusy; |
2876 | ||
15693458 SRRH |
2877 | if (buffer->irq_work.waiters_pending) { |
2878 | buffer->irq_work.waiters_pending = false; | |
2879 | /* irq_work_queue() supplies it's own memory barriers */ | |
2880 | irq_work_queue(&buffer->irq_work.work); | |
2881 | } | |
2882 | ||
2883 | if (cpu_buffer->irq_work.waiters_pending) { | |
2884 | cpu_buffer->irq_work.waiters_pending = false; | |
2885 | /* irq_work_queue() supplies it's own memory barriers */ | |
2886 | irq_work_queue(&cpu_buffer->irq_work.work); | |
2887 | } | |
1e0d6714 SRRH |
2888 | |
2889 | pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page; | |
2890 | ||
2891 | if (!pagebusy && cpu_buffer->irq_work.full_waiters_pending) { | |
2892 | cpu_buffer->irq_work.wakeup_full = true; | |
2893 | cpu_buffer->irq_work.full_waiters_pending = false; | |
2894 | /* irq_work_queue() supplies it's own memory barriers */ | |
2895 | irq_work_queue(&cpu_buffer->irq_work.work); | |
2896 | } | |
15693458 SRRH |
2897 | } |
2898 | ||
7a8e76a3 SR |
2899 | /** |
2900 | * ring_buffer_unlock_commit - commit a reserved | |
2901 | * @buffer: The buffer to commit to | |
2902 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
2903 | * |
2904 | * This commits the data to the ring buffer, and releases any locks held. | |
2905 | * | |
2906 | * Must be paired with ring_buffer_lock_reserve. | |
2907 | */ | |
2908 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 2909 | struct ring_buffer_event *event) |
7a8e76a3 SR |
2910 | { |
2911 | struct ring_buffer_per_cpu *cpu_buffer; | |
2912 | int cpu = raw_smp_processor_id(); | |
2913 | ||
2914 | cpu_buffer = buffer->buffers[cpu]; | |
2915 | ||
7a8e76a3 SR |
2916 | rb_commit(cpu_buffer, event); |
2917 | ||
15693458 SRRH |
2918 | rb_wakeups(buffer, cpu_buffer); |
2919 | ||
58a09ec6 | 2920 | trace_recursive_unlock(cpu_buffer); |
261842b7 | 2921 | |
5168ae50 | 2922 | preempt_enable_notrace(); |
7a8e76a3 SR |
2923 | |
2924 | return 0; | |
2925 | } | |
c4f50183 | 2926 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 2927 | |
f3b9aae1 FW |
2928 | static inline void rb_event_discard(struct ring_buffer_event *event) |
2929 | { | |
69d1b839 SR |
2930 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) |
2931 | event = skip_time_extend(event); | |
2932 | ||
334d4169 LJ |
2933 | /* array[0] holds the actual length for the discarded event */ |
2934 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
2935 | event->type_len = RINGBUF_TYPE_PADDING; | |
f3b9aae1 FW |
2936 | /* time delta must be non zero */ |
2937 | if (!event->time_delta) | |
2938 | event->time_delta = 1; | |
2939 | } | |
2940 | ||
a1863c21 SR |
2941 | /* |
2942 | * Decrement the entries to the page that an event is on. | |
2943 | * The event does not even need to exist, only the pointer | |
2944 | * to the page it is on. This may only be called before the commit | |
2945 | * takes place. | |
2946 | */ | |
2947 | static inline void | |
2948 | rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer, | |
2949 | struct ring_buffer_event *event) | |
2950 | { | |
2951 | unsigned long addr = (unsigned long)event; | |
2952 | struct buffer_page *bpage = cpu_buffer->commit_page; | |
2953 | struct buffer_page *start; | |
2954 | ||
2955 | addr &= PAGE_MASK; | |
2956 | ||
2957 | /* Do the likely case first */ | |
2958 | if (likely(bpage->page == (void *)addr)) { | |
2959 | local_dec(&bpage->entries); | |
2960 | return; | |
2961 | } | |
2962 | ||
2963 | /* | |
2964 | * Because the commit page may be on the reader page we | |
2965 | * start with the next page and check the end loop there. | |
2966 | */ | |
2967 | rb_inc_page(cpu_buffer, &bpage); | |
2968 | start = bpage; | |
2969 | do { | |
2970 | if (bpage->page == (void *)addr) { | |
2971 | local_dec(&bpage->entries); | |
2972 | return; | |
2973 | } | |
2974 | rb_inc_page(cpu_buffer, &bpage); | |
2975 | } while (bpage != start); | |
2976 | ||
2977 | /* commit not part of this buffer?? */ | |
2978 | RB_WARN_ON(cpu_buffer, 1); | |
2979 | } | |
2980 | ||
fa1b47dd SR |
2981 | /** |
2982 | * ring_buffer_commit_discard - discard an event that has not been committed | |
2983 | * @buffer: the ring buffer | |
2984 | * @event: non committed event to discard | |
2985 | * | |
dc892f73 SR |
2986 | * Sometimes an event that is in the ring buffer needs to be ignored. |
2987 | * This function lets the user discard an event in the ring buffer | |
2988 | * and then that event will not be read later. | |
2989 | * | |
2990 | * This function only works if it is called before the the item has been | |
2991 | * committed. It will try to free the event from the ring buffer | |
fa1b47dd SR |
2992 | * if another event has not been added behind it. |
2993 | * | |
2994 | * If another event has been added behind it, it will set the event | |
2995 | * up as discarded, and perform the commit. | |
2996 | * | |
2997 | * If this function is called, do not call ring_buffer_unlock_commit on | |
2998 | * the event. | |
2999 | */ | |
3000 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
3001 | struct ring_buffer_event *event) | |
3002 | { | |
3003 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
3004 | int cpu; |
3005 | ||
3006 | /* The event is discarded regardless */ | |
f3b9aae1 | 3007 | rb_event_discard(event); |
fa1b47dd | 3008 | |
fa743953 SR |
3009 | cpu = smp_processor_id(); |
3010 | cpu_buffer = buffer->buffers[cpu]; | |
3011 | ||
fa1b47dd SR |
3012 | /* |
3013 | * This must only be called if the event has not been | |
3014 | * committed yet. Thus we can assume that preemption | |
3015 | * is still disabled. | |
3016 | */ | |
fa743953 | 3017 | RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing)); |
fa1b47dd | 3018 | |
a1863c21 | 3019 | rb_decrement_entry(cpu_buffer, event); |
0f2541d2 | 3020 | if (rb_try_to_discard(cpu_buffer, event)) |
edd813bf | 3021 | goto out; |
fa1b47dd SR |
3022 | |
3023 | /* | |
3024 | * The commit is still visible by the reader, so we | |
a1863c21 | 3025 | * must still update the timestamp. |
fa1b47dd | 3026 | */ |
a1863c21 | 3027 | rb_update_write_stamp(cpu_buffer, event); |
fa1b47dd | 3028 | out: |
fa743953 | 3029 | rb_end_commit(cpu_buffer); |
fa1b47dd | 3030 | |
58a09ec6 | 3031 | trace_recursive_unlock(cpu_buffer); |
f3b9aae1 | 3032 | |
5168ae50 | 3033 | preempt_enable_notrace(); |
fa1b47dd SR |
3034 | |
3035 | } | |
3036 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
3037 | ||
7a8e76a3 SR |
3038 | /** |
3039 | * ring_buffer_write - write data to the buffer without reserving | |
3040 | * @buffer: The ring buffer to write to. | |
3041 | * @length: The length of the data being written (excluding the event header) | |
3042 | * @data: The data to write to the buffer. | |
3043 | * | |
3044 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
3045 | * one function. If you already have the data to write to the buffer, it | |
3046 | * may be easier to simply call this function. | |
3047 | * | |
3048 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
3049 | * and not the length of the event which would hold the header. | |
3050 | */ | |
3051 | int ring_buffer_write(struct ring_buffer *buffer, | |
01e3e710 DS |
3052 | unsigned long length, |
3053 | void *data) | |
7a8e76a3 SR |
3054 | { |
3055 | struct ring_buffer_per_cpu *cpu_buffer; | |
3056 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
3057 | void *body; |
3058 | int ret = -EBUSY; | |
5168ae50 | 3059 | int cpu; |
7a8e76a3 | 3060 | |
5168ae50 | 3061 | preempt_disable_notrace(); |
bf41a158 | 3062 | |
52fbe9cd LJ |
3063 | if (atomic_read(&buffer->record_disabled)) |
3064 | goto out; | |
3065 | ||
7a8e76a3 SR |
3066 | cpu = raw_smp_processor_id(); |
3067 | ||
9e01c1b7 | 3068 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 3069 | goto out; |
7a8e76a3 SR |
3070 | |
3071 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
3072 | |
3073 | if (atomic_read(&cpu_buffer->record_disabled)) | |
3074 | goto out; | |
3075 | ||
be957c44 SR |
3076 | if (length > BUF_MAX_DATA_SIZE) |
3077 | goto out; | |
3078 | ||
985e871b SRRH |
3079 | if (unlikely(trace_recursive_lock(cpu_buffer))) |
3080 | goto out; | |
3081 | ||
62f0b3eb | 3082 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 3083 | if (!event) |
985e871b | 3084 | goto out_unlock; |
7a8e76a3 SR |
3085 | |
3086 | body = rb_event_data(event); | |
3087 | ||
3088 | memcpy(body, data, length); | |
3089 | ||
3090 | rb_commit(cpu_buffer, event); | |
3091 | ||
15693458 SRRH |
3092 | rb_wakeups(buffer, cpu_buffer); |
3093 | ||
7a8e76a3 | 3094 | ret = 0; |
985e871b SRRH |
3095 | |
3096 | out_unlock: | |
3097 | trace_recursive_unlock(cpu_buffer); | |
3098 | ||
7a8e76a3 | 3099 | out: |
5168ae50 | 3100 | preempt_enable_notrace(); |
7a8e76a3 SR |
3101 | |
3102 | return ret; | |
3103 | } | |
c4f50183 | 3104 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 3105 | |
34a148bf | 3106 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
3107 | { |
3108 | struct buffer_page *reader = cpu_buffer->reader_page; | |
77ae365e | 3109 | struct buffer_page *head = rb_set_head_page(cpu_buffer); |
bf41a158 SR |
3110 | struct buffer_page *commit = cpu_buffer->commit_page; |
3111 | ||
77ae365e SR |
3112 | /* In case of error, head will be NULL */ |
3113 | if (unlikely(!head)) | |
3114 | return 1; | |
3115 | ||
bf41a158 SR |
3116 | return reader->read == rb_page_commit(reader) && |
3117 | (commit == reader || | |
3118 | (commit == head && | |
3119 | head->read == rb_page_commit(commit))); | |
3120 | } | |
3121 | ||
7a8e76a3 SR |
3122 | /** |
3123 | * ring_buffer_record_disable - stop all writes into the buffer | |
3124 | * @buffer: The ring buffer to stop writes to. | |
3125 | * | |
3126 | * This prevents all writes to the buffer. Any attempt to write | |
3127 | * to the buffer after this will fail and return NULL. | |
3128 | * | |
3129 | * The caller should call synchronize_sched() after this. | |
3130 | */ | |
3131 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
3132 | { | |
3133 | atomic_inc(&buffer->record_disabled); | |
3134 | } | |
c4f50183 | 3135 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
3136 | |
3137 | /** | |
3138 | * ring_buffer_record_enable - enable writes to the buffer | |
3139 | * @buffer: The ring buffer to enable writes | |
3140 | * | |
3141 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 3142 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
3143 | */ |
3144 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
3145 | { | |
3146 | atomic_dec(&buffer->record_disabled); | |
3147 | } | |
c4f50183 | 3148 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 | 3149 | |
499e5470 SR |
3150 | /** |
3151 | * ring_buffer_record_off - stop all writes into the buffer | |
3152 | * @buffer: The ring buffer to stop writes to. | |
3153 | * | |
3154 | * This prevents all writes to the buffer. Any attempt to write | |
3155 | * to the buffer after this will fail and return NULL. | |
3156 | * | |
3157 | * This is different than ring_buffer_record_disable() as | |
87abb3b1 | 3158 | * it works like an on/off switch, where as the disable() version |
499e5470 SR |
3159 | * must be paired with a enable(). |
3160 | */ | |
3161 | void ring_buffer_record_off(struct ring_buffer *buffer) | |
3162 | { | |
3163 | unsigned int rd; | |
3164 | unsigned int new_rd; | |
3165 | ||
3166 | do { | |
3167 | rd = atomic_read(&buffer->record_disabled); | |
3168 | new_rd = rd | RB_BUFFER_OFF; | |
3169 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
3170 | } | |
3171 | EXPORT_SYMBOL_GPL(ring_buffer_record_off); | |
3172 | ||
3173 | /** | |
3174 | * ring_buffer_record_on - restart writes into the buffer | |
3175 | * @buffer: The ring buffer to start writes to. | |
3176 | * | |
3177 | * This enables all writes to the buffer that was disabled by | |
3178 | * ring_buffer_record_off(). | |
3179 | * | |
3180 | * This is different than ring_buffer_record_enable() as | |
87abb3b1 | 3181 | * it works like an on/off switch, where as the enable() version |
499e5470 SR |
3182 | * must be paired with a disable(). |
3183 | */ | |
3184 | void ring_buffer_record_on(struct ring_buffer *buffer) | |
3185 | { | |
3186 | unsigned int rd; | |
3187 | unsigned int new_rd; | |
3188 | ||
3189 | do { | |
3190 | rd = atomic_read(&buffer->record_disabled); | |
3191 | new_rd = rd & ~RB_BUFFER_OFF; | |
3192 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
3193 | } | |
3194 | EXPORT_SYMBOL_GPL(ring_buffer_record_on); | |
3195 | ||
3196 | /** | |
3197 | * ring_buffer_record_is_on - return true if the ring buffer can write | |
3198 | * @buffer: The ring buffer to see if write is enabled | |
3199 | * | |
3200 | * Returns true if the ring buffer is in a state that it accepts writes. | |
3201 | */ | |
3202 | int ring_buffer_record_is_on(struct ring_buffer *buffer) | |
3203 | { | |
3204 | return !atomic_read(&buffer->record_disabled); | |
3205 | } | |
3206 | ||
7a8e76a3 SR |
3207 | /** |
3208 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
3209 | * @buffer: The ring buffer to stop writes to. | |
3210 | * @cpu: The CPU buffer to stop | |
3211 | * | |
3212 | * This prevents all writes to the buffer. Any attempt to write | |
3213 | * to the buffer after this will fail and return NULL. | |
3214 | * | |
3215 | * The caller should call synchronize_sched() after this. | |
3216 | */ | |
3217 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
3218 | { | |
3219 | struct ring_buffer_per_cpu *cpu_buffer; | |
3220 | ||
9e01c1b7 | 3221 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3222 | return; |
7a8e76a3 SR |
3223 | |
3224 | cpu_buffer = buffer->buffers[cpu]; | |
3225 | atomic_inc(&cpu_buffer->record_disabled); | |
3226 | } | |
c4f50183 | 3227 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
3228 | |
3229 | /** | |
3230 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
3231 | * @buffer: The ring buffer to enable writes | |
3232 | * @cpu: The CPU to enable. | |
3233 | * | |
3234 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 3235 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
3236 | */ |
3237 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
3238 | { | |
3239 | struct ring_buffer_per_cpu *cpu_buffer; | |
3240 | ||
9e01c1b7 | 3241 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3242 | return; |
7a8e76a3 SR |
3243 | |
3244 | cpu_buffer = buffer->buffers[cpu]; | |
3245 | atomic_dec(&cpu_buffer->record_disabled); | |
3246 | } | |
c4f50183 | 3247 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 | 3248 | |
f6195aa0 SR |
3249 | /* |
3250 | * The total entries in the ring buffer is the running counter | |
3251 | * of entries entered into the ring buffer, minus the sum of | |
3252 | * the entries read from the ring buffer and the number of | |
3253 | * entries that were overwritten. | |
3254 | */ | |
3255 | static inline unsigned long | |
3256 | rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer) | |
3257 | { | |
3258 | return local_read(&cpu_buffer->entries) - | |
3259 | (local_read(&cpu_buffer->overrun) + cpu_buffer->read); | |
3260 | } | |
3261 | ||
c64e148a VN |
3262 | /** |
3263 | * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer | |
3264 | * @buffer: The ring buffer | |
3265 | * @cpu: The per CPU buffer to read from. | |
3266 | */ | |
50ecf2c3 | 3267 | u64 ring_buffer_oldest_event_ts(struct ring_buffer *buffer, int cpu) |
c64e148a VN |
3268 | { |
3269 | unsigned long flags; | |
3270 | struct ring_buffer_per_cpu *cpu_buffer; | |
3271 | struct buffer_page *bpage; | |
da830e58 | 3272 | u64 ret = 0; |
c64e148a VN |
3273 | |
3274 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3275 | return 0; | |
3276 | ||
3277 | cpu_buffer = buffer->buffers[cpu]; | |
7115e3fc | 3278 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
3279 | /* |
3280 | * if the tail is on reader_page, oldest time stamp is on the reader | |
3281 | * page | |
3282 | */ | |
3283 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
3284 | bpage = cpu_buffer->reader_page; | |
3285 | else | |
3286 | bpage = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
3287 | if (bpage) |
3288 | ret = bpage->page->time_stamp; | |
7115e3fc | 3289 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
3290 | |
3291 | return ret; | |
3292 | } | |
3293 | EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts); | |
3294 | ||
3295 | /** | |
3296 | * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer | |
3297 | * @buffer: The ring buffer | |
3298 | * @cpu: The per CPU buffer to read from. | |
3299 | */ | |
3300 | unsigned long ring_buffer_bytes_cpu(struct ring_buffer *buffer, int cpu) | |
3301 | { | |
3302 | struct ring_buffer_per_cpu *cpu_buffer; | |
3303 | unsigned long ret; | |
3304 | ||
3305 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3306 | return 0; | |
3307 | ||
3308 | cpu_buffer = buffer->buffers[cpu]; | |
3309 | ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes; | |
3310 | ||
3311 | return ret; | |
3312 | } | |
3313 | EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu); | |
3314 | ||
7a8e76a3 SR |
3315 | /** |
3316 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
3317 | * @buffer: The ring buffer | |
3318 | * @cpu: The per CPU buffer to get the entries from. | |
3319 | */ | |
3320 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
3321 | { | |
3322 | struct ring_buffer_per_cpu *cpu_buffer; | |
3323 | ||
9e01c1b7 | 3324 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3325 | return 0; |
7a8e76a3 SR |
3326 | |
3327 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 3328 | |
f6195aa0 | 3329 | return rb_num_of_entries(cpu_buffer); |
7a8e76a3 | 3330 | } |
c4f50183 | 3331 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
3332 | |
3333 | /** | |
884bfe89 SP |
3334 | * ring_buffer_overrun_cpu - get the number of overruns caused by the ring |
3335 | * buffer wrapping around (only if RB_FL_OVERWRITE is on). | |
7a8e76a3 SR |
3336 | * @buffer: The ring buffer |
3337 | * @cpu: The per CPU buffer to get the number of overruns from | |
3338 | */ | |
3339 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3340 | { | |
3341 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3342 | unsigned long ret; |
7a8e76a3 | 3343 | |
9e01c1b7 | 3344 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3345 | return 0; |
7a8e76a3 SR |
3346 | |
3347 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3348 | ret = local_read(&cpu_buffer->overrun); |
554f786e SR |
3349 | |
3350 | return ret; | |
7a8e76a3 | 3351 | } |
c4f50183 | 3352 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 3353 | |
f0d2c681 | 3354 | /** |
884bfe89 SP |
3355 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by |
3356 | * commits failing due to the buffer wrapping around while there are uncommitted | |
3357 | * events, such as during an interrupt storm. | |
f0d2c681 SR |
3358 | * @buffer: The ring buffer |
3359 | * @cpu: The per CPU buffer to get the number of overruns from | |
3360 | */ | |
3361 | unsigned long | |
3362 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3363 | { | |
3364 | struct ring_buffer_per_cpu *cpu_buffer; | |
3365 | unsigned long ret; | |
3366 | ||
3367 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3368 | return 0; | |
3369 | ||
3370 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3371 | ret = local_read(&cpu_buffer->commit_overrun); |
f0d2c681 SR |
3372 | |
3373 | return ret; | |
3374 | } | |
3375 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
3376 | ||
884bfe89 SP |
3377 | /** |
3378 | * ring_buffer_dropped_events_cpu - get the number of dropped events caused by | |
3379 | * the ring buffer filling up (only if RB_FL_OVERWRITE is off). | |
3380 | * @buffer: The ring buffer | |
3381 | * @cpu: The per CPU buffer to get the number of overruns from | |
3382 | */ | |
3383 | unsigned long | |
3384 | ring_buffer_dropped_events_cpu(struct ring_buffer *buffer, int cpu) | |
3385 | { | |
3386 | struct ring_buffer_per_cpu *cpu_buffer; | |
3387 | unsigned long ret; | |
3388 | ||
3389 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3390 | return 0; | |
3391 | ||
3392 | cpu_buffer = buffer->buffers[cpu]; | |
3393 | ret = local_read(&cpu_buffer->dropped_events); | |
3394 | ||
3395 | return ret; | |
3396 | } | |
3397 | EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu); | |
3398 | ||
ad964704 SRRH |
3399 | /** |
3400 | * ring_buffer_read_events_cpu - get the number of events successfully read | |
3401 | * @buffer: The ring buffer | |
3402 | * @cpu: The per CPU buffer to get the number of events read | |
3403 | */ | |
3404 | unsigned long | |
3405 | ring_buffer_read_events_cpu(struct ring_buffer *buffer, int cpu) | |
3406 | { | |
3407 | struct ring_buffer_per_cpu *cpu_buffer; | |
3408 | ||
3409 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3410 | return 0; | |
3411 | ||
3412 | cpu_buffer = buffer->buffers[cpu]; | |
3413 | return cpu_buffer->read; | |
3414 | } | |
3415 | EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu); | |
3416 | ||
7a8e76a3 SR |
3417 | /** |
3418 | * ring_buffer_entries - get the number of entries in a buffer | |
3419 | * @buffer: The ring buffer | |
3420 | * | |
3421 | * Returns the total number of entries in the ring buffer | |
3422 | * (all CPU entries) | |
3423 | */ | |
3424 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
3425 | { | |
3426 | struct ring_buffer_per_cpu *cpu_buffer; | |
3427 | unsigned long entries = 0; | |
3428 | int cpu; | |
3429 | ||
3430 | /* if you care about this being correct, lock the buffer */ | |
3431 | for_each_buffer_cpu(buffer, cpu) { | |
3432 | cpu_buffer = buffer->buffers[cpu]; | |
f6195aa0 | 3433 | entries += rb_num_of_entries(cpu_buffer); |
7a8e76a3 SR |
3434 | } |
3435 | ||
3436 | return entries; | |
3437 | } | |
c4f50183 | 3438 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
3439 | |
3440 | /** | |
67b394f7 | 3441 | * ring_buffer_overruns - get the number of overruns in buffer |
7a8e76a3 SR |
3442 | * @buffer: The ring buffer |
3443 | * | |
3444 | * Returns the total number of overruns in the ring buffer | |
3445 | * (all CPU entries) | |
3446 | */ | |
3447 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
3448 | { | |
3449 | struct ring_buffer_per_cpu *cpu_buffer; | |
3450 | unsigned long overruns = 0; | |
3451 | int cpu; | |
3452 | ||
3453 | /* if you care about this being correct, lock the buffer */ | |
3454 | for_each_buffer_cpu(buffer, cpu) { | |
3455 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3456 | overruns += local_read(&cpu_buffer->overrun); |
7a8e76a3 SR |
3457 | } |
3458 | ||
3459 | return overruns; | |
3460 | } | |
c4f50183 | 3461 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 3462 | |
642edba5 | 3463 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
3464 | { |
3465 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3466 | ||
d769041f | 3467 | /* Iterator usage is expected to have record disabled */ |
651e22f2 SRRH |
3468 | iter->head_page = cpu_buffer->reader_page; |
3469 | iter->head = cpu_buffer->reader_page->read; | |
3470 | ||
3471 | iter->cache_reader_page = iter->head_page; | |
24607f11 | 3472 | iter->cache_read = cpu_buffer->read; |
651e22f2 | 3473 | |
d769041f SR |
3474 | if (iter->head) |
3475 | iter->read_stamp = cpu_buffer->read_stamp; | |
3476 | else | |
abc9b56d | 3477 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 3478 | } |
f83c9d0f | 3479 | |
642edba5 SR |
3480 | /** |
3481 | * ring_buffer_iter_reset - reset an iterator | |
3482 | * @iter: The iterator to reset | |
3483 | * | |
3484 | * Resets the iterator, so that it will start from the beginning | |
3485 | * again. | |
3486 | */ | |
3487 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
3488 | { | |
554f786e | 3489 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
3490 | unsigned long flags; |
3491 | ||
554f786e SR |
3492 | if (!iter) |
3493 | return; | |
3494 | ||
3495 | cpu_buffer = iter->cpu_buffer; | |
3496 | ||
5389f6fa | 3497 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
642edba5 | 3498 | rb_iter_reset(iter); |
5389f6fa | 3499 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 3500 | } |
c4f50183 | 3501 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
3502 | |
3503 | /** | |
3504 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
3505 | * @iter: The iterator to check | |
3506 | */ | |
3507 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
3508 | { | |
3509 | struct ring_buffer_per_cpu *cpu_buffer; | |
3510 | ||
3511 | cpu_buffer = iter->cpu_buffer; | |
3512 | ||
bf41a158 SR |
3513 | return iter->head_page == cpu_buffer->commit_page && |
3514 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 3515 | } |
c4f50183 | 3516 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
3517 | |
3518 | static void | |
3519 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
3520 | struct ring_buffer_event *event) | |
3521 | { | |
3522 | u64 delta; | |
3523 | ||
334d4169 | 3524 | switch (event->type_len) { |
7a8e76a3 SR |
3525 | case RINGBUF_TYPE_PADDING: |
3526 | return; | |
3527 | ||
3528 | case RINGBUF_TYPE_TIME_EXTEND: | |
3529 | delta = event->array[0]; | |
3530 | delta <<= TS_SHIFT; | |
3531 | delta += event->time_delta; | |
3532 | cpu_buffer->read_stamp += delta; | |
3533 | return; | |
3534 | ||
3535 | case RINGBUF_TYPE_TIME_STAMP: | |
3536 | /* FIXME: not implemented */ | |
3537 | return; | |
3538 | ||
3539 | case RINGBUF_TYPE_DATA: | |
3540 | cpu_buffer->read_stamp += event->time_delta; | |
3541 | return; | |
3542 | ||
3543 | default: | |
3544 | BUG(); | |
3545 | } | |
3546 | return; | |
3547 | } | |
3548 | ||
3549 | static void | |
3550 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
3551 | struct ring_buffer_event *event) | |
3552 | { | |
3553 | u64 delta; | |
3554 | ||
334d4169 | 3555 | switch (event->type_len) { |
7a8e76a3 SR |
3556 | case RINGBUF_TYPE_PADDING: |
3557 | return; | |
3558 | ||
3559 | case RINGBUF_TYPE_TIME_EXTEND: | |
3560 | delta = event->array[0]; | |
3561 | delta <<= TS_SHIFT; | |
3562 | delta += event->time_delta; | |
3563 | iter->read_stamp += delta; | |
3564 | return; | |
3565 | ||
3566 | case RINGBUF_TYPE_TIME_STAMP: | |
3567 | /* FIXME: not implemented */ | |
3568 | return; | |
3569 | ||
3570 | case RINGBUF_TYPE_DATA: | |
3571 | iter->read_stamp += event->time_delta; | |
3572 | return; | |
3573 | ||
3574 | default: | |
3575 | BUG(); | |
3576 | } | |
3577 | return; | |
3578 | } | |
3579 | ||
d769041f SR |
3580 | static struct buffer_page * |
3581 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 3582 | { |
d769041f | 3583 | struct buffer_page *reader = NULL; |
66a8cb95 | 3584 | unsigned long overwrite; |
d769041f | 3585 | unsigned long flags; |
818e3dd3 | 3586 | int nr_loops = 0; |
77ae365e | 3587 | int ret; |
d769041f | 3588 | |
3e03fb7f | 3589 | local_irq_save(flags); |
0199c4e6 | 3590 | arch_spin_lock(&cpu_buffer->lock); |
d769041f SR |
3591 | |
3592 | again: | |
818e3dd3 SR |
3593 | /* |
3594 | * This should normally only loop twice. But because the | |
3595 | * start of the reader inserts an empty page, it causes | |
3596 | * a case where we will loop three times. There should be no | |
3597 | * reason to loop four times (that I know of). | |
3598 | */ | |
3e89c7bb | 3599 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
3600 | reader = NULL; |
3601 | goto out; | |
3602 | } | |
3603 | ||
d769041f SR |
3604 | reader = cpu_buffer->reader_page; |
3605 | ||
3606 | /* If there's more to read, return this page */ | |
bf41a158 | 3607 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
3608 | goto out; |
3609 | ||
3610 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
3611 | if (RB_WARN_ON(cpu_buffer, |
3612 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
3613 | goto out; | |
d769041f SR |
3614 | |
3615 | /* check if we caught up to the tail */ | |
3616 | reader = NULL; | |
bf41a158 | 3617 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 3618 | goto out; |
7a8e76a3 | 3619 | |
a5fb8331 SR |
3620 | /* Don't bother swapping if the ring buffer is empty */ |
3621 | if (rb_num_of_entries(cpu_buffer) == 0) | |
3622 | goto out; | |
3623 | ||
7a8e76a3 | 3624 | /* |
d769041f | 3625 | * Reset the reader page to size zero. |
7a8e76a3 | 3626 | */ |
77ae365e SR |
3627 | local_set(&cpu_buffer->reader_page->write, 0); |
3628 | local_set(&cpu_buffer->reader_page->entries, 0); | |
3629 | local_set(&cpu_buffer->reader_page->page->commit, 0); | |
ff0ff84a | 3630 | cpu_buffer->reader_page->real_end = 0; |
7a8e76a3 | 3631 | |
77ae365e SR |
3632 | spin: |
3633 | /* | |
3634 | * Splice the empty reader page into the list around the head. | |
3635 | */ | |
3636 | reader = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
3637 | if (!reader) |
3638 | goto out; | |
0e1ff5d7 | 3639 | cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next); |
d769041f | 3640 | cpu_buffer->reader_page->list.prev = reader->list.prev; |
bf41a158 | 3641 | |
3adc54fa SR |
3642 | /* |
3643 | * cpu_buffer->pages just needs to point to the buffer, it | |
3644 | * has no specific buffer page to point to. Lets move it out | |
25985edc | 3645 | * of our way so we don't accidentally swap it. |
3adc54fa SR |
3646 | */ |
3647 | cpu_buffer->pages = reader->list.prev; | |
3648 | ||
77ae365e SR |
3649 | /* The reader page will be pointing to the new head */ |
3650 | rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list); | |
7a8e76a3 | 3651 | |
66a8cb95 SR |
3652 | /* |
3653 | * We want to make sure we read the overruns after we set up our | |
3654 | * pointers to the next object. The writer side does a | |
3655 | * cmpxchg to cross pages which acts as the mb on the writer | |
3656 | * side. Note, the reader will constantly fail the swap | |
3657 | * while the writer is updating the pointers, so this | |
3658 | * guarantees that the overwrite recorded here is the one we | |
3659 | * want to compare with the last_overrun. | |
3660 | */ | |
3661 | smp_mb(); | |
3662 | overwrite = local_read(&(cpu_buffer->overrun)); | |
3663 | ||
77ae365e SR |
3664 | /* |
3665 | * Here's the tricky part. | |
3666 | * | |
3667 | * We need to move the pointer past the header page. | |
3668 | * But we can only do that if a writer is not currently | |
3669 | * moving it. The page before the header page has the | |
3670 | * flag bit '1' set if it is pointing to the page we want. | |
3671 | * but if the writer is in the process of moving it | |
3672 | * than it will be '2' or already moved '0'. | |
3673 | */ | |
3674 | ||
3675 | ret = rb_head_page_replace(reader, cpu_buffer->reader_page); | |
7a8e76a3 SR |
3676 | |
3677 | /* | |
77ae365e | 3678 | * If we did not convert it, then we must try again. |
7a8e76a3 | 3679 | */ |
77ae365e SR |
3680 | if (!ret) |
3681 | goto spin; | |
7a8e76a3 | 3682 | |
77ae365e SR |
3683 | /* |
3684 | * Yeah! We succeeded in replacing the page. | |
3685 | * | |
3686 | * Now make the new head point back to the reader page. | |
3687 | */ | |
5ded3dc6 | 3688 | rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list; |
77ae365e | 3689 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
d769041f SR |
3690 | |
3691 | /* Finally update the reader page to the new head */ | |
3692 | cpu_buffer->reader_page = reader; | |
3693 | rb_reset_reader_page(cpu_buffer); | |
3694 | ||
66a8cb95 SR |
3695 | if (overwrite != cpu_buffer->last_overrun) { |
3696 | cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun; | |
3697 | cpu_buffer->last_overrun = overwrite; | |
3698 | } | |
3699 | ||
d769041f SR |
3700 | goto again; |
3701 | ||
3702 | out: | |
0199c4e6 | 3703 | arch_spin_unlock(&cpu_buffer->lock); |
3e03fb7f | 3704 | local_irq_restore(flags); |
d769041f SR |
3705 | |
3706 | return reader; | |
3707 | } | |
3708 | ||
3709 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
3710 | { | |
3711 | struct ring_buffer_event *event; | |
3712 | struct buffer_page *reader; | |
3713 | unsigned length; | |
3714 | ||
3715 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 3716 | |
d769041f | 3717 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
3718 | if (RB_WARN_ON(cpu_buffer, !reader)) |
3719 | return; | |
7a8e76a3 | 3720 | |
d769041f SR |
3721 | event = rb_reader_event(cpu_buffer); |
3722 | ||
a1863c21 | 3723 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
e4906eff | 3724 | cpu_buffer->read++; |
d769041f SR |
3725 | |
3726 | rb_update_read_stamp(cpu_buffer, event); | |
3727 | ||
3728 | length = rb_event_length(event); | |
6f807acd | 3729 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
3730 | } |
3731 | ||
3732 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
3733 | { | |
7a8e76a3 SR |
3734 | struct ring_buffer_per_cpu *cpu_buffer; |
3735 | struct ring_buffer_event *event; | |
3736 | unsigned length; | |
3737 | ||
3738 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 SR |
3739 | |
3740 | /* | |
3741 | * Check if we are at the end of the buffer. | |
3742 | */ | |
bf41a158 | 3743 | if (iter->head >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
3744 | /* discarded commits can make the page empty */ |
3745 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 3746 | return; |
d769041f | 3747 | rb_inc_iter(iter); |
7a8e76a3 SR |
3748 | return; |
3749 | } | |
3750 | ||
3751 | event = rb_iter_head_event(iter); | |
3752 | ||
3753 | length = rb_event_length(event); | |
3754 | ||
3755 | /* | |
3756 | * This should not be called to advance the header if we are | |
3757 | * at the tail of the buffer. | |
3758 | */ | |
3e89c7bb | 3759 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 3760 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
3761 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
3762 | return; | |
7a8e76a3 SR |
3763 | |
3764 | rb_update_iter_read_stamp(iter, event); | |
3765 | ||
3766 | iter->head += length; | |
3767 | ||
3768 | /* check for end of page padding */ | |
bf41a158 SR |
3769 | if ((iter->head >= rb_page_size(iter->head_page)) && |
3770 | (iter->head_page != cpu_buffer->commit_page)) | |
771e0384 | 3771 | rb_inc_iter(iter); |
7a8e76a3 SR |
3772 | } |
3773 | ||
66a8cb95 SR |
3774 | static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer) |
3775 | { | |
3776 | return cpu_buffer->lost_events; | |
3777 | } | |
3778 | ||
f83c9d0f | 3779 | static struct ring_buffer_event * |
66a8cb95 SR |
3780 | rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts, |
3781 | unsigned long *lost_events) | |
7a8e76a3 | 3782 | { |
7a8e76a3 | 3783 | struct ring_buffer_event *event; |
d769041f | 3784 | struct buffer_page *reader; |
818e3dd3 | 3785 | int nr_loops = 0; |
7a8e76a3 | 3786 | |
7a8e76a3 | 3787 | again: |
818e3dd3 | 3788 | /* |
69d1b839 SR |
3789 | * We repeat when a time extend is encountered. |
3790 | * Since the time extend is always attached to a data event, | |
3791 | * we should never loop more than once. | |
3792 | * (We never hit the following condition more than twice). | |
818e3dd3 | 3793 | */ |
69d1b839 | 3794 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 3795 | return NULL; |
818e3dd3 | 3796 | |
d769041f SR |
3797 | reader = rb_get_reader_page(cpu_buffer); |
3798 | if (!reader) | |
7a8e76a3 SR |
3799 | return NULL; |
3800 | ||
d769041f | 3801 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 3802 | |
334d4169 | 3803 | switch (event->type_len) { |
7a8e76a3 | 3804 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3805 | if (rb_null_event(event)) |
3806 | RB_WARN_ON(cpu_buffer, 1); | |
3807 | /* | |
3808 | * Because the writer could be discarding every | |
3809 | * event it creates (which would probably be bad) | |
3810 | * if we were to go back to "again" then we may never | |
3811 | * catch up, and will trigger the warn on, or lock | |
3812 | * the box. Return the padding, and we will release | |
3813 | * the current locks, and try again. | |
3814 | */ | |
2d622719 | 3815 | return event; |
7a8e76a3 SR |
3816 | |
3817 | case RINGBUF_TYPE_TIME_EXTEND: | |
3818 | /* Internal data, OK to advance */ | |
d769041f | 3819 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3820 | goto again; |
3821 | ||
3822 | case RINGBUF_TYPE_TIME_STAMP: | |
3823 | /* FIXME: not implemented */ | |
d769041f | 3824 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3825 | goto again; |
3826 | ||
3827 | case RINGBUF_TYPE_DATA: | |
3828 | if (ts) { | |
3829 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
d8eeb2d3 | 3830 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, |
37886f6a | 3831 | cpu_buffer->cpu, ts); |
7a8e76a3 | 3832 | } |
66a8cb95 SR |
3833 | if (lost_events) |
3834 | *lost_events = rb_lost_events(cpu_buffer); | |
7a8e76a3 SR |
3835 | return event; |
3836 | ||
3837 | default: | |
3838 | BUG(); | |
3839 | } | |
3840 | ||
3841 | return NULL; | |
3842 | } | |
c4f50183 | 3843 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 3844 | |
f83c9d0f SR |
3845 | static struct ring_buffer_event * |
3846 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
3847 | { |
3848 | struct ring_buffer *buffer; | |
3849 | struct ring_buffer_per_cpu *cpu_buffer; | |
3850 | struct ring_buffer_event *event; | |
818e3dd3 | 3851 | int nr_loops = 0; |
7a8e76a3 | 3852 | |
7a8e76a3 SR |
3853 | cpu_buffer = iter->cpu_buffer; |
3854 | buffer = cpu_buffer->buffer; | |
3855 | ||
492a74f4 SR |
3856 | /* |
3857 | * Check if someone performed a consuming read to | |
3858 | * the buffer. A consuming read invalidates the iterator | |
3859 | * and we need to reset the iterator in this case. | |
3860 | */ | |
3861 | if (unlikely(iter->cache_read != cpu_buffer->read || | |
3862 | iter->cache_reader_page != cpu_buffer->reader_page)) | |
3863 | rb_iter_reset(iter); | |
3864 | ||
7a8e76a3 | 3865 | again: |
3c05d748 SR |
3866 | if (ring_buffer_iter_empty(iter)) |
3867 | return NULL; | |
3868 | ||
818e3dd3 | 3869 | /* |
021de3d9 SRRH |
3870 | * We repeat when a time extend is encountered or we hit |
3871 | * the end of the page. Since the time extend is always attached | |
3872 | * to a data event, we should never loop more than three times. | |
3873 | * Once for going to next page, once on time extend, and | |
3874 | * finally once to get the event. | |
3875 | * (We never hit the following condition more than thrice). | |
818e3dd3 | 3876 | */ |
021de3d9 | 3877 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) |
818e3dd3 | 3878 | return NULL; |
818e3dd3 | 3879 | |
7a8e76a3 SR |
3880 | if (rb_per_cpu_empty(cpu_buffer)) |
3881 | return NULL; | |
3882 | ||
10e83fd0 | 3883 | if (iter->head >= rb_page_size(iter->head_page)) { |
3c05d748 SR |
3884 | rb_inc_iter(iter); |
3885 | goto again; | |
3886 | } | |
3887 | ||
7a8e76a3 SR |
3888 | event = rb_iter_head_event(iter); |
3889 | ||
334d4169 | 3890 | switch (event->type_len) { |
7a8e76a3 | 3891 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3892 | if (rb_null_event(event)) { |
3893 | rb_inc_iter(iter); | |
3894 | goto again; | |
3895 | } | |
3896 | rb_advance_iter(iter); | |
3897 | return event; | |
7a8e76a3 SR |
3898 | |
3899 | case RINGBUF_TYPE_TIME_EXTEND: | |
3900 | /* Internal data, OK to advance */ | |
3901 | rb_advance_iter(iter); | |
3902 | goto again; | |
3903 | ||
3904 | case RINGBUF_TYPE_TIME_STAMP: | |
3905 | /* FIXME: not implemented */ | |
3906 | rb_advance_iter(iter); | |
3907 | goto again; | |
3908 | ||
3909 | case RINGBUF_TYPE_DATA: | |
3910 | if (ts) { | |
3911 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
3912 | ring_buffer_normalize_time_stamp(buffer, |
3913 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
3914 | } |
3915 | return event; | |
3916 | ||
3917 | default: | |
3918 | BUG(); | |
3919 | } | |
3920 | ||
3921 | return NULL; | |
3922 | } | |
c4f50183 | 3923 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 3924 | |
289a5a25 | 3925 | static inline bool rb_reader_lock(struct ring_buffer_per_cpu *cpu_buffer) |
8d707e8e | 3926 | { |
289a5a25 SRRH |
3927 | if (likely(!in_nmi())) { |
3928 | raw_spin_lock(&cpu_buffer->reader_lock); | |
3929 | return true; | |
3930 | } | |
3931 | ||
8d707e8e SR |
3932 | /* |
3933 | * If an NMI die dumps out the content of the ring buffer | |
289a5a25 SRRH |
3934 | * trylock must be used to prevent a deadlock if the NMI |
3935 | * preempted a task that holds the ring buffer locks. If | |
3936 | * we get the lock then all is fine, if not, then continue | |
3937 | * to do the read, but this can corrupt the ring buffer, | |
3938 | * so it must be permanently disabled from future writes. | |
3939 | * Reading from NMI is a oneshot deal. | |
8d707e8e | 3940 | */ |
289a5a25 SRRH |
3941 | if (raw_spin_trylock(&cpu_buffer->reader_lock)) |
3942 | return true; | |
8d707e8e | 3943 | |
289a5a25 SRRH |
3944 | /* Continue without locking, but disable the ring buffer */ |
3945 | atomic_inc(&cpu_buffer->record_disabled); | |
3946 | return false; | |
3947 | } | |
3948 | ||
3949 | static inline void | |
3950 | rb_reader_unlock(struct ring_buffer_per_cpu *cpu_buffer, bool locked) | |
3951 | { | |
3952 | if (likely(locked)) | |
3953 | raw_spin_unlock(&cpu_buffer->reader_lock); | |
3954 | return; | |
8d707e8e SR |
3955 | } |
3956 | ||
f83c9d0f SR |
3957 | /** |
3958 | * ring_buffer_peek - peek at the next event to be read | |
3959 | * @buffer: The ring buffer to read | |
3960 | * @cpu: The cpu to peak at | |
3961 | * @ts: The timestamp counter of this event. | |
66a8cb95 | 3962 | * @lost_events: a variable to store if events were lost (may be NULL) |
f83c9d0f SR |
3963 | * |
3964 | * This will return the event that will be read next, but does | |
3965 | * not consume the data. | |
3966 | */ | |
3967 | struct ring_buffer_event * | |
66a8cb95 SR |
3968 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts, |
3969 | unsigned long *lost_events) | |
f83c9d0f SR |
3970 | { |
3971 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 3972 | struct ring_buffer_event *event; |
f83c9d0f | 3973 | unsigned long flags; |
289a5a25 | 3974 | bool dolock; |
f83c9d0f | 3975 | |
554f786e | 3976 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3977 | return NULL; |
554f786e | 3978 | |
2d622719 | 3979 | again: |
8d707e8e | 3980 | local_irq_save(flags); |
289a5a25 | 3981 | dolock = rb_reader_lock(cpu_buffer); |
66a8cb95 | 3982 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
469535a5 RR |
3983 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
3984 | rb_advance_reader(cpu_buffer); | |
289a5a25 | 3985 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e | 3986 | local_irq_restore(flags); |
f83c9d0f | 3987 | |
1b959e18 | 3988 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3989 | goto again; |
2d622719 | 3990 | |
f83c9d0f SR |
3991 | return event; |
3992 | } | |
3993 | ||
3994 | /** | |
3995 | * ring_buffer_iter_peek - peek at the next event to be read | |
3996 | * @iter: The ring buffer iterator | |
3997 | * @ts: The timestamp counter of this event. | |
3998 | * | |
3999 | * This will return the event that will be read next, but does | |
4000 | * not increment the iterator. | |
4001 | */ | |
4002 | struct ring_buffer_event * | |
4003 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
4004 | { | |
4005 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
4006 | struct ring_buffer_event *event; | |
4007 | unsigned long flags; | |
4008 | ||
2d622719 | 4009 | again: |
5389f6fa | 4010 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 4011 | event = rb_iter_peek(iter, ts); |
5389f6fa | 4012 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 4013 | |
1b959e18 | 4014 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 4015 | goto again; |
2d622719 | 4016 | |
f83c9d0f SR |
4017 | return event; |
4018 | } | |
4019 | ||
7a8e76a3 SR |
4020 | /** |
4021 | * ring_buffer_consume - return an event and consume it | |
4022 | * @buffer: The ring buffer to get the next event from | |
66a8cb95 SR |
4023 | * @cpu: the cpu to read the buffer from |
4024 | * @ts: a variable to store the timestamp (may be NULL) | |
4025 | * @lost_events: a variable to store if events were lost (may be NULL) | |
7a8e76a3 SR |
4026 | * |
4027 | * Returns the next event in the ring buffer, and that event is consumed. | |
4028 | * Meaning, that sequential reads will keep returning a different event, | |
4029 | * and eventually empty the ring buffer if the producer is slower. | |
4030 | */ | |
4031 | struct ring_buffer_event * | |
66a8cb95 SR |
4032 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts, |
4033 | unsigned long *lost_events) | |
7a8e76a3 | 4034 | { |
554f786e SR |
4035 | struct ring_buffer_per_cpu *cpu_buffer; |
4036 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 4037 | unsigned long flags; |
289a5a25 | 4038 | bool dolock; |
7a8e76a3 | 4039 | |
2d622719 | 4040 | again: |
554f786e SR |
4041 | /* might be called in atomic */ |
4042 | preempt_disable(); | |
4043 | ||
9e01c1b7 | 4044 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 4045 | goto out; |
7a8e76a3 | 4046 | |
554f786e | 4047 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e | 4048 | local_irq_save(flags); |
289a5a25 | 4049 | dolock = rb_reader_lock(cpu_buffer); |
f83c9d0f | 4050 | |
66a8cb95 SR |
4051 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
4052 | if (event) { | |
4053 | cpu_buffer->lost_events = 0; | |
469535a5 | 4054 | rb_advance_reader(cpu_buffer); |
66a8cb95 | 4055 | } |
7a8e76a3 | 4056 | |
289a5a25 | 4057 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e | 4058 | local_irq_restore(flags); |
f83c9d0f | 4059 | |
554f786e SR |
4060 | out: |
4061 | preempt_enable(); | |
4062 | ||
1b959e18 | 4063 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 4064 | goto again; |
2d622719 | 4065 | |
7a8e76a3 SR |
4066 | return event; |
4067 | } | |
c4f50183 | 4068 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
4069 | |
4070 | /** | |
72c9ddfd | 4071 | * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer |
7a8e76a3 SR |
4072 | * @buffer: The ring buffer to read from |
4073 | * @cpu: The cpu buffer to iterate over | |
4074 | * | |
72c9ddfd DM |
4075 | * This performs the initial preparations necessary to iterate |
4076 | * through the buffer. Memory is allocated, buffer recording | |
4077 | * is disabled, and the iterator pointer is returned to the caller. | |
7a8e76a3 | 4078 | * |
72c9ddfd DM |
4079 | * Disabling buffer recordng prevents the reading from being |
4080 | * corrupted. This is not a consuming read, so a producer is not | |
4081 | * expected. | |
4082 | * | |
4083 | * After a sequence of ring_buffer_read_prepare calls, the user is | |
d611851b | 4084 | * expected to make at least one call to ring_buffer_read_prepare_sync. |
72c9ddfd DM |
4085 | * Afterwards, ring_buffer_read_start is invoked to get things going |
4086 | * for real. | |
4087 | * | |
d611851b | 4088 | * This overall must be paired with ring_buffer_read_finish. |
7a8e76a3 SR |
4089 | */ |
4090 | struct ring_buffer_iter * | |
72c9ddfd | 4091 | ring_buffer_read_prepare(struct ring_buffer *buffer, int cpu) |
7a8e76a3 SR |
4092 | { |
4093 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 4094 | struct ring_buffer_iter *iter; |
7a8e76a3 | 4095 | |
9e01c1b7 | 4096 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4097 | return NULL; |
7a8e76a3 SR |
4098 | |
4099 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
4100 | if (!iter) | |
8aabee57 | 4101 | return NULL; |
7a8e76a3 SR |
4102 | |
4103 | cpu_buffer = buffer->buffers[cpu]; | |
4104 | ||
4105 | iter->cpu_buffer = cpu_buffer; | |
4106 | ||
83f40318 | 4107 | atomic_inc(&buffer->resize_disabled); |
7a8e76a3 | 4108 | atomic_inc(&cpu_buffer->record_disabled); |
72c9ddfd DM |
4109 | |
4110 | return iter; | |
4111 | } | |
4112 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare); | |
4113 | ||
4114 | /** | |
4115 | * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls | |
4116 | * | |
4117 | * All previously invoked ring_buffer_read_prepare calls to prepare | |
4118 | * iterators will be synchronized. Afterwards, read_buffer_read_start | |
4119 | * calls on those iterators are allowed. | |
4120 | */ | |
4121 | void | |
4122 | ring_buffer_read_prepare_sync(void) | |
4123 | { | |
7a8e76a3 | 4124 | synchronize_sched(); |
72c9ddfd DM |
4125 | } |
4126 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync); | |
4127 | ||
4128 | /** | |
4129 | * ring_buffer_read_start - start a non consuming read of the buffer | |
4130 | * @iter: The iterator returned by ring_buffer_read_prepare | |
4131 | * | |
4132 | * This finalizes the startup of an iteration through the buffer. | |
4133 | * The iterator comes from a call to ring_buffer_read_prepare and | |
4134 | * an intervening ring_buffer_read_prepare_sync must have been | |
4135 | * performed. | |
4136 | * | |
d611851b | 4137 | * Must be paired with ring_buffer_read_finish. |
72c9ddfd DM |
4138 | */ |
4139 | void | |
4140 | ring_buffer_read_start(struct ring_buffer_iter *iter) | |
4141 | { | |
4142 | struct ring_buffer_per_cpu *cpu_buffer; | |
4143 | unsigned long flags; | |
4144 | ||
4145 | if (!iter) | |
4146 | return; | |
4147 | ||
4148 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 | 4149 | |
5389f6fa | 4150 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
0199c4e6 | 4151 | arch_spin_lock(&cpu_buffer->lock); |
642edba5 | 4152 | rb_iter_reset(iter); |
0199c4e6 | 4153 | arch_spin_unlock(&cpu_buffer->lock); |
5389f6fa | 4154 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 4155 | } |
c4f50183 | 4156 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
4157 | |
4158 | /** | |
d611851b | 4159 | * ring_buffer_read_finish - finish reading the iterator of the buffer |
7a8e76a3 SR |
4160 | * @iter: The iterator retrieved by ring_buffer_start |
4161 | * | |
4162 | * This re-enables the recording to the buffer, and frees the | |
4163 | * iterator. | |
4164 | */ | |
4165 | void | |
4166 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
4167 | { | |
4168 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
9366c1ba | 4169 | unsigned long flags; |
7a8e76a3 | 4170 | |
659f451f SR |
4171 | /* |
4172 | * Ring buffer is disabled from recording, here's a good place | |
9366c1ba SR |
4173 | * to check the integrity of the ring buffer. |
4174 | * Must prevent readers from trying to read, as the check | |
4175 | * clears the HEAD page and readers require it. | |
659f451f | 4176 | */ |
9366c1ba | 4177 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
659f451f | 4178 | rb_check_pages(cpu_buffer); |
9366c1ba | 4179 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
659f451f | 4180 | |
7a8e76a3 | 4181 | atomic_dec(&cpu_buffer->record_disabled); |
83f40318 | 4182 | atomic_dec(&cpu_buffer->buffer->resize_disabled); |
7a8e76a3 SR |
4183 | kfree(iter); |
4184 | } | |
c4f50183 | 4185 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
4186 | |
4187 | /** | |
4188 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
4189 | * @iter: The ring buffer iterator | |
4190 | * @ts: The time stamp of the event read. | |
4191 | * | |
4192 | * This reads the next event in the ring buffer and increments the iterator. | |
4193 | */ | |
4194 | struct ring_buffer_event * | |
4195 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
4196 | { | |
4197 | struct ring_buffer_event *event; | |
f83c9d0f SR |
4198 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
4199 | unsigned long flags; | |
7a8e76a3 | 4200 | |
5389f6fa | 4201 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
7e9391cf | 4202 | again: |
f83c9d0f | 4203 | event = rb_iter_peek(iter, ts); |
7a8e76a3 | 4204 | if (!event) |
f83c9d0f | 4205 | goto out; |
7a8e76a3 | 4206 | |
7e9391cf SR |
4207 | if (event->type_len == RINGBUF_TYPE_PADDING) |
4208 | goto again; | |
4209 | ||
7a8e76a3 | 4210 | rb_advance_iter(iter); |
f83c9d0f | 4211 | out: |
5389f6fa | 4212 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
4213 | |
4214 | return event; | |
4215 | } | |
c4f50183 | 4216 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
4217 | |
4218 | /** | |
4219 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
4220 | * @buffer: The ring buffer. | |
4221 | */ | |
438ced17 | 4222 | unsigned long ring_buffer_size(struct ring_buffer *buffer, int cpu) |
7a8e76a3 | 4223 | { |
438ced17 VN |
4224 | /* |
4225 | * Earlier, this method returned | |
4226 | * BUF_PAGE_SIZE * buffer->nr_pages | |
4227 | * Since the nr_pages field is now removed, we have converted this to | |
4228 | * return the per cpu buffer value. | |
4229 | */ | |
4230 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
4231 | return 0; | |
4232 | ||
4233 | return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages; | |
7a8e76a3 | 4234 | } |
c4f50183 | 4235 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
4236 | |
4237 | static void | |
4238 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
4239 | { | |
77ae365e SR |
4240 | rb_head_page_deactivate(cpu_buffer); |
4241 | ||
7a8e76a3 | 4242 | cpu_buffer->head_page |
3adc54fa | 4243 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 4244 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 4245 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 4246 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 4247 | |
6f807acd | 4248 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
4249 | |
4250 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
4251 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
4252 | ||
4253 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
5040b4b7 | 4254 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
bf41a158 | 4255 | local_set(&cpu_buffer->reader_page->write, 0); |
778c55d4 | 4256 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 4257 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 4258 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 4259 | |
c64e148a | 4260 | local_set(&cpu_buffer->entries_bytes, 0); |
77ae365e | 4261 | local_set(&cpu_buffer->overrun, 0); |
884bfe89 SP |
4262 | local_set(&cpu_buffer->commit_overrun, 0); |
4263 | local_set(&cpu_buffer->dropped_events, 0); | |
e4906eff | 4264 | local_set(&cpu_buffer->entries, 0); |
fa743953 SR |
4265 | local_set(&cpu_buffer->committing, 0); |
4266 | local_set(&cpu_buffer->commits, 0); | |
77ae365e | 4267 | cpu_buffer->read = 0; |
c64e148a | 4268 | cpu_buffer->read_bytes = 0; |
69507c06 SR |
4269 | |
4270 | cpu_buffer->write_stamp = 0; | |
4271 | cpu_buffer->read_stamp = 0; | |
77ae365e | 4272 | |
66a8cb95 SR |
4273 | cpu_buffer->lost_events = 0; |
4274 | cpu_buffer->last_overrun = 0; | |
4275 | ||
77ae365e | 4276 | rb_head_page_activate(cpu_buffer); |
7a8e76a3 SR |
4277 | } |
4278 | ||
4279 | /** | |
4280 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
4281 | * @buffer: The ring buffer to reset a per cpu buffer of | |
4282 | * @cpu: The CPU buffer to be reset | |
4283 | */ | |
4284 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
4285 | { | |
4286 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
4287 | unsigned long flags; | |
4288 | ||
9e01c1b7 | 4289 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4290 | return; |
7a8e76a3 | 4291 | |
83f40318 | 4292 | atomic_inc(&buffer->resize_disabled); |
41ede23e SR |
4293 | atomic_inc(&cpu_buffer->record_disabled); |
4294 | ||
83f40318 VN |
4295 | /* Make sure all commits have finished */ |
4296 | synchronize_sched(); | |
4297 | ||
5389f6fa | 4298 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 4299 | |
41b6a95d SR |
4300 | if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing))) |
4301 | goto out; | |
4302 | ||
0199c4e6 | 4303 | arch_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
4304 | |
4305 | rb_reset_cpu(cpu_buffer); | |
4306 | ||
0199c4e6 | 4307 | arch_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 4308 | |
41b6a95d | 4309 | out: |
5389f6fa | 4310 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
41ede23e SR |
4311 | |
4312 | atomic_dec(&cpu_buffer->record_disabled); | |
83f40318 | 4313 | atomic_dec(&buffer->resize_disabled); |
7a8e76a3 | 4314 | } |
c4f50183 | 4315 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
4316 | |
4317 | /** | |
4318 | * ring_buffer_reset - reset a ring buffer | |
4319 | * @buffer: The ring buffer to reset all cpu buffers | |
4320 | */ | |
4321 | void ring_buffer_reset(struct ring_buffer *buffer) | |
4322 | { | |
7a8e76a3 SR |
4323 | int cpu; |
4324 | ||
7a8e76a3 | 4325 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 4326 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 4327 | } |
c4f50183 | 4328 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
4329 | |
4330 | /** | |
4331 | * rind_buffer_empty - is the ring buffer empty? | |
4332 | * @buffer: The ring buffer to test | |
4333 | */ | |
4334 | int ring_buffer_empty(struct ring_buffer *buffer) | |
4335 | { | |
4336 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 4337 | unsigned long flags; |
289a5a25 | 4338 | bool dolock; |
7a8e76a3 | 4339 | int cpu; |
d4788207 | 4340 | int ret; |
7a8e76a3 SR |
4341 | |
4342 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
4343 | for_each_buffer_cpu(buffer, cpu) { | |
4344 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e | 4345 | local_irq_save(flags); |
289a5a25 | 4346 | dolock = rb_reader_lock(cpu_buffer); |
d4788207 | 4347 | ret = rb_per_cpu_empty(cpu_buffer); |
289a5a25 | 4348 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e SR |
4349 | local_irq_restore(flags); |
4350 | ||
d4788207 | 4351 | if (!ret) |
7a8e76a3 SR |
4352 | return 0; |
4353 | } | |
554f786e | 4354 | |
7a8e76a3 SR |
4355 | return 1; |
4356 | } | |
c4f50183 | 4357 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
4358 | |
4359 | /** | |
4360 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
4361 | * @buffer: The ring buffer | |
4362 | * @cpu: The CPU buffer to test | |
4363 | */ | |
4364 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
4365 | { | |
4366 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 4367 | unsigned long flags; |
289a5a25 | 4368 | bool dolock; |
8aabee57 | 4369 | int ret; |
7a8e76a3 | 4370 | |
9e01c1b7 | 4371 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4372 | return 1; |
7a8e76a3 SR |
4373 | |
4374 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e | 4375 | local_irq_save(flags); |
289a5a25 | 4376 | dolock = rb_reader_lock(cpu_buffer); |
554f786e | 4377 | ret = rb_per_cpu_empty(cpu_buffer); |
289a5a25 | 4378 | rb_reader_unlock(cpu_buffer, dolock); |
8d707e8e | 4379 | local_irq_restore(flags); |
554f786e SR |
4380 | |
4381 | return ret; | |
7a8e76a3 | 4382 | } |
c4f50183 | 4383 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 | 4384 | |
85bac32c | 4385 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
7a8e76a3 SR |
4386 | /** |
4387 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
4388 | * @buffer_a: One buffer to swap with | |
4389 | * @buffer_b: The other buffer to swap with | |
4390 | * | |
4391 | * This function is useful for tracers that want to take a "snapshot" | |
4392 | * of a CPU buffer and has another back up buffer lying around. | |
4393 | * it is expected that the tracer handles the cpu buffer not being | |
4394 | * used at the moment. | |
4395 | */ | |
4396 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
4397 | struct ring_buffer *buffer_b, int cpu) | |
4398 | { | |
4399 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
4400 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
4401 | int ret = -EINVAL; |
4402 | ||
9e01c1b7 RR |
4403 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
4404 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 4405 | goto out; |
7a8e76a3 | 4406 | |
438ced17 VN |
4407 | cpu_buffer_a = buffer_a->buffers[cpu]; |
4408 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
4409 | ||
7a8e76a3 | 4410 | /* At least make sure the two buffers are somewhat the same */ |
438ced17 | 4411 | if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages) |
554f786e SR |
4412 | goto out; |
4413 | ||
4414 | ret = -EAGAIN; | |
7a8e76a3 | 4415 | |
97b17efe | 4416 | if (atomic_read(&buffer_a->record_disabled)) |
554f786e | 4417 | goto out; |
97b17efe SR |
4418 | |
4419 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 4420 | goto out; |
97b17efe | 4421 | |
97b17efe | 4422 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 4423 | goto out; |
97b17efe SR |
4424 | |
4425 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 4426 | goto out; |
97b17efe | 4427 | |
7a8e76a3 SR |
4428 | /* |
4429 | * We can't do a synchronize_sched here because this | |
4430 | * function can be called in atomic context. | |
4431 | * Normally this will be called from the same CPU as cpu. | |
4432 | * If not it's up to the caller to protect this. | |
4433 | */ | |
4434 | atomic_inc(&cpu_buffer_a->record_disabled); | |
4435 | atomic_inc(&cpu_buffer_b->record_disabled); | |
4436 | ||
98277991 SR |
4437 | ret = -EBUSY; |
4438 | if (local_read(&cpu_buffer_a->committing)) | |
4439 | goto out_dec; | |
4440 | if (local_read(&cpu_buffer_b->committing)) | |
4441 | goto out_dec; | |
4442 | ||
7a8e76a3 SR |
4443 | buffer_a->buffers[cpu] = cpu_buffer_b; |
4444 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
4445 | ||
4446 | cpu_buffer_b->buffer = buffer_a; | |
4447 | cpu_buffer_a->buffer = buffer_b; | |
4448 | ||
98277991 SR |
4449 | ret = 0; |
4450 | ||
4451 | out_dec: | |
7a8e76a3 SR |
4452 | atomic_dec(&cpu_buffer_a->record_disabled); |
4453 | atomic_dec(&cpu_buffer_b->record_disabled); | |
554f786e | 4454 | out: |
554f786e | 4455 | return ret; |
7a8e76a3 | 4456 | } |
c4f50183 | 4457 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
85bac32c | 4458 | #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */ |
7a8e76a3 | 4459 | |
8789a9e7 SR |
4460 | /** |
4461 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
4462 | * @buffer: the buffer to allocate for. | |
d611851b | 4463 | * @cpu: the cpu buffer to allocate. |
8789a9e7 SR |
4464 | * |
4465 | * This function is used in conjunction with ring_buffer_read_page. | |
4466 | * When reading a full page from the ring buffer, these functions | |
4467 | * can be used to speed up the process. The calling function should | |
4468 | * allocate a few pages first with this function. Then when it | |
4469 | * needs to get pages from the ring buffer, it passes the result | |
4470 | * of this function into ring_buffer_read_page, which will swap | |
4471 | * the page that was allocated, with the read page of the buffer. | |
4472 | * | |
4473 | * Returns: | |
4474 | * The page allocated, or NULL on error. | |
4475 | */ | |
7ea59064 | 4476 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer, int cpu) |
8789a9e7 | 4477 | { |
044fa782 | 4478 | struct buffer_data_page *bpage; |
7ea59064 | 4479 | struct page *page; |
8789a9e7 | 4480 | |
d7ec4bfe VN |
4481 | page = alloc_pages_node(cpu_to_node(cpu), |
4482 | GFP_KERNEL | __GFP_NORETRY, 0); | |
7ea59064 | 4483 | if (!page) |
8789a9e7 SR |
4484 | return NULL; |
4485 | ||
7ea59064 | 4486 | bpage = page_address(page); |
8789a9e7 | 4487 | |
ef7a4a16 SR |
4488 | rb_init_page(bpage); |
4489 | ||
044fa782 | 4490 | return bpage; |
8789a9e7 | 4491 | } |
d6ce96da | 4492 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
4493 | |
4494 | /** | |
4495 | * ring_buffer_free_read_page - free an allocated read page | |
4496 | * @buffer: the buffer the page was allocate for | |
4497 | * @data: the page to free | |
4498 | * | |
4499 | * Free a page allocated from ring_buffer_alloc_read_page. | |
4500 | */ | |
4501 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
4502 | { | |
4503 | free_page((unsigned long)data); | |
4504 | } | |
d6ce96da | 4505 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
4506 | |
4507 | /** | |
4508 | * ring_buffer_read_page - extract a page from the ring buffer | |
4509 | * @buffer: buffer to extract from | |
4510 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 4511 | * @len: amount to extract |
8789a9e7 SR |
4512 | * @cpu: the cpu of the buffer to extract |
4513 | * @full: should the extraction only happen when the page is full. | |
4514 | * | |
4515 | * This function will pull out a page from the ring buffer and consume it. | |
4516 | * @data_page must be the address of the variable that was returned | |
4517 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
4518 | * to swap with a page in the ring buffer. | |
4519 | * | |
4520 | * for example: | |
d611851b | 4521 | * rpage = ring_buffer_alloc_read_page(buffer, cpu); |
8789a9e7 SR |
4522 | * if (!rpage) |
4523 | * return error; | |
ef7a4a16 | 4524 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
4525 | * if (ret >= 0) |
4526 | * process_page(rpage, ret); | |
8789a9e7 SR |
4527 | * |
4528 | * When @full is set, the function will not return true unless | |
4529 | * the writer is off the reader page. | |
4530 | * | |
4531 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
4532 | * The ring buffer can be used anywhere in the kernel and can not | |
4533 | * blindly call wake_up. The layer that uses the ring buffer must be | |
4534 | * responsible for that. | |
4535 | * | |
4536 | * Returns: | |
667d2412 LJ |
4537 | * >=0 if data has been transferred, returns the offset of consumed data. |
4538 | * <0 if no data has been transferred. | |
8789a9e7 SR |
4539 | */ |
4540 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 4541 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
4542 | { |
4543 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
4544 | struct ring_buffer_event *event; | |
044fa782 | 4545 | struct buffer_data_page *bpage; |
ef7a4a16 | 4546 | struct buffer_page *reader; |
ff0ff84a | 4547 | unsigned long missed_events; |
8789a9e7 | 4548 | unsigned long flags; |
ef7a4a16 | 4549 | unsigned int commit; |
667d2412 | 4550 | unsigned int read; |
4f3640f8 | 4551 | u64 save_timestamp; |
667d2412 | 4552 | int ret = -1; |
8789a9e7 | 4553 | |
554f786e SR |
4554 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
4555 | goto out; | |
4556 | ||
474d32b6 SR |
4557 | /* |
4558 | * If len is not big enough to hold the page header, then | |
4559 | * we can not copy anything. | |
4560 | */ | |
4561 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 4562 | goto out; |
474d32b6 SR |
4563 | |
4564 | len -= BUF_PAGE_HDR_SIZE; | |
4565 | ||
8789a9e7 | 4566 | if (!data_page) |
554f786e | 4567 | goto out; |
8789a9e7 | 4568 | |
044fa782 SR |
4569 | bpage = *data_page; |
4570 | if (!bpage) | |
554f786e | 4571 | goto out; |
8789a9e7 | 4572 | |
5389f6fa | 4573 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4574 | |
ef7a4a16 SR |
4575 | reader = rb_get_reader_page(cpu_buffer); |
4576 | if (!reader) | |
554f786e | 4577 | goto out_unlock; |
8789a9e7 | 4578 | |
ef7a4a16 SR |
4579 | event = rb_reader_event(cpu_buffer); |
4580 | ||
4581 | read = reader->read; | |
4582 | commit = rb_page_commit(reader); | |
667d2412 | 4583 | |
66a8cb95 | 4584 | /* Check if any events were dropped */ |
ff0ff84a | 4585 | missed_events = cpu_buffer->lost_events; |
66a8cb95 | 4586 | |
8789a9e7 | 4587 | /* |
474d32b6 SR |
4588 | * If this page has been partially read or |
4589 | * if len is not big enough to read the rest of the page or | |
4590 | * a writer is still on the page, then | |
4591 | * we must copy the data from the page to the buffer. | |
4592 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 4593 | */ |
474d32b6 | 4594 | if (read || (len < (commit - read)) || |
ef7a4a16 | 4595 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 4596 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
4597 | unsigned int rpos = read; |
4598 | unsigned int pos = 0; | |
ef7a4a16 | 4599 | unsigned int size; |
8789a9e7 SR |
4600 | |
4601 | if (full) | |
554f786e | 4602 | goto out_unlock; |
8789a9e7 | 4603 | |
ef7a4a16 SR |
4604 | if (len > (commit - read)) |
4605 | len = (commit - read); | |
4606 | ||
69d1b839 SR |
4607 | /* Always keep the time extend and data together */ |
4608 | size = rb_event_ts_length(event); | |
ef7a4a16 SR |
4609 | |
4610 | if (len < size) | |
554f786e | 4611 | goto out_unlock; |
ef7a4a16 | 4612 | |
4f3640f8 SR |
4613 | /* save the current timestamp, since the user will need it */ |
4614 | save_timestamp = cpu_buffer->read_stamp; | |
4615 | ||
ef7a4a16 SR |
4616 | /* Need to copy one event at a time */ |
4617 | do { | |
e1e35927 DS |
4618 | /* We need the size of one event, because |
4619 | * rb_advance_reader only advances by one event, | |
4620 | * whereas rb_event_ts_length may include the size of | |
4621 | * one or two events. | |
4622 | * We have already ensured there's enough space if this | |
4623 | * is a time extend. */ | |
4624 | size = rb_event_length(event); | |
474d32b6 | 4625 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
4626 | |
4627 | len -= size; | |
4628 | ||
4629 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
4630 | rpos = reader->read; |
4631 | pos += size; | |
ef7a4a16 | 4632 | |
18fab912 HY |
4633 | if (rpos >= commit) |
4634 | break; | |
4635 | ||
ef7a4a16 | 4636 | event = rb_reader_event(cpu_buffer); |
69d1b839 SR |
4637 | /* Always keep the time extend and data together */ |
4638 | size = rb_event_ts_length(event); | |
e1e35927 | 4639 | } while (len >= size); |
667d2412 LJ |
4640 | |
4641 | /* update bpage */ | |
ef7a4a16 | 4642 | local_set(&bpage->commit, pos); |
4f3640f8 | 4643 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 4644 | |
474d32b6 SR |
4645 | /* we copied everything to the beginning */ |
4646 | read = 0; | |
8789a9e7 | 4647 | } else { |
afbab76a | 4648 | /* update the entry counter */ |
77ae365e | 4649 | cpu_buffer->read += rb_page_entries(reader); |
c64e148a | 4650 | cpu_buffer->read_bytes += BUF_PAGE_SIZE; |
afbab76a | 4651 | |
8789a9e7 | 4652 | /* swap the pages */ |
044fa782 | 4653 | rb_init_page(bpage); |
ef7a4a16 SR |
4654 | bpage = reader->page; |
4655 | reader->page = *data_page; | |
4656 | local_set(&reader->write, 0); | |
778c55d4 | 4657 | local_set(&reader->entries, 0); |
ef7a4a16 | 4658 | reader->read = 0; |
044fa782 | 4659 | *data_page = bpage; |
ff0ff84a SR |
4660 | |
4661 | /* | |
4662 | * Use the real_end for the data size, | |
4663 | * This gives us a chance to store the lost events | |
4664 | * on the page. | |
4665 | */ | |
4666 | if (reader->real_end) | |
4667 | local_set(&bpage->commit, reader->real_end); | |
8789a9e7 | 4668 | } |
667d2412 | 4669 | ret = read; |
8789a9e7 | 4670 | |
66a8cb95 | 4671 | cpu_buffer->lost_events = 0; |
2711ca23 SR |
4672 | |
4673 | commit = local_read(&bpage->commit); | |
66a8cb95 SR |
4674 | /* |
4675 | * Set a flag in the commit field if we lost events | |
4676 | */ | |
ff0ff84a | 4677 | if (missed_events) { |
ff0ff84a SR |
4678 | /* If there is room at the end of the page to save the |
4679 | * missed events, then record it there. | |
4680 | */ | |
4681 | if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) { | |
4682 | memcpy(&bpage->data[commit], &missed_events, | |
4683 | sizeof(missed_events)); | |
4684 | local_add(RB_MISSED_STORED, &bpage->commit); | |
2711ca23 | 4685 | commit += sizeof(missed_events); |
ff0ff84a | 4686 | } |
66a8cb95 | 4687 | local_add(RB_MISSED_EVENTS, &bpage->commit); |
ff0ff84a | 4688 | } |
66a8cb95 | 4689 | |
2711ca23 SR |
4690 | /* |
4691 | * This page may be off to user land. Zero it out here. | |
4692 | */ | |
4693 | if (commit < BUF_PAGE_SIZE) | |
4694 | memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit); | |
4695 | ||
554f786e | 4696 | out_unlock: |
5389f6fa | 4697 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4698 | |
554f786e | 4699 | out: |
8789a9e7 SR |
4700 | return ret; |
4701 | } | |
d6ce96da | 4702 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 4703 | |
59222efe | 4704 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
4705 | static int rb_cpu_notify(struct notifier_block *self, |
4706 | unsigned long action, void *hcpu) | |
554f786e SR |
4707 | { |
4708 | struct ring_buffer *buffer = | |
4709 | container_of(self, struct ring_buffer, cpu_notify); | |
4710 | long cpu = (long)hcpu; | |
438ced17 VN |
4711 | int cpu_i, nr_pages_same; |
4712 | unsigned int nr_pages; | |
554f786e SR |
4713 | |
4714 | switch (action) { | |
4715 | case CPU_UP_PREPARE: | |
4716 | case CPU_UP_PREPARE_FROZEN: | |
3f237a79 | 4717 | if (cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e SR |
4718 | return NOTIFY_OK; |
4719 | ||
438ced17 VN |
4720 | nr_pages = 0; |
4721 | nr_pages_same = 1; | |
4722 | /* check if all cpu sizes are same */ | |
4723 | for_each_buffer_cpu(buffer, cpu_i) { | |
4724 | /* fill in the size from first enabled cpu */ | |
4725 | if (nr_pages == 0) | |
4726 | nr_pages = buffer->buffers[cpu_i]->nr_pages; | |
4727 | if (nr_pages != buffer->buffers[cpu_i]->nr_pages) { | |
4728 | nr_pages_same = 0; | |
4729 | break; | |
4730 | } | |
4731 | } | |
4732 | /* allocate minimum pages, user can later expand it */ | |
4733 | if (!nr_pages_same) | |
4734 | nr_pages = 2; | |
554f786e | 4735 | buffer->buffers[cpu] = |
438ced17 | 4736 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
554f786e SR |
4737 | if (!buffer->buffers[cpu]) { |
4738 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
4739 | cpu); | |
4740 | return NOTIFY_OK; | |
4741 | } | |
4742 | smp_wmb(); | |
3f237a79 | 4743 | cpumask_set_cpu(cpu, buffer->cpumask); |
554f786e SR |
4744 | break; |
4745 | case CPU_DOWN_PREPARE: | |
4746 | case CPU_DOWN_PREPARE_FROZEN: | |
4747 | /* | |
4748 | * Do nothing. | |
4749 | * If we were to free the buffer, then the user would | |
4750 | * lose any trace that was in the buffer. | |
4751 | */ | |
4752 | break; | |
4753 | default: | |
4754 | break; | |
4755 | } | |
4756 | return NOTIFY_OK; | |
4757 | } | |
4758 | #endif | |
6c43e554 SRRH |
4759 | |
4760 | #ifdef CONFIG_RING_BUFFER_STARTUP_TEST | |
4761 | /* | |
4762 | * This is a basic integrity check of the ring buffer. | |
4763 | * Late in the boot cycle this test will run when configured in. | |
4764 | * It will kick off a thread per CPU that will go into a loop | |
4765 | * writing to the per cpu ring buffer various sizes of data. | |
4766 | * Some of the data will be large items, some small. | |
4767 | * | |
4768 | * Another thread is created that goes into a spin, sending out | |
4769 | * IPIs to the other CPUs to also write into the ring buffer. | |
4770 | * this is to test the nesting ability of the buffer. | |
4771 | * | |
4772 | * Basic stats are recorded and reported. If something in the | |
4773 | * ring buffer should happen that's not expected, a big warning | |
4774 | * is displayed and all ring buffers are disabled. | |
4775 | */ | |
4776 | static struct task_struct *rb_threads[NR_CPUS] __initdata; | |
4777 | ||
4778 | struct rb_test_data { | |
4779 | struct ring_buffer *buffer; | |
4780 | unsigned long events; | |
4781 | unsigned long bytes_written; | |
4782 | unsigned long bytes_alloc; | |
4783 | unsigned long bytes_dropped; | |
4784 | unsigned long events_nested; | |
4785 | unsigned long bytes_written_nested; | |
4786 | unsigned long bytes_alloc_nested; | |
4787 | unsigned long bytes_dropped_nested; | |
4788 | int min_size_nested; | |
4789 | int max_size_nested; | |
4790 | int max_size; | |
4791 | int min_size; | |
4792 | int cpu; | |
4793 | int cnt; | |
4794 | }; | |
4795 | ||
4796 | static struct rb_test_data rb_data[NR_CPUS] __initdata; | |
4797 | ||
4798 | /* 1 meg per cpu */ | |
4799 | #define RB_TEST_BUFFER_SIZE 1048576 | |
4800 | ||
4801 | static char rb_string[] __initdata = | |
4802 | "abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\" | |
4803 | "?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890" | |
4804 | "!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv"; | |
4805 | ||
4806 | static bool rb_test_started __initdata; | |
4807 | ||
4808 | struct rb_item { | |
4809 | int size; | |
4810 | char str[]; | |
4811 | }; | |
4812 | ||
4813 | static __init int rb_write_something(struct rb_test_data *data, bool nested) | |
4814 | { | |
4815 | struct ring_buffer_event *event; | |
4816 | struct rb_item *item; | |
4817 | bool started; | |
4818 | int event_len; | |
4819 | int size; | |
4820 | int len; | |
4821 | int cnt; | |
4822 | ||
4823 | /* Have nested writes different that what is written */ | |
4824 | cnt = data->cnt + (nested ? 27 : 0); | |
4825 | ||
4826 | /* Multiply cnt by ~e, to make some unique increment */ | |
4827 | size = (data->cnt * 68 / 25) % (sizeof(rb_string) - 1); | |
4828 | ||
4829 | len = size + sizeof(struct rb_item); | |
4830 | ||
4831 | started = rb_test_started; | |
4832 | /* read rb_test_started before checking buffer enabled */ | |
4833 | smp_rmb(); | |
4834 | ||
4835 | event = ring_buffer_lock_reserve(data->buffer, len); | |
4836 | if (!event) { | |
4837 | /* Ignore dropped events before test starts. */ | |
4838 | if (started) { | |
4839 | if (nested) | |
4840 | data->bytes_dropped += len; | |
4841 | else | |
4842 | data->bytes_dropped_nested += len; | |
4843 | } | |
4844 | return len; | |
4845 | } | |
4846 | ||
4847 | event_len = ring_buffer_event_length(event); | |
4848 | ||
4849 | if (RB_WARN_ON(data->buffer, event_len < len)) | |
4850 | goto out; | |
4851 | ||
4852 | item = ring_buffer_event_data(event); | |
4853 | item->size = size; | |
4854 | memcpy(item->str, rb_string, size); | |
4855 | ||
4856 | if (nested) { | |
4857 | data->bytes_alloc_nested += event_len; | |
4858 | data->bytes_written_nested += len; | |
4859 | data->events_nested++; | |
4860 | if (!data->min_size_nested || len < data->min_size_nested) | |
4861 | data->min_size_nested = len; | |
4862 | if (len > data->max_size_nested) | |
4863 | data->max_size_nested = len; | |
4864 | } else { | |
4865 | data->bytes_alloc += event_len; | |
4866 | data->bytes_written += len; | |
4867 | data->events++; | |
4868 | if (!data->min_size || len < data->min_size) | |
4869 | data->max_size = len; | |
4870 | if (len > data->max_size) | |
4871 | data->max_size = len; | |
4872 | } | |
4873 | ||
4874 | out: | |
4875 | ring_buffer_unlock_commit(data->buffer, event); | |
4876 | ||
4877 | return 0; | |
4878 | } | |
4879 | ||
4880 | static __init int rb_test(void *arg) | |
4881 | { | |
4882 | struct rb_test_data *data = arg; | |
4883 | ||
4884 | while (!kthread_should_stop()) { | |
4885 | rb_write_something(data, false); | |
4886 | data->cnt++; | |
4887 | ||
4888 | set_current_state(TASK_INTERRUPTIBLE); | |
4889 | /* Now sleep between a min of 100-300us and a max of 1ms */ | |
4890 | usleep_range(((data->cnt % 3) + 1) * 100, 1000); | |
4891 | } | |
4892 | ||
4893 | return 0; | |
4894 | } | |
4895 | ||
4896 | static __init void rb_ipi(void *ignore) | |
4897 | { | |
4898 | struct rb_test_data *data; | |
4899 | int cpu = smp_processor_id(); | |
4900 | ||
4901 | data = &rb_data[cpu]; | |
4902 | rb_write_something(data, true); | |
4903 | } | |
4904 | ||
4905 | static __init int rb_hammer_test(void *arg) | |
4906 | { | |
4907 | while (!kthread_should_stop()) { | |
4908 | ||
4909 | /* Send an IPI to all cpus to write data! */ | |
4910 | smp_call_function(rb_ipi, NULL, 1); | |
4911 | /* No sleep, but for non preempt, let others run */ | |
4912 | schedule(); | |
4913 | } | |
4914 | ||
4915 | return 0; | |
4916 | } | |
4917 | ||
4918 | static __init int test_ringbuffer(void) | |
4919 | { | |
4920 | struct task_struct *rb_hammer; | |
4921 | struct ring_buffer *buffer; | |
4922 | int cpu; | |
4923 | int ret = 0; | |
4924 | ||
4925 | pr_info("Running ring buffer tests...\n"); | |
4926 | ||
4927 | buffer = ring_buffer_alloc(RB_TEST_BUFFER_SIZE, RB_FL_OVERWRITE); | |
4928 | if (WARN_ON(!buffer)) | |
4929 | return 0; | |
4930 | ||
4931 | /* Disable buffer so that threads can't write to it yet */ | |
4932 | ring_buffer_record_off(buffer); | |
4933 | ||
4934 | for_each_online_cpu(cpu) { | |
4935 | rb_data[cpu].buffer = buffer; | |
4936 | rb_data[cpu].cpu = cpu; | |
4937 | rb_data[cpu].cnt = cpu; | |
4938 | rb_threads[cpu] = kthread_create(rb_test, &rb_data[cpu], | |
4939 | "rbtester/%d", cpu); | |
4940 | if (WARN_ON(!rb_threads[cpu])) { | |
4941 | pr_cont("FAILED\n"); | |
4942 | ret = -1; | |
4943 | goto out_free; | |
4944 | } | |
4945 | ||
4946 | kthread_bind(rb_threads[cpu], cpu); | |
4947 | wake_up_process(rb_threads[cpu]); | |
4948 | } | |
4949 | ||
4950 | /* Now create the rb hammer! */ | |
4951 | rb_hammer = kthread_run(rb_hammer_test, NULL, "rbhammer"); | |
4952 | if (WARN_ON(!rb_hammer)) { | |
4953 | pr_cont("FAILED\n"); | |
4954 | ret = -1; | |
4955 | goto out_free; | |
4956 | } | |
4957 | ||
4958 | ring_buffer_record_on(buffer); | |
4959 | /* | |
4960 | * Show buffer is enabled before setting rb_test_started. | |
4961 | * Yes there's a small race window where events could be | |
4962 | * dropped and the thread wont catch it. But when a ring | |
4963 | * buffer gets enabled, there will always be some kind of | |
4964 | * delay before other CPUs see it. Thus, we don't care about | |
4965 | * those dropped events. We care about events dropped after | |
4966 | * the threads see that the buffer is active. | |
4967 | */ | |
4968 | smp_wmb(); | |
4969 | rb_test_started = true; | |
4970 | ||
4971 | set_current_state(TASK_INTERRUPTIBLE); | |
4972 | /* Just run for 10 seconds */; | |
4973 | schedule_timeout(10 * HZ); | |
4974 | ||
4975 | kthread_stop(rb_hammer); | |
4976 | ||
4977 | out_free: | |
4978 | for_each_online_cpu(cpu) { | |
4979 | if (!rb_threads[cpu]) | |
4980 | break; | |
4981 | kthread_stop(rb_threads[cpu]); | |
4982 | } | |
4983 | if (ret) { | |
4984 | ring_buffer_free(buffer); | |
4985 | return ret; | |
4986 | } | |
4987 | ||
4988 | /* Report! */ | |
4989 | pr_info("finished\n"); | |
4990 | for_each_online_cpu(cpu) { | |
4991 | struct ring_buffer_event *event; | |
4992 | struct rb_test_data *data = &rb_data[cpu]; | |
4993 | struct rb_item *item; | |
4994 | unsigned long total_events; | |
4995 | unsigned long total_dropped; | |
4996 | unsigned long total_written; | |
4997 | unsigned long total_alloc; | |
4998 | unsigned long total_read = 0; | |
4999 | unsigned long total_size = 0; | |
5000 | unsigned long total_len = 0; | |
5001 | unsigned long total_lost = 0; | |
5002 | unsigned long lost; | |
5003 | int big_event_size; | |
5004 | int small_event_size; | |
5005 | ||
5006 | ret = -1; | |
5007 | ||
5008 | total_events = data->events + data->events_nested; | |
5009 | total_written = data->bytes_written + data->bytes_written_nested; | |
5010 | total_alloc = data->bytes_alloc + data->bytes_alloc_nested; | |
5011 | total_dropped = data->bytes_dropped + data->bytes_dropped_nested; | |
5012 | ||
5013 | big_event_size = data->max_size + data->max_size_nested; | |
5014 | small_event_size = data->min_size + data->min_size_nested; | |
5015 | ||
5016 | pr_info("CPU %d:\n", cpu); | |
5017 | pr_info(" events: %ld\n", total_events); | |
5018 | pr_info(" dropped bytes: %ld\n", total_dropped); | |
5019 | pr_info(" alloced bytes: %ld\n", total_alloc); | |
5020 | pr_info(" written bytes: %ld\n", total_written); | |
5021 | pr_info(" biggest event: %d\n", big_event_size); | |
5022 | pr_info(" smallest event: %d\n", small_event_size); | |
5023 | ||
5024 | if (RB_WARN_ON(buffer, total_dropped)) | |
5025 | break; | |
5026 | ||
5027 | ret = 0; | |
5028 | ||
5029 | while ((event = ring_buffer_consume(buffer, cpu, NULL, &lost))) { | |
5030 | total_lost += lost; | |
5031 | item = ring_buffer_event_data(event); | |
5032 | total_len += ring_buffer_event_length(event); | |
5033 | total_size += item->size + sizeof(struct rb_item); | |
5034 | if (memcmp(&item->str[0], rb_string, item->size) != 0) { | |
5035 | pr_info("FAILED!\n"); | |
5036 | pr_info("buffer had: %.*s\n", item->size, item->str); | |
5037 | pr_info("expected: %.*s\n", item->size, rb_string); | |
5038 | RB_WARN_ON(buffer, 1); | |
5039 | ret = -1; | |
5040 | break; | |
5041 | } | |
5042 | total_read++; | |
5043 | } | |
5044 | if (ret) | |
5045 | break; | |
5046 | ||
5047 | ret = -1; | |
5048 | ||
5049 | pr_info(" read events: %ld\n", total_read); | |
5050 | pr_info(" lost events: %ld\n", total_lost); | |
5051 | pr_info(" total events: %ld\n", total_lost + total_read); | |
5052 | pr_info(" recorded len bytes: %ld\n", total_len); | |
5053 | pr_info(" recorded size bytes: %ld\n", total_size); | |
5054 | if (total_lost) | |
5055 | pr_info(" With dropped events, record len and size may not match\n" | |
5056 | " alloced and written from above\n"); | |
5057 | if (!total_lost) { | |
5058 | if (RB_WARN_ON(buffer, total_len != total_alloc || | |
5059 | total_size != total_written)) | |
5060 | break; | |
5061 | } | |
5062 | if (RB_WARN_ON(buffer, total_lost + total_read != total_events)) | |
5063 | break; | |
5064 | ||
5065 | ret = 0; | |
5066 | } | |
5067 | if (!ret) | |
5068 | pr_info("Ring buffer PASSED!\n"); | |
5069 | ||
5070 | ring_buffer_free(buffer); | |
5071 | return 0; | |
5072 | } | |
5073 | ||
5074 | late_initcall(test_ringbuffer); | |
5075 | #endif /* CONFIG_RING_BUFFER_STARTUP_TEST */ |