Commit | Line | Data |
---|---|---|
7a8e76a3 SR |
1 | /* |
2 | * Generic ring buffer | |
3 | * | |
4 | * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com> | |
5 | */ | |
0b07436d | 6 | #include <linux/ftrace_event.h> |
7a8e76a3 | 7 | #include <linux/ring_buffer.h> |
14131f2f | 8 | #include <linux/trace_clock.h> |
0b07436d | 9 | #include <linux/trace_seq.h> |
7a8e76a3 | 10 | #include <linux/spinlock.h> |
15693458 | 11 | #include <linux/irq_work.h> |
7a8e76a3 | 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 | ||
033601a3 SR |
118 | /* |
119 | * A fast way to enable or disable all ring buffers is to | |
120 | * call tracing_on or tracing_off. Turning off the ring buffers | |
121 | * prevents all ring buffers from being recorded to. | |
122 | * Turning this switch on, makes it OK to write to the | |
123 | * ring buffer, if the ring buffer is enabled itself. | |
124 | * | |
125 | * There's three layers that must be on in order to write | |
126 | * to the ring buffer. | |
127 | * | |
128 | * 1) This global flag must be set. | |
129 | * 2) The ring buffer must be enabled for recording. | |
130 | * 3) The per cpu buffer must be enabled for recording. | |
131 | * | |
132 | * In case of an anomaly, this global flag has a bit set that | |
133 | * will permantly disable all ring buffers. | |
134 | */ | |
135 | ||
136 | /* | |
137 | * Global flag to disable all recording to ring buffers | |
138 | * This has two bits: ON, DISABLED | |
139 | * | |
140 | * ON DISABLED | |
141 | * ---- ---------- | |
142 | * 0 0 : ring buffers are off | |
143 | * 1 0 : ring buffers are on | |
144 | * X 1 : ring buffers are permanently disabled | |
145 | */ | |
146 | ||
147 | enum { | |
148 | RB_BUFFERS_ON_BIT = 0, | |
149 | RB_BUFFERS_DISABLED_BIT = 1, | |
150 | }; | |
151 | ||
152 | enum { | |
153 | RB_BUFFERS_ON = 1 << RB_BUFFERS_ON_BIT, | |
154 | RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT, | |
155 | }; | |
156 | ||
5e39841c | 157 | static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON; |
a3583244 | 158 | |
499e5470 SR |
159 | /* Used for individual buffers (after the counter) */ |
160 | #define RB_BUFFER_OFF (1 << 20) | |
a3583244 | 161 | |
499e5470 | 162 | #define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data) |
033601a3 SR |
163 | |
164 | /** | |
165 | * tracing_off_permanent - permanently disable ring buffers | |
166 | * | |
167 | * This function, once called, will disable all ring buffers | |
c3706f00 | 168 | * permanently. |
033601a3 SR |
169 | */ |
170 | void tracing_off_permanent(void) | |
171 | { | |
172 | set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags); | |
a3583244 SR |
173 | } |
174 | ||
e3d6bf0a | 175 | #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array)) |
67d34724 | 176 | #define RB_ALIGNMENT 4U |
334d4169 | 177 | #define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
c7b09308 | 178 | #define RB_EVNT_MIN_SIZE 8U /* two 32bit words */ |
334d4169 | 179 | |
649508f6 | 180 | #ifndef CONFIG_HAVE_64BIT_ALIGNED_ACCESS |
2271048d SR |
181 | # define RB_FORCE_8BYTE_ALIGNMENT 0 |
182 | # define RB_ARCH_ALIGNMENT RB_ALIGNMENT | |
183 | #else | |
184 | # define RB_FORCE_8BYTE_ALIGNMENT 1 | |
185 | # define RB_ARCH_ALIGNMENT 8U | |
186 | #endif | |
187 | ||
649508f6 JH |
188 | #define RB_ALIGN_DATA __aligned(RB_ARCH_ALIGNMENT) |
189 | ||
334d4169 LJ |
190 | /* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */ |
191 | #define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX | |
7a8e76a3 SR |
192 | |
193 | enum { | |
194 | RB_LEN_TIME_EXTEND = 8, | |
195 | RB_LEN_TIME_STAMP = 16, | |
196 | }; | |
197 | ||
69d1b839 SR |
198 | #define skip_time_extend(event) \ |
199 | ((struct ring_buffer_event *)((char *)event + RB_LEN_TIME_EXTEND)) | |
200 | ||
2d622719 TZ |
201 | static inline int rb_null_event(struct ring_buffer_event *event) |
202 | { | |
a1863c21 | 203 | return event->type_len == RINGBUF_TYPE_PADDING && !event->time_delta; |
2d622719 TZ |
204 | } |
205 | ||
206 | static void rb_event_set_padding(struct ring_buffer_event *event) | |
207 | { | |
a1863c21 | 208 | /* padding has a NULL time_delta */ |
334d4169 | 209 | event->type_len = RINGBUF_TYPE_PADDING; |
2d622719 TZ |
210 | event->time_delta = 0; |
211 | } | |
212 | ||
34a148bf | 213 | static unsigned |
2d622719 | 214 | rb_event_data_length(struct ring_buffer_event *event) |
7a8e76a3 SR |
215 | { |
216 | unsigned length; | |
217 | ||
334d4169 LJ |
218 | if (event->type_len) |
219 | length = event->type_len * RB_ALIGNMENT; | |
2d622719 TZ |
220 | else |
221 | length = event->array[0]; | |
222 | return length + RB_EVNT_HDR_SIZE; | |
223 | } | |
224 | ||
69d1b839 SR |
225 | /* |
226 | * Return the length of the given event. Will return | |
227 | * the length of the time extend if the event is a | |
228 | * time extend. | |
229 | */ | |
230 | static inline unsigned | |
2d622719 TZ |
231 | rb_event_length(struct ring_buffer_event *event) |
232 | { | |
334d4169 | 233 | switch (event->type_len) { |
7a8e76a3 | 234 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
235 | if (rb_null_event(event)) |
236 | /* undefined */ | |
237 | return -1; | |
334d4169 | 238 | return event->array[0] + RB_EVNT_HDR_SIZE; |
7a8e76a3 SR |
239 | |
240 | case RINGBUF_TYPE_TIME_EXTEND: | |
241 | return RB_LEN_TIME_EXTEND; | |
242 | ||
243 | case RINGBUF_TYPE_TIME_STAMP: | |
244 | return RB_LEN_TIME_STAMP; | |
245 | ||
246 | case RINGBUF_TYPE_DATA: | |
2d622719 | 247 | return rb_event_data_length(event); |
7a8e76a3 SR |
248 | default: |
249 | BUG(); | |
250 | } | |
251 | /* not hit */ | |
252 | return 0; | |
253 | } | |
254 | ||
69d1b839 SR |
255 | /* |
256 | * Return total length of time extend and data, | |
257 | * or just the event length for all other events. | |
258 | */ | |
259 | static inline unsigned | |
260 | rb_event_ts_length(struct ring_buffer_event *event) | |
261 | { | |
262 | unsigned len = 0; | |
263 | ||
264 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
265 | /* time extends include the data event after it */ | |
266 | len = RB_LEN_TIME_EXTEND; | |
267 | event = skip_time_extend(event); | |
268 | } | |
269 | return len + rb_event_length(event); | |
270 | } | |
271 | ||
7a8e76a3 SR |
272 | /** |
273 | * ring_buffer_event_length - return the length of the event | |
274 | * @event: the event to get the length of | |
69d1b839 SR |
275 | * |
276 | * Returns the size of the data load of a data event. | |
277 | * If the event is something other than a data event, it | |
278 | * returns the size of the event itself. With the exception | |
279 | * of a TIME EXTEND, where it still returns the size of the | |
280 | * data load of the data event after it. | |
7a8e76a3 SR |
281 | */ |
282 | unsigned ring_buffer_event_length(struct ring_buffer_event *event) | |
283 | { | |
69d1b839 SR |
284 | unsigned length; |
285 | ||
286 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) | |
287 | event = skip_time_extend(event); | |
288 | ||
289 | length = rb_event_length(event); | |
334d4169 | 290 | if (event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
465634ad RR |
291 | return length; |
292 | length -= RB_EVNT_HDR_SIZE; | |
293 | if (length > RB_MAX_SMALL_DATA + sizeof(event->array[0])) | |
294 | length -= sizeof(event->array[0]); | |
295 | return length; | |
7a8e76a3 | 296 | } |
c4f50183 | 297 | EXPORT_SYMBOL_GPL(ring_buffer_event_length); |
7a8e76a3 SR |
298 | |
299 | /* inline for ring buffer fast paths */ | |
34a148bf | 300 | static void * |
7a8e76a3 SR |
301 | rb_event_data(struct ring_buffer_event *event) |
302 | { | |
69d1b839 SR |
303 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) |
304 | event = skip_time_extend(event); | |
334d4169 | 305 | BUG_ON(event->type_len > RINGBUF_TYPE_DATA_TYPE_LEN_MAX); |
7a8e76a3 | 306 | /* If length is in len field, then array[0] has the data */ |
334d4169 | 307 | if (event->type_len) |
7a8e76a3 SR |
308 | return (void *)&event->array[0]; |
309 | /* Otherwise length is in array[0] and array[1] has the data */ | |
310 | return (void *)&event->array[1]; | |
311 | } | |
312 | ||
313 | /** | |
314 | * ring_buffer_event_data - return the data of the event | |
315 | * @event: the event to get the data from | |
316 | */ | |
317 | void *ring_buffer_event_data(struct ring_buffer_event *event) | |
318 | { | |
319 | return rb_event_data(event); | |
320 | } | |
c4f50183 | 321 | EXPORT_SYMBOL_GPL(ring_buffer_event_data); |
7a8e76a3 SR |
322 | |
323 | #define for_each_buffer_cpu(buffer, cpu) \ | |
9e01c1b7 | 324 | for_each_cpu(cpu, buffer->cpumask) |
7a8e76a3 SR |
325 | |
326 | #define TS_SHIFT 27 | |
327 | #define TS_MASK ((1ULL << TS_SHIFT) - 1) | |
328 | #define TS_DELTA_TEST (~TS_MASK) | |
329 | ||
66a8cb95 SR |
330 | /* Flag when events were overwritten */ |
331 | #define RB_MISSED_EVENTS (1 << 31) | |
ff0ff84a SR |
332 | /* Missed count stored at end */ |
333 | #define RB_MISSED_STORED (1 << 30) | |
66a8cb95 | 334 | |
abc9b56d | 335 | struct buffer_data_page { |
e4c2ce82 | 336 | u64 time_stamp; /* page time stamp */ |
c3706f00 | 337 | local_t commit; /* write committed index */ |
649508f6 | 338 | unsigned char data[] RB_ALIGN_DATA; /* data of buffer page */ |
abc9b56d SR |
339 | }; |
340 | ||
77ae365e SR |
341 | /* |
342 | * Note, the buffer_page list must be first. The buffer pages | |
343 | * are allocated in cache lines, which means that each buffer | |
344 | * page will be at the beginning of a cache line, and thus | |
345 | * the least significant bits will be zero. We use this to | |
346 | * add flags in the list struct pointers, to make the ring buffer | |
347 | * lockless. | |
348 | */ | |
abc9b56d | 349 | struct buffer_page { |
778c55d4 | 350 | struct list_head list; /* list of buffer pages */ |
abc9b56d | 351 | local_t write; /* index for next write */ |
6f807acd | 352 | unsigned read; /* index for next read */ |
778c55d4 | 353 | local_t entries; /* entries on this page */ |
ff0ff84a | 354 | unsigned long real_end; /* real end of data */ |
abc9b56d | 355 | struct buffer_data_page *page; /* Actual data page */ |
7a8e76a3 SR |
356 | }; |
357 | ||
77ae365e SR |
358 | /* |
359 | * The buffer page counters, write and entries, must be reset | |
360 | * atomically when crossing page boundaries. To synchronize this | |
361 | * update, two counters are inserted into the number. One is | |
362 | * the actual counter for the write position or count on the page. | |
363 | * | |
364 | * The other is a counter of updaters. Before an update happens | |
365 | * the update partition of the counter is incremented. This will | |
366 | * allow the updater to update the counter atomically. | |
367 | * | |
368 | * The counter is 20 bits, and the state data is 12. | |
369 | */ | |
370 | #define RB_WRITE_MASK 0xfffff | |
371 | #define RB_WRITE_INTCNT (1 << 20) | |
372 | ||
044fa782 | 373 | static void rb_init_page(struct buffer_data_page *bpage) |
abc9b56d | 374 | { |
044fa782 | 375 | local_set(&bpage->commit, 0); |
abc9b56d SR |
376 | } |
377 | ||
474d32b6 SR |
378 | /** |
379 | * ring_buffer_page_len - the size of data on the page. | |
380 | * @page: The page to read | |
381 | * | |
382 | * Returns the amount of data on the page, including buffer page header. | |
383 | */ | |
ef7a4a16 SR |
384 | size_t ring_buffer_page_len(void *page) |
385 | { | |
474d32b6 SR |
386 | return local_read(&((struct buffer_data_page *)page)->commit) |
387 | + BUF_PAGE_HDR_SIZE; | |
ef7a4a16 SR |
388 | } |
389 | ||
ed56829c SR |
390 | /* |
391 | * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing | |
392 | * this issue out. | |
393 | */ | |
34a148bf | 394 | static void free_buffer_page(struct buffer_page *bpage) |
ed56829c | 395 | { |
34a148bf | 396 | free_page((unsigned long)bpage->page); |
e4c2ce82 | 397 | kfree(bpage); |
ed56829c SR |
398 | } |
399 | ||
7a8e76a3 SR |
400 | /* |
401 | * We need to fit the time_stamp delta into 27 bits. | |
402 | */ | |
403 | static inline int test_time_stamp(u64 delta) | |
404 | { | |
405 | if (delta & TS_DELTA_TEST) | |
406 | return 1; | |
407 | return 0; | |
408 | } | |
409 | ||
474d32b6 | 410 | #define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE) |
7a8e76a3 | 411 | |
be957c44 SR |
412 | /* Max payload is BUF_PAGE_SIZE - header (8bytes) */ |
413 | #define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) | |
414 | ||
d1b182a8 SR |
415 | int ring_buffer_print_page_header(struct trace_seq *s) |
416 | { | |
417 | struct buffer_data_page field; | |
c0cd93aa SRRH |
418 | |
419 | trace_seq_printf(s, "\tfield: u64 timestamp;\t" | |
420 | "offset:0;\tsize:%u;\tsigned:%u;\n", | |
421 | (unsigned int)sizeof(field.time_stamp), | |
422 | (unsigned int)is_signed_type(u64)); | |
423 | ||
424 | trace_seq_printf(s, "\tfield: local_t commit;\t" | |
425 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
426 | (unsigned int)offsetof(typeof(field), commit), | |
427 | (unsigned int)sizeof(field.commit), | |
428 | (unsigned int)is_signed_type(long)); | |
429 | ||
430 | trace_seq_printf(s, "\tfield: int overwrite;\t" | |
431 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
432 | (unsigned int)offsetof(typeof(field), commit), | |
433 | 1, | |
434 | (unsigned int)is_signed_type(long)); | |
435 | ||
436 | trace_seq_printf(s, "\tfield: char data;\t" | |
437 | "offset:%u;\tsize:%u;\tsigned:%u;\n", | |
438 | (unsigned int)offsetof(typeof(field), data), | |
439 | (unsigned int)BUF_PAGE_SIZE, | |
440 | (unsigned int)is_signed_type(char)); | |
441 | ||
442 | return !trace_seq_has_overflowed(s); | |
d1b182a8 SR |
443 | } |
444 | ||
15693458 SRRH |
445 | struct rb_irq_work { |
446 | struct irq_work work; | |
447 | wait_queue_head_t waiters; | |
1e0d6714 | 448 | wait_queue_head_t full_waiters; |
15693458 | 449 | bool waiters_pending; |
1e0d6714 SRRH |
450 | bool full_waiters_pending; |
451 | bool wakeup_full; | |
15693458 SRRH |
452 | }; |
453 | ||
7a8e76a3 SR |
454 | /* |
455 | * head_page == tail_page && head == tail then buffer is empty. | |
456 | */ | |
457 | struct ring_buffer_per_cpu { | |
458 | int cpu; | |
985023de | 459 | atomic_t record_disabled; |
7a8e76a3 | 460 | struct ring_buffer *buffer; |
5389f6fa | 461 | raw_spinlock_t reader_lock; /* serialize readers */ |
445c8951 | 462 | arch_spinlock_t lock; |
7a8e76a3 | 463 | struct lock_class_key lock_key; |
438ced17 | 464 | unsigned int nr_pages; |
3adc54fa | 465 | struct list_head *pages; |
6f807acd SR |
466 | struct buffer_page *head_page; /* read from head */ |
467 | struct buffer_page *tail_page; /* write to tail */ | |
c3706f00 | 468 | struct buffer_page *commit_page; /* committed pages */ |
d769041f | 469 | struct buffer_page *reader_page; |
66a8cb95 SR |
470 | unsigned long lost_events; |
471 | unsigned long last_overrun; | |
c64e148a | 472 | local_t entries_bytes; |
e4906eff | 473 | local_t entries; |
884bfe89 SP |
474 | local_t overrun; |
475 | local_t commit_overrun; | |
476 | local_t dropped_events; | |
fa743953 SR |
477 | local_t committing; |
478 | local_t commits; | |
77ae365e | 479 | unsigned long read; |
c64e148a | 480 | unsigned long read_bytes; |
7a8e76a3 SR |
481 | u64 write_stamp; |
482 | u64 read_stamp; | |
438ced17 VN |
483 | /* ring buffer pages to update, > 0 to add, < 0 to remove */ |
484 | int nr_pages_to_update; | |
485 | struct list_head new_pages; /* new pages to add */ | |
83f40318 | 486 | struct work_struct update_pages_work; |
05fdd70d | 487 | struct completion update_done; |
15693458 SRRH |
488 | |
489 | struct rb_irq_work irq_work; | |
7a8e76a3 SR |
490 | }; |
491 | ||
492 | struct ring_buffer { | |
7a8e76a3 SR |
493 | unsigned flags; |
494 | int cpus; | |
7a8e76a3 | 495 | atomic_t record_disabled; |
83f40318 | 496 | atomic_t resize_disabled; |
00f62f61 | 497 | cpumask_var_t cpumask; |
7a8e76a3 | 498 | |
1f8a6a10 PZ |
499 | struct lock_class_key *reader_lock_key; |
500 | ||
7a8e76a3 SR |
501 | struct mutex mutex; |
502 | ||
503 | struct ring_buffer_per_cpu **buffers; | |
554f786e | 504 | |
59222efe | 505 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
506 | struct notifier_block cpu_notify; |
507 | #endif | |
37886f6a | 508 | u64 (*clock)(void); |
15693458 SRRH |
509 | |
510 | struct rb_irq_work irq_work; | |
7a8e76a3 SR |
511 | }; |
512 | ||
513 | struct ring_buffer_iter { | |
514 | struct ring_buffer_per_cpu *cpu_buffer; | |
515 | unsigned long head; | |
516 | struct buffer_page *head_page; | |
492a74f4 SR |
517 | struct buffer_page *cache_reader_page; |
518 | unsigned long cache_read; | |
7a8e76a3 SR |
519 | u64 read_stamp; |
520 | }; | |
521 | ||
15693458 SRRH |
522 | /* |
523 | * rb_wake_up_waiters - wake up tasks waiting for ring buffer input | |
524 | * | |
525 | * Schedules a delayed work to wake up any task that is blocked on the | |
526 | * ring buffer waiters queue. | |
527 | */ | |
528 | static void rb_wake_up_waiters(struct irq_work *work) | |
529 | { | |
530 | struct rb_irq_work *rbwork = container_of(work, struct rb_irq_work, work); | |
531 | ||
532 | wake_up_all(&rbwork->waiters); | |
1e0d6714 SRRH |
533 | if (rbwork->wakeup_full) { |
534 | rbwork->wakeup_full = false; | |
535 | wake_up_all(&rbwork->full_waiters); | |
536 | } | |
15693458 SRRH |
537 | } |
538 | ||
539 | /** | |
540 | * ring_buffer_wait - wait for input to the ring buffer | |
541 | * @buffer: buffer to wait on | |
542 | * @cpu: the cpu buffer to wait on | |
e30f53aa | 543 | * @full: wait until a full page is available, if @cpu != RING_BUFFER_ALL_CPUS |
15693458 SRRH |
544 | * |
545 | * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon | |
546 | * as data is added to any of the @buffer's cpu buffers. Otherwise | |
547 | * it will wait for data to be added to a specific cpu buffer. | |
548 | */ | |
e30f53aa | 549 | int ring_buffer_wait(struct ring_buffer *buffer, int cpu, bool full) |
15693458 | 550 | { |
e30f53aa | 551 | struct ring_buffer_per_cpu *uninitialized_var(cpu_buffer); |
15693458 SRRH |
552 | DEFINE_WAIT(wait); |
553 | struct rb_irq_work *work; | |
e30f53aa | 554 | int ret = 0; |
15693458 SRRH |
555 | |
556 | /* | |
557 | * Depending on what the caller is waiting for, either any | |
558 | * data in any cpu buffer, or a specific buffer, put the | |
559 | * caller on the appropriate wait queue. | |
560 | */ | |
1e0d6714 | 561 | if (cpu == RING_BUFFER_ALL_CPUS) { |
15693458 | 562 | work = &buffer->irq_work; |
1e0d6714 SRRH |
563 | /* Full only makes sense on per cpu reads */ |
564 | full = false; | |
565 | } else { | |
8b8b3683 SRRH |
566 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
567 | return -ENODEV; | |
15693458 SRRH |
568 | cpu_buffer = buffer->buffers[cpu]; |
569 | work = &cpu_buffer->irq_work; | |
570 | } | |
571 | ||
572 | ||
e30f53aa | 573 | while (true) { |
1e0d6714 SRRH |
574 | if (full) |
575 | prepare_to_wait(&work->full_waiters, &wait, TASK_INTERRUPTIBLE); | |
576 | else | |
577 | prepare_to_wait(&work->waiters, &wait, TASK_INTERRUPTIBLE); | |
e30f53aa RV |
578 | |
579 | /* | |
580 | * The events can happen in critical sections where | |
581 | * checking a work queue can cause deadlocks. | |
582 | * After adding a task to the queue, this flag is set | |
583 | * only to notify events to try to wake up the queue | |
584 | * using irq_work. | |
585 | * | |
586 | * We don't clear it even if the buffer is no longer | |
587 | * empty. The flag only causes the next event to run | |
588 | * irq_work to do the work queue wake up. The worse | |
589 | * that can happen if we race with !trace_empty() is that | |
590 | * an event will cause an irq_work to try to wake up | |
591 | * an empty queue. | |
592 | * | |
593 | * There's no reason to protect this flag either, as | |
594 | * the work queue and irq_work logic will do the necessary | |
595 | * synchronization for the wake ups. The only thing | |
596 | * that is necessary is that the wake up happens after | |
597 | * a task has been queued. It's OK for spurious wake ups. | |
598 | */ | |
1e0d6714 SRRH |
599 | if (full) |
600 | work->full_waiters_pending = true; | |
601 | else | |
602 | work->waiters_pending = true; | |
e30f53aa RV |
603 | |
604 | if (signal_pending(current)) { | |
605 | ret = -EINTR; | |
606 | break; | |
607 | } | |
608 | ||
609 | if (cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) | |
610 | break; | |
611 | ||
612 | if (cpu != RING_BUFFER_ALL_CPUS && | |
613 | !ring_buffer_empty_cpu(buffer, cpu)) { | |
614 | unsigned long flags; | |
615 | bool pagebusy; | |
616 | ||
617 | if (!full) | |
618 | break; | |
619 | ||
620 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); | |
621 | pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page; | |
622 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); | |
623 | ||
624 | if (!pagebusy) | |
625 | break; | |
626 | } | |
15693458 | 627 | |
15693458 | 628 | schedule(); |
e30f53aa | 629 | } |
15693458 | 630 | |
1e0d6714 SRRH |
631 | if (full) |
632 | finish_wait(&work->full_waiters, &wait); | |
633 | else | |
634 | finish_wait(&work->waiters, &wait); | |
e30f53aa RV |
635 | |
636 | return ret; | |
15693458 SRRH |
637 | } |
638 | ||
639 | /** | |
640 | * ring_buffer_poll_wait - poll on buffer input | |
641 | * @buffer: buffer to wait on | |
642 | * @cpu: the cpu buffer to wait on | |
643 | * @filp: the file descriptor | |
644 | * @poll_table: The poll descriptor | |
645 | * | |
646 | * If @cpu == RING_BUFFER_ALL_CPUS then the task will wake up as soon | |
647 | * as data is added to any of the @buffer's cpu buffers. Otherwise | |
648 | * it will wait for data to be added to a specific cpu buffer. | |
649 | * | |
650 | * Returns POLLIN | POLLRDNORM if data exists in the buffers, | |
651 | * zero otherwise. | |
652 | */ | |
653 | int ring_buffer_poll_wait(struct ring_buffer *buffer, int cpu, | |
654 | struct file *filp, poll_table *poll_table) | |
655 | { | |
656 | struct ring_buffer_per_cpu *cpu_buffer; | |
657 | struct rb_irq_work *work; | |
658 | ||
15693458 SRRH |
659 | if (cpu == RING_BUFFER_ALL_CPUS) |
660 | work = &buffer->irq_work; | |
661 | else { | |
6721cb60 SRRH |
662 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
663 | return -EINVAL; | |
664 | ||
15693458 SRRH |
665 | cpu_buffer = buffer->buffers[cpu]; |
666 | work = &cpu_buffer->irq_work; | |
667 | } | |
668 | ||
15693458 | 669 | poll_wait(filp, &work->waiters, poll_table); |
4ce97dbf JB |
670 | work->waiters_pending = true; |
671 | /* | |
672 | * There's a tight race between setting the waiters_pending and | |
673 | * checking if the ring buffer is empty. Once the waiters_pending bit | |
674 | * is set, the next event will wake the task up, but we can get stuck | |
675 | * if there's only a single event in. | |
676 | * | |
677 | * FIXME: Ideally, we need a memory barrier on the writer side as well, | |
678 | * but adding a memory barrier to all events will cause too much of a | |
679 | * performance hit in the fast path. We only need a memory barrier when | |
680 | * the buffer goes from empty to having content. But as this race is | |
681 | * extremely small, and it's not a problem if another event comes in, we | |
682 | * will fix it later. | |
683 | */ | |
684 | smp_mb(); | |
15693458 SRRH |
685 | |
686 | if ((cpu == RING_BUFFER_ALL_CPUS && !ring_buffer_empty(buffer)) || | |
687 | (cpu != RING_BUFFER_ALL_CPUS && !ring_buffer_empty_cpu(buffer, cpu))) | |
688 | return POLLIN | POLLRDNORM; | |
689 | return 0; | |
690 | } | |
691 | ||
f536aafc | 692 | /* buffer may be either ring_buffer or ring_buffer_per_cpu */ |
077c5407 SR |
693 | #define RB_WARN_ON(b, cond) \ |
694 | ({ \ | |
695 | int _____ret = unlikely(cond); \ | |
696 | if (_____ret) { \ | |
697 | if (__same_type(*(b), struct ring_buffer_per_cpu)) { \ | |
698 | struct ring_buffer_per_cpu *__b = \ | |
699 | (void *)b; \ | |
700 | atomic_inc(&__b->buffer->record_disabled); \ | |
701 | } else \ | |
702 | atomic_inc(&b->record_disabled); \ | |
703 | WARN_ON(1); \ | |
704 | } \ | |
705 | _____ret; \ | |
3e89c7bb | 706 | }) |
f536aafc | 707 | |
37886f6a SR |
708 | /* Up this if you want to test the TIME_EXTENTS and normalization */ |
709 | #define DEBUG_SHIFT 0 | |
710 | ||
6d3f1e12 | 711 | static inline u64 rb_time_stamp(struct ring_buffer *buffer) |
88eb0125 SR |
712 | { |
713 | /* shift to debug/test normalization and TIME_EXTENTS */ | |
714 | return buffer->clock() << DEBUG_SHIFT; | |
715 | } | |
716 | ||
37886f6a SR |
717 | u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu) |
718 | { | |
719 | u64 time; | |
720 | ||
721 | preempt_disable_notrace(); | |
6d3f1e12 | 722 | time = rb_time_stamp(buffer); |
37886f6a SR |
723 | preempt_enable_no_resched_notrace(); |
724 | ||
725 | return time; | |
726 | } | |
727 | EXPORT_SYMBOL_GPL(ring_buffer_time_stamp); | |
728 | ||
729 | void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer, | |
730 | int cpu, u64 *ts) | |
731 | { | |
732 | /* Just stupid testing the normalize function and deltas */ | |
733 | *ts >>= DEBUG_SHIFT; | |
734 | } | |
735 | EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp); | |
736 | ||
77ae365e SR |
737 | /* |
738 | * Making the ring buffer lockless makes things tricky. | |
739 | * Although writes only happen on the CPU that they are on, | |
740 | * and they only need to worry about interrupts. Reads can | |
741 | * happen on any CPU. | |
742 | * | |
743 | * The reader page is always off the ring buffer, but when the | |
744 | * reader finishes with a page, it needs to swap its page with | |
745 | * a new one from the buffer. The reader needs to take from | |
746 | * the head (writes go to the tail). But if a writer is in overwrite | |
747 | * mode and wraps, it must push the head page forward. | |
748 | * | |
749 | * Here lies the problem. | |
750 | * | |
751 | * The reader must be careful to replace only the head page, and | |
752 | * not another one. As described at the top of the file in the | |
753 | * ASCII art, the reader sets its old page to point to the next | |
754 | * page after head. It then sets the page after head to point to | |
755 | * the old reader page. But if the writer moves the head page | |
756 | * during this operation, the reader could end up with the tail. | |
757 | * | |
758 | * We use cmpxchg to help prevent this race. We also do something | |
759 | * special with the page before head. We set the LSB to 1. | |
760 | * | |
761 | * When the writer must push the page forward, it will clear the | |
762 | * bit that points to the head page, move the head, and then set | |
763 | * the bit that points to the new head page. | |
764 | * | |
765 | * We also don't want an interrupt coming in and moving the head | |
766 | * page on another writer. Thus we use the second LSB to catch | |
767 | * that too. Thus: | |
768 | * | |
769 | * head->list->prev->next bit 1 bit 0 | |
770 | * ------- ------- | |
771 | * Normal page 0 0 | |
772 | * Points to head page 0 1 | |
773 | * New head page 1 0 | |
774 | * | |
775 | * Note we can not trust the prev pointer of the head page, because: | |
776 | * | |
777 | * +----+ +-----+ +-----+ | |
778 | * | |------>| T |---X--->| N | | |
779 | * | |<------| | | | | |
780 | * +----+ +-----+ +-----+ | |
781 | * ^ ^ | | |
782 | * | +-----+ | | | |
783 | * +----------| R |----------+ | | |
784 | * | |<-----------+ | |
785 | * +-----+ | |
786 | * | |
787 | * Key: ---X--> HEAD flag set in pointer | |
788 | * T Tail page | |
789 | * R Reader page | |
790 | * N Next page | |
791 | * | |
792 | * (see __rb_reserve_next() to see where this happens) | |
793 | * | |
794 | * What the above shows is that the reader just swapped out | |
795 | * the reader page with a page in the buffer, but before it | |
796 | * could make the new header point back to the new page added | |
797 | * it was preempted by a writer. The writer moved forward onto | |
798 | * the new page added by the reader and is about to move forward | |
799 | * again. | |
800 | * | |
801 | * You can see, it is legitimate for the previous pointer of | |
802 | * the head (or any page) not to point back to itself. But only | |
803 | * temporarially. | |
804 | */ | |
805 | ||
806 | #define RB_PAGE_NORMAL 0UL | |
807 | #define RB_PAGE_HEAD 1UL | |
808 | #define RB_PAGE_UPDATE 2UL | |
809 | ||
810 | ||
811 | #define RB_FLAG_MASK 3UL | |
812 | ||
813 | /* PAGE_MOVED is not part of the mask */ | |
814 | #define RB_PAGE_MOVED 4UL | |
815 | ||
816 | /* | |
817 | * rb_list_head - remove any bit | |
818 | */ | |
819 | static struct list_head *rb_list_head(struct list_head *list) | |
820 | { | |
821 | unsigned long val = (unsigned long)list; | |
822 | ||
823 | return (struct list_head *)(val & ~RB_FLAG_MASK); | |
824 | } | |
825 | ||
826 | /* | |
6d3f1e12 | 827 | * rb_is_head_page - test if the given page is the head page |
77ae365e SR |
828 | * |
829 | * Because the reader may move the head_page pointer, we can | |
830 | * not trust what the head page is (it may be pointing to | |
831 | * the reader page). But if the next page is a header page, | |
832 | * its flags will be non zero. | |
833 | */ | |
42b16b3f | 834 | static inline int |
77ae365e SR |
835 | rb_is_head_page(struct ring_buffer_per_cpu *cpu_buffer, |
836 | struct buffer_page *page, struct list_head *list) | |
837 | { | |
838 | unsigned long val; | |
839 | ||
840 | val = (unsigned long)list->next; | |
841 | ||
842 | if ((val & ~RB_FLAG_MASK) != (unsigned long)&page->list) | |
843 | return RB_PAGE_MOVED; | |
844 | ||
845 | return val & RB_FLAG_MASK; | |
846 | } | |
847 | ||
848 | /* | |
849 | * rb_is_reader_page | |
850 | * | |
851 | * The unique thing about the reader page, is that, if the | |
852 | * writer is ever on it, the previous pointer never points | |
853 | * back to the reader page. | |
854 | */ | |
855 | static int rb_is_reader_page(struct buffer_page *page) | |
856 | { | |
857 | struct list_head *list = page->list.prev; | |
858 | ||
859 | return rb_list_head(list->next) != &page->list; | |
860 | } | |
861 | ||
862 | /* | |
863 | * rb_set_list_to_head - set a list_head to be pointing to head. | |
864 | */ | |
865 | static void rb_set_list_to_head(struct ring_buffer_per_cpu *cpu_buffer, | |
866 | struct list_head *list) | |
867 | { | |
868 | unsigned long *ptr; | |
869 | ||
870 | ptr = (unsigned long *)&list->next; | |
871 | *ptr |= RB_PAGE_HEAD; | |
872 | *ptr &= ~RB_PAGE_UPDATE; | |
873 | } | |
874 | ||
875 | /* | |
876 | * rb_head_page_activate - sets up head page | |
877 | */ | |
878 | static void rb_head_page_activate(struct ring_buffer_per_cpu *cpu_buffer) | |
879 | { | |
880 | struct buffer_page *head; | |
881 | ||
882 | head = cpu_buffer->head_page; | |
883 | if (!head) | |
884 | return; | |
885 | ||
886 | /* | |
887 | * Set the previous list pointer to have the HEAD flag. | |
888 | */ | |
889 | rb_set_list_to_head(cpu_buffer, head->list.prev); | |
890 | } | |
891 | ||
892 | static void rb_list_head_clear(struct list_head *list) | |
893 | { | |
894 | unsigned long *ptr = (unsigned long *)&list->next; | |
895 | ||
896 | *ptr &= ~RB_FLAG_MASK; | |
897 | } | |
898 | ||
899 | /* | |
900 | * rb_head_page_dactivate - clears head page ptr (for free list) | |
901 | */ | |
902 | static void | |
903 | rb_head_page_deactivate(struct ring_buffer_per_cpu *cpu_buffer) | |
904 | { | |
905 | struct list_head *hd; | |
906 | ||
907 | /* Go through the whole list and clear any pointers found. */ | |
908 | rb_list_head_clear(cpu_buffer->pages); | |
909 | ||
910 | list_for_each(hd, cpu_buffer->pages) | |
911 | rb_list_head_clear(hd); | |
912 | } | |
913 | ||
914 | static int rb_head_page_set(struct ring_buffer_per_cpu *cpu_buffer, | |
915 | struct buffer_page *head, | |
916 | struct buffer_page *prev, | |
917 | int old_flag, int new_flag) | |
918 | { | |
919 | struct list_head *list; | |
920 | unsigned long val = (unsigned long)&head->list; | |
921 | unsigned long ret; | |
922 | ||
923 | list = &prev->list; | |
924 | ||
925 | val &= ~RB_FLAG_MASK; | |
926 | ||
08a40816 SR |
927 | ret = cmpxchg((unsigned long *)&list->next, |
928 | val | old_flag, val | new_flag); | |
77ae365e SR |
929 | |
930 | /* check if the reader took the page */ | |
931 | if ((ret & ~RB_FLAG_MASK) != val) | |
932 | return RB_PAGE_MOVED; | |
933 | ||
934 | return ret & RB_FLAG_MASK; | |
935 | } | |
936 | ||
937 | static int rb_head_page_set_update(struct ring_buffer_per_cpu *cpu_buffer, | |
938 | struct buffer_page *head, | |
939 | struct buffer_page *prev, | |
940 | int old_flag) | |
941 | { | |
942 | return rb_head_page_set(cpu_buffer, head, prev, | |
943 | old_flag, RB_PAGE_UPDATE); | |
944 | } | |
945 | ||
946 | static int rb_head_page_set_head(struct ring_buffer_per_cpu *cpu_buffer, | |
947 | struct buffer_page *head, | |
948 | struct buffer_page *prev, | |
949 | int old_flag) | |
950 | { | |
951 | return rb_head_page_set(cpu_buffer, head, prev, | |
952 | old_flag, RB_PAGE_HEAD); | |
953 | } | |
954 | ||
955 | static int rb_head_page_set_normal(struct ring_buffer_per_cpu *cpu_buffer, | |
956 | struct buffer_page *head, | |
957 | struct buffer_page *prev, | |
958 | int old_flag) | |
959 | { | |
960 | return rb_head_page_set(cpu_buffer, head, prev, | |
961 | old_flag, RB_PAGE_NORMAL); | |
962 | } | |
963 | ||
964 | static inline void rb_inc_page(struct ring_buffer_per_cpu *cpu_buffer, | |
965 | struct buffer_page **bpage) | |
966 | { | |
967 | struct list_head *p = rb_list_head((*bpage)->list.next); | |
968 | ||
969 | *bpage = list_entry(p, struct buffer_page, list); | |
970 | } | |
971 | ||
972 | static struct buffer_page * | |
973 | rb_set_head_page(struct ring_buffer_per_cpu *cpu_buffer) | |
974 | { | |
975 | struct buffer_page *head; | |
976 | struct buffer_page *page; | |
977 | struct list_head *list; | |
978 | int i; | |
979 | ||
980 | if (RB_WARN_ON(cpu_buffer, !cpu_buffer->head_page)) | |
981 | return NULL; | |
982 | ||
983 | /* sanity check */ | |
984 | list = cpu_buffer->pages; | |
985 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev->next) != list)) | |
986 | return NULL; | |
987 | ||
988 | page = head = cpu_buffer->head_page; | |
989 | /* | |
990 | * It is possible that the writer moves the header behind | |
991 | * where we started, and we miss in one loop. | |
992 | * A second loop should grab the header, but we'll do | |
993 | * three loops just because I'm paranoid. | |
994 | */ | |
995 | for (i = 0; i < 3; i++) { | |
996 | do { | |
997 | if (rb_is_head_page(cpu_buffer, page, page->list.prev)) { | |
998 | cpu_buffer->head_page = page; | |
999 | return page; | |
1000 | } | |
1001 | rb_inc_page(cpu_buffer, &page); | |
1002 | } while (page != head); | |
1003 | } | |
1004 | ||
1005 | RB_WARN_ON(cpu_buffer, 1); | |
1006 | ||
1007 | return NULL; | |
1008 | } | |
1009 | ||
1010 | static int rb_head_page_replace(struct buffer_page *old, | |
1011 | struct buffer_page *new) | |
1012 | { | |
1013 | unsigned long *ptr = (unsigned long *)&old->list.prev->next; | |
1014 | unsigned long val; | |
1015 | unsigned long ret; | |
1016 | ||
1017 | val = *ptr & ~RB_FLAG_MASK; | |
1018 | val |= RB_PAGE_HEAD; | |
1019 | ||
08a40816 | 1020 | ret = cmpxchg(ptr, val, (unsigned long)&new->list); |
77ae365e SR |
1021 | |
1022 | return ret == val; | |
1023 | } | |
1024 | ||
1025 | /* | |
1026 | * rb_tail_page_update - move the tail page forward | |
1027 | * | |
1028 | * Returns 1 if moved tail page, 0 if someone else did. | |
1029 | */ | |
1030 | static int rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer, | |
1031 | struct buffer_page *tail_page, | |
1032 | struct buffer_page *next_page) | |
1033 | { | |
1034 | struct buffer_page *old_tail; | |
1035 | unsigned long old_entries; | |
1036 | unsigned long old_write; | |
1037 | int ret = 0; | |
1038 | ||
1039 | /* | |
1040 | * The tail page now needs to be moved forward. | |
1041 | * | |
1042 | * We need to reset the tail page, but without messing | |
1043 | * with possible erasing of data brought in by interrupts | |
1044 | * that have moved the tail page and are currently on it. | |
1045 | * | |
1046 | * We add a counter to the write field to denote this. | |
1047 | */ | |
1048 | old_write = local_add_return(RB_WRITE_INTCNT, &next_page->write); | |
1049 | old_entries = local_add_return(RB_WRITE_INTCNT, &next_page->entries); | |
1050 | ||
1051 | /* | |
1052 | * Just make sure we have seen our old_write and synchronize | |
1053 | * with any interrupts that come in. | |
1054 | */ | |
1055 | barrier(); | |
1056 | ||
1057 | /* | |
1058 | * If the tail page is still the same as what we think | |
1059 | * it is, then it is up to us to update the tail | |
1060 | * pointer. | |
1061 | */ | |
1062 | if (tail_page == cpu_buffer->tail_page) { | |
1063 | /* Zero the write counter */ | |
1064 | unsigned long val = old_write & ~RB_WRITE_MASK; | |
1065 | unsigned long eval = old_entries & ~RB_WRITE_MASK; | |
1066 | ||
1067 | /* | |
1068 | * This will only succeed if an interrupt did | |
1069 | * not come in and change it. In which case, we | |
1070 | * do not want to modify it. | |
da706d8b LJ |
1071 | * |
1072 | * We add (void) to let the compiler know that we do not care | |
1073 | * about the return value of these functions. We use the | |
1074 | * cmpxchg to only update if an interrupt did not already | |
1075 | * do it for us. If the cmpxchg fails, we don't care. | |
77ae365e | 1076 | */ |
da706d8b LJ |
1077 | (void)local_cmpxchg(&next_page->write, old_write, val); |
1078 | (void)local_cmpxchg(&next_page->entries, old_entries, eval); | |
77ae365e SR |
1079 | |
1080 | /* | |
1081 | * No need to worry about races with clearing out the commit. | |
1082 | * it only can increment when a commit takes place. But that | |
1083 | * only happens in the outer most nested commit. | |
1084 | */ | |
1085 | local_set(&next_page->page->commit, 0); | |
1086 | ||
1087 | old_tail = cmpxchg(&cpu_buffer->tail_page, | |
1088 | tail_page, next_page); | |
1089 | ||
1090 | if (old_tail == tail_page) | |
1091 | ret = 1; | |
1092 | } | |
1093 | ||
1094 | return ret; | |
1095 | } | |
1096 | ||
1097 | static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer, | |
1098 | struct buffer_page *bpage) | |
1099 | { | |
1100 | unsigned long val = (unsigned long)bpage; | |
1101 | ||
1102 | if (RB_WARN_ON(cpu_buffer, val & RB_FLAG_MASK)) | |
1103 | return 1; | |
1104 | ||
1105 | return 0; | |
1106 | } | |
1107 | ||
1108 | /** | |
1109 | * rb_check_list - make sure a pointer to a list has the last bits zero | |
1110 | */ | |
1111 | static int rb_check_list(struct ring_buffer_per_cpu *cpu_buffer, | |
1112 | struct list_head *list) | |
1113 | { | |
1114 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->prev) != list->prev)) | |
1115 | return 1; | |
1116 | if (RB_WARN_ON(cpu_buffer, rb_list_head(list->next) != list->next)) | |
1117 | return 1; | |
1118 | return 0; | |
1119 | } | |
1120 | ||
7a8e76a3 | 1121 | /** |
d611851b | 1122 | * rb_check_pages - integrity check of buffer pages |
7a8e76a3 SR |
1123 | * @cpu_buffer: CPU buffer with pages to test |
1124 | * | |
c3706f00 | 1125 | * As a safety measure we check to make sure the data pages have not |
7a8e76a3 SR |
1126 | * been corrupted. |
1127 | */ | |
1128 | static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
1129 | { | |
3adc54fa | 1130 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1131 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1132 | |
308f7eeb SR |
1133 | /* Reset the head page if it exists */ |
1134 | if (cpu_buffer->head_page) | |
1135 | rb_set_head_page(cpu_buffer); | |
1136 | ||
77ae365e SR |
1137 | rb_head_page_deactivate(cpu_buffer); |
1138 | ||
3e89c7bb SR |
1139 | if (RB_WARN_ON(cpu_buffer, head->next->prev != head)) |
1140 | return -1; | |
1141 | if (RB_WARN_ON(cpu_buffer, head->prev->next != head)) | |
1142 | return -1; | |
7a8e76a3 | 1143 | |
77ae365e SR |
1144 | if (rb_check_list(cpu_buffer, head)) |
1145 | return -1; | |
1146 | ||
044fa782 | 1147 | list_for_each_entry_safe(bpage, tmp, head, list) { |
3e89c7bb | 1148 | if (RB_WARN_ON(cpu_buffer, |
044fa782 | 1149 | bpage->list.next->prev != &bpage->list)) |
3e89c7bb SR |
1150 | return -1; |
1151 | if (RB_WARN_ON(cpu_buffer, | |
044fa782 | 1152 | bpage->list.prev->next != &bpage->list)) |
3e89c7bb | 1153 | return -1; |
77ae365e SR |
1154 | if (rb_check_list(cpu_buffer, &bpage->list)) |
1155 | return -1; | |
7a8e76a3 SR |
1156 | } |
1157 | ||
77ae365e SR |
1158 | rb_head_page_activate(cpu_buffer); |
1159 | ||
7a8e76a3 SR |
1160 | return 0; |
1161 | } | |
1162 | ||
438ced17 | 1163 | static int __rb_allocate_pages(int nr_pages, struct list_head *pages, int cpu) |
7a8e76a3 | 1164 | { |
438ced17 | 1165 | int i; |
044fa782 | 1166 | struct buffer_page *bpage, *tmp; |
3adc54fa | 1167 | |
7a8e76a3 | 1168 | for (i = 0; i < nr_pages; i++) { |
7ea59064 | 1169 | struct page *page; |
d7ec4bfe VN |
1170 | /* |
1171 | * __GFP_NORETRY flag makes sure that the allocation fails | |
1172 | * gracefully without invoking oom-killer and the system is | |
1173 | * not destabilized. | |
1174 | */ | |
044fa782 | 1175 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
d7ec4bfe | 1176 | GFP_KERNEL | __GFP_NORETRY, |
438ced17 | 1177 | cpu_to_node(cpu)); |
044fa782 | 1178 | if (!bpage) |
e4c2ce82 | 1179 | goto free_pages; |
77ae365e | 1180 | |
438ced17 | 1181 | list_add(&bpage->list, pages); |
77ae365e | 1182 | |
438ced17 | 1183 | page = alloc_pages_node(cpu_to_node(cpu), |
d7ec4bfe | 1184 | GFP_KERNEL | __GFP_NORETRY, 0); |
7ea59064 | 1185 | if (!page) |
7a8e76a3 | 1186 | goto free_pages; |
7ea59064 | 1187 | bpage->page = page_address(page); |
044fa782 | 1188 | rb_init_page(bpage->page); |
7a8e76a3 SR |
1189 | } |
1190 | ||
438ced17 VN |
1191 | return 0; |
1192 | ||
1193 | free_pages: | |
1194 | list_for_each_entry_safe(bpage, tmp, pages, list) { | |
1195 | list_del_init(&bpage->list); | |
1196 | free_buffer_page(bpage); | |
1197 | } | |
1198 | ||
1199 | return -ENOMEM; | |
1200 | } | |
1201 | ||
1202 | static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer, | |
1203 | unsigned nr_pages) | |
1204 | { | |
1205 | LIST_HEAD(pages); | |
1206 | ||
1207 | WARN_ON(!nr_pages); | |
1208 | ||
1209 | if (__rb_allocate_pages(nr_pages, &pages, cpu_buffer->cpu)) | |
1210 | return -ENOMEM; | |
1211 | ||
3adc54fa SR |
1212 | /* |
1213 | * The ring buffer page list is a circular list that does not | |
1214 | * start and end with a list head. All page list items point to | |
1215 | * other pages. | |
1216 | */ | |
1217 | cpu_buffer->pages = pages.next; | |
1218 | list_del(&pages); | |
7a8e76a3 | 1219 | |
438ced17 VN |
1220 | cpu_buffer->nr_pages = nr_pages; |
1221 | ||
7a8e76a3 SR |
1222 | rb_check_pages(cpu_buffer); |
1223 | ||
1224 | return 0; | |
7a8e76a3 SR |
1225 | } |
1226 | ||
1227 | static struct ring_buffer_per_cpu * | |
438ced17 | 1228 | rb_allocate_cpu_buffer(struct ring_buffer *buffer, int nr_pages, int cpu) |
7a8e76a3 SR |
1229 | { |
1230 | struct ring_buffer_per_cpu *cpu_buffer; | |
044fa782 | 1231 | struct buffer_page *bpage; |
7ea59064 | 1232 | struct page *page; |
7a8e76a3 SR |
1233 | int ret; |
1234 | ||
1235 | cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()), | |
1236 | GFP_KERNEL, cpu_to_node(cpu)); | |
1237 | if (!cpu_buffer) | |
1238 | return NULL; | |
1239 | ||
1240 | cpu_buffer->cpu = cpu; | |
1241 | cpu_buffer->buffer = buffer; | |
5389f6fa | 1242 | raw_spin_lock_init(&cpu_buffer->reader_lock); |
1f8a6a10 | 1243 | lockdep_set_class(&cpu_buffer->reader_lock, buffer->reader_lock_key); |
edc35bd7 | 1244 | cpu_buffer->lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; |
83f40318 | 1245 | INIT_WORK(&cpu_buffer->update_pages_work, update_pages_handler); |
05fdd70d | 1246 | init_completion(&cpu_buffer->update_done); |
15693458 | 1247 | init_irq_work(&cpu_buffer->irq_work.work, rb_wake_up_waiters); |
f1dc6725 | 1248 | init_waitqueue_head(&cpu_buffer->irq_work.waiters); |
1e0d6714 | 1249 | init_waitqueue_head(&cpu_buffer->irq_work.full_waiters); |
7a8e76a3 | 1250 | |
044fa782 | 1251 | bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()), |
e4c2ce82 | 1252 | GFP_KERNEL, cpu_to_node(cpu)); |
044fa782 | 1253 | if (!bpage) |
e4c2ce82 SR |
1254 | goto fail_free_buffer; |
1255 | ||
77ae365e SR |
1256 | rb_check_bpage(cpu_buffer, bpage); |
1257 | ||
044fa782 | 1258 | cpu_buffer->reader_page = bpage; |
7ea59064 VN |
1259 | page = alloc_pages_node(cpu_to_node(cpu), GFP_KERNEL, 0); |
1260 | if (!page) | |
e4c2ce82 | 1261 | goto fail_free_reader; |
7ea59064 | 1262 | bpage->page = page_address(page); |
044fa782 | 1263 | rb_init_page(bpage->page); |
e4c2ce82 | 1264 | |
d769041f | 1265 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); |
44b99462 | 1266 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
d769041f | 1267 | |
438ced17 | 1268 | ret = rb_allocate_pages(cpu_buffer, nr_pages); |
7a8e76a3 | 1269 | if (ret < 0) |
d769041f | 1270 | goto fail_free_reader; |
7a8e76a3 SR |
1271 | |
1272 | cpu_buffer->head_page | |
3adc54fa | 1273 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 1274 | cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page; |
7a8e76a3 | 1275 | |
77ae365e SR |
1276 | rb_head_page_activate(cpu_buffer); |
1277 | ||
7a8e76a3 SR |
1278 | return cpu_buffer; |
1279 | ||
d769041f SR |
1280 | fail_free_reader: |
1281 | free_buffer_page(cpu_buffer->reader_page); | |
1282 | ||
7a8e76a3 SR |
1283 | fail_free_buffer: |
1284 | kfree(cpu_buffer); | |
1285 | return NULL; | |
1286 | } | |
1287 | ||
1288 | static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) | |
1289 | { | |
3adc54fa | 1290 | struct list_head *head = cpu_buffer->pages; |
044fa782 | 1291 | struct buffer_page *bpage, *tmp; |
7a8e76a3 | 1292 | |
d769041f SR |
1293 | free_buffer_page(cpu_buffer->reader_page); |
1294 | ||
77ae365e SR |
1295 | rb_head_page_deactivate(cpu_buffer); |
1296 | ||
3adc54fa SR |
1297 | if (head) { |
1298 | list_for_each_entry_safe(bpage, tmp, head, list) { | |
1299 | list_del_init(&bpage->list); | |
1300 | free_buffer_page(bpage); | |
1301 | } | |
1302 | bpage = list_entry(head, struct buffer_page, list); | |
044fa782 | 1303 | free_buffer_page(bpage); |
7a8e76a3 | 1304 | } |
3adc54fa | 1305 | |
7a8e76a3 SR |
1306 | kfree(cpu_buffer); |
1307 | } | |
1308 | ||
59222efe | 1309 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
1310 | static int rb_cpu_notify(struct notifier_block *self, |
1311 | unsigned long action, void *hcpu); | |
554f786e SR |
1312 | #endif |
1313 | ||
7a8e76a3 | 1314 | /** |
d611851b | 1315 | * __ring_buffer_alloc - allocate a new ring_buffer |
68814b58 | 1316 | * @size: the size in bytes per cpu that is needed. |
7a8e76a3 SR |
1317 | * @flags: attributes to set for the ring buffer. |
1318 | * | |
1319 | * Currently the only flag that is available is the RB_FL_OVERWRITE | |
1320 | * flag. This flag means that the buffer will overwrite old data | |
1321 | * when the buffer wraps. If this flag is not set, the buffer will | |
1322 | * drop data when the tail hits the head. | |
1323 | */ | |
1f8a6a10 PZ |
1324 | struct ring_buffer *__ring_buffer_alloc(unsigned long size, unsigned flags, |
1325 | struct lock_class_key *key) | |
7a8e76a3 SR |
1326 | { |
1327 | struct ring_buffer *buffer; | |
1328 | int bsize; | |
438ced17 | 1329 | int cpu, nr_pages; |
7a8e76a3 SR |
1330 | |
1331 | /* keep it in its own cache line */ | |
1332 | buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()), | |
1333 | GFP_KERNEL); | |
1334 | if (!buffer) | |
1335 | return NULL; | |
1336 | ||
9e01c1b7 RR |
1337 | if (!alloc_cpumask_var(&buffer->cpumask, GFP_KERNEL)) |
1338 | goto fail_free_buffer; | |
1339 | ||
438ced17 | 1340 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
7a8e76a3 | 1341 | buffer->flags = flags; |
37886f6a | 1342 | buffer->clock = trace_clock_local; |
1f8a6a10 | 1343 | buffer->reader_lock_key = key; |
7a8e76a3 | 1344 | |
15693458 | 1345 | init_irq_work(&buffer->irq_work.work, rb_wake_up_waiters); |
f1dc6725 | 1346 | init_waitqueue_head(&buffer->irq_work.waiters); |
15693458 | 1347 | |
7a8e76a3 | 1348 | /* need at least two pages */ |
438ced17 VN |
1349 | if (nr_pages < 2) |
1350 | nr_pages = 2; | |
7a8e76a3 | 1351 | |
3bf832ce FW |
1352 | /* |
1353 | * In case of non-hotplug cpu, if the ring-buffer is allocated | |
1354 | * in early initcall, it will not be notified of secondary cpus. | |
1355 | * In that off case, we need to allocate for all possible cpus. | |
1356 | */ | |
1357 | #ifdef CONFIG_HOTPLUG_CPU | |
d39ad278 | 1358 | cpu_notifier_register_begin(); |
554f786e | 1359 | cpumask_copy(buffer->cpumask, cpu_online_mask); |
3bf832ce FW |
1360 | #else |
1361 | cpumask_copy(buffer->cpumask, cpu_possible_mask); | |
1362 | #endif | |
7a8e76a3 SR |
1363 | buffer->cpus = nr_cpu_ids; |
1364 | ||
1365 | bsize = sizeof(void *) * nr_cpu_ids; | |
1366 | buffer->buffers = kzalloc(ALIGN(bsize, cache_line_size()), | |
1367 | GFP_KERNEL); | |
1368 | if (!buffer->buffers) | |
9e01c1b7 | 1369 | goto fail_free_cpumask; |
7a8e76a3 SR |
1370 | |
1371 | for_each_buffer_cpu(buffer, cpu) { | |
1372 | buffer->buffers[cpu] = | |
438ced17 | 1373 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
7a8e76a3 SR |
1374 | if (!buffer->buffers[cpu]) |
1375 | goto fail_free_buffers; | |
1376 | } | |
1377 | ||
59222efe | 1378 | #ifdef CONFIG_HOTPLUG_CPU |
554f786e SR |
1379 | buffer->cpu_notify.notifier_call = rb_cpu_notify; |
1380 | buffer->cpu_notify.priority = 0; | |
d39ad278 SB |
1381 | __register_cpu_notifier(&buffer->cpu_notify); |
1382 | cpu_notifier_register_done(); | |
554f786e SR |
1383 | #endif |
1384 | ||
7a8e76a3 SR |
1385 | mutex_init(&buffer->mutex); |
1386 | ||
1387 | return buffer; | |
1388 | ||
1389 | fail_free_buffers: | |
1390 | for_each_buffer_cpu(buffer, cpu) { | |
1391 | if (buffer->buffers[cpu]) | |
1392 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1393 | } | |
1394 | kfree(buffer->buffers); | |
1395 | ||
9e01c1b7 RR |
1396 | fail_free_cpumask: |
1397 | free_cpumask_var(buffer->cpumask); | |
d39ad278 SB |
1398 | #ifdef CONFIG_HOTPLUG_CPU |
1399 | cpu_notifier_register_done(); | |
1400 | #endif | |
9e01c1b7 | 1401 | |
7a8e76a3 SR |
1402 | fail_free_buffer: |
1403 | kfree(buffer); | |
1404 | return NULL; | |
1405 | } | |
1f8a6a10 | 1406 | EXPORT_SYMBOL_GPL(__ring_buffer_alloc); |
7a8e76a3 SR |
1407 | |
1408 | /** | |
1409 | * ring_buffer_free - free a ring buffer. | |
1410 | * @buffer: the buffer to free. | |
1411 | */ | |
1412 | void | |
1413 | ring_buffer_free(struct ring_buffer *buffer) | |
1414 | { | |
1415 | int cpu; | |
1416 | ||
59222efe | 1417 | #ifdef CONFIG_HOTPLUG_CPU |
d39ad278 SB |
1418 | cpu_notifier_register_begin(); |
1419 | __unregister_cpu_notifier(&buffer->cpu_notify); | |
554f786e SR |
1420 | #endif |
1421 | ||
7a8e76a3 SR |
1422 | for_each_buffer_cpu(buffer, cpu) |
1423 | rb_free_cpu_buffer(buffer->buffers[cpu]); | |
1424 | ||
d39ad278 SB |
1425 | #ifdef CONFIG_HOTPLUG_CPU |
1426 | cpu_notifier_register_done(); | |
1427 | #endif | |
554f786e | 1428 | |
bd3f0221 | 1429 | kfree(buffer->buffers); |
9e01c1b7 RR |
1430 | free_cpumask_var(buffer->cpumask); |
1431 | ||
7a8e76a3 SR |
1432 | kfree(buffer); |
1433 | } | |
c4f50183 | 1434 | EXPORT_SYMBOL_GPL(ring_buffer_free); |
7a8e76a3 | 1435 | |
37886f6a SR |
1436 | void ring_buffer_set_clock(struct ring_buffer *buffer, |
1437 | u64 (*clock)(void)) | |
1438 | { | |
1439 | buffer->clock = clock; | |
1440 | } | |
1441 | ||
7a8e76a3 SR |
1442 | static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); |
1443 | ||
83f40318 VN |
1444 | static inline unsigned long rb_page_entries(struct buffer_page *bpage) |
1445 | { | |
1446 | return local_read(&bpage->entries) & RB_WRITE_MASK; | |
1447 | } | |
1448 | ||
1449 | static inline unsigned long rb_page_write(struct buffer_page *bpage) | |
1450 | { | |
1451 | return local_read(&bpage->write) & RB_WRITE_MASK; | |
1452 | } | |
1453 | ||
5040b4b7 | 1454 | static int |
83f40318 | 1455 | rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned int nr_pages) |
7a8e76a3 | 1456 | { |
83f40318 VN |
1457 | struct list_head *tail_page, *to_remove, *next_page; |
1458 | struct buffer_page *to_remove_page, *tmp_iter_page; | |
1459 | struct buffer_page *last_page, *first_page; | |
1460 | unsigned int nr_removed; | |
1461 | unsigned long head_bit; | |
1462 | int page_entries; | |
1463 | ||
1464 | head_bit = 0; | |
7a8e76a3 | 1465 | |
5389f6fa | 1466 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1467 | atomic_inc(&cpu_buffer->record_disabled); |
1468 | /* | |
1469 | * We don't race with the readers since we have acquired the reader | |
1470 | * lock. We also don't race with writers after disabling recording. | |
1471 | * This makes it easy to figure out the first and the last page to be | |
1472 | * removed from the list. We unlink all the pages in between including | |
1473 | * the first and last pages. This is done in a busy loop so that we | |
1474 | * lose the least number of traces. | |
1475 | * The pages are freed after we restart recording and unlock readers. | |
1476 | */ | |
1477 | tail_page = &cpu_buffer->tail_page->list; | |
77ae365e | 1478 | |
83f40318 VN |
1479 | /* |
1480 | * tail page might be on reader page, we remove the next page | |
1481 | * from the ring buffer | |
1482 | */ | |
1483 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
1484 | tail_page = rb_list_head(tail_page->next); | |
1485 | to_remove = tail_page; | |
1486 | ||
1487 | /* start of pages to remove */ | |
1488 | first_page = list_entry(rb_list_head(to_remove->next), | |
1489 | struct buffer_page, list); | |
1490 | ||
1491 | for (nr_removed = 0; nr_removed < nr_pages; nr_removed++) { | |
1492 | to_remove = rb_list_head(to_remove)->next; | |
1493 | head_bit |= (unsigned long)to_remove & RB_PAGE_HEAD; | |
7a8e76a3 | 1494 | } |
7a8e76a3 | 1495 | |
83f40318 | 1496 | next_page = rb_list_head(to_remove)->next; |
7a8e76a3 | 1497 | |
83f40318 VN |
1498 | /* |
1499 | * Now we remove all pages between tail_page and next_page. | |
1500 | * Make sure that we have head_bit value preserved for the | |
1501 | * next page | |
1502 | */ | |
1503 | tail_page->next = (struct list_head *)((unsigned long)next_page | | |
1504 | head_bit); | |
1505 | next_page = rb_list_head(next_page); | |
1506 | next_page->prev = tail_page; | |
1507 | ||
1508 | /* make sure pages points to a valid page in the ring buffer */ | |
1509 | cpu_buffer->pages = next_page; | |
1510 | ||
1511 | /* update head page */ | |
1512 | if (head_bit) | |
1513 | cpu_buffer->head_page = list_entry(next_page, | |
1514 | struct buffer_page, list); | |
1515 | ||
1516 | /* | |
1517 | * change read pointer to make sure any read iterators reset | |
1518 | * themselves | |
1519 | */ | |
1520 | cpu_buffer->read = 0; | |
1521 | ||
1522 | /* pages are removed, resume tracing and then free the pages */ | |
1523 | atomic_dec(&cpu_buffer->record_disabled); | |
5389f6fa | 1524 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
83f40318 VN |
1525 | |
1526 | RB_WARN_ON(cpu_buffer, list_empty(cpu_buffer->pages)); | |
1527 | ||
1528 | /* last buffer page to remove */ | |
1529 | last_page = list_entry(rb_list_head(to_remove), struct buffer_page, | |
1530 | list); | |
1531 | tmp_iter_page = first_page; | |
1532 | ||
1533 | do { | |
1534 | to_remove_page = tmp_iter_page; | |
1535 | rb_inc_page(cpu_buffer, &tmp_iter_page); | |
1536 | ||
1537 | /* update the counters */ | |
1538 | page_entries = rb_page_entries(to_remove_page); | |
1539 | if (page_entries) { | |
1540 | /* | |
1541 | * If something was added to this page, it was full | |
1542 | * since it is not the tail page. So we deduct the | |
1543 | * bytes consumed in ring buffer from here. | |
48fdc72f | 1544 | * Increment overrun to account for the lost events. |
83f40318 | 1545 | */ |
48fdc72f | 1546 | local_add(page_entries, &cpu_buffer->overrun); |
83f40318 VN |
1547 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
1548 | } | |
1549 | ||
1550 | /* | |
1551 | * We have already removed references to this list item, just | |
1552 | * free up the buffer_page and its page | |
1553 | */ | |
1554 | free_buffer_page(to_remove_page); | |
1555 | nr_removed--; | |
1556 | ||
1557 | } while (to_remove_page != last_page); | |
1558 | ||
1559 | RB_WARN_ON(cpu_buffer, nr_removed); | |
5040b4b7 VN |
1560 | |
1561 | return nr_removed == 0; | |
7a8e76a3 SR |
1562 | } |
1563 | ||
5040b4b7 VN |
1564 | static int |
1565 | rb_insert_pages(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 1566 | { |
5040b4b7 VN |
1567 | struct list_head *pages = &cpu_buffer->new_pages; |
1568 | int retries, success; | |
7a8e76a3 | 1569 | |
5389f6fa | 1570 | raw_spin_lock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1571 | /* |
1572 | * We are holding the reader lock, so the reader page won't be swapped | |
1573 | * in the ring buffer. Now we are racing with the writer trying to | |
1574 | * move head page and the tail page. | |
1575 | * We are going to adapt the reader page update process where: | |
1576 | * 1. We first splice the start and end of list of new pages between | |
1577 | * the head page and its previous page. | |
1578 | * 2. We cmpxchg the prev_page->next to point from head page to the | |
1579 | * start of new pages list. | |
1580 | * 3. Finally, we update the head->prev to the end of new list. | |
1581 | * | |
1582 | * We will try this process 10 times, to make sure that we don't keep | |
1583 | * spinning. | |
1584 | */ | |
1585 | retries = 10; | |
1586 | success = 0; | |
1587 | while (retries--) { | |
1588 | struct list_head *head_page, *prev_page, *r; | |
1589 | struct list_head *last_page, *first_page; | |
1590 | struct list_head *head_page_with_bit; | |
77ae365e | 1591 | |
5040b4b7 | 1592 | head_page = &rb_set_head_page(cpu_buffer)->list; |
54f7be5b SR |
1593 | if (!head_page) |
1594 | break; | |
5040b4b7 VN |
1595 | prev_page = head_page->prev; |
1596 | ||
1597 | first_page = pages->next; | |
1598 | last_page = pages->prev; | |
1599 | ||
1600 | head_page_with_bit = (struct list_head *) | |
1601 | ((unsigned long)head_page | RB_PAGE_HEAD); | |
1602 | ||
1603 | last_page->next = head_page_with_bit; | |
1604 | first_page->prev = prev_page; | |
1605 | ||
1606 | r = cmpxchg(&prev_page->next, head_page_with_bit, first_page); | |
1607 | ||
1608 | if (r == head_page_with_bit) { | |
1609 | /* | |
1610 | * yay, we replaced the page pointer to our new list, | |
1611 | * now, we just have to update to head page's prev | |
1612 | * pointer to point to end of list | |
1613 | */ | |
1614 | head_page->prev = last_page; | |
1615 | success = 1; | |
1616 | break; | |
1617 | } | |
7a8e76a3 | 1618 | } |
7a8e76a3 | 1619 | |
5040b4b7 VN |
1620 | if (success) |
1621 | INIT_LIST_HEAD(pages); | |
1622 | /* | |
1623 | * If we weren't successful in adding in new pages, warn and stop | |
1624 | * tracing | |
1625 | */ | |
1626 | RB_WARN_ON(cpu_buffer, !success); | |
5389f6fa | 1627 | raw_spin_unlock_irq(&cpu_buffer->reader_lock); |
5040b4b7 VN |
1628 | |
1629 | /* free pages if they weren't inserted */ | |
1630 | if (!success) { | |
1631 | struct buffer_page *bpage, *tmp; | |
1632 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, | |
1633 | list) { | |
1634 | list_del_init(&bpage->list); | |
1635 | free_buffer_page(bpage); | |
1636 | } | |
1637 | } | |
1638 | return success; | |
7a8e76a3 SR |
1639 | } |
1640 | ||
83f40318 | 1641 | static void rb_update_pages(struct ring_buffer_per_cpu *cpu_buffer) |
438ced17 | 1642 | { |
5040b4b7 VN |
1643 | int success; |
1644 | ||
438ced17 | 1645 | if (cpu_buffer->nr_pages_to_update > 0) |
5040b4b7 | 1646 | success = rb_insert_pages(cpu_buffer); |
438ced17 | 1647 | else |
5040b4b7 VN |
1648 | success = rb_remove_pages(cpu_buffer, |
1649 | -cpu_buffer->nr_pages_to_update); | |
83f40318 | 1650 | |
5040b4b7 VN |
1651 | if (success) |
1652 | cpu_buffer->nr_pages += cpu_buffer->nr_pages_to_update; | |
83f40318 VN |
1653 | } |
1654 | ||
1655 | static void update_pages_handler(struct work_struct *work) | |
1656 | { | |
1657 | struct ring_buffer_per_cpu *cpu_buffer = container_of(work, | |
1658 | struct ring_buffer_per_cpu, update_pages_work); | |
1659 | rb_update_pages(cpu_buffer); | |
05fdd70d | 1660 | complete(&cpu_buffer->update_done); |
438ced17 VN |
1661 | } |
1662 | ||
7a8e76a3 SR |
1663 | /** |
1664 | * ring_buffer_resize - resize the ring buffer | |
1665 | * @buffer: the buffer to resize. | |
1666 | * @size: the new size. | |
d611851b | 1667 | * @cpu_id: the cpu buffer to resize |
7a8e76a3 | 1668 | * |
7a8e76a3 SR |
1669 | * Minimum size is 2 * BUF_PAGE_SIZE. |
1670 | * | |
83f40318 | 1671 | * Returns 0 on success and < 0 on failure. |
7a8e76a3 | 1672 | */ |
438ced17 VN |
1673 | int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size, |
1674 | int cpu_id) | |
7a8e76a3 SR |
1675 | { |
1676 | struct ring_buffer_per_cpu *cpu_buffer; | |
438ced17 | 1677 | unsigned nr_pages; |
83f40318 | 1678 | int cpu, err = 0; |
7a8e76a3 | 1679 | |
ee51a1de IM |
1680 | /* |
1681 | * Always succeed at resizing a non-existent buffer: | |
1682 | */ | |
1683 | if (!buffer) | |
1684 | return size; | |
1685 | ||
6a31e1f1 SR |
1686 | /* Make sure the requested buffer exists */ |
1687 | if (cpu_id != RING_BUFFER_ALL_CPUS && | |
1688 | !cpumask_test_cpu(cpu_id, buffer->cpumask)) | |
1689 | return size; | |
1690 | ||
7a8e76a3 SR |
1691 | size = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
1692 | size *= BUF_PAGE_SIZE; | |
7a8e76a3 SR |
1693 | |
1694 | /* we need a minimum of two pages */ | |
1695 | if (size < BUF_PAGE_SIZE * 2) | |
1696 | size = BUF_PAGE_SIZE * 2; | |
1697 | ||
83f40318 | 1698 | nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); |
18421015 | 1699 | |
83f40318 VN |
1700 | /* |
1701 | * Don't succeed if resizing is disabled, as a reader might be | |
1702 | * manipulating the ring buffer and is expecting a sane state while | |
1703 | * this is true. | |
1704 | */ | |
1705 | if (atomic_read(&buffer->resize_disabled)) | |
1706 | return -EBUSY; | |
18421015 | 1707 | |
83f40318 | 1708 | /* prevent another thread from changing buffer sizes */ |
7a8e76a3 | 1709 | mutex_lock(&buffer->mutex); |
7a8e76a3 | 1710 | |
438ced17 VN |
1711 | if (cpu_id == RING_BUFFER_ALL_CPUS) { |
1712 | /* calculate the pages to update */ | |
7a8e76a3 SR |
1713 | for_each_buffer_cpu(buffer, cpu) { |
1714 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 | 1715 | |
438ced17 VN |
1716 | cpu_buffer->nr_pages_to_update = nr_pages - |
1717 | cpu_buffer->nr_pages; | |
438ced17 VN |
1718 | /* |
1719 | * nothing more to do for removing pages or no update | |
1720 | */ | |
1721 | if (cpu_buffer->nr_pages_to_update <= 0) | |
1722 | continue; | |
d7ec4bfe | 1723 | /* |
438ced17 VN |
1724 | * to add pages, make sure all new pages can be |
1725 | * allocated without receiving ENOMEM | |
d7ec4bfe | 1726 | */ |
438ced17 VN |
1727 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
1728 | if (__rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 | 1729 | &cpu_buffer->new_pages, cpu)) { |
438ced17 | 1730 | /* not enough memory for new pages */ |
83f40318 VN |
1731 | err = -ENOMEM; |
1732 | goto out_err; | |
1733 | } | |
1734 | } | |
1735 | ||
1736 | get_online_cpus(); | |
1737 | /* | |
1738 | * Fire off all the required work handlers | |
05fdd70d | 1739 | * We can't schedule on offline CPUs, but it's not necessary |
83f40318 VN |
1740 | * since we can change their buffer sizes without any race. |
1741 | */ | |
1742 | for_each_buffer_cpu(buffer, cpu) { | |
1743 | cpu_buffer = buffer->buffers[cpu]; | |
05fdd70d | 1744 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
1745 | continue; |
1746 | ||
021c5b34 CM |
1747 | /* Can't run something on an offline CPU. */ |
1748 | if (!cpu_online(cpu)) { | |
f5eb5588 SRRH |
1749 | rb_update_pages(cpu_buffer); |
1750 | cpu_buffer->nr_pages_to_update = 0; | |
1751 | } else { | |
05fdd70d VN |
1752 | schedule_work_on(cpu, |
1753 | &cpu_buffer->update_pages_work); | |
f5eb5588 | 1754 | } |
7a8e76a3 | 1755 | } |
7a8e76a3 | 1756 | |
438ced17 VN |
1757 | /* wait for all the updates to complete */ |
1758 | for_each_buffer_cpu(buffer, cpu) { | |
1759 | cpu_buffer = buffer->buffers[cpu]; | |
05fdd70d | 1760 | if (!cpu_buffer->nr_pages_to_update) |
83f40318 VN |
1761 | continue; |
1762 | ||
05fdd70d VN |
1763 | if (cpu_online(cpu)) |
1764 | wait_for_completion(&cpu_buffer->update_done); | |
83f40318 | 1765 | cpu_buffer->nr_pages_to_update = 0; |
438ced17 | 1766 | } |
83f40318 VN |
1767 | |
1768 | put_online_cpus(); | |
438ced17 | 1769 | } else { |
8e49f418 VN |
1770 | /* Make sure this CPU has been intitialized */ |
1771 | if (!cpumask_test_cpu(cpu_id, buffer->cpumask)) | |
1772 | goto out; | |
1773 | ||
438ced17 | 1774 | cpu_buffer = buffer->buffers[cpu_id]; |
83f40318 | 1775 | |
438ced17 VN |
1776 | if (nr_pages == cpu_buffer->nr_pages) |
1777 | goto out; | |
7a8e76a3 | 1778 | |
438ced17 VN |
1779 | cpu_buffer->nr_pages_to_update = nr_pages - |
1780 | cpu_buffer->nr_pages; | |
1781 | ||
1782 | INIT_LIST_HEAD(&cpu_buffer->new_pages); | |
1783 | if (cpu_buffer->nr_pages_to_update > 0 && | |
1784 | __rb_allocate_pages(cpu_buffer->nr_pages_to_update, | |
83f40318 VN |
1785 | &cpu_buffer->new_pages, cpu_id)) { |
1786 | err = -ENOMEM; | |
1787 | goto out_err; | |
1788 | } | |
438ced17 | 1789 | |
83f40318 VN |
1790 | get_online_cpus(); |
1791 | ||
021c5b34 CM |
1792 | /* Can't run something on an offline CPU. */ |
1793 | if (!cpu_online(cpu_id)) | |
f5eb5588 SRRH |
1794 | rb_update_pages(cpu_buffer); |
1795 | else { | |
83f40318 VN |
1796 | schedule_work_on(cpu_id, |
1797 | &cpu_buffer->update_pages_work); | |
05fdd70d | 1798 | wait_for_completion(&cpu_buffer->update_done); |
f5eb5588 | 1799 | } |
83f40318 | 1800 | |
83f40318 | 1801 | cpu_buffer->nr_pages_to_update = 0; |
05fdd70d | 1802 | put_online_cpus(); |
438ced17 | 1803 | } |
7a8e76a3 SR |
1804 | |
1805 | out: | |
659f451f SR |
1806 | /* |
1807 | * The ring buffer resize can happen with the ring buffer | |
1808 | * enabled, so that the update disturbs the tracing as little | |
1809 | * as possible. But if the buffer is disabled, we do not need | |
1810 | * to worry about that, and we can take the time to verify | |
1811 | * that the buffer is not corrupt. | |
1812 | */ | |
1813 | if (atomic_read(&buffer->record_disabled)) { | |
1814 | atomic_inc(&buffer->record_disabled); | |
1815 | /* | |
1816 | * Even though the buffer was disabled, we must make sure | |
1817 | * that it is truly disabled before calling rb_check_pages. | |
1818 | * There could have been a race between checking | |
1819 | * record_disable and incrementing it. | |
1820 | */ | |
1821 | synchronize_sched(); | |
1822 | for_each_buffer_cpu(buffer, cpu) { | |
1823 | cpu_buffer = buffer->buffers[cpu]; | |
1824 | rb_check_pages(cpu_buffer); | |
1825 | } | |
1826 | atomic_dec(&buffer->record_disabled); | |
1827 | } | |
1828 | ||
7a8e76a3 | 1829 | mutex_unlock(&buffer->mutex); |
7a8e76a3 SR |
1830 | return size; |
1831 | ||
83f40318 | 1832 | out_err: |
438ced17 VN |
1833 | for_each_buffer_cpu(buffer, cpu) { |
1834 | struct buffer_page *bpage, *tmp; | |
83f40318 | 1835 | |
438ced17 | 1836 | cpu_buffer = buffer->buffers[cpu]; |
438ced17 | 1837 | cpu_buffer->nr_pages_to_update = 0; |
83f40318 | 1838 | |
438ced17 VN |
1839 | if (list_empty(&cpu_buffer->new_pages)) |
1840 | continue; | |
83f40318 | 1841 | |
438ced17 VN |
1842 | list_for_each_entry_safe(bpage, tmp, &cpu_buffer->new_pages, |
1843 | list) { | |
1844 | list_del_init(&bpage->list); | |
1845 | free_buffer_page(bpage); | |
1846 | } | |
7a8e76a3 | 1847 | } |
641d2f63 | 1848 | mutex_unlock(&buffer->mutex); |
83f40318 | 1849 | return err; |
7a8e76a3 | 1850 | } |
c4f50183 | 1851 | EXPORT_SYMBOL_GPL(ring_buffer_resize); |
7a8e76a3 | 1852 | |
750912fa DS |
1853 | void ring_buffer_change_overwrite(struct ring_buffer *buffer, int val) |
1854 | { | |
1855 | mutex_lock(&buffer->mutex); | |
1856 | if (val) | |
1857 | buffer->flags |= RB_FL_OVERWRITE; | |
1858 | else | |
1859 | buffer->flags &= ~RB_FL_OVERWRITE; | |
1860 | mutex_unlock(&buffer->mutex); | |
1861 | } | |
1862 | EXPORT_SYMBOL_GPL(ring_buffer_change_overwrite); | |
1863 | ||
8789a9e7 | 1864 | static inline void * |
044fa782 | 1865 | __rb_data_page_index(struct buffer_data_page *bpage, unsigned index) |
8789a9e7 | 1866 | { |
044fa782 | 1867 | return bpage->data + index; |
8789a9e7 SR |
1868 | } |
1869 | ||
044fa782 | 1870 | static inline void *__rb_page_index(struct buffer_page *bpage, unsigned index) |
7a8e76a3 | 1871 | { |
044fa782 | 1872 | return bpage->page->data + index; |
7a8e76a3 SR |
1873 | } |
1874 | ||
1875 | static inline struct ring_buffer_event * | |
d769041f | 1876 | rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1877 | { |
6f807acd SR |
1878 | return __rb_page_index(cpu_buffer->reader_page, |
1879 | cpu_buffer->reader_page->read); | |
1880 | } | |
1881 | ||
7a8e76a3 SR |
1882 | static inline struct ring_buffer_event * |
1883 | rb_iter_head_event(struct ring_buffer_iter *iter) | |
1884 | { | |
6f807acd | 1885 | return __rb_page_index(iter->head_page, iter->head); |
7a8e76a3 SR |
1886 | } |
1887 | ||
bf41a158 SR |
1888 | static inline unsigned rb_page_commit(struct buffer_page *bpage) |
1889 | { | |
abc9b56d | 1890 | return local_read(&bpage->page->commit); |
bf41a158 SR |
1891 | } |
1892 | ||
25985edc | 1893 | /* Size is determined by what has been committed */ |
bf41a158 SR |
1894 | static inline unsigned rb_page_size(struct buffer_page *bpage) |
1895 | { | |
1896 | return rb_page_commit(bpage); | |
1897 | } | |
1898 | ||
1899 | static inline unsigned | |
1900 | rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer) | |
1901 | { | |
1902 | return rb_page_commit(cpu_buffer->commit_page); | |
1903 | } | |
1904 | ||
bf41a158 SR |
1905 | static inline unsigned |
1906 | rb_event_index(struct ring_buffer_event *event) | |
1907 | { | |
1908 | unsigned long addr = (unsigned long)event; | |
1909 | ||
22f470f8 | 1910 | return (addr & ~PAGE_MASK) - BUF_PAGE_HDR_SIZE; |
bf41a158 SR |
1911 | } |
1912 | ||
0f0c85fc | 1913 | static inline int |
fa743953 SR |
1914 | rb_event_is_commit(struct ring_buffer_per_cpu *cpu_buffer, |
1915 | struct ring_buffer_event *event) | |
bf41a158 SR |
1916 | { |
1917 | unsigned long addr = (unsigned long)event; | |
1918 | unsigned long index; | |
1919 | ||
1920 | index = rb_event_index(event); | |
1921 | addr &= PAGE_MASK; | |
1922 | ||
1923 | return cpu_buffer->commit_page->page == (void *)addr && | |
1924 | rb_commit_index(cpu_buffer) == index; | |
1925 | } | |
1926 | ||
34a148bf | 1927 | static void |
bf41a158 | 1928 | rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1929 | { |
77ae365e SR |
1930 | unsigned long max_count; |
1931 | ||
bf41a158 SR |
1932 | /* |
1933 | * We only race with interrupts and NMIs on this CPU. | |
1934 | * If we own the commit event, then we can commit | |
1935 | * all others that interrupted us, since the interruptions | |
1936 | * are in stack format (they finish before they come | |
1937 | * back to us). This allows us to do a simple loop to | |
1938 | * assign the commit to the tail. | |
1939 | */ | |
a8ccf1d6 | 1940 | again: |
438ced17 | 1941 | max_count = cpu_buffer->nr_pages * 100; |
77ae365e | 1942 | |
bf41a158 | 1943 | while (cpu_buffer->commit_page != cpu_buffer->tail_page) { |
77ae365e SR |
1944 | if (RB_WARN_ON(cpu_buffer, !(--max_count))) |
1945 | return; | |
1946 | if (RB_WARN_ON(cpu_buffer, | |
1947 | rb_is_reader_page(cpu_buffer->tail_page))) | |
1948 | return; | |
1949 | local_set(&cpu_buffer->commit_page->page->commit, | |
1950 | rb_page_write(cpu_buffer->commit_page)); | |
bf41a158 | 1951 | rb_inc_page(cpu_buffer, &cpu_buffer->commit_page); |
abc9b56d SR |
1952 | cpu_buffer->write_stamp = |
1953 | cpu_buffer->commit_page->page->time_stamp; | |
bf41a158 SR |
1954 | /* add barrier to keep gcc from optimizing too much */ |
1955 | barrier(); | |
1956 | } | |
1957 | while (rb_commit_index(cpu_buffer) != | |
1958 | rb_page_write(cpu_buffer->commit_page)) { | |
77ae365e SR |
1959 | |
1960 | local_set(&cpu_buffer->commit_page->page->commit, | |
1961 | rb_page_write(cpu_buffer->commit_page)); | |
1962 | RB_WARN_ON(cpu_buffer, | |
1963 | local_read(&cpu_buffer->commit_page->page->commit) & | |
1964 | ~RB_WRITE_MASK); | |
bf41a158 SR |
1965 | barrier(); |
1966 | } | |
a8ccf1d6 SR |
1967 | |
1968 | /* again, keep gcc from optimizing */ | |
1969 | barrier(); | |
1970 | ||
1971 | /* | |
1972 | * If an interrupt came in just after the first while loop | |
1973 | * and pushed the tail page forward, we will be left with | |
1974 | * a dangling commit that will never go forward. | |
1975 | */ | |
1976 | if (unlikely(cpu_buffer->commit_page != cpu_buffer->tail_page)) | |
1977 | goto again; | |
7a8e76a3 SR |
1978 | } |
1979 | ||
d769041f | 1980 | static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer) |
7a8e76a3 | 1981 | { |
abc9b56d | 1982 | cpu_buffer->read_stamp = cpu_buffer->reader_page->page->time_stamp; |
6f807acd | 1983 | cpu_buffer->reader_page->read = 0; |
d769041f SR |
1984 | } |
1985 | ||
34a148bf | 1986 | static void rb_inc_iter(struct ring_buffer_iter *iter) |
d769041f SR |
1987 | { |
1988 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
1989 | ||
1990 | /* | |
1991 | * The iterator could be on the reader page (it starts there). | |
1992 | * But the head could have moved, since the reader was | |
1993 | * found. Check for this case and assign the iterator | |
1994 | * to the head page instead of next. | |
1995 | */ | |
1996 | if (iter->head_page == cpu_buffer->reader_page) | |
77ae365e | 1997 | iter->head_page = rb_set_head_page(cpu_buffer); |
d769041f SR |
1998 | else |
1999 | rb_inc_page(cpu_buffer, &iter->head_page); | |
2000 | ||
abc9b56d | 2001 | iter->read_stamp = iter->head_page->page->time_stamp; |
7a8e76a3 SR |
2002 | iter->head = 0; |
2003 | } | |
2004 | ||
69d1b839 SR |
2005 | /* Slow path, do not inline */ |
2006 | static noinline struct ring_buffer_event * | |
2007 | rb_add_time_stamp(struct ring_buffer_event *event, u64 delta) | |
2008 | { | |
2009 | event->type_len = RINGBUF_TYPE_TIME_EXTEND; | |
2010 | ||
2011 | /* Not the first event on the page? */ | |
2012 | if (rb_event_index(event)) { | |
2013 | event->time_delta = delta & TS_MASK; | |
2014 | event->array[0] = delta >> TS_SHIFT; | |
2015 | } else { | |
2016 | /* nope, just zero it */ | |
2017 | event->time_delta = 0; | |
2018 | event->array[0] = 0; | |
2019 | } | |
2020 | ||
2021 | return skip_time_extend(event); | |
2022 | } | |
2023 | ||
7a8e76a3 | 2024 | /** |
01e3e710 | 2025 | * rb_update_event - update event type and data |
021de3d9 | 2026 | * @event: the event to update |
7a8e76a3 SR |
2027 | * @type: the type of event |
2028 | * @length: the size of the event field in the ring buffer | |
2029 | * | |
2030 | * Update the type and data fields of the event. The length | |
2031 | * is the actual size that is written to the ring buffer, | |
2032 | * and with this, we can determine what to place into the | |
2033 | * data field. | |
2034 | */ | |
34a148bf | 2035 | static void |
69d1b839 SR |
2036 | rb_update_event(struct ring_buffer_per_cpu *cpu_buffer, |
2037 | struct ring_buffer_event *event, unsigned length, | |
2038 | int add_timestamp, u64 delta) | |
7a8e76a3 | 2039 | { |
69d1b839 SR |
2040 | /* Only a commit updates the timestamp */ |
2041 | if (unlikely(!rb_event_is_commit(cpu_buffer, event))) | |
2042 | delta = 0; | |
7a8e76a3 | 2043 | |
69d1b839 SR |
2044 | /* |
2045 | * If we need to add a timestamp, then we | |
2046 | * add it to the start of the resevered space. | |
2047 | */ | |
2048 | if (unlikely(add_timestamp)) { | |
2049 | event = rb_add_time_stamp(event, delta); | |
2050 | length -= RB_LEN_TIME_EXTEND; | |
2051 | delta = 0; | |
7a8e76a3 | 2052 | } |
69d1b839 SR |
2053 | |
2054 | event->time_delta = delta; | |
2055 | length -= RB_EVNT_HDR_SIZE; | |
2056 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) { | |
2057 | event->type_len = 0; | |
2058 | event->array[0] = length; | |
2059 | } else | |
2060 | event->type_len = DIV_ROUND_UP(length, RB_ALIGNMENT); | |
7a8e76a3 SR |
2061 | } |
2062 | ||
77ae365e SR |
2063 | /* |
2064 | * rb_handle_head_page - writer hit the head page | |
2065 | * | |
2066 | * Returns: +1 to retry page | |
2067 | * 0 to continue | |
2068 | * -1 on error | |
2069 | */ | |
2070 | static int | |
2071 | rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer, | |
2072 | struct buffer_page *tail_page, | |
2073 | struct buffer_page *next_page) | |
2074 | { | |
2075 | struct buffer_page *new_head; | |
2076 | int entries; | |
2077 | int type; | |
2078 | int ret; | |
2079 | ||
2080 | entries = rb_page_entries(next_page); | |
2081 | ||
2082 | /* | |
2083 | * The hard part is here. We need to move the head | |
2084 | * forward, and protect against both readers on | |
2085 | * other CPUs and writers coming in via interrupts. | |
2086 | */ | |
2087 | type = rb_head_page_set_update(cpu_buffer, next_page, tail_page, | |
2088 | RB_PAGE_HEAD); | |
2089 | ||
2090 | /* | |
2091 | * type can be one of four: | |
2092 | * NORMAL - an interrupt already moved it for us | |
2093 | * HEAD - we are the first to get here. | |
2094 | * UPDATE - we are the interrupt interrupting | |
2095 | * a current move. | |
2096 | * MOVED - a reader on another CPU moved the next | |
2097 | * pointer to its reader page. Give up | |
2098 | * and try again. | |
2099 | */ | |
2100 | ||
2101 | switch (type) { | |
2102 | case RB_PAGE_HEAD: | |
2103 | /* | |
2104 | * We changed the head to UPDATE, thus | |
2105 | * it is our responsibility to update | |
2106 | * the counters. | |
2107 | */ | |
2108 | local_add(entries, &cpu_buffer->overrun); | |
c64e148a | 2109 | local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); |
77ae365e SR |
2110 | |
2111 | /* | |
2112 | * The entries will be zeroed out when we move the | |
2113 | * tail page. | |
2114 | */ | |
2115 | ||
2116 | /* still more to do */ | |
2117 | break; | |
2118 | ||
2119 | case RB_PAGE_UPDATE: | |
2120 | /* | |
2121 | * This is an interrupt that interrupt the | |
2122 | * previous update. Still more to do. | |
2123 | */ | |
2124 | break; | |
2125 | case RB_PAGE_NORMAL: | |
2126 | /* | |
2127 | * An interrupt came in before the update | |
2128 | * and processed this for us. | |
2129 | * Nothing left to do. | |
2130 | */ | |
2131 | return 1; | |
2132 | case RB_PAGE_MOVED: | |
2133 | /* | |
2134 | * The reader is on another CPU and just did | |
2135 | * a swap with our next_page. | |
2136 | * Try again. | |
2137 | */ | |
2138 | return 1; | |
2139 | default: | |
2140 | RB_WARN_ON(cpu_buffer, 1); /* WTF??? */ | |
2141 | return -1; | |
2142 | } | |
2143 | ||
2144 | /* | |
2145 | * Now that we are here, the old head pointer is | |
2146 | * set to UPDATE. This will keep the reader from | |
2147 | * swapping the head page with the reader page. | |
2148 | * The reader (on another CPU) will spin till | |
2149 | * we are finished. | |
2150 | * | |
2151 | * We just need to protect against interrupts | |
2152 | * doing the job. We will set the next pointer | |
2153 | * to HEAD. After that, we set the old pointer | |
2154 | * to NORMAL, but only if it was HEAD before. | |
2155 | * otherwise we are an interrupt, and only | |
2156 | * want the outer most commit to reset it. | |
2157 | */ | |
2158 | new_head = next_page; | |
2159 | rb_inc_page(cpu_buffer, &new_head); | |
2160 | ||
2161 | ret = rb_head_page_set_head(cpu_buffer, new_head, next_page, | |
2162 | RB_PAGE_NORMAL); | |
2163 | ||
2164 | /* | |
2165 | * Valid returns are: | |
2166 | * HEAD - an interrupt came in and already set it. | |
2167 | * NORMAL - One of two things: | |
2168 | * 1) We really set it. | |
2169 | * 2) A bunch of interrupts came in and moved | |
2170 | * the page forward again. | |
2171 | */ | |
2172 | switch (ret) { | |
2173 | case RB_PAGE_HEAD: | |
2174 | case RB_PAGE_NORMAL: | |
2175 | /* OK */ | |
2176 | break; | |
2177 | default: | |
2178 | RB_WARN_ON(cpu_buffer, 1); | |
2179 | return -1; | |
2180 | } | |
2181 | ||
2182 | /* | |
2183 | * It is possible that an interrupt came in, | |
2184 | * set the head up, then more interrupts came in | |
2185 | * and moved it again. When we get back here, | |
2186 | * the page would have been set to NORMAL but we | |
2187 | * just set it back to HEAD. | |
2188 | * | |
2189 | * How do you detect this? Well, if that happened | |
2190 | * the tail page would have moved. | |
2191 | */ | |
2192 | if (ret == RB_PAGE_NORMAL) { | |
2193 | /* | |
2194 | * If the tail had moved passed next, then we need | |
2195 | * to reset the pointer. | |
2196 | */ | |
2197 | if (cpu_buffer->tail_page != tail_page && | |
2198 | cpu_buffer->tail_page != next_page) | |
2199 | rb_head_page_set_normal(cpu_buffer, new_head, | |
2200 | next_page, | |
2201 | RB_PAGE_HEAD); | |
2202 | } | |
2203 | ||
2204 | /* | |
2205 | * If this was the outer most commit (the one that | |
2206 | * changed the original pointer from HEAD to UPDATE), | |
2207 | * then it is up to us to reset it to NORMAL. | |
2208 | */ | |
2209 | if (type == RB_PAGE_HEAD) { | |
2210 | ret = rb_head_page_set_normal(cpu_buffer, next_page, | |
2211 | tail_page, | |
2212 | RB_PAGE_UPDATE); | |
2213 | if (RB_WARN_ON(cpu_buffer, | |
2214 | ret != RB_PAGE_UPDATE)) | |
2215 | return -1; | |
2216 | } | |
2217 | ||
2218 | return 0; | |
2219 | } | |
2220 | ||
34a148bf | 2221 | static unsigned rb_calculate_event_length(unsigned length) |
7a8e76a3 SR |
2222 | { |
2223 | struct ring_buffer_event event; /* Used only for sizeof array */ | |
2224 | ||
2225 | /* zero length can cause confusions */ | |
2226 | if (!length) | |
2227 | length = 1; | |
2228 | ||
2271048d | 2229 | if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) |
7a8e76a3 SR |
2230 | length += sizeof(event.array[0]); |
2231 | ||
2232 | length += RB_EVNT_HDR_SIZE; | |
2271048d | 2233 | length = ALIGN(length, RB_ARCH_ALIGNMENT); |
7a8e76a3 SR |
2234 | |
2235 | return length; | |
2236 | } | |
2237 | ||
c7b09308 SR |
2238 | static inline void |
2239 | rb_reset_tail(struct ring_buffer_per_cpu *cpu_buffer, | |
2240 | struct buffer_page *tail_page, | |
2241 | unsigned long tail, unsigned long length) | |
2242 | { | |
2243 | struct ring_buffer_event *event; | |
2244 | ||
2245 | /* | |
2246 | * Only the event that crossed the page boundary | |
2247 | * must fill the old tail_page with padding. | |
2248 | */ | |
2249 | if (tail >= BUF_PAGE_SIZE) { | |
b3230c8b SR |
2250 | /* |
2251 | * If the page was filled, then we still need | |
2252 | * to update the real_end. Reset it to zero | |
2253 | * and the reader will ignore it. | |
2254 | */ | |
2255 | if (tail == BUF_PAGE_SIZE) | |
2256 | tail_page->real_end = 0; | |
2257 | ||
c7b09308 SR |
2258 | local_sub(length, &tail_page->write); |
2259 | return; | |
2260 | } | |
2261 | ||
2262 | event = __rb_page_index(tail_page, tail); | |
b0b7065b | 2263 | kmemcheck_annotate_bitfield(event, bitfield); |
c7b09308 | 2264 | |
c64e148a VN |
2265 | /* account for padding bytes */ |
2266 | local_add(BUF_PAGE_SIZE - tail, &cpu_buffer->entries_bytes); | |
2267 | ||
ff0ff84a SR |
2268 | /* |
2269 | * Save the original length to the meta data. | |
2270 | * This will be used by the reader to add lost event | |
2271 | * counter. | |
2272 | */ | |
2273 | tail_page->real_end = tail; | |
2274 | ||
c7b09308 SR |
2275 | /* |
2276 | * If this event is bigger than the minimum size, then | |
2277 | * we need to be careful that we don't subtract the | |
2278 | * write counter enough to allow another writer to slip | |
2279 | * in on this page. | |
2280 | * We put in a discarded commit instead, to make sure | |
2281 | * that this space is not used again. | |
2282 | * | |
2283 | * If we are less than the minimum size, we don't need to | |
2284 | * worry about it. | |
2285 | */ | |
2286 | if (tail > (BUF_PAGE_SIZE - RB_EVNT_MIN_SIZE)) { | |
2287 | /* No room for any events */ | |
2288 | ||
2289 | /* Mark the rest of the page with padding */ | |
2290 | rb_event_set_padding(event); | |
2291 | ||
2292 | /* Set the write back to the previous setting */ | |
2293 | local_sub(length, &tail_page->write); | |
2294 | return; | |
2295 | } | |
2296 | ||
2297 | /* Put in a discarded event */ | |
2298 | event->array[0] = (BUF_PAGE_SIZE - tail) - RB_EVNT_HDR_SIZE; | |
2299 | event->type_len = RINGBUF_TYPE_PADDING; | |
2300 | /* time delta must be non zero */ | |
2301 | event->time_delta = 1; | |
c7b09308 SR |
2302 | |
2303 | /* Set write to end of buffer */ | |
2304 | length = (tail + length) - BUF_PAGE_SIZE; | |
2305 | local_sub(length, &tail_page->write); | |
2306 | } | |
6634ff26 | 2307 | |
747e94ae SR |
2308 | /* |
2309 | * This is the slow path, force gcc not to inline it. | |
2310 | */ | |
2311 | static noinline struct ring_buffer_event * | |
6634ff26 SR |
2312 | rb_move_tail(struct ring_buffer_per_cpu *cpu_buffer, |
2313 | unsigned long length, unsigned long tail, | |
e8bc43e8 | 2314 | struct buffer_page *tail_page, u64 ts) |
7a8e76a3 | 2315 | { |
5a50e33c | 2316 | struct buffer_page *commit_page = cpu_buffer->commit_page; |
7a8e76a3 | 2317 | struct ring_buffer *buffer = cpu_buffer->buffer; |
77ae365e SR |
2318 | struct buffer_page *next_page; |
2319 | int ret; | |
aa20ae84 SR |
2320 | |
2321 | next_page = tail_page; | |
2322 | ||
aa20ae84 SR |
2323 | rb_inc_page(cpu_buffer, &next_page); |
2324 | ||
aa20ae84 SR |
2325 | /* |
2326 | * If for some reason, we had an interrupt storm that made | |
2327 | * it all the way around the buffer, bail, and warn | |
2328 | * about it. | |
2329 | */ | |
2330 | if (unlikely(next_page == commit_page)) { | |
77ae365e | 2331 | local_inc(&cpu_buffer->commit_overrun); |
aa20ae84 SR |
2332 | goto out_reset; |
2333 | } | |
2334 | ||
77ae365e SR |
2335 | /* |
2336 | * This is where the fun begins! | |
2337 | * | |
2338 | * We are fighting against races between a reader that | |
2339 | * could be on another CPU trying to swap its reader | |
2340 | * page with the buffer head. | |
2341 | * | |
2342 | * We are also fighting against interrupts coming in and | |
2343 | * moving the head or tail on us as well. | |
2344 | * | |
2345 | * If the next page is the head page then we have filled | |
2346 | * the buffer, unless the commit page is still on the | |
2347 | * reader page. | |
2348 | */ | |
2349 | if (rb_is_head_page(cpu_buffer, next_page, &tail_page->list)) { | |
aa20ae84 | 2350 | |
77ae365e SR |
2351 | /* |
2352 | * If the commit is not on the reader page, then | |
2353 | * move the header page. | |
2354 | */ | |
2355 | if (!rb_is_reader_page(cpu_buffer->commit_page)) { | |
2356 | /* | |
2357 | * If we are not in overwrite mode, | |
2358 | * this is easy, just stop here. | |
2359 | */ | |
884bfe89 SP |
2360 | if (!(buffer->flags & RB_FL_OVERWRITE)) { |
2361 | local_inc(&cpu_buffer->dropped_events); | |
77ae365e | 2362 | goto out_reset; |
884bfe89 | 2363 | } |
77ae365e SR |
2364 | |
2365 | ret = rb_handle_head_page(cpu_buffer, | |
2366 | tail_page, | |
2367 | next_page); | |
2368 | if (ret < 0) | |
2369 | goto out_reset; | |
2370 | if (ret) | |
2371 | goto out_again; | |
2372 | } else { | |
2373 | /* | |
2374 | * We need to be careful here too. The | |
2375 | * commit page could still be on the reader | |
2376 | * page. We could have a small buffer, and | |
2377 | * have filled up the buffer with events | |
2378 | * from interrupts and such, and wrapped. | |
2379 | * | |
2380 | * Note, if the tail page is also the on the | |
2381 | * reader_page, we let it move out. | |
2382 | */ | |
2383 | if (unlikely((cpu_buffer->commit_page != | |
2384 | cpu_buffer->tail_page) && | |
2385 | (cpu_buffer->commit_page == | |
2386 | cpu_buffer->reader_page))) { | |
2387 | local_inc(&cpu_buffer->commit_overrun); | |
2388 | goto out_reset; | |
2389 | } | |
aa20ae84 SR |
2390 | } |
2391 | } | |
2392 | ||
77ae365e SR |
2393 | ret = rb_tail_page_update(cpu_buffer, tail_page, next_page); |
2394 | if (ret) { | |
2395 | /* | |
2396 | * Nested commits always have zero deltas, so | |
2397 | * just reread the time stamp | |
2398 | */ | |
e8bc43e8 SR |
2399 | ts = rb_time_stamp(buffer); |
2400 | next_page->page->time_stamp = ts; | |
aa20ae84 SR |
2401 | } |
2402 | ||
77ae365e | 2403 | out_again: |
aa20ae84 | 2404 | |
77ae365e | 2405 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
aa20ae84 SR |
2406 | |
2407 | /* fail and let the caller try again */ | |
2408 | return ERR_PTR(-EAGAIN); | |
2409 | ||
45141d46 | 2410 | out_reset: |
6f3b3440 | 2411 | /* reset write */ |
c7b09308 | 2412 | rb_reset_tail(cpu_buffer, tail_page, tail, length); |
6f3b3440 | 2413 | |
bf41a158 | 2414 | return NULL; |
7a8e76a3 SR |
2415 | } |
2416 | ||
6634ff26 SR |
2417 | static struct ring_buffer_event * |
2418 | __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer, | |
69d1b839 SR |
2419 | unsigned long length, u64 ts, |
2420 | u64 delta, int add_timestamp) | |
6634ff26 | 2421 | { |
5a50e33c | 2422 | struct buffer_page *tail_page; |
6634ff26 SR |
2423 | struct ring_buffer_event *event; |
2424 | unsigned long tail, write; | |
2425 | ||
69d1b839 SR |
2426 | /* |
2427 | * If the time delta since the last event is too big to | |
2428 | * hold in the time field of the event, then we append a | |
2429 | * TIME EXTEND event ahead of the data event. | |
2430 | */ | |
2431 | if (unlikely(add_timestamp)) | |
2432 | length += RB_LEN_TIME_EXTEND; | |
2433 | ||
6634ff26 SR |
2434 | tail_page = cpu_buffer->tail_page; |
2435 | write = local_add_return(length, &tail_page->write); | |
77ae365e SR |
2436 | |
2437 | /* set write to only the index of the write */ | |
2438 | write &= RB_WRITE_MASK; | |
6634ff26 SR |
2439 | tail = write - length; |
2440 | ||
d651aa1d SRRH |
2441 | /* |
2442 | * If this is the first commit on the page, then it has the same | |
2443 | * timestamp as the page itself. | |
2444 | */ | |
2445 | if (!tail) | |
2446 | delta = 0; | |
2447 | ||
6634ff26 | 2448 | /* See if we shot pass the end of this buffer page */ |
747e94ae | 2449 | if (unlikely(write > BUF_PAGE_SIZE)) |
6634ff26 | 2450 | return rb_move_tail(cpu_buffer, length, tail, |
5a50e33c | 2451 | tail_page, ts); |
6634ff26 SR |
2452 | |
2453 | /* We reserved something on the buffer */ | |
2454 | ||
6634ff26 | 2455 | event = __rb_page_index(tail_page, tail); |
1744a21d | 2456 | kmemcheck_annotate_bitfield(event, bitfield); |
69d1b839 | 2457 | rb_update_event(cpu_buffer, event, length, add_timestamp, delta); |
6634ff26 | 2458 | |
69d1b839 | 2459 | local_inc(&tail_page->entries); |
6634ff26 SR |
2460 | |
2461 | /* | |
fa743953 SR |
2462 | * If this is the first commit on the page, then update |
2463 | * its timestamp. | |
6634ff26 | 2464 | */ |
fa743953 | 2465 | if (!tail) |
e8bc43e8 | 2466 | tail_page->page->time_stamp = ts; |
6634ff26 | 2467 | |
c64e148a VN |
2468 | /* account for these added bytes */ |
2469 | local_add(length, &cpu_buffer->entries_bytes); | |
2470 | ||
6634ff26 SR |
2471 | return event; |
2472 | } | |
2473 | ||
edd813bf SR |
2474 | static inline int |
2475 | rb_try_to_discard(struct ring_buffer_per_cpu *cpu_buffer, | |
2476 | struct ring_buffer_event *event) | |
2477 | { | |
2478 | unsigned long new_index, old_index; | |
2479 | struct buffer_page *bpage; | |
2480 | unsigned long index; | |
2481 | unsigned long addr; | |
2482 | ||
2483 | new_index = rb_event_index(event); | |
69d1b839 | 2484 | old_index = new_index + rb_event_ts_length(event); |
edd813bf SR |
2485 | addr = (unsigned long)event; |
2486 | addr &= PAGE_MASK; | |
2487 | ||
2488 | bpage = cpu_buffer->tail_page; | |
2489 | ||
2490 | if (bpage->page == (void *)addr && rb_page_write(bpage) == old_index) { | |
77ae365e SR |
2491 | unsigned long write_mask = |
2492 | local_read(&bpage->write) & ~RB_WRITE_MASK; | |
c64e148a | 2493 | unsigned long event_length = rb_event_length(event); |
edd813bf SR |
2494 | /* |
2495 | * This is on the tail page. It is possible that | |
2496 | * a write could come in and move the tail page | |
2497 | * and write to the next page. That is fine | |
2498 | * because we just shorten what is on this page. | |
2499 | */ | |
77ae365e SR |
2500 | old_index += write_mask; |
2501 | new_index += write_mask; | |
edd813bf | 2502 | index = local_cmpxchg(&bpage->write, old_index, new_index); |
c64e148a VN |
2503 | if (index == old_index) { |
2504 | /* update counters */ | |
2505 | local_sub(event_length, &cpu_buffer->entries_bytes); | |
edd813bf | 2506 | return 1; |
c64e148a | 2507 | } |
edd813bf SR |
2508 | } |
2509 | ||
2510 | /* could not discard */ | |
2511 | return 0; | |
2512 | } | |
2513 | ||
fa743953 SR |
2514 | static void rb_start_commit(struct ring_buffer_per_cpu *cpu_buffer) |
2515 | { | |
2516 | local_inc(&cpu_buffer->committing); | |
2517 | local_inc(&cpu_buffer->commits); | |
2518 | } | |
2519 | ||
d9abde21 | 2520 | static inline void rb_end_commit(struct ring_buffer_per_cpu *cpu_buffer) |
fa743953 SR |
2521 | { |
2522 | unsigned long commits; | |
2523 | ||
2524 | if (RB_WARN_ON(cpu_buffer, | |
2525 | !local_read(&cpu_buffer->committing))) | |
2526 | return; | |
2527 | ||
2528 | again: | |
2529 | commits = local_read(&cpu_buffer->commits); | |
2530 | /* synchronize with interrupts */ | |
2531 | barrier(); | |
2532 | if (local_read(&cpu_buffer->committing) == 1) | |
2533 | rb_set_commit_to_write(cpu_buffer); | |
2534 | ||
2535 | local_dec(&cpu_buffer->committing); | |
2536 | ||
2537 | /* synchronize with interrupts */ | |
2538 | barrier(); | |
2539 | ||
2540 | /* | |
2541 | * Need to account for interrupts coming in between the | |
2542 | * updating of the commit page and the clearing of the | |
2543 | * committing counter. | |
2544 | */ | |
2545 | if (unlikely(local_read(&cpu_buffer->commits) != commits) && | |
2546 | !local_read(&cpu_buffer->committing)) { | |
2547 | local_inc(&cpu_buffer->committing); | |
2548 | goto again; | |
2549 | } | |
2550 | } | |
2551 | ||
7a8e76a3 | 2552 | static struct ring_buffer_event * |
62f0b3eb SR |
2553 | rb_reserve_next_event(struct ring_buffer *buffer, |
2554 | struct ring_buffer_per_cpu *cpu_buffer, | |
1cd8d735 | 2555 | unsigned long length) |
7a8e76a3 SR |
2556 | { |
2557 | struct ring_buffer_event *event; | |
69d1b839 | 2558 | u64 ts, delta; |
818e3dd3 | 2559 | int nr_loops = 0; |
69d1b839 | 2560 | int add_timestamp; |
140ff891 | 2561 | u64 diff; |
7a8e76a3 | 2562 | |
fa743953 SR |
2563 | rb_start_commit(cpu_buffer); |
2564 | ||
85bac32c | 2565 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
62f0b3eb SR |
2566 | /* |
2567 | * Due to the ability to swap a cpu buffer from a buffer | |
2568 | * it is possible it was swapped before we committed. | |
2569 | * (committing stops a swap). We check for it here and | |
2570 | * if it happened, we have to fail the write. | |
2571 | */ | |
2572 | barrier(); | |
2573 | if (unlikely(ACCESS_ONCE(cpu_buffer->buffer) != buffer)) { | |
2574 | local_dec(&cpu_buffer->committing); | |
2575 | local_dec(&cpu_buffer->commits); | |
2576 | return NULL; | |
2577 | } | |
85bac32c | 2578 | #endif |
62f0b3eb | 2579 | |
be957c44 | 2580 | length = rb_calculate_event_length(length); |
bf41a158 | 2581 | again: |
69d1b839 SR |
2582 | add_timestamp = 0; |
2583 | delta = 0; | |
2584 | ||
818e3dd3 SR |
2585 | /* |
2586 | * We allow for interrupts to reenter here and do a trace. | |
2587 | * If one does, it will cause this original code to loop | |
2588 | * back here. Even with heavy interrupts happening, this | |
2589 | * should only happen a few times in a row. If this happens | |
2590 | * 1000 times in a row, there must be either an interrupt | |
2591 | * storm or we have something buggy. | |
2592 | * Bail! | |
2593 | */ | |
3e89c7bb | 2594 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) |
fa743953 | 2595 | goto out_fail; |
818e3dd3 | 2596 | |
6d3f1e12 | 2597 | ts = rb_time_stamp(cpu_buffer->buffer); |
140ff891 | 2598 | diff = ts - cpu_buffer->write_stamp; |
7a8e76a3 | 2599 | |
140ff891 SR |
2600 | /* make sure this diff is calculated here */ |
2601 | barrier(); | |
bf41a158 | 2602 | |
140ff891 SR |
2603 | /* Did the write stamp get updated already? */ |
2604 | if (likely(ts >= cpu_buffer->write_stamp)) { | |
168b6b1d SR |
2605 | delta = diff; |
2606 | if (unlikely(test_time_stamp(delta))) { | |
31274d72 JO |
2607 | int local_clock_stable = 1; |
2608 | #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK | |
35af99e6 | 2609 | local_clock_stable = sched_clock_stable(); |
31274d72 | 2610 | #endif |
69d1b839 | 2611 | WARN_ONCE(delta > (1ULL << 59), |
31274d72 | 2612 | KERN_WARNING "Delta way too big! %llu ts=%llu write stamp = %llu\n%s", |
69d1b839 SR |
2613 | (unsigned long long)delta, |
2614 | (unsigned long long)ts, | |
31274d72 JO |
2615 | (unsigned long long)cpu_buffer->write_stamp, |
2616 | local_clock_stable ? "" : | |
2617 | "If you just came from a suspend/resume,\n" | |
2618 | "please switch to the trace global clock:\n" | |
2619 | " echo global > /sys/kernel/debug/tracing/trace_clock\n"); | |
69d1b839 | 2620 | add_timestamp = 1; |
7a8e76a3 | 2621 | } |
168b6b1d | 2622 | } |
7a8e76a3 | 2623 | |
69d1b839 SR |
2624 | event = __rb_reserve_next(cpu_buffer, length, ts, |
2625 | delta, add_timestamp); | |
168b6b1d | 2626 | if (unlikely(PTR_ERR(event) == -EAGAIN)) |
bf41a158 SR |
2627 | goto again; |
2628 | ||
fa743953 SR |
2629 | if (!event) |
2630 | goto out_fail; | |
7a8e76a3 | 2631 | |
7a8e76a3 | 2632 | return event; |
fa743953 SR |
2633 | |
2634 | out_fail: | |
2635 | rb_end_commit(cpu_buffer); | |
2636 | return NULL; | |
7a8e76a3 SR |
2637 | } |
2638 | ||
1155de47 PM |
2639 | #ifdef CONFIG_TRACING |
2640 | ||
567cd4da SR |
2641 | /* |
2642 | * The lock and unlock are done within a preempt disable section. | |
2643 | * The current_context per_cpu variable can only be modified | |
2644 | * by the current task between lock and unlock. But it can | |
2645 | * be modified more than once via an interrupt. To pass this | |
2646 | * information from the lock to the unlock without having to | |
2647 | * access the 'in_interrupt()' functions again (which do show | |
2648 | * a bit of overhead in something as critical as function tracing, | |
2649 | * we use a bitmask trick. | |
2650 | * | |
2651 | * bit 0 = NMI context | |
2652 | * bit 1 = IRQ context | |
2653 | * bit 2 = SoftIRQ context | |
2654 | * bit 3 = normal context. | |
2655 | * | |
2656 | * This works because this is the order of contexts that can | |
2657 | * preempt other contexts. A SoftIRQ never preempts an IRQ | |
2658 | * context. | |
2659 | * | |
2660 | * When the context is determined, the corresponding bit is | |
2661 | * checked and set (if it was set, then a recursion of that context | |
2662 | * happened). | |
2663 | * | |
2664 | * On unlock, we need to clear this bit. To do so, just subtract | |
2665 | * 1 from the current_context and AND it to itself. | |
2666 | * | |
2667 | * (binary) | |
2668 | * 101 - 1 = 100 | |
2669 | * 101 & 100 = 100 (clearing bit zero) | |
2670 | * | |
2671 | * 1010 - 1 = 1001 | |
2672 | * 1010 & 1001 = 1000 (clearing bit 1) | |
2673 | * | |
2674 | * The least significant bit can be cleared this way, and it | |
2675 | * just so happens that it is the same bit corresponding to | |
2676 | * the current context. | |
2677 | */ | |
2678 | static DEFINE_PER_CPU(unsigned int, current_context); | |
261842b7 | 2679 | |
567cd4da | 2680 | static __always_inline int trace_recursive_lock(void) |
261842b7 | 2681 | { |
567cd4da SR |
2682 | unsigned int val = this_cpu_read(current_context); |
2683 | int bit; | |
d9abde21 | 2684 | |
567cd4da SR |
2685 | if (in_interrupt()) { |
2686 | if (in_nmi()) | |
2687 | bit = 0; | |
2688 | else if (in_irq()) | |
2689 | bit = 1; | |
2690 | else | |
2691 | bit = 2; | |
2692 | } else | |
2693 | bit = 3; | |
d9abde21 | 2694 | |
567cd4da SR |
2695 | if (unlikely(val & (1 << bit))) |
2696 | return 1; | |
d9abde21 | 2697 | |
567cd4da SR |
2698 | val |= (1 << bit); |
2699 | this_cpu_write(current_context, val); | |
d9abde21 | 2700 | |
567cd4da | 2701 | return 0; |
261842b7 SR |
2702 | } |
2703 | ||
567cd4da | 2704 | static __always_inline void trace_recursive_unlock(void) |
261842b7 | 2705 | { |
567cd4da | 2706 | unsigned int val = this_cpu_read(current_context); |
261842b7 | 2707 | |
567cd4da SR |
2708 | val--; |
2709 | val &= this_cpu_read(current_context); | |
2710 | this_cpu_write(current_context, val); | |
261842b7 SR |
2711 | } |
2712 | ||
1155de47 PM |
2713 | #else |
2714 | ||
2715 | #define trace_recursive_lock() (0) | |
2716 | #define trace_recursive_unlock() do { } while (0) | |
2717 | ||
2718 | #endif | |
2719 | ||
7a8e76a3 SR |
2720 | /** |
2721 | * ring_buffer_lock_reserve - reserve a part of the buffer | |
2722 | * @buffer: the ring buffer to reserve from | |
2723 | * @length: the length of the data to reserve (excluding event header) | |
7a8e76a3 SR |
2724 | * |
2725 | * Returns a reseverd event on the ring buffer to copy directly to. | |
2726 | * The user of this interface will need to get the body to write into | |
2727 | * and can use the ring_buffer_event_data() interface. | |
2728 | * | |
2729 | * The length is the length of the data needed, not the event length | |
2730 | * which also includes the event header. | |
2731 | * | |
2732 | * Must be paired with ring_buffer_unlock_commit, unless NULL is returned. | |
2733 | * If NULL is returned, then nothing has been allocated or locked. | |
2734 | */ | |
2735 | struct ring_buffer_event * | |
0a987751 | 2736 | ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length) |
7a8e76a3 SR |
2737 | { |
2738 | struct ring_buffer_per_cpu *cpu_buffer; | |
2739 | struct ring_buffer_event *event; | |
5168ae50 | 2740 | int cpu; |
7a8e76a3 | 2741 | |
033601a3 | 2742 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
2743 | return NULL; |
2744 | ||
bf41a158 | 2745 | /* If we are tracing schedule, we don't want to recurse */ |
5168ae50 | 2746 | preempt_disable_notrace(); |
bf41a158 | 2747 | |
52fbe9cd LJ |
2748 | if (atomic_read(&buffer->record_disabled)) |
2749 | goto out_nocheck; | |
2750 | ||
261842b7 SR |
2751 | if (trace_recursive_lock()) |
2752 | goto out_nocheck; | |
2753 | ||
7a8e76a3 SR |
2754 | cpu = raw_smp_processor_id(); |
2755 | ||
9e01c1b7 | 2756 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 2757 | goto out; |
7a8e76a3 SR |
2758 | |
2759 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
2760 | |
2761 | if (atomic_read(&cpu_buffer->record_disabled)) | |
d769041f | 2762 | goto out; |
7a8e76a3 | 2763 | |
be957c44 | 2764 | if (length > BUF_MAX_DATA_SIZE) |
bf41a158 | 2765 | goto out; |
7a8e76a3 | 2766 | |
62f0b3eb | 2767 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 | 2768 | if (!event) |
d769041f | 2769 | goto out; |
7a8e76a3 SR |
2770 | |
2771 | return event; | |
2772 | ||
d769041f | 2773 | out: |
261842b7 SR |
2774 | trace_recursive_unlock(); |
2775 | ||
2776 | out_nocheck: | |
5168ae50 | 2777 | preempt_enable_notrace(); |
7a8e76a3 SR |
2778 | return NULL; |
2779 | } | |
c4f50183 | 2780 | EXPORT_SYMBOL_GPL(ring_buffer_lock_reserve); |
7a8e76a3 | 2781 | |
a1863c21 SR |
2782 | static void |
2783 | rb_update_write_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
7a8e76a3 SR |
2784 | struct ring_buffer_event *event) |
2785 | { | |
69d1b839 SR |
2786 | u64 delta; |
2787 | ||
fa743953 SR |
2788 | /* |
2789 | * The event first in the commit queue updates the | |
2790 | * time stamp. | |
2791 | */ | |
69d1b839 SR |
2792 | if (rb_event_is_commit(cpu_buffer, event)) { |
2793 | /* | |
2794 | * A commit event that is first on a page | |
2795 | * updates the write timestamp with the page stamp | |
2796 | */ | |
2797 | if (!rb_event_index(event)) | |
2798 | cpu_buffer->write_stamp = | |
2799 | cpu_buffer->commit_page->page->time_stamp; | |
2800 | else if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) { | |
2801 | delta = event->array[0]; | |
2802 | delta <<= TS_SHIFT; | |
2803 | delta += event->time_delta; | |
2804 | cpu_buffer->write_stamp += delta; | |
2805 | } else | |
2806 | cpu_buffer->write_stamp += event->time_delta; | |
2807 | } | |
a1863c21 | 2808 | } |
bf41a158 | 2809 | |
a1863c21 SR |
2810 | static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer, |
2811 | struct ring_buffer_event *event) | |
2812 | { | |
2813 | local_inc(&cpu_buffer->entries); | |
2814 | rb_update_write_stamp(cpu_buffer, event); | |
fa743953 | 2815 | rb_end_commit(cpu_buffer); |
7a8e76a3 SR |
2816 | } |
2817 | ||
15693458 SRRH |
2818 | static __always_inline void |
2819 | rb_wakeups(struct ring_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer) | |
2820 | { | |
1e0d6714 SRRH |
2821 | bool pagebusy; |
2822 | ||
15693458 SRRH |
2823 | if (buffer->irq_work.waiters_pending) { |
2824 | buffer->irq_work.waiters_pending = false; | |
2825 | /* irq_work_queue() supplies it's own memory barriers */ | |
2826 | irq_work_queue(&buffer->irq_work.work); | |
2827 | } | |
2828 | ||
2829 | if (cpu_buffer->irq_work.waiters_pending) { | |
2830 | cpu_buffer->irq_work.waiters_pending = false; | |
2831 | /* irq_work_queue() supplies it's own memory barriers */ | |
2832 | irq_work_queue(&cpu_buffer->irq_work.work); | |
2833 | } | |
1e0d6714 SRRH |
2834 | |
2835 | pagebusy = cpu_buffer->reader_page == cpu_buffer->commit_page; | |
2836 | ||
2837 | if (!pagebusy && cpu_buffer->irq_work.full_waiters_pending) { | |
2838 | cpu_buffer->irq_work.wakeup_full = true; | |
2839 | cpu_buffer->irq_work.full_waiters_pending = false; | |
2840 | /* irq_work_queue() supplies it's own memory barriers */ | |
2841 | irq_work_queue(&cpu_buffer->irq_work.work); | |
2842 | } | |
15693458 SRRH |
2843 | } |
2844 | ||
7a8e76a3 SR |
2845 | /** |
2846 | * ring_buffer_unlock_commit - commit a reserved | |
2847 | * @buffer: The buffer to commit to | |
2848 | * @event: The event pointer to commit. | |
7a8e76a3 SR |
2849 | * |
2850 | * This commits the data to the ring buffer, and releases any locks held. | |
2851 | * | |
2852 | * Must be paired with ring_buffer_lock_reserve. | |
2853 | */ | |
2854 | int ring_buffer_unlock_commit(struct ring_buffer *buffer, | |
0a987751 | 2855 | struct ring_buffer_event *event) |
7a8e76a3 SR |
2856 | { |
2857 | struct ring_buffer_per_cpu *cpu_buffer; | |
2858 | int cpu = raw_smp_processor_id(); | |
2859 | ||
2860 | cpu_buffer = buffer->buffers[cpu]; | |
2861 | ||
7a8e76a3 SR |
2862 | rb_commit(cpu_buffer, event); |
2863 | ||
15693458 SRRH |
2864 | rb_wakeups(buffer, cpu_buffer); |
2865 | ||
261842b7 SR |
2866 | trace_recursive_unlock(); |
2867 | ||
5168ae50 | 2868 | preempt_enable_notrace(); |
7a8e76a3 SR |
2869 | |
2870 | return 0; | |
2871 | } | |
c4f50183 | 2872 | EXPORT_SYMBOL_GPL(ring_buffer_unlock_commit); |
7a8e76a3 | 2873 | |
f3b9aae1 FW |
2874 | static inline void rb_event_discard(struct ring_buffer_event *event) |
2875 | { | |
69d1b839 SR |
2876 | if (event->type_len == RINGBUF_TYPE_TIME_EXTEND) |
2877 | event = skip_time_extend(event); | |
2878 | ||
334d4169 LJ |
2879 | /* array[0] holds the actual length for the discarded event */ |
2880 | event->array[0] = rb_event_data_length(event) - RB_EVNT_HDR_SIZE; | |
2881 | event->type_len = RINGBUF_TYPE_PADDING; | |
f3b9aae1 FW |
2882 | /* time delta must be non zero */ |
2883 | if (!event->time_delta) | |
2884 | event->time_delta = 1; | |
2885 | } | |
2886 | ||
a1863c21 SR |
2887 | /* |
2888 | * Decrement the entries to the page that an event is on. | |
2889 | * The event does not even need to exist, only the pointer | |
2890 | * to the page it is on. This may only be called before the commit | |
2891 | * takes place. | |
2892 | */ | |
2893 | static inline void | |
2894 | rb_decrement_entry(struct ring_buffer_per_cpu *cpu_buffer, | |
2895 | struct ring_buffer_event *event) | |
2896 | { | |
2897 | unsigned long addr = (unsigned long)event; | |
2898 | struct buffer_page *bpage = cpu_buffer->commit_page; | |
2899 | struct buffer_page *start; | |
2900 | ||
2901 | addr &= PAGE_MASK; | |
2902 | ||
2903 | /* Do the likely case first */ | |
2904 | if (likely(bpage->page == (void *)addr)) { | |
2905 | local_dec(&bpage->entries); | |
2906 | return; | |
2907 | } | |
2908 | ||
2909 | /* | |
2910 | * Because the commit page may be on the reader page we | |
2911 | * start with the next page and check the end loop there. | |
2912 | */ | |
2913 | rb_inc_page(cpu_buffer, &bpage); | |
2914 | start = bpage; | |
2915 | do { | |
2916 | if (bpage->page == (void *)addr) { | |
2917 | local_dec(&bpage->entries); | |
2918 | return; | |
2919 | } | |
2920 | rb_inc_page(cpu_buffer, &bpage); | |
2921 | } while (bpage != start); | |
2922 | ||
2923 | /* commit not part of this buffer?? */ | |
2924 | RB_WARN_ON(cpu_buffer, 1); | |
2925 | } | |
2926 | ||
fa1b47dd SR |
2927 | /** |
2928 | * ring_buffer_commit_discard - discard an event that has not been committed | |
2929 | * @buffer: the ring buffer | |
2930 | * @event: non committed event to discard | |
2931 | * | |
dc892f73 SR |
2932 | * Sometimes an event that is in the ring buffer needs to be ignored. |
2933 | * This function lets the user discard an event in the ring buffer | |
2934 | * and then that event will not be read later. | |
2935 | * | |
2936 | * This function only works if it is called before the the item has been | |
2937 | * committed. It will try to free the event from the ring buffer | |
fa1b47dd SR |
2938 | * if another event has not been added behind it. |
2939 | * | |
2940 | * If another event has been added behind it, it will set the event | |
2941 | * up as discarded, and perform the commit. | |
2942 | * | |
2943 | * If this function is called, do not call ring_buffer_unlock_commit on | |
2944 | * the event. | |
2945 | */ | |
2946 | void ring_buffer_discard_commit(struct ring_buffer *buffer, | |
2947 | struct ring_buffer_event *event) | |
2948 | { | |
2949 | struct ring_buffer_per_cpu *cpu_buffer; | |
fa1b47dd SR |
2950 | int cpu; |
2951 | ||
2952 | /* The event is discarded regardless */ | |
f3b9aae1 | 2953 | rb_event_discard(event); |
fa1b47dd | 2954 | |
fa743953 SR |
2955 | cpu = smp_processor_id(); |
2956 | cpu_buffer = buffer->buffers[cpu]; | |
2957 | ||
fa1b47dd SR |
2958 | /* |
2959 | * This must only be called if the event has not been | |
2960 | * committed yet. Thus we can assume that preemption | |
2961 | * is still disabled. | |
2962 | */ | |
fa743953 | 2963 | RB_WARN_ON(buffer, !local_read(&cpu_buffer->committing)); |
fa1b47dd | 2964 | |
a1863c21 | 2965 | rb_decrement_entry(cpu_buffer, event); |
0f2541d2 | 2966 | if (rb_try_to_discard(cpu_buffer, event)) |
edd813bf | 2967 | goto out; |
fa1b47dd SR |
2968 | |
2969 | /* | |
2970 | * The commit is still visible by the reader, so we | |
a1863c21 | 2971 | * must still update the timestamp. |
fa1b47dd | 2972 | */ |
a1863c21 | 2973 | rb_update_write_stamp(cpu_buffer, event); |
fa1b47dd | 2974 | out: |
fa743953 | 2975 | rb_end_commit(cpu_buffer); |
fa1b47dd | 2976 | |
f3b9aae1 FW |
2977 | trace_recursive_unlock(); |
2978 | ||
5168ae50 | 2979 | preempt_enable_notrace(); |
fa1b47dd SR |
2980 | |
2981 | } | |
2982 | EXPORT_SYMBOL_GPL(ring_buffer_discard_commit); | |
2983 | ||
7a8e76a3 SR |
2984 | /** |
2985 | * ring_buffer_write - write data to the buffer without reserving | |
2986 | * @buffer: The ring buffer to write to. | |
2987 | * @length: The length of the data being written (excluding the event header) | |
2988 | * @data: The data to write to the buffer. | |
2989 | * | |
2990 | * This is like ring_buffer_lock_reserve and ring_buffer_unlock_commit as | |
2991 | * one function. If you already have the data to write to the buffer, it | |
2992 | * may be easier to simply call this function. | |
2993 | * | |
2994 | * Note, like ring_buffer_lock_reserve, the length is the length of the data | |
2995 | * and not the length of the event which would hold the header. | |
2996 | */ | |
2997 | int ring_buffer_write(struct ring_buffer *buffer, | |
01e3e710 DS |
2998 | unsigned long length, |
2999 | void *data) | |
7a8e76a3 SR |
3000 | { |
3001 | struct ring_buffer_per_cpu *cpu_buffer; | |
3002 | struct ring_buffer_event *event; | |
7a8e76a3 SR |
3003 | void *body; |
3004 | int ret = -EBUSY; | |
5168ae50 | 3005 | int cpu; |
7a8e76a3 | 3006 | |
033601a3 | 3007 | if (ring_buffer_flags != RB_BUFFERS_ON) |
a3583244 SR |
3008 | return -EBUSY; |
3009 | ||
5168ae50 | 3010 | preempt_disable_notrace(); |
bf41a158 | 3011 | |
52fbe9cd LJ |
3012 | if (atomic_read(&buffer->record_disabled)) |
3013 | goto out; | |
3014 | ||
7a8e76a3 SR |
3015 | cpu = raw_smp_processor_id(); |
3016 | ||
9e01c1b7 | 3017 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
d769041f | 3018 | goto out; |
7a8e76a3 SR |
3019 | |
3020 | cpu_buffer = buffer->buffers[cpu]; | |
7a8e76a3 SR |
3021 | |
3022 | if (atomic_read(&cpu_buffer->record_disabled)) | |
3023 | goto out; | |
3024 | ||
be957c44 SR |
3025 | if (length > BUF_MAX_DATA_SIZE) |
3026 | goto out; | |
3027 | ||
62f0b3eb | 3028 | event = rb_reserve_next_event(buffer, cpu_buffer, length); |
7a8e76a3 SR |
3029 | if (!event) |
3030 | goto out; | |
3031 | ||
3032 | body = rb_event_data(event); | |
3033 | ||
3034 | memcpy(body, data, length); | |
3035 | ||
3036 | rb_commit(cpu_buffer, event); | |
3037 | ||
15693458 SRRH |
3038 | rb_wakeups(buffer, cpu_buffer); |
3039 | ||
7a8e76a3 SR |
3040 | ret = 0; |
3041 | out: | |
5168ae50 | 3042 | preempt_enable_notrace(); |
7a8e76a3 SR |
3043 | |
3044 | return ret; | |
3045 | } | |
c4f50183 | 3046 | EXPORT_SYMBOL_GPL(ring_buffer_write); |
7a8e76a3 | 3047 | |
34a148bf | 3048 | static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) |
bf41a158 SR |
3049 | { |
3050 | struct buffer_page *reader = cpu_buffer->reader_page; | |
77ae365e | 3051 | struct buffer_page *head = rb_set_head_page(cpu_buffer); |
bf41a158 SR |
3052 | struct buffer_page *commit = cpu_buffer->commit_page; |
3053 | ||
77ae365e SR |
3054 | /* In case of error, head will be NULL */ |
3055 | if (unlikely(!head)) | |
3056 | return 1; | |
3057 | ||
bf41a158 SR |
3058 | return reader->read == rb_page_commit(reader) && |
3059 | (commit == reader || | |
3060 | (commit == head && | |
3061 | head->read == rb_page_commit(commit))); | |
3062 | } | |
3063 | ||
7a8e76a3 SR |
3064 | /** |
3065 | * ring_buffer_record_disable - stop all writes into the buffer | |
3066 | * @buffer: The ring buffer to stop writes to. | |
3067 | * | |
3068 | * This prevents all writes to the buffer. Any attempt to write | |
3069 | * to the buffer after this will fail and return NULL. | |
3070 | * | |
3071 | * The caller should call synchronize_sched() after this. | |
3072 | */ | |
3073 | void ring_buffer_record_disable(struct ring_buffer *buffer) | |
3074 | { | |
3075 | atomic_inc(&buffer->record_disabled); | |
3076 | } | |
c4f50183 | 3077 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable); |
7a8e76a3 SR |
3078 | |
3079 | /** | |
3080 | * ring_buffer_record_enable - enable writes to the buffer | |
3081 | * @buffer: The ring buffer to enable writes | |
3082 | * | |
3083 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 3084 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
3085 | */ |
3086 | void ring_buffer_record_enable(struct ring_buffer *buffer) | |
3087 | { | |
3088 | atomic_dec(&buffer->record_disabled); | |
3089 | } | |
c4f50183 | 3090 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable); |
7a8e76a3 | 3091 | |
499e5470 SR |
3092 | /** |
3093 | * ring_buffer_record_off - stop all writes into the buffer | |
3094 | * @buffer: The ring buffer to stop writes to. | |
3095 | * | |
3096 | * This prevents all writes to the buffer. Any attempt to write | |
3097 | * to the buffer after this will fail and return NULL. | |
3098 | * | |
3099 | * This is different than ring_buffer_record_disable() as | |
87abb3b1 | 3100 | * it works like an on/off switch, where as the disable() version |
499e5470 SR |
3101 | * must be paired with a enable(). |
3102 | */ | |
3103 | void ring_buffer_record_off(struct ring_buffer *buffer) | |
3104 | { | |
3105 | unsigned int rd; | |
3106 | unsigned int new_rd; | |
3107 | ||
3108 | do { | |
3109 | rd = atomic_read(&buffer->record_disabled); | |
3110 | new_rd = rd | RB_BUFFER_OFF; | |
3111 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
3112 | } | |
3113 | EXPORT_SYMBOL_GPL(ring_buffer_record_off); | |
3114 | ||
3115 | /** | |
3116 | * ring_buffer_record_on - restart writes into the buffer | |
3117 | * @buffer: The ring buffer to start writes to. | |
3118 | * | |
3119 | * This enables all writes to the buffer that was disabled by | |
3120 | * ring_buffer_record_off(). | |
3121 | * | |
3122 | * This is different than ring_buffer_record_enable() as | |
87abb3b1 | 3123 | * it works like an on/off switch, where as the enable() version |
499e5470 SR |
3124 | * must be paired with a disable(). |
3125 | */ | |
3126 | void ring_buffer_record_on(struct ring_buffer *buffer) | |
3127 | { | |
3128 | unsigned int rd; | |
3129 | unsigned int new_rd; | |
3130 | ||
3131 | do { | |
3132 | rd = atomic_read(&buffer->record_disabled); | |
3133 | new_rd = rd & ~RB_BUFFER_OFF; | |
3134 | } while (atomic_cmpxchg(&buffer->record_disabled, rd, new_rd) != rd); | |
3135 | } | |
3136 | EXPORT_SYMBOL_GPL(ring_buffer_record_on); | |
3137 | ||
3138 | /** | |
3139 | * ring_buffer_record_is_on - return true if the ring buffer can write | |
3140 | * @buffer: The ring buffer to see if write is enabled | |
3141 | * | |
3142 | * Returns true if the ring buffer is in a state that it accepts writes. | |
3143 | */ | |
3144 | int ring_buffer_record_is_on(struct ring_buffer *buffer) | |
3145 | { | |
3146 | return !atomic_read(&buffer->record_disabled); | |
3147 | } | |
3148 | ||
7a8e76a3 SR |
3149 | /** |
3150 | * ring_buffer_record_disable_cpu - stop all writes into the cpu_buffer | |
3151 | * @buffer: The ring buffer to stop writes to. | |
3152 | * @cpu: The CPU buffer to stop | |
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 | * The caller should call synchronize_sched() after this. | |
3158 | */ | |
3159 | void ring_buffer_record_disable_cpu(struct ring_buffer *buffer, int cpu) | |
3160 | { | |
3161 | struct ring_buffer_per_cpu *cpu_buffer; | |
3162 | ||
9e01c1b7 | 3163 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3164 | return; |
7a8e76a3 SR |
3165 | |
3166 | cpu_buffer = buffer->buffers[cpu]; | |
3167 | atomic_inc(&cpu_buffer->record_disabled); | |
3168 | } | |
c4f50183 | 3169 | EXPORT_SYMBOL_GPL(ring_buffer_record_disable_cpu); |
7a8e76a3 SR |
3170 | |
3171 | /** | |
3172 | * ring_buffer_record_enable_cpu - enable writes to the buffer | |
3173 | * @buffer: The ring buffer to enable writes | |
3174 | * @cpu: The CPU to enable. | |
3175 | * | |
3176 | * Note, multiple disables will need the same number of enables | |
c41b20e7 | 3177 | * to truly enable the writing (much like preempt_disable). |
7a8e76a3 SR |
3178 | */ |
3179 | void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu) | |
3180 | { | |
3181 | struct ring_buffer_per_cpu *cpu_buffer; | |
3182 | ||
9e01c1b7 | 3183 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3184 | return; |
7a8e76a3 SR |
3185 | |
3186 | cpu_buffer = buffer->buffers[cpu]; | |
3187 | atomic_dec(&cpu_buffer->record_disabled); | |
3188 | } | |
c4f50183 | 3189 | EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); |
7a8e76a3 | 3190 | |
f6195aa0 SR |
3191 | /* |
3192 | * The total entries in the ring buffer is the running counter | |
3193 | * of entries entered into the ring buffer, minus the sum of | |
3194 | * the entries read from the ring buffer and the number of | |
3195 | * entries that were overwritten. | |
3196 | */ | |
3197 | static inline unsigned long | |
3198 | rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer) | |
3199 | { | |
3200 | return local_read(&cpu_buffer->entries) - | |
3201 | (local_read(&cpu_buffer->overrun) + cpu_buffer->read); | |
3202 | } | |
3203 | ||
c64e148a VN |
3204 | /** |
3205 | * ring_buffer_oldest_event_ts - get the oldest event timestamp from the buffer | |
3206 | * @buffer: The ring buffer | |
3207 | * @cpu: The per CPU buffer to read from. | |
3208 | */ | |
50ecf2c3 | 3209 | u64 ring_buffer_oldest_event_ts(struct ring_buffer *buffer, int cpu) |
c64e148a VN |
3210 | { |
3211 | unsigned long flags; | |
3212 | struct ring_buffer_per_cpu *cpu_buffer; | |
3213 | struct buffer_page *bpage; | |
da830e58 | 3214 | u64 ret = 0; |
c64e148a VN |
3215 | |
3216 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3217 | return 0; | |
3218 | ||
3219 | cpu_buffer = buffer->buffers[cpu]; | |
7115e3fc | 3220 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
3221 | /* |
3222 | * if the tail is on reader_page, oldest time stamp is on the reader | |
3223 | * page | |
3224 | */ | |
3225 | if (cpu_buffer->tail_page == cpu_buffer->reader_page) | |
3226 | bpage = cpu_buffer->reader_page; | |
3227 | else | |
3228 | bpage = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
3229 | if (bpage) |
3230 | ret = bpage->page->time_stamp; | |
7115e3fc | 3231 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
c64e148a VN |
3232 | |
3233 | return ret; | |
3234 | } | |
3235 | EXPORT_SYMBOL_GPL(ring_buffer_oldest_event_ts); | |
3236 | ||
3237 | /** | |
3238 | * ring_buffer_bytes_cpu - get the number of bytes consumed in a cpu buffer | |
3239 | * @buffer: The ring buffer | |
3240 | * @cpu: The per CPU buffer to read from. | |
3241 | */ | |
3242 | unsigned long ring_buffer_bytes_cpu(struct ring_buffer *buffer, int cpu) | |
3243 | { | |
3244 | struct ring_buffer_per_cpu *cpu_buffer; | |
3245 | unsigned long ret; | |
3246 | ||
3247 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3248 | return 0; | |
3249 | ||
3250 | cpu_buffer = buffer->buffers[cpu]; | |
3251 | ret = local_read(&cpu_buffer->entries_bytes) - cpu_buffer->read_bytes; | |
3252 | ||
3253 | return ret; | |
3254 | } | |
3255 | EXPORT_SYMBOL_GPL(ring_buffer_bytes_cpu); | |
3256 | ||
7a8e76a3 SR |
3257 | /** |
3258 | * ring_buffer_entries_cpu - get the number of entries in a cpu buffer | |
3259 | * @buffer: The ring buffer | |
3260 | * @cpu: The per CPU buffer to get the entries from. | |
3261 | */ | |
3262 | unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) | |
3263 | { | |
3264 | struct ring_buffer_per_cpu *cpu_buffer; | |
3265 | ||
9e01c1b7 | 3266 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3267 | return 0; |
7a8e76a3 SR |
3268 | |
3269 | cpu_buffer = buffer->buffers[cpu]; | |
554f786e | 3270 | |
f6195aa0 | 3271 | return rb_num_of_entries(cpu_buffer); |
7a8e76a3 | 3272 | } |
c4f50183 | 3273 | EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); |
7a8e76a3 SR |
3274 | |
3275 | /** | |
884bfe89 SP |
3276 | * ring_buffer_overrun_cpu - get the number of overruns caused by the ring |
3277 | * buffer wrapping around (only if RB_FL_OVERWRITE is on). | |
7a8e76a3 SR |
3278 | * @buffer: The ring buffer |
3279 | * @cpu: The per CPU buffer to get the number of overruns from | |
3280 | */ | |
3281 | unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3282 | { | |
3283 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 3284 | unsigned long ret; |
7a8e76a3 | 3285 | |
9e01c1b7 | 3286 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3287 | return 0; |
7a8e76a3 SR |
3288 | |
3289 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3290 | ret = local_read(&cpu_buffer->overrun); |
554f786e SR |
3291 | |
3292 | return ret; | |
7a8e76a3 | 3293 | } |
c4f50183 | 3294 | EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); |
7a8e76a3 | 3295 | |
f0d2c681 | 3296 | /** |
884bfe89 SP |
3297 | * ring_buffer_commit_overrun_cpu - get the number of overruns caused by |
3298 | * commits failing due to the buffer wrapping around while there are uncommitted | |
3299 | * events, such as during an interrupt storm. | |
f0d2c681 SR |
3300 | * @buffer: The ring buffer |
3301 | * @cpu: The per CPU buffer to get the number of overruns from | |
3302 | */ | |
3303 | unsigned long | |
3304 | ring_buffer_commit_overrun_cpu(struct ring_buffer *buffer, int cpu) | |
3305 | { | |
3306 | struct ring_buffer_per_cpu *cpu_buffer; | |
3307 | unsigned long ret; | |
3308 | ||
3309 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3310 | return 0; | |
3311 | ||
3312 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3313 | ret = local_read(&cpu_buffer->commit_overrun); |
f0d2c681 SR |
3314 | |
3315 | return ret; | |
3316 | } | |
3317 | EXPORT_SYMBOL_GPL(ring_buffer_commit_overrun_cpu); | |
3318 | ||
884bfe89 SP |
3319 | /** |
3320 | * ring_buffer_dropped_events_cpu - get the number of dropped events caused by | |
3321 | * the ring buffer filling up (only if RB_FL_OVERWRITE is off). | |
3322 | * @buffer: The ring buffer | |
3323 | * @cpu: The per CPU buffer to get the number of overruns from | |
3324 | */ | |
3325 | unsigned long | |
3326 | ring_buffer_dropped_events_cpu(struct ring_buffer *buffer, int cpu) | |
3327 | { | |
3328 | struct ring_buffer_per_cpu *cpu_buffer; | |
3329 | unsigned long ret; | |
3330 | ||
3331 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3332 | return 0; | |
3333 | ||
3334 | cpu_buffer = buffer->buffers[cpu]; | |
3335 | ret = local_read(&cpu_buffer->dropped_events); | |
3336 | ||
3337 | return ret; | |
3338 | } | |
3339 | EXPORT_SYMBOL_GPL(ring_buffer_dropped_events_cpu); | |
3340 | ||
ad964704 SRRH |
3341 | /** |
3342 | * ring_buffer_read_events_cpu - get the number of events successfully read | |
3343 | * @buffer: The ring buffer | |
3344 | * @cpu: The per CPU buffer to get the number of events read | |
3345 | */ | |
3346 | unsigned long | |
3347 | ring_buffer_read_events_cpu(struct ring_buffer *buffer, int cpu) | |
3348 | { | |
3349 | struct ring_buffer_per_cpu *cpu_buffer; | |
3350 | ||
3351 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
3352 | return 0; | |
3353 | ||
3354 | cpu_buffer = buffer->buffers[cpu]; | |
3355 | return cpu_buffer->read; | |
3356 | } | |
3357 | EXPORT_SYMBOL_GPL(ring_buffer_read_events_cpu); | |
3358 | ||
7a8e76a3 SR |
3359 | /** |
3360 | * ring_buffer_entries - get the number of entries in a buffer | |
3361 | * @buffer: The ring buffer | |
3362 | * | |
3363 | * Returns the total number of entries in the ring buffer | |
3364 | * (all CPU entries) | |
3365 | */ | |
3366 | unsigned long ring_buffer_entries(struct ring_buffer *buffer) | |
3367 | { | |
3368 | struct ring_buffer_per_cpu *cpu_buffer; | |
3369 | unsigned long entries = 0; | |
3370 | int cpu; | |
3371 | ||
3372 | /* if you care about this being correct, lock the buffer */ | |
3373 | for_each_buffer_cpu(buffer, cpu) { | |
3374 | cpu_buffer = buffer->buffers[cpu]; | |
f6195aa0 | 3375 | entries += rb_num_of_entries(cpu_buffer); |
7a8e76a3 SR |
3376 | } |
3377 | ||
3378 | return entries; | |
3379 | } | |
c4f50183 | 3380 | EXPORT_SYMBOL_GPL(ring_buffer_entries); |
7a8e76a3 SR |
3381 | |
3382 | /** | |
67b394f7 | 3383 | * ring_buffer_overruns - get the number of overruns in buffer |
7a8e76a3 SR |
3384 | * @buffer: The ring buffer |
3385 | * | |
3386 | * Returns the total number of overruns in the ring buffer | |
3387 | * (all CPU entries) | |
3388 | */ | |
3389 | unsigned long ring_buffer_overruns(struct ring_buffer *buffer) | |
3390 | { | |
3391 | struct ring_buffer_per_cpu *cpu_buffer; | |
3392 | unsigned long overruns = 0; | |
3393 | int cpu; | |
3394 | ||
3395 | /* if you care about this being correct, lock the buffer */ | |
3396 | for_each_buffer_cpu(buffer, cpu) { | |
3397 | cpu_buffer = buffer->buffers[cpu]; | |
77ae365e | 3398 | overruns += local_read(&cpu_buffer->overrun); |
7a8e76a3 SR |
3399 | } |
3400 | ||
3401 | return overruns; | |
3402 | } | |
c4f50183 | 3403 | EXPORT_SYMBOL_GPL(ring_buffer_overruns); |
7a8e76a3 | 3404 | |
642edba5 | 3405 | static void rb_iter_reset(struct ring_buffer_iter *iter) |
7a8e76a3 SR |
3406 | { |
3407 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3408 | ||
d769041f | 3409 | /* Iterator usage is expected to have record disabled */ |
651e22f2 SRRH |
3410 | iter->head_page = cpu_buffer->reader_page; |
3411 | iter->head = cpu_buffer->reader_page->read; | |
3412 | ||
3413 | iter->cache_reader_page = iter->head_page; | |
24607f11 | 3414 | iter->cache_read = cpu_buffer->read; |
651e22f2 | 3415 | |
d769041f SR |
3416 | if (iter->head) |
3417 | iter->read_stamp = cpu_buffer->read_stamp; | |
3418 | else | |
abc9b56d | 3419 | iter->read_stamp = iter->head_page->page->time_stamp; |
642edba5 | 3420 | } |
f83c9d0f | 3421 | |
642edba5 SR |
3422 | /** |
3423 | * ring_buffer_iter_reset - reset an iterator | |
3424 | * @iter: The iterator to reset | |
3425 | * | |
3426 | * Resets the iterator, so that it will start from the beginning | |
3427 | * again. | |
3428 | */ | |
3429 | void ring_buffer_iter_reset(struct ring_buffer_iter *iter) | |
3430 | { | |
554f786e | 3431 | struct ring_buffer_per_cpu *cpu_buffer; |
642edba5 SR |
3432 | unsigned long flags; |
3433 | ||
554f786e SR |
3434 | if (!iter) |
3435 | return; | |
3436 | ||
3437 | cpu_buffer = iter->cpu_buffer; | |
3438 | ||
5389f6fa | 3439 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
642edba5 | 3440 | rb_iter_reset(iter); |
5389f6fa | 3441 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 3442 | } |
c4f50183 | 3443 | EXPORT_SYMBOL_GPL(ring_buffer_iter_reset); |
7a8e76a3 SR |
3444 | |
3445 | /** | |
3446 | * ring_buffer_iter_empty - check if an iterator has no more to read | |
3447 | * @iter: The iterator to check | |
3448 | */ | |
3449 | int ring_buffer_iter_empty(struct ring_buffer_iter *iter) | |
3450 | { | |
3451 | struct ring_buffer_per_cpu *cpu_buffer; | |
3452 | ||
3453 | cpu_buffer = iter->cpu_buffer; | |
3454 | ||
bf41a158 SR |
3455 | return iter->head_page == cpu_buffer->commit_page && |
3456 | iter->head == rb_commit_index(cpu_buffer); | |
7a8e76a3 | 3457 | } |
c4f50183 | 3458 | EXPORT_SYMBOL_GPL(ring_buffer_iter_empty); |
7a8e76a3 SR |
3459 | |
3460 | static void | |
3461 | rb_update_read_stamp(struct ring_buffer_per_cpu *cpu_buffer, | |
3462 | struct ring_buffer_event *event) | |
3463 | { | |
3464 | u64 delta; | |
3465 | ||
334d4169 | 3466 | switch (event->type_len) { |
7a8e76a3 SR |
3467 | case RINGBUF_TYPE_PADDING: |
3468 | return; | |
3469 | ||
3470 | case RINGBUF_TYPE_TIME_EXTEND: | |
3471 | delta = event->array[0]; | |
3472 | delta <<= TS_SHIFT; | |
3473 | delta += event->time_delta; | |
3474 | cpu_buffer->read_stamp += delta; | |
3475 | return; | |
3476 | ||
3477 | case RINGBUF_TYPE_TIME_STAMP: | |
3478 | /* FIXME: not implemented */ | |
3479 | return; | |
3480 | ||
3481 | case RINGBUF_TYPE_DATA: | |
3482 | cpu_buffer->read_stamp += event->time_delta; | |
3483 | return; | |
3484 | ||
3485 | default: | |
3486 | BUG(); | |
3487 | } | |
3488 | return; | |
3489 | } | |
3490 | ||
3491 | static void | |
3492 | rb_update_iter_read_stamp(struct ring_buffer_iter *iter, | |
3493 | struct ring_buffer_event *event) | |
3494 | { | |
3495 | u64 delta; | |
3496 | ||
334d4169 | 3497 | switch (event->type_len) { |
7a8e76a3 SR |
3498 | case RINGBUF_TYPE_PADDING: |
3499 | return; | |
3500 | ||
3501 | case RINGBUF_TYPE_TIME_EXTEND: | |
3502 | delta = event->array[0]; | |
3503 | delta <<= TS_SHIFT; | |
3504 | delta += event->time_delta; | |
3505 | iter->read_stamp += delta; | |
3506 | return; | |
3507 | ||
3508 | case RINGBUF_TYPE_TIME_STAMP: | |
3509 | /* FIXME: not implemented */ | |
3510 | return; | |
3511 | ||
3512 | case RINGBUF_TYPE_DATA: | |
3513 | iter->read_stamp += event->time_delta; | |
3514 | return; | |
3515 | ||
3516 | default: | |
3517 | BUG(); | |
3518 | } | |
3519 | return; | |
3520 | } | |
3521 | ||
d769041f SR |
3522 | static struct buffer_page * |
3523 | rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer) | |
7a8e76a3 | 3524 | { |
d769041f | 3525 | struct buffer_page *reader = NULL; |
66a8cb95 | 3526 | unsigned long overwrite; |
d769041f | 3527 | unsigned long flags; |
818e3dd3 | 3528 | int nr_loops = 0; |
77ae365e | 3529 | int ret; |
d769041f | 3530 | |
3e03fb7f | 3531 | local_irq_save(flags); |
0199c4e6 | 3532 | arch_spin_lock(&cpu_buffer->lock); |
d769041f SR |
3533 | |
3534 | again: | |
818e3dd3 SR |
3535 | /* |
3536 | * This should normally only loop twice. But because the | |
3537 | * start of the reader inserts an empty page, it causes | |
3538 | * a case where we will loop three times. There should be no | |
3539 | * reason to loop four times (that I know of). | |
3540 | */ | |
3e89c7bb | 3541 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) { |
818e3dd3 SR |
3542 | reader = NULL; |
3543 | goto out; | |
3544 | } | |
3545 | ||
d769041f SR |
3546 | reader = cpu_buffer->reader_page; |
3547 | ||
3548 | /* If there's more to read, return this page */ | |
bf41a158 | 3549 | if (cpu_buffer->reader_page->read < rb_page_size(reader)) |
d769041f SR |
3550 | goto out; |
3551 | ||
3552 | /* Never should we have an index greater than the size */ | |
3e89c7bb SR |
3553 | if (RB_WARN_ON(cpu_buffer, |
3554 | cpu_buffer->reader_page->read > rb_page_size(reader))) | |
3555 | goto out; | |
d769041f SR |
3556 | |
3557 | /* check if we caught up to the tail */ | |
3558 | reader = NULL; | |
bf41a158 | 3559 | if (cpu_buffer->commit_page == cpu_buffer->reader_page) |
d769041f | 3560 | goto out; |
7a8e76a3 | 3561 | |
a5fb8331 SR |
3562 | /* Don't bother swapping if the ring buffer is empty */ |
3563 | if (rb_num_of_entries(cpu_buffer) == 0) | |
3564 | goto out; | |
3565 | ||
7a8e76a3 | 3566 | /* |
d769041f | 3567 | * Reset the reader page to size zero. |
7a8e76a3 | 3568 | */ |
77ae365e SR |
3569 | local_set(&cpu_buffer->reader_page->write, 0); |
3570 | local_set(&cpu_buffer->reader_page->entries, 0); | |
3571 | local_set(&cpu_buffer->reader_page->page->commit, 0); | |
ff0ff84a | 3572 | cpu_buffer->reader_page->real_end = 0; |
7a8e76a3 | 3573 | |
77ae365e SR |
3574 | spin: |
3575 | /* | |
3576 | * Splice the empty reader page into the list around the head. | |
3577 | */ | |
3578 | reader = rb_set_head_page(cpu_buffer); | |
54f7be5b SR |
3579 | if (!reader) |
3580 | goto out; | |
0e1ff5d7 | 3581 | cpu_buffer->reader_page->list.next = rb_list_head(reader->list.next); |
d769041f | 3582 | cpu_buffer->reader_page->list.prev = reader->list.prev; |
bf41a158 | 3583 | |
3adc54fa SR |
3584 | /* |
3585 | * cpu_buffer->pages just needs to point to the buffer, it | |
3586 | * has no specific buffer page to point to. Lets move it out | |
25985edc | 3587 | * of our way so we don't accidentally swap it. |
3adc54fa SR |
3588 | */ |
3589 | cpu_buffer->pages = reader->list.prev; | |
3590 | ||
77ae365e SR |
3591 | /* The reader page will be pointing to the new head */ |
3592 | rb_set_list_to_head(cpu_buffer, &cpu_buffer->reader_page->list); | |
7a8e76a3 | 3593 | |
66a8cb95 SR |
3594 | /* |
3595 | * We want to make sure we read the overruns after we set up our | |
3596 | * pointers to the next object. The writer side does a | |
3597 | * cmpxchg to cross pages which acts as the mb on the writer | |
3598 | * side. Note, the reader will constantly fail the swap | |
3599 | * while the writer is updating the pointers, so this | |
3600 | * guarantees that the overwrite recorded here is the one we | |
3601 | * want to compare with the last_overrun. | |
3602 | */ | |
3603 | smp_mb(); | |
3604 | overwrite = local_read(&(cpu_buffer->overrun)); | |
3605 | ||
77ae365e SR |
3606 | /* |
3607 | * Here's the tricky part. | |
3608 | * | |
3609 | * We need to move the pointer past the header page. | |
3610 | * But we can only do that if a writer is not currently | |
3611 | * moving it. The page before the header page has the | |
3612 | * flag bit '1' set if it is pointing to the page we want. | |
3613 | * but if the writer is in the process of moving it | |
3614 | * than it will be '2' or already moved '0'. | |
3615 | */ | |
3616 | ||
3617 | ret = rb_head_page_replace(reader, cpu_buffer->reader_page); | |
7a8e76a3 SR |
3618 | |
3619 | /* | |
77ae365e | 3620 | * If we did not convert it, then we must try again. |
7a8e76a3 | 3621 | */ |
77ae365e SR |
3622 | if (!ret) |
3623 | goto spin; | |
7a8e76a3 | 3624 | |
77ae365e SR |
3625 | /* |
3626 | * Yeah! We succeeded in replacing the page. | |
3627 | * | |
3628 | * Now make the new head point back to the reader page. | |
3629 | */ | |
5ded3dc6 | 3630 | rb_list_head(reader->list.next)->prev = &cpu_buffer->reader_page->list; |
77ae365e | 3631 | rb_inc_page(cpu_buffer, &cpu_buffer->head_page); |
d769041f SR |
3632 | |
3633 | /* Finally update the reader page to the new head */ | |
3634 | cpu_buffer->reader_page = reader; | |
3635 | rb_reset_reader_page(cpu_buffer); | |
3636 | ||
66a8cb95 SR |
3637 | if (overwrite != cpu_buffer->last_overrun) { |
3638 | cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun; | |
3639 | cpu_buffer->last_overrun = overwrite; | |
3640 | } | |
3641 | ||
d769041f SR |
3642 | goto again; |
3643 | ||
3644 | out: | |
0199c4e6 | 3645 | arch_spin_unlock(&cpu_buffer->lock); |
3e03fb7f | 3646 | local_irq_restore(flags); |
d769041f SR |
3647 | |
3648 | return reader; | |
3649 | } | |
3650 | ||
3651 | static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer) | |
3652 | { | |
3653 | struct ring_buffer_event *event; | |
3654 | struct buffer_page *reader; | |
3655 | unsigned length; | |
3656 | ||
3657 | reader = rb_get_reader_page(cpu_buffer); | |
7a8e76a3 | 3658 | |
d769041f | 3659 | /* This function should not be called when buffer is empty */ |
3e89c7bb SR |
3660 | if (RB_WARN_ON(cpu_buffer, !reader)) |
3661 | return; | |
7a8e76a3 | 3662 | |
d769041f SR |
3663 | event = rb_reader_event(cpu_buffer); |
3664 | ||
a1863c21 | 3665 | if (event->type_len <= RINGBUF_TYPE_DATA_TYPE_LEN_MAX) |
e4906eff | 3666 | cpu_buffer->read++; |
d769041f SR |
3667 | |
3668 | rb_update_read_stamp(cpu_buffer, event); | |
3669 | ||
3670 | length = rb_event_length(event); | |
6f807acd | 3671 | cpu_buffer->reader_page->read += length; |
7a8e76a3 SR |
3672 | } |
3673 | ||
3674 | static void rb_advance_iter(struct ring_buffer_iter *iter) | |
3675 | { | |
7a8e76a3 SR |
3676 | struct ring_buffer_per_cpu *cpu_buffer; |
3677 | struct ring_buffer_event *event; | |
3678 | unsigned length; | |
3679 | ||
3680 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 SR |
3681 | |
3682 | /* | |
3683 | * Check if we are at the end of the buffer. | |
3684 | */ | |
bf41a158 | 3685 | if (iter->head >= rb_page_size(iter->head_page)) { |
ea05b57c SR |
3686 | /* discarded commits can make the page empty */ |
3687 | if (iter->head_page == cpu_buffer->commit_page) | |
3e89c7bb | 3688 | return; |
d769041f | 3689 | rb_inc_iter(iter); |
7a8e76a3 SR |
3690 | return; |
3691 | } | |
3692 | ||
3693 | event = rb_iter_head_event(iter); | |
3694 | ||
3695 | length = rb_event_length(event); | |
3696 | ||
3697 | /* | |
3698 | * This should not be called to advance the header if we are | |
3699 | * at the tail of the buffer. | |
3700 | */ | |
3e89c7bb | 3701 | if (RB_WARN_ON(cpu_buffer, |
f536aafc | 3702 | (iter->head_page == cpu_buffer->commit_page) && |
3e89c7bb SR |
3703 | (iter->head + length > rb_commit_index(cpu_buffer)))) |
3704 | return; | |
7a8e76a3 SR |
3705 | |
3706 | rb_update_iter_read_stamp(iter, event); | |
3707 | ||
3708 | iter->head += length; | |
3709 | ||
3710 | /* check for end of page padding */ | |
bf41a158 SR |
3711 | if ((iter->head >= rb_page_size(iter->head_page)) && |
3712 | (iter->head_page != cpu_buffer->commit_page)) | |
771e0384 | 3713 | rb_inc_iter(iter); |
7a8e76a3 SR |
3714 | } |
3715 | ||
66a8cb95 SR |
3716 | static int rb_lost_events(struct ring_buffer_per_cpu *cpu_buffer) |
3717 | { | |
3718 | return cpu_buffer->lost_events; | |
3719 | } | |
3720 | ||
f83c9d0f | 3721 | static struct ring_buffer_event * |
66a8cb95 SR |
3722 | rb_buffer_peek(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts, |
3723 | unsigned long *lost_events) | |
7a8e76a3 | 3724 | { |
7a8e76a3 | 3725 | struct ring_buffer_event *event; |
d769041f | 3726 | struct buffer_page *reader; |
818e3dd3 | 3727 | int nr_loops = 0; |
7a8e76a3 | 3728 | |
7a8e76a3 | 3729 | again: |
818e3dd3 | 3730 | /* |
69d1b839 SR |
3731 | * We repeat when a time extend is encountered. |
3732 | * Since the time extend is always attached to a data event, | |
3733 | * we should never loop more than once. | |
3734 | * (We never hit the following condition more than twice). | |
818e3dd3 | 3735 | */ |
69d1b839 | 3736 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 2)) |
818e3dd3 | 3737 | return NULL; |
818e3dd3 | 3738 | |
d769041f SR |
3739 | reader = rb_get_reader_page(cpu_buffer); |
3740 | if (!reader) | |
7a8e76a3 SR |
3741 | return NULL; |
3742 | ||
d769041f | 3743 | event = rb_reader_event(cpu_buffer); |
7a8e76a3 | 3744 | |
334d4169 | 3745 | switch (event->type_len) { |
7a8e76a3 | 3746 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3747 | if (rb_null_event(event)) |
3748 | RB_WARN_ON(cpu_buffer, 1); | |
3749 | /* | |
3750 | * Because the writer could be discarding every | |
3751 | * event it creates (which would probably be bad) | |
3752 | * if we were to go back to "again" then we may never | |
3753 | * catch up, and will trigger the warn on, or lock | |
3754 | * the box. Return the padding, and we will release | |
3755 | * the current locks, and try again. | |
3756 | */ | |
2d622719 | 3757 | return event; |
7a8e76a3 SR |
3758 | |
3759 | case RINGBUF_TYPE_TIME_EXTEND: | |
3760 | /* Internal data, OK to advance */ | |
d769041f | 3761 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3762 | goto again; |
3763 | ||
3764 | case RINGBUF_TYPE_TIME_STAMP: | |
3765 | /* FIXME: not implemented */ | |
d769041f | 3766 | rb_advance_reader(cpu_buffer); |
7a8e76a3 SR |
3767 | goto again; |
3768 | ||
3769 | case RINGBUF_TYPE_DATA: | |
3770 | if (ts) { | |
3771 | *ts = cpu_buffer->read_stamp + event->time_delta; | |
d8eeb2d3 | 3772 | ring_buffer_normalize_time_stamp(cpu_buffer->buffer, |
37886f6a | 3773 | cpu_buffer->cpu, ts); |
7a8e76a3 | 3774 | } |
66a8cb95 SR |
3775 | if (lost_events) |
3776 | *lost_events = rb_lost_events(cpu_buffer); | |
7a8e76a3 SR |
3777 | return event; |
3778 | ||
3779 | default: | |
3780 | BUG(); | |
3781 | } | |
3782 | ||
3783 | return NULL; | |
3784 | } | |
c4f50183 | 3785 | EXPORT_SYMBOL_GPL(ring_buffer_peek); |
7a8e76a3 | 3786 | |
f83c9d0f SR |
3787 | static struct ring_buffer_event * |
3788 | rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
7a8e76a3 SR |
3789 | { |
3790 | struct ring_buffer *buffer; | |
3791 | struct ring_buffer_per_cpu *cpu_buffer; | |
3792 | struct ring_buffer_event *event; | |
818e3dd3 | 3793 | int nr_loops = 0; |
7a8e76a3 | 3794 | |
7a8e76a3 SR |
3795 | cpu_buffer = iter->cpu_buffer; |
3796 | buffer = cpu_buffer->buffer; | |
3797 | ||
492a74f4 SR |
3798 | /* |
3799 | * Check if someone performed a consuming read to | |
3800 | * the buffer. A consuming read invalidates the iterator | |
3801 | * and we need to reset the iterator in this case. | |
3802 | */ | |
3803 | if (unlikely(iter->cache_read != cpu_buffer->read || | |
3804 | iter->cache_reader_page != cpu_buffer->reader_page)) | |
3805 | rb_iter_reset(iter); | |
3806 | ||
7a8e76a3 | 3807 | again: |
3c05d748 SR |
3808 | if (ring_buffer_iter_empty(iter)) |
3809 | return NULL; | |
3810 | ||
818e3dd3 | 3811 | /* |
021de3d9 SRRH |
3812 | * We repeat when a time extend is encountered or we hit |
3813 | * the end of the page. Since the time extend is always attached | |
3814 | * to a data event, we should never loop more than three times. | |
3815 | * Once for going to next page, once on time extend, and | |
3816 | * finally once to get the event. | |
3817 | * (We never hit the following condition more than thrice). | |
818e3dd3 | 3818 | */ |
021de3d9 | 3819 | if (RB_WARN_ON(cpu_buffer, ++nr_loops > 3)) |
818e3dd3 | 3820 | return NULL; |
818e3dd3 | 3821 | |
7a8e76a3 SR |
3822 | if (rb_per_cpu_empty(cpu_buffer)) |
3823 | return NULL; | |
3824 | ||
10e83fd0 | 3825 | if (iter->head >= rb_page_size(iter->head_page)) { |
3c05d748 SR |
3826 | rb_inc_iter(iter); |
3827 | goto again; | |
3828 | } | |
3829 | ||
7a8e76a3 SR |
3830 | event = rb_iter_head_event(iter); |
3831 | ||
334d4169 | 3832 | switch (event->type_len) { |
7a8e76a3 | 3833 | case RINGBUF_TYPE_PADDING: |
2d622719 TZ |
3834 | if (rb_null_event(event)) { |
3835 | rb_inc_iter(iter); | |
3836 | goto again; | |
3837 | } | |
3838 | rb_advance_iter(iter); | |
3839 | return event; | |
7a8e76a3 SR |
3840 | |
3841 | case RINGBUF_TYPE_TIME_EXTEND: | |
3842 | /* Internal data, OK to advance */ | |
3843 | rb_advance_iter(iter); | |
3844 | goto again; | |
3845 | ||
3846 | case RINGBUF_TYPE_TIME_STAMP: | |
3847 | /* FIXME: not implemented */ | |
3848 | rb_advance_iter(iter); | |
3849 | goto again; | |
3850 | ||
3851 | case RINGBUF_TYPE_DATA: | |
3852 | if (ts) { | |
3853 | *ts = iter->read_stamp + event->time_delta; | |
37886f6a SR |
3854 | ring_buffer_normalize_time_stamp(buffer, |
3855 | cpu_buffer->cpu, ts); | |
7a8e76a3 SR |
3856 | } |
3857 | return event; | |
3858 | ||
3859 | default: | |
3860 | BUG(); | |
3861 | } | |
3862 | ||
3863 | return NULL; | |
3864 | } | |
c4f50183 | 3865 | EXPORT_SYMBOL_GPL(ring_buffer_iter_peek); |
7a8e76a3 | 3866 | |
8d707e8e SR |
3867 | static inline int rb_ok_to_lock(void) |
3868 | { | |
3869 | /* | |
3870 | * If an NMI die dumps out the content of the ring buffer | |
3871 | * do not grab locks. We also permanently disable the ring | |
3872 | * buffer too. A one time deal is all you get from reading | |
3873 | * the ring buffer from an NMI. | |
3874 | */ | |
464e85eb | 3875 | if (likely(!in_nmi())) |
8d707e8e SR |
3876 | return 1; |
3877 | ||
3878 | tracing_off_permanent(); | |
3879 | return 0; | |
3880 | } | |
3881 | ||
f83c9d0f SR |
3882 | /** |
3883 | * ring_buffer_peek - peek at the next event to be read | |
3884 | * @buffer: The ring buffer to read | |
3885 | * @cpu: The cpu to peak at | |
3886 | * @ts: The timestamp counter of this event. | |
66a8cb95 | 3887 | * @lost_events: a variable to store if events were lost (may be NULL) |
f83c9d0f SR |
3888 | * |
3889 | * This will return the event that will be read next, but does | |
3890 | * not consume the data. | |
3891 | */ | |
3892 | struct ring_buffer_event * | |
66a8cb95 SR |
3893 | ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts, |
3894 | unsigned long *lost_events) | |
f83c9d0f SR |
3895 | { |
3896 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
8aabee57 | 3897 | struct ring_buffer_event *event; |
f83c9d0f | 3898 | unsigned long flags; |
8d707e8e | 3899 | int dolock; |
f83c9d0f | 3900 | |
554f786e | 3901 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 3902 | return NULL; |
554f786e | 3903 | |
8d707e8e | 3904 | dolock = rb_ok_to_lock(); |
2d622719 | 3905 | again: |
8d707e8e SR |
3906 | local_irq_save(flags); |
3907 | if (dolock) | |
5389f6fa | 3908 | raw_spin_lock(&cpu_buffer->reader_lock); |
66a8cb95 | 3909 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
469535a5 RR |
3910 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
3911 | rb_advance_reader(cpu_buffer); | |
8d707e8e | 3912 | if (dolock) |
5389f6fa | 3913 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3914 | local_irq_restore(flags); |
f83c9d0f | 3915 | |
1b959e18 | 3916 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3917 | goto again; |
2d622719 | 3918 | |
f83c9d0f SR |
3919 | return event; |
3920 | } | |
3921 | ||
3922 | /** | |
3923 | * ring_buffer_iter_peek - peek at the next event to be read | |
3924 | * @iter: The ring buffer iterator | |
3925 | * @ts: The timestamp counter of this event. | |
3926 | * | |
3927 | * This will return the event that will be read next, but does | |
3928 | * not increment the iterator. | |
3929 | */ | |
3930 | struct ring_buffer_event * | |
3931 | ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts) | |
3932 | { | |
3933 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
3934 | struct ring_buffer_event *event; | |
3935 | unsigned long flags; | |
3936 | ||
2d622719 | 3937 | again: |
5389f6fa | 3938 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3939 | event = rb_iter_peek(iter, ts); |
5389f6fa | 3940 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 3941 | |
1b959e18 | 3942 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3943 | goto again; |
2d622719 | 3944 | |
f83c9d0f SR |
3945 | return event; |
3946 | } | |
3947 | ||
7a8e76a3 SR |
3948 | /** |
3949 | * ring_buffer_consume - return an event and consume it | |
3950 | * @buffer: The ring buffer to get the next event from | |
66a8cb95 SR |
3951 | * @cpu: the cpu to read the buffer from |
3952 | * @ts: a variable to store the timestamp (may be NULL) | |
3953 | * @lost_events: a variable to store if events were lost (may be NULL) | |
7a8e76a3 SR |
3954 | * |
3955 | * Returns the next event in the ring buffer, and that event is consumed. | |
3956 | * Meaning, that sequential reads will keep returning a different event, | |
3957 | * and eventually empty the ring buffer if the producer is slower. | |
3958 | */ | |
3959 | struct ring_buffer_event * | |
66a8cb95 SR |
3960 | ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts, |
3961 | unsigned long *lost_events) | |
7a8e76a3 | 3962 | { |
554f786e SR |
3963 | struct ring_buffer_per_cpu *cpu_buffer; |
3964 | struct ring_buffer_event *event = NULL; | |
f83c9d0f | 3965 | unsigned long flags; |
8d707e8e SR |
3966 | int dolock; |
3967 | ||
3968 | dolock = rb_ok_to_lock(); | |
7a8e76a3 | 3969 | |
2d622719 | 3970 | again: |
554f786e SR |
3971 | /* might be called in atomic */ |
3972 | preempt_disable(); | |
3973 | ||
9e01c1b7 | 3974 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e | 3975 | goto out; |
7a8e76a3 | 3976 | |
554f786e | 3977 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
3978 | local_irq_save(flags); |
3979 | if (dolock) | |
5389f6fa | 3980 | raw_spin_lock(&cpu_buffer->reader_lock); |
f83c9d0f | 3981 | |
66a8cb95 SR |
3982 | event = rb_buffer_peek(cpu_buffer, ts, lost_events); |
3983 | if (event) { | |
3984 | cpu_buffer->lost_events = 0; | |
469535a5 | 3985 | rb_advance_reader(cpu_buffer); |
66a8cb95 | 3986 | } |
7a8e76a3 | 3987 | |
8d707e8e | 3988 | if (dolock) |
5389f6fa | 3989 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 3990 | local_irq_restore(flags); |
f83c9d0f | 3991 | |
554f786e SR |
3992 | out: |
3993 | preempt_enable(); | |
3994 | ||
1b959e18 | 3995 | if (event && event->type_len == RINGBUF_TYPE_PADDING) |
2d622719 | 3996 | goto again; |
2d622719 | 3997 | |
7a8e76a3 SR |
3998 | return event; |
3999 | } | |
c4f50183 | 4000 | EXPORT_SYMBOL_GPL(ring_buffer_consume); |
7a8e76a3 SR |
4001 | |
4002 | /** | |
72c9ddfd | 4003 | * ring_buffer_read_prepare - Prepare for a non consuming read of the buffer |
7a8e76a3 SR |
4004 | * @buffer: The ring buffer to read from |
4005 | * @cpu: The cpu buffer to iterate over | |
4006 | * | |
72c9ddfd DM |
4007 | * This performs the initial preparations necessary to iterate |
4008 | * through the buffer. Memory is allocated, buffer recording | |
4009 | * is disabled, and the iterator pointer is returned to the caller. | |
7a8e76a3 | 4010 | * |
72c9ddfd DM |
4011 | * Disabling buffer recordng prevents the reading from being |
4012 | * corrupted. This is not a consuming read, so a producer is not | |
4013 | * expected. | |
4014 | * | |
4015 | * After a sequence of ring_buffer_read_prepare calls, the user is | |
d611851b | 4016 | * expected to make at least one call to ring_buffer_read_prepare_sync. |
72c9ddfd DM |
4017 | * Afterwards, ring_buffer_read_start is invoked to get things going |
4018 | * for real. | |
4019 | * | |
d611851b | 4020 | * This overall must be paired with ring_buffer_read_finish. |
7a8e76a3 SR |
4021 | */ |
4022 | struct ring_buffer_iter * | |
72c9ddfd | 4023 | ring_buffer_read_prepare(struct ring_buffer *buffer, int cpu) |
7a8e76a3 SR |
4024 | { |
4025 | struct ring_buffer_per_cpu *cpu_buffer; | |
8aabee57 | 4026 | struct ring_buffer_iter *iter; |
7a8e76a3 | 4027 | |
9e01c1b7 | 4028 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4029 | return NULL; |
7a8e76a3 SR |
4030 | |
4031 | iter = kmalloc(sizeof(*iter), GFP_KERNEL); | |
4032 | if (!iter) | |
8aabee57 | 4033 | return NULL; |
7a8e76a3 SR |
4034 | |
4035 | cpu_buffer = buffer->buffers[cpu]; | |
4036 | ||
4037 | iter->cpu_buffer = cpu_buffer; | |
4038 | ||
83f40318 | 4039 | atomic_inc(&buffer->resize_disabled); |
7a8e76a3 | 4040 | atomic_inc(&cpu_buffer->record_disabled); |
72c9ddfd DM |
4041 | |
4042 | return iter; | |
4043 | } | |
4044 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare); | |
4045 | ||
4046 | /** | |
4047 | * ring_buffer_read_prepare_sync - Synchronize a set of prepare calls | |
4048 | * | |
4049 | * All previously invoked ring_buffer_read_prepare calls to prepare | |
4050 | * iterators will be synchronized. Afterwards, read_buffer_read_start | |
4051 | * calls on those iterators are allowed. | |
4052 | */ | |
4053 | void | |
4054 | ring_buffer_read_prepare_sync(void) | |
4055 | { | |
7a8e76a3 | 4056 | synchronize_sched(); |
72c9ddfd DM |
4057 | } |
4058 | EXPORT_SYMBOL_GPL(ring_buffer_read_prepare_sync); | |
4059 | ||
4060 | /** | |
4061 | * ring_buffer_read_start - start a non consuming read of the buffer | |
4062 | * @iter: The iterator returned by ring_buffer_read_prepare | |
4063 | * | |
4064 | * This finalizes the startup of an iteration through the buffer. | |
4065 | * The iterator comes from a call to ring_buffer_read_prepare and | |
4066 | * an intervening ring_buffer_read_prepare_sync must have been | |
4067 | * performed. | |
4068 | * | |
d611851b | 4069 | * Must be paired with ring_buffer_read_finish. |
72c9ddfd DM |
4070 | */ |
4071 | void | |
4072 | ring_buffer_read_start(struct ring_buffer_iter *iter) | |
4073 | { | |
4074 | struct ring_buffer_per_cpu *cpu_buffer; | |
4075 | unsigned long flags; | |
4076 | ||
4077 | if (!iter) | |
4078 | return; | |
4079 | ||
4080 | cpu_buffer = iter->cpu_buffer; | |
7a8e76a3 | 4081 | |
5389f6fa | 4082 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
0199c4e6 | 4083 | arch_spin_lock(&cpu_buffer->lock); |
642edba5 | 4084 | rb_iter_reset(iter); |
0199c4e6 | 4085 | arch_spin_unlock(&cpu_buffer->lock); |
5389f6fa | 4086 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 | 4087 | } |
c4f50183 | 4088 | EXPORT_SYMBOL_GPL(ring_buffer_read_start); |
7a8e76a3 SR |
4089 | |
4090 | /** | |
d611851b | 4091 | * ring_buffer_read_finish - finish reading the iterator of the buffer |
7a8e76a3 SR |
4092 | * @iter: The iterator retrieved by ring_buffer_start |
4093 | * | |
4094 | * This re-enables the recording to the buffer, and frees the | |
4095 | * iterator. | |
4096 | */ | |
4097 | void | |
4098 | ring_buffer_read_finish(struct ring_buffer_iter *iter) | |
4099 | { | |
4100 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; | |
9366c1ba | 4101 | unsigned long flags; |
7a8e76a3 | 4102 | |
659f451f SR |
4103 | /* |
4104 | * Ring buffer is disabled from recording, here's a good place | |
9366c1ba SR |
4105 | * to check the integrity of the ring buffer. |
4106 | * Must prevent readers from trying to read, as the check | |
4107 | * clears the HEAD page and readers require it. | |
659f451f | 4108 | */ |
9366c1ba | 4109 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
659f451f | 4110 | rb_check_pages(cpu_buffer); |
9366c1ba | 4111 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
659f451f | 4112 | |
7a8e76a3 | 4113 | atomic_dec(&cpu_buffer->record_disabled); |
83f40318 | 4114 | atomic_dec(&cpu_buffer->buffer->resize_disabled); |
7a8e76a3 SR |
4115 | kfree(iter); |
4116 | } | |
c4f50183 | 4117 | EXPORT_SYMBOL_GPL(ring_buffer_read_finish); |
7a8e76a3 SR |
4118 | |
4119 | /** | |
4120 | * ring_buffer_read - read the next item in the ring buffer by the iterator | |
4121 | * @iter: The ring buffer iterator | |
4122 | * @ts: The time stamp of the event read. | |
4123 | * | |
4124 | * This reads the next event in the ring buffer and increments the iterator. | |
4125 | */ | |
4126 | struct ring_buffer_event * | |
4127 | ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts) | |
4128 | { | |
4129 | struct ring_buffer_event *event; | |
f83c9d0f SR |
4130 | struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; |
4131 | unsigned long flags; | |
7a8e76a3 | 4132 | |
5389f6fa | 4133 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
7e9391cf | 4134 | again: |
f83c9d0f | 4135 | event = rb_iter_peek(iter, ts); |
7a8e76a3 | 4136 | if (!event) |
f83c9d0f | 4137 | goto out; |
7a8e76a3 | 4138 | |
7e9391cf SR |
4139 | if (event->type_len == RINGBUF_TYPE_PADDING) |
4140 | goto again; | |
4141 | ||
7a8e76a3 | 4142 | rb_advance_iter(iter); |
f83c9d0f | 4143 | out: |
5389f6fa | 4144 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
7a8e76a3 SR |
4145 | |
4146 | return event; | |
4147 | } | |
c4f50183 | 4148 | EXPORT_SYMBOL_GPL(ring_buffer_read); |
7a8e76a3 SR |
4149 | |
4150 | /** | |
4151 | * ring_buffer_size - return the size of the ring buffer (in bytes) | |
4152 | * @buffer: The ring buffer. | |
4153 | */ | |
438ced17 | 4154 | unsigned long ring_buffer_size(struct ring_buffer *buffer, int cpu) |
7a8e76a3 | 4155 | { |
438ced17 VN |
4156 | /* |
4157 | * Earlier, this method returned | |
4158 | * BUF_PAGE_SIZE * buffer->nr_pages | |
4159 | * Since the nr_pages field is now removed, we have converted this to | |
4160 | * return the per cpu buffer value. | |
4161 | */ | |
4162 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) | |
4163 | return 0; | |
4164 | ||
4165 | return BUF_PAGE_SIZE * buffer->buffers[cpu]->nr_pages; | |
7a8e76a3 | 4166 | } |
c4f50183 | 4167 | EXPORT_SYMBOL_GPL(ring_buffer_size); |
7a8e76a3 SR |
4168 | |
4169 | static void | |
4170 | rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) | |
4171 | { | |
77ae365e SR |
4172 | rb_head_page_deactivate(cpu_buffer); |
4173 | ||
7a8e76a3 | 4174 | cpu_buffer->head_page |
3adc54fa | 4175 | = list_entry(cpu_buffer->pages, struct buffer_page, list); |
bf41a158 | 4176 | local_set(&cpu_buffer->head_page->write, 0); |
778c55d4 | 4177 | local_set(&cpu_buffer->head_page->entries, 0); |
abc9b56d | 4178 | local_set(&cpu_buffer->head_page->page->commit, 0); |
d769041f | 4179 | |
6f807acd | 4180 | cpu_buffer->head_page->read = 0; |
bf41a158 SR |
4181 | |
4182 | cpu_buffer->tail_page = cpu_buffer->head_page; | |
4183 | cpu_buffer->commit_page = cpu_buffer->head_page; | |
4184 | ||
4185 | INIT_LIST_HEAD(&cpu_buffer->reader_page->list); | |
5040b4b7 | 4186 | INIT_LIST_HEAD(&cpu_buffer->new_pages); |
bf41a158 | 4187 | local_set(&cpu_buffer->reader_page->write, 0); |
778c55d4 | 4188 | local_set(&cpu_buffer->reader_page->entries, 0); |
abc9b56d | 4189 | local_set(&cpu_buffer->reader_page->page->commit, 0); |
6f807acd | 4190 | cpu_buffer->reader_page->read = 0; |
7a8e76a3 | 4191 | |
c64e148a | 4192 | local_set(&cpu_buffer->entries_bytes, 0); |
77ae365e | 4193 | local_set(&cpu_buffer->overrun, 0); |
884bfe89 SP |
4194 | local_set(&cpu_buffer->commit_overrun, 0); |
4195 | local_set(&cpu_buffer->dropped_events, 0); | |
e4906eff | 4196 | local_set(&cpu_buffer->entries, 0); |
fa743953 SR |
4197 | local_set(&cpu_buffer->committing, 0); |
4198 | local_set(&cpu_buffer->commits, 0); | |
77ae365e | 4199 | cpu_buffer->read = 0; |
c64e148a | 4200 | cpu_buffer->read_bytes = 0; |
69507c06 SR |
4201 | |
4202 | cpu_buffer->write_stamp = 0; | |
4203 | cpu_buffer->read_stamp = 0; | |
77ae365e | 4204 | |
66a8cb95 SR |
4205 | cpu_buffer->lost_events = 0; |
4206 | cpu_buffer->last_overrun = 0; | |
4207 | ||
77ae365e | 4208 | rb_head_page_activate(cpu_buffer); |
7a8e76a3 SR |
4209 | } |
4210 | ||
4211 | /** | |
4212 | * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer | |
4213 | * @buffer: The ring buffer to reset a per cpu buffer of | |
4214 | * @cpu: The CPU buffer to be reset | |
4215 | */ | |
4216 | void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu) | |
4217 | { | |
4218 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
4219 | unsigned long flags; | |
4220 | ||
9e01c1b7 | 4221 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4222 | return; |
7a8e76a3 | 4223 | |
83f40318 | 4224 | atomic_inc(&buffer->resize_disabled); |
41ede23e SR |
4225 | atomic_inc(&cpu_buffer->record_disabled); |
4226 | ||
83f40318 VN |
4227 | /* Make sure all commits have finished */ |
4228 | synchronize_sched(); | |
4229 | ||
5389f6fa | 4230 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
f83c9d0f | 4231 | |
41b6a95d SR |
4232 | if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing))) |
4233 | goto out; | |
4234 | ||
0199c4e6 | 4235 | arch_spin_lock(&cpu_buffer->lock); |
7a8e76a3 SR |
4236 | |
4237 | rb_reset_cpu(cpu_buffer); | |
4238 | ||
0199c4e6 | 4239 | arch_spin_unlock(&cpu_buffer->lock); |
f83c9d0f | 4240 | |
41b6a95d | 4241 | out: |
5389f6fa | 4242 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
41ede23e SR |
4243 | |
4244 | atomic_dec(&cpu_buffer->record_disabled); | |
83f40318 | 4245 | atomic_dec(&buffer->resize_disabled); |
7a8e76a3 | 4246 | } |
c4f50183 | 4247 | EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); |
7a8e76a3 SR |
4248 | |
4249 | /** | |
4250 | * ring_buffer_reset - reset a ring buffer | |
4251 | * @buffer: The ring buffer to reset all cpu buffers | |
4252 | */ | |
4253 | void ring_buffer_reset(struct ring_buffer *buffer) | |
4254 | { | |
7a8e76a3 SR |
4255 | int cpu; |
4256 | ||
7a8e76a3 | 4257 | for_each_buffer_cpu(buffer, cpu) |
d769041f | 4258 | ring_buffer_reset_cpu(buffer, cpu); |
7a8e76a3 | 4259 | } |
c4f50183 | 4260 | EXPORT_SYMBOL_GPL(ring_buffer_reset); |
7a8e76a3 SR |
4261 | |
4262 | /** | |
4263 | * rind_buffer_empty - is the ring buffer empty? | |
4264 | * @buffer: The ring buffer to test | |
4265 | */ | |
4266 | int ring_buffer_empty(struct ring_buffer *buffer) | |
4267 | { | |
4268 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 4269 | unsigned long flags; |
8d707e8e | 4270 | int dolock; |
7a8e76a3 | 4271 | int cpu; |
d4788207 | 4272 | int ret; |
7a8e76a3 | 4273 | |
8d707e8e | 4274 | dolock = rb_ok_to_lock(); |
7a8e76a3 SR |
4275 | |
4276 | /* yes this is racy, but if you don't like the race, lock the buffer */ | |
4277 | for_each_buffer_cpu(buffer, cpu) { | |
4278 | cpu_buffer = buffer->buffers[cpu]; | |
8d707e8e SR |
4279 | local_irq_save(flags); |
4280 | if (dolock) | |
5389f6fa | 4281 | raw_spin_lock(&cpu_buffer->reader_lock); |
d4788207 | 4282 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e | 4283 | if (dolock) |
5389f6fa | 4284 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e SR |
4285 | local_irq_restore(flags); |
4286 | ||
d4788207 | 4287 | if (!ret) |
7a8e76a3 SR |
4288 | return 0; |
4289 | } | |
554f786e | 4290 | |
7a8e76a3 SR |
4291 | return 1; |
4292 | } | |
c4f50183 | 4293 | EXPORT_SYMBOL_GPL(ring_buffer_empty); |
7a8e76a3 SR |
4294 | |
4295 | /** | |
4296 | * ring_buffer_empty_cpu - is a cpu buffer of a ring buffer empty? | |
4297 | * @buffer: The ring buffer | |
4298 | * @cpu: The CPU buffer to test | |
4299 | */ | |
4300 | int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) | |
4301 | { | |
4302 | struct ring_buffer_per_cpu *cpu_buffer; | |
d4788207 | 4303 | unsigned long flags; |
8d707e8e | 4304 | int dolock; |
8aabee57 | 4305 | int ret; |
7a8e76a3 | 4306 | |
9e01c1b7 | 4307 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
8aabee57 | 4308 | return 1; |
7a8e76a3 | 4309 | |
8d707e8e SR |
4310 | dolock = rb_ok_to_lock(); |
4311 | ||
7a8e76a3 | 4312 | cpu_buffer = buffer->buffers[cpu]; |
8d707e8e SR |
4313 | local_irq_save(flags); |
4314 | if (dolock) | |
5389f6fa | 4315 | raw_spin_lock(&cpu_buffer->reader_lock); |
554f786e | 4316 | ret = rb_per_cpu_empty(cpu_buffer); |
8d707e8e | 4317 | if (dolock) |
5389f6fa | 4318 | raw_spin_unlock(&cpu_buffer->reader_lock); |
8d707e8e | 4319 | local_irq_restore(flags); |
554f786e SR |
4320 | |
4321 | return ret; | |
7a8e76a3 | 4322 | } |
c4f50183 | 4323 | EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); |
7a8e76a3 | 4324 | |
85bac32c | 4325 | #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP |
7a8e76a3 SR |
4326 | /** |
4327 | * ring_buffer_swap_cpu - swap a CPU buffer between two ring buffers | |
4328 | * @buffer_a: One buffer to swap with | |
4329 | * @buffer_b: The other buffer to swap with | |
4330 | * | |
4331 | * This function is useful for tracers that want to take a "snapshot" | |
4332 | * of a CPU buffer and has another back up buffer lying around. | |
4333 | * it is expected that the tracer handles the cpu buffer not being | |
4334 | * used at the moment. | |
4335 | */ | |
4336 | int ring_buffer_swap_cpu(struct ring_buffer *buffer_a, | |
4337 | struct ring_buffer *buffer_b, int cpu) | |
4338 | { | |
4339 | struct ring_buffer_per_cpu *cpu_buffer_a; | |
4340 | struct ring_buffer_per_cpu *cpu_buffer_b; | |
554f786e SR |
4341 | int ret = -EINVAL; |
4342 | ||
9e01c1b7 RR |
4343 | if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || |
4344 | !cpumask_test_cpu(cpu, buffer_b->cpumask)) | |
554f786e | 4345 | goto out; |
7a8e76a3 | 4346 | |
438ced17 VN |
4347 | cpu_buffer_a = buffer_a->buffers[cpu]; |
4348 | cpu_buffer_b = buffer_b->buffers[cpu]; | |
4349 | ||
7a8e76a3 | 4350 | /* At least make sure the two buffers are somewhat the same */ |
438ced17 | 4351 | if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages) |
554f786e SR |
4352 | goto out; |
4353 | ||
4354 | ret = -EAGAIN; | |
7a8e76a3 | 4355 | |
97b17efe | 4356 | if (ring_buffer_flags != RB_BUFFERS_ON) |
554f786e | 4357 | goto out; |
97b17efe SR |
4358 | |
4359 | if (atomic_read(&buffer_a->record_disabled)) | |
554f786e | 4360 | goto out; |
97b17efe SR |
4361 | |
4362 | if (atomic_read(&buffer_b->record_disabled)) | |
554f786e | 4363 | goto out; |
97b17efe | 4364 | |
97b17efe | 4365 | if (atomic_read(&cpu_buffer_a->record_disabled)) |
554f786e | 4366 | goto out; |
97b17efe SR |
4367 | |
4368 | if (atomic_read(&cpu_buffer_b->record_disabled)) | |
554f786e | 4369 | goto out; |
97b17efe | 4370 | |
7a8e76a3 SR |
4371 | /* |
4372 | * We can't do a synchronize_sched here because this | |
4373 | * function can be called in atomic context. | |
4374 | * Normally this will be called from the same CPU as cpu. | |
4375 | * If not it's up to the caller to protect this. | |
4376 | */ | |
4377 | atomic_inc(&cpu_buffer_a->record_disabled); | |
4378 | atomic_inc(&cpu_buffer_b->record_disabled); | |
4379 | ||
98277991 SR |
4380 | ret = -EBUSY; |
4381 | if (local_read(&cpu_buffer_a->committing)) | |
4382 | goto out_dec; | |
4383 | if (local_read(&cpu_buffer_b->committing)) | |
4384 | goto out_dec; | |
4385 | ||
7a8e76a3 SR |
4386 | buffer_a->buffers[cpu] = cpu_buffer_b; |
4387 | buffer_b->buffers[cpu] = cpu_buffer_a; | |
4388 | ||
4389 | cpu_buffer_b->buffer = buffer_a; | |
4390 | cpu_buffer_a->buffer = buffer_b; | |
4391 | ||
98277991 SR |
4392 | ret = 0; |
4393 | ||
4394 | out_dec: | |
7a8e76a3 SR |
4395 | atomic_dec(&cpu_buffer_a->record_disabled); |
4396 | atomic_dec(&cpu_buffer_b->record_disabled); | |
554f786e | 4397 | out: |
554f786e | 4398 | return ret; |
7a8e76a3 | 4399 | } |
c4f50183 | 4400 | EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); |
85bac32c | 4401 | #endif /* CONFIG_RING_BUFFER_ALLOW_SWAP */ |
7a8e76a3 | 4402 | |
8789a9e7 SR |
4403 | /** |
4404 | * ring_buffer_alloc_read_page - allocate a page to read from buffer | |
4405 | * @buffer: the buffer to allocate for. | |
d611851b | 4406 | * @cpu: the cpu buffer to allocate. |
8789a9e7 SR |
4407 | * |
4408 | * This function is used in conjunction with ring_buffer_read_page. | |
4409 | * When reading a full page from the ring buffer, these functions | |
4410 | * can be used to speed up the process. The calling function should | |
4411 | * allocate a few pages first with this function. Then when it | |
4412 | * needs to get pages from the ring buffer, it passes the result | |
4413 | * of this function into ring_buffer_read_page, which will swap | |
4414 | * the page that was allocated, with the read page of the buffer. | |
4415 | * | |
4416 | * Returns: | |
4417 | * The page allocated, or NULL on error. | |
4418 | */ | |
7ea59064 | 4419 | void *ring_buffer_alloc_read_page(struct ring_buffer *buffer, int cpu) |
8789a9e7 | 4420 | { |
044fa782 | 4421 | struct buffer_data_page *bpage; |
7ea59064 | 4422 | struct page *page; |
8789a9e7 | 4423 | |
d7ec4bfe VN |
4424 | page = alloc_pages_node(cpu_to_node(cpu), |
4425 | GFP_KERNEL | __GFP_NORETRY, 0); | |
7ea59064 | 4426 | if (!page) |
8789a9e7 SR |
4427 | return NULL; |
4428 | ||
7ea59064 | 4429 | bpage = page_address(page); |
8789a9e7 | 4430 | |
ef7a4a16 SR |
4431 | rb_init_page(bpage); |
4432 | ||
044fa782 | 4433 | return bpage; |
8789a9e7 | 4434 | } |
d6ce96da | 4435 | EXPORT_SYMBOL_GPL(ring_buffer_alloc_read_page); |
8789a9e7 SR |
4436 | |
4437 | /** | |
4438 | * ring_buffer_free_read_page - free an allocated read page | |
4439 | * @buffer: the buffer the page was allocate for | |
4440 | * @data: the page to free | |
4441 | * | |
4442 | * Free a page allocated from ring_buffer_alloc_read_page. | |
4443 | */ | |
4444 | void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data) | |
4445 | { | |
4446 | free_page((unsigned long)data); | |
4447 | } | |
d6ce96da | 4448 | EXPORT_SYMBOL_GPL(ring_buffer_free_read_page); |
8789a9e7 SR |
4449 | |
4450 | /** | |
4451 | * ring_buffer_read_page - extract a page from the ring buffer | |
4452 | * @buffer: buffer to extract from | |
4453 | * @data_page: the page to use allocated from ring_buffer_alloc_read_page | |
ef7a4a16 | 4454 | * @len: amount to extract |
8789a9e7 SR |
4455 | * @cpu: the cpu of the buffer to extract |
4456 | * @full: should the extraction only happen when the page is full. | |
4457 | * | |
4458 | * This function will pull out a page from the ring buffer and consume it. | |
4459 | * @data_page must be the address of the variable that was returned | |
4460 | * from ring_buffer_alloc_read_page. This is because the page might be used | |
4461 | * to swap with a page in the ring buffer. | |
4462 | * | |
4463 | * for example: | |
d611851b | 4464 | * rpage = ring_buffer_alloc_read_page(buffer, cpu); |
8789a9e7 SR |
4465 | * if (!rpage) |
4466 | * return error; | |
ef7a4a16 | 4467 | * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0); |
667d2412 LJ |
4468 | * if (ret >= 0) |
4469 | * process_page(rpage, ret); | |
8789a9e7 SR |
4470 | * |
4471 | * When @full is set, the function will not return true unless | |
4472 | * the writer is off the reader page. | |
4473 | * | |
4474 | * Note: it is up to the calling functions to handle sleeps and wakeups. | |
4475 | * The ring buffer can be used anywhere in the kernel and can not | |
4476 | * blindly call wake_up. The layer that uses the ring buffer must be | |
4477 | * responsible for that. | |
4478 | * | |
4479 | * Returns: | |
667d2412 LJ |
4480 | * >=0 if data has been transferred, returns the offset of consumed data. |
4481 | * <0 if no data has been transferred. | |
8789a9e7 SR |
4482 | */ |
4483 | int ring_buffer_read_page(struct ring_buffer *buffer, | |
ef7a4a16 | 4484 | void **data_page, size_t len, int cpu, int full) |
8789a9e7 SR |
4485 | { |
4486 | struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; | |
4487 | struct ring_buffer_event *event; | |
044fa782 | 4488 | struct buffer_data_page *bpage; |
ef7a4a16 | 4489 | struct buffer_page *reader; |
ff0ff84a | 4490 | unsigned long missed_events; |
8789a9e7 | 4491 | unsigned long flags; |
ef7a4a16 | 4492 | unsigned int commit; |
667d2412 | 4493 | unsigned int read; |
4f3640f8 | 4494 | u64 save_timestamp; |
667d2412 | 4495 | int ret = -1; |
8789a9e7 | 4496 | |
554f786e SR |
4497 | if (!cpumask_test_cpu(cpu, buffer->cpumask)) |
4498 | goto out; | |
4499 | ||
474d32b6 SR |
4500 | /* |
4501 | * If len is not big enough to hold the page header, then | |
4502 | * we can not copy anything. | |
4503 | */ | |
4504 | if (len <= BUF_PAGE_HDR_SIZE) | |
554f786e | 4505 | goto out; |
474d32b6 SR |
4506 | |
4507 | len -= BUF_PAGE_HDR_SIZE; | |
4508 | ||
8789a9e7 | 4509 | if (!data_page) |
554f786e | 4510 | goto out; |
8789a9e7 | 4511 | |
044fa782 SR |
4512 | bpage = *data_page; |
4513 | if (!bpage) | |
554f786e | 4514 | goto out; |
8789a9e7 | 4515 | |
5389f6fa | 4516 | raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4517 | |
ef7a4a16 SR |
4518 | reader = rb_get_reader_page(cpu_buffer); |
4519 | if (!reader) | |
554f786e | 4520 | goto out_unlock; |
8789a9e7 | 4521 | |
ef7a4a16 SR |
4522 | event = rb_reader_event(cpu_buffer); |
4523 | ||
4524 | read = reader->read; | |
4525 | commit = rb_page_commit(reader); | |
667d2412 | 4526 | |
66a8cb95 | 4527 | /* Check if any events were dropped */ |
ff0ff84a | 4528 | missed_events = cpu_buffer->lost_events; |
66a8cb95 | 4529 | |
8789a9e7 | 4530 | /* |
474d32b6 SR |
4531 | * If this page has been partially read or |
4532 | * if len is not big enough to read the rest of the page or | |
4533 | * a writer is still on the page, then | |
4534 | * we must copy the data from the page to the buffer. | |
4535 | * Otherwise, we can simply swap the page with the one passed in. | |
8789a9e7 | 4536 | */ |
474d32b6 | 4537 | if (read || (len < (commit - read)) || |
ef7a4a16 | 4538 | cpu_buffer->reader_page == cpu_buffer->commit_page) { |
667d2412 | 4539 | struct buffer_data_page *rpage = cpu_buffer->reader_page->page; |
474d32b6 SR |
4540 | unsigned int rpos = read; |
4541 | unsigned int pos = 0; | |
ef7a4a16 | 4542 | unsigned int size; |
8789a9e7 SR |
4543 | |
4544 | if (full) | |
554f786e | 4545 | goto out_unlock; |
8789a9e7 | 4546 | |
ef7a4a16 SR |
4547 | if (len > (commit - read)) |
4548 | len = (commit - read); | |
4549 | ||
69d1b839 SR |
4550 | /* Always keep the time extend and data together */ |
4551 | size = rb_event_ts_length(event); | |
ef7a4a16 SR |
4552 | |
4553 | if (len < size) | |
554f786e | 4554 | goto out_unlock; |
ef7a4a16 | 4555 | |
4f3640f8 SR |
4556 | /* save the current timestamp, since the user will need it */ |
4557 | save_timestamp = cpu_buffer->read_stamp; | |
4558 | ||
ef7a4a16 SR |
4559 | /* Need to copy one event at a time */ |
4560 | do { | |
e1e35927 DS |
4561 | /* We need the size of one event, because |
4562 | * rb_advance_reader only advances by one event, | |
4563 | * whereas rb_event_ts_length may include the size of | |
4564 | * one or two events. | |
4565 | * We have already ensured there's enough space if this | |
4566 | * is a time extend. */ | |
4567 | size = rb_event_length(event); | |
474d32b6 | 4568 | memcpy(bpage->data + pos, rpage->data + rpos, size); |
ef7a4a16 SR |
4569 | |
4570 | len -= size; | |
4571 | ||
4572 | rb_advance_reader(cpu_buffer); | |
474d32b6 SR |
4573 | rpos = reader->read; |
4574 | pos += size; | |
ef7a4a16 | 4575 | |
18fab912 HY |
4576 | if (rpos >= commit) |
4577 | break; | |
4578 | ||
ef7a4a16 | 4579 | event = rb_reader_event(cpu_buffer); |
69d1b839 SR |
4580 | /* Always keep the time extend and data together */ |
4581 | size = rb_event_ts_length(event); | |
e1e35927 | 4582 | } while (len >= size); |
667d2412 LJ |
4583 | |
4584 | /* update bpage */ | |
ef7a4a16 | 4585 | local_set(&bpage->commit, pos); |
4f3640f8 | 4586 | bpage->time_stamp = save_timestamp; |
ef7a4a16 | 4587 | |
474d32b6 SR |
4588 | /* we copied everything to the beginning */ |
4589 | read = 0; | |
8789a9e7 | 4590 | } else { |
afbab76a | 4591 | /* update the entry counter */ |
77ae365e | 4592 | cpu_buffer->read += rb_page_entries(reader); |
c64e148a | 4593 | cpu_buffer->read_bytes += BUF_PAGE_SIZE; |
afbab76a | 4594 | |
8789a9e7 | 4595 | /* swap the pages */ |
044fa782 | 4596 | rb_init_page(bpage); |
ef7a4a16 SR |
4597 | bpage = reader->page; |
4598 | reader->page = *data_page; | |
4599 | local_set(&reader->write, 0); | |
778c55d4 | 4600 | local_set(&reader->entries, 0); |
ef7a4a16 | 4601 | reader->read = 0; |
044fa782 | 4602 | *data_page = bpage; |
ff0ff84a SR |
4603 | |
4604 | /* | |
4605 | * Use the real_end for the data size, | |
4606 | * This gives us a chance to store the lost events | |
4607 | * on the page. | |
4608 | */ | |
4609 | if (reader->real_end) | |
4610 | local_set(&bpage->commit, reader->real_end); | |
8789a9e7 | 4611 | } |
667d2412 | 4612 | ret = read; |
8789a9e7 | 4613 | |
66a8cb95 | 4614 | cpu_buffer->lost_events = 0; |
2711ca23 SR |
4615 | |
4616 | commit = local_read(&bpage->commit); | |
66a8cb95 SR |
4617 | /* |
4618 | * Set a flag in the commit field if we lost events | |
4619 | */ | |
ff0ff84a | 4620 | if (missed_events) { |
ff0ff84a SR |
4621 | /* If there is room at the end of the page to save the |
4622 | * missed events, then record it there. | |
4623 | */ | |
4624 | if (BUF_PAGE_SIZE - commit >= sizeof(missed_events)) { | |
4625 | memcpy(&bpage->data[commit], &missed_events, | |
4626 | sizeof(missed_events)); | |
4627 | local_add(RB_MISSED_STORED, &bpage->commit); | |
2711ca23 | 4628 | commit += sizeof(missed_events); |
ff0ff84a | 4629 | } |
66a8cb95 | 4630 | local_add(RB_MISSED_EVENTS, &bpage->commit); |
ff0ff84a | 4631 | } |
66a8cb95 | 4632 | |
2711ca23 SR |
4633 | /* |
4634 | * This page may be off to user land. Zero it out here. | |
4635 | */ | |
4636 | if (commit < BUF_PAGE_SIZE) | |
4637 | memset(&bpage->data[commit], 0, BUF_PAGE_SIZE - commit); | |
4638 | ||
554f786e | 4639 | out_unlock: |
5389f6fa | 4640 | raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); |
8789a9e7 | 4641 | |
554f786e | 4642 | out: |
8789a9e7 SR |
4643 | return ret; |
4644 | } | |
d6ce96da | 4645 | EXPORT_SYMBOL_GPL(ring_buffer_read_page); |
8789a9e7 | 4646 | |
59222efe | 4647 | #ifdef CONFIG_HOTPLUG_CPU |
09c9e84d FW |
4648 | static int rb_cpu_notify(struct notifier_block *self, |
4649 | unsigned long action, void *hcpu) | |
554f786e SR |
4650 | { |
4651 | struct ring_buffer *buffer = | |
4652 | container_of(self, struct ring_buffer, cpu_notify); | |
4653 | long cpu = (long)hcpu; | |
438ced17 VN |
4654 | int cpu_i, nr_pages_same; |
4655 | unsigned int nr_pages; | |
554f786e SR |
4656 | |
4657 | switch (action) { | |
4658 | case CPU_UP_PREPARE: | |
4659 | case CPU_UP_PREPARE_FROZEN: | |
3f237a79 | 4660 | if (cpumask_test_cpu(cpu, buffer->cpumask)) |
554f786e SR |
4661 | return NOTIFY_OK; |
4662 | ||
438ced17 VN |
4663 | nr_pages = 0; |
4664 | nr_pages_same = 1; | |
4665 | /* check if all cpu sizes are same */ | |
4666 | for_each_buffer_cpu(buffer, cpu_i) { | |
4667 | /* fill in the size from first enabled cpu */ | |
4668 | if (nr_pages == 0) | |
4669 | nr_pages = buffer->buffers[cpu_i]->nr_pages; | |
4670 | if (nr_pages != buffer->buffers[cpu_i]->nr_pages) { | |
4671 | nr_pages_same = 0; | |
4672 | break; | |
4673 | } | |
4674 | } | |
4675 | /* allocate minimum pages, user can later expand it */ | |
4676 | if (!nr_pages_same) | |
4677 | nr_pages = 2; | |
554f786e | 4678 | buffer->buffers[cpu] = |
438ced17 | 4679 | rb_allocate_cpu_buffer(buffer, nr_pages, cpu); |
554f786e SR |
4680 | if (!buffer->buffers[cpu]) { |
4681 | WARN(1, "failed to allocate ring buffer on CPU %ld\n", | |
4682 | cpu); | |
4683 | return NOTIFY_OK; | |
4684 | } | |
4685 | smp_wmb(); | |
3f237a79 | 4686 | cpumask_set_cpu(cpu, buffer->cpumask); |
554f786e SR |
4687 | break; |
4688 | case CPU_DOWN_PREPARE: | |
4689 | case CPU_DOWN_PREPARE_FROZEN: | |
4690 | /* | |
4691 | * Do nothing. | |
4692 | * If we were to free the buffer, then the user would | |
4693 | * lose any trace that was in the buffer. | |
4694 | */ | |
4695 | break; | |
4696 | default: | |
4697 | break; | |
4698 | } | |
4699 | return NOTIFY_OK; | |
4700 | } | |
4701 | #endif | |
6c43e554 SRRH |
4702 | |
4703 | #ifdef CONFIG_RING_BUFFER_STARTUP_TEST | |
4704 | /* | |
4705 | * This is a basic integrity check of the ring buffer. | |
4706 | * Late in the boot cycle this test will run when configured in. | |
4707 | * It will kick off a thread per CPU that will go into a loop | |
4708 | * writing to the per cpu ring buffer various sizes of data. | |
4709 | * Some of the data will be large items, some small. | |
4710 | * | |
4711 | * Another thread is created that goes into a spin, sending out | |
4712 | * IPIs to the other CPUs to also write into the ring buffer. | |
4713 | * this is to test the nesting ability of the buffer. | |
4714 | * | |
4715 | * Basic stats are recorded and reported. If something in the | |
4716 | * ring buffer should happen that's not expected, a big warning | |
4717 | * is displayed and all ring buffers are disabled. | |
4718 | */ | |
4719 | static struct task_struct *rb_threads[NR_CPUS] __initdata; | |
4720 | ||
4721 | struct rb_test_data { | |
4722 | struct ring_buffer *buffer; | |
4723 | unsigned long events; | |
4724 | unsigned long bytes_written; | |
4725 | unsigned long bytes_alloc; | |
4726 | unsigned long bytes_dropped; | |
4727 | unsigned long events_nested; | |
4728 | unsigned long bytes_written_nested; | |
4729 | unsigned long bytes_alloc_nested; | |
4730 | unsigned long bytes_dropped_nested; | |
4731 | int min_size_nested; | |
4732 | int max_size_nested; | |
4733 | int max_size; | |
4734 | int min_size; | |
4735 | int cpu; | |
4736 | int cnt; | |
4737 | }; | |
4738 | ||
4739 | static struct rb_test_data rb_data[NR_CPUS] __initdata; | |
4740 | ||
4741 | /* 1 meg per cpu */ | |
4742 | #define RB_TEST_BUFFER_SIZE 1048576 | |
4743 | ||
4744 | static char rb_string[] __initdata = | |
4745 | "abcdefghijklmnopqrstuvwxyz1234567890!@#$%^&*()?+\\" | |
4746 | "?+|:';\",.<>/?abcdefghijklmnopqrstuvwxyz1234567890" | |
4747 | "!@#$%^&*()?+\\?+|:';\",.<>/?abcdefghijklmnopqrstuv"; | |
4748 | ||
4749 | static bool rb_test_started __initdata; | |
4750 | ||
4751 | struct rb_item { | |
4752 | int size; | |
4753 | char str[]; | |
4754 | }; | |
4755 | ||
4756 | static __init int rb_write_something(struct rb_test_data *data, bool nested) | |
4757 | { | |
4758 | struct ring_buffer_event *event; | |
4759 | struct rb_item *item; | |
4760 | bool started; | |
4761 | int event_len; | |
4762 | int size; | |
4763 | int len; | |
4764 | int cnt; | |
4765 | ||
4766 | /* Have nested writes different that what is written */ | |
4767 | cnt = data->cnt + (nested ? 27 : 0); | |
4768 | ||
4769 | /* Multiply cnt by ~e, to make some unique increment */ | |
4770 | size = (data->cnt * 68 / 25) % (sizeof(rb_string) - 1); | |
4771 | ||
4772 | len = size + sizeof(struct rb_item); | |
4773 | ||
4774 | started = rb_test_started; | |
4775 | /* read rb_test_started before checking buffer enabled */ | |
4776 | smp_rmb(); | |
4777 | ||
4778 | event = ring_buffer_lock_reserve(data->buffer, len); | |
4779 | if (!event) { | |
4780 | /* Ignore dropped events before test starts. */ | |
4781 | if (started) { | |
4782 | if (nested) | |
4783 | data->bytes_dropped += len; | |
4784 | else | |
4785 | data->bytes_dropped_nested += len; | |
4786 | } | |
4787 | return len; | |
4788 | } | |
4789 | ||
4790 | event_len = ring_buffer_event_length(event); | |
4791 | ||
4792 | if (RB_WARN_ON(data->buffer, event_len < len)) | |
4793 | goto out; | |
4794 | ||
4795 | item = ring_buffer_event_data(event); | |
4796 | item->size = size; | |
4797 | memcpy(item->str, rb_string, size); | |
4798 | ||
4799 | if (nested) { | |
4800 | data->bytes_alloc_nested += event_len; | |
4801 | data->bytes_written_nested += len; | |
4802 | data->events_nested++; | |
4803 | if (!data->min_size_nested || len < data->min_size_nested) | |
4804 | data->min_size_nested = len; | |
4805 | if (len > data->max_size_nested) | |
4806 | data->max_size_nested = len; | |
4807 | } else { | |
4808 | data->bytes_alloc += event_len; | |
4809 | data->bytes_written += len; | |
4810 | data->events++; | |
4811 | if (!data->min_size || len < data->min_size) | |
4812 | data->max_size = len; | |
4813 | if (len > data->max_size) | |
4814 | data->max_size = len; | |
4815 | } | |
4816 | ||
4817 | out: | |
4818 | ring_buffer_unlock_commit(data->buffer, event); | |
4819 | ||
4820 | return 0; | |
4821 | } | |
4822 | ||
4823 | static __init int rb_test(void *arg) | |
4824 | { | |
4825 | struct rb_test_data *data = arg; | |
4826 | ||
4827 | while (!kthread_should_stop()) { | |
4828 | rb_write_something(data, false); | |
4829 | data->cnt++; | |
4830 | ||
4831 | set_current_state(TASK_INTERRUPTIBLE); | |
4832 | /* Now sleep between a min of 100-300us and a max of 1ms */ | |
4833 | usleep_range(((data->cnt % 3) + 1) * 100, 1000); | |
4834 | } | |
4835 | ||
4836 | return 0; | |
4837 | } | |
4838 | ||
4839 | static __init void rb_ipi(void *ignore) | |
4840 | { | |
4841 | struct rb_test_data *data; | |
4842 | int cpu = smp_processor_id(); | |
4843 | ||
4844 | data = &rb_data[cpu]; | |
4845 | rb_write_something(data, true); | |
4846 | } | |
4847 | ||
4848 | static __init int rb_hammer_test(void *arg) | |
4849 | { | |
4850 | while (!kthread_should_stop()) { | |
4851 | ||
4852 | /* Send an IPI to all cpus to write data! */ | |
4853 | smp_call_function(rb_ipi, NULL, 1); | |
4854 | /* No sleep, but for non preempt, let others run */ | |
4855 | schedule(); | |
4856 | } | |
4857 | ||
4858 | return 0; | |
4859 | } | |
4860 | ||
4861 | static __init int test_ringbuffer(void) | |
4862 | { | |
4863 | struct task_struct *rb_hammer; | |
4864 | struct ring_buffer *buffer; | |
4865 | int cpu; | |
4866 | int ret = 0; | |
4867 | ||
4868 | pr_info("Running ring buffer tests...\n"); | |
4869 | ||
4870 | buffer = ring_buffer_alloc(RB_TEST_BUFFER_SIZE, RB_FL_OVERWRITE); | |
4871 | if (WARN_ON(!buffer)) | |
4872 | return 0; | |
4873 | ||
4874 | /* Disable buffer so that threads can't write to it yet */ | |
4875 | ring_buffer_record_off(buffer); | |
4876 | ||
4877 | for_each_online_cpu(cpu) { | |
4878 | rb_data[cpu].buffer = buffer; | |
4879 | rb_data[cpu].cpu = cpu; | |
4880 | rb_data[cpu].cnt = cpu; | |
4881 | rb_threads[cpu] = kthread_create(rb_test, &rb_data[cpu], | |
4882 | "rbtester/%d", cpu); | |
4883 | if (WARN_ON(!rb_threads[cpu])) { | |
4884 | pr_cont("FAILED\n"); | |
4885 | ret = -1; | |
4886 | goto out_free; | |
4887 | } | |
4888 | ||
4889 | kthread_bind(rb_threads[cpu], cpu); | |
4890 | wake_up_process(rb_threads[cpu]); | |
4891 | } | |
4892 | ||
4893 | /* Now create the rb hammer! */ | |
4894 | rb_hammer = kthread_run(rb_hammer_test, NULL, "rbhammer"); | |
4895 | if (WARN_ON(!rb_hammer)) { | |
4896 | pr_cont("FAILED\n"); | |
4897 | ret = -1; | |
4898 | goto out_free; | |
4899 | } | |
4900 | ||
4901 | ring_buffer_record_on(buffer); | |
4902 | /* | |
4903 | * Show buffer is enabled before setting rb_test_started. | |
4904 | * Yes there's a small race window where events could be | |
4905 | * dropped and the thread wont catch it. But when a ring | |
4906 | * buffer gets enabled, there will always be some kind of | |
4907 | * delay before other CPUs see it. Thus, we don't care about | |
4908 | * those dropped events. We care about events dropped after | |
4909 | * the threads see that the buffer is active. | |
4910 | */ | |
4911 | smp_wmb(); | |
4912 | rb_test_started = true; | |
4913 | ||
4914 | set_current_state(TASK_INTERRUPTIBLE); | |
4915 | /* Just run for 10 seconds */; | |
4916 | schedule_timeout(10 * HZ); | |
4917 | ||
4918 | kthread_stop(rb_hammer); | |
4919 | ||
4920 | out_free: | |
4921 | for_each_online_cpu(cpu) { | |
4922 | if (!rb_threads[cpu]) | |
4923 | break; | |
4924 | kthread_stop(rb_threads[cpu]); | |
4925 | } | |
4926 | if (ret) { | |
4927 | ring_buffer_free(buffer); | |
4928 | return ret; | |
4929 | } | |
4930 | ||
4931 | /* Report! */ | |
4932 | pr_info("finished\n"); | |
4933 | for_each_online_cpu(cpu) { | |
4934 | struct ring_buffer_event *event; | |
4935 | struct rb_test_data *data = &rb_data[cpu]; | |
4936 | struct rb_item *item; | |
4937 | unsigned long total_events; | |
4938 | unsigned long total_dropped; | |
4939 | unsigned long total_written; | |
4940 | unsigned long total_alloc; | |
4941 | unsigned long total_read = 0; | |
4942 | unsigned long total_size = 0; | |
4943 | unsigned long total_len = 0; | |
4944 | unsigned long total_lost = 0; | |
4945 | unsigned long lost; | |
4946 | int big_event_size; | |
4947 | int small_event_size; | |
4948 | ||
4949 | ret = -1; | |
4950 | ||
4951 | total_events = data->events + data->events_nested; | |
4952 | total_written = data->bytes_written + data->bytes_written_nested; | |
4953 | total_alloc = data->bytes_alloc + data->bytes_alloc_nested; | |
4954 | total_dropped = data->bytes_dropped + data->bytes_dropped_nested; | |
4955 | ||
4956 | big_event_size = data->max_size + data->max_size_nested; | |
4957 | small_event_size = data->min_size + data->min_size_nested; | |
4958 | ||
4959 | pr_info("CPU %d:\n", cpu); | |
4960 | pr_info(" events: %ld\n", total_events); | |
4961 | pr_info(" dropped bytes: %ld\n", total_dropped); | |
4962 | pr_info(" alloced bytes: %ld\n", total_alloc); | |
4963 | pr_info(" written bytes: %ld\n", total_written); | |
4964 | pr_info(" biggest event: %d\n", big_event_size); | |
4965 | pr_info(" smallest event: %d\n", small_event_size); | |
4966 | ||
4967 | if (RB_WARN_ON(buffer, total_dropped)) | |
4968 | break; | |
4969 | ||
4970 | ret = 0; | |
4971 | ||
4972 | while ((event = ring_buffer_consume(buffer, cpu, NULL, &lost))) { | |
4973 | total_lost += lost; | |
4974 | item = ring_buffer_event_data(event); | |
4975 | total_len += ring_buffer_event_length(event); | |
4976 | total_size += item->size + sizeof(struct rb_item); | |
4977 | if (memcmp(&item->str[0], rb_string, item->size) != 0) { | |
4978 | pr_info("FAILED!\n"); | |
4979 | pr_info("buffer had: %.*s\n", item->size, item->str); | |
4980 | pr_info("expected: %.*s\n", item->size, rb_string); | |
4981 | RB_WARN_ON(buffer, 1); | |
4982 | ret = -1; | |
4983 | break; | |
4984 | } | |
4985 | total_read++; | |
4986 | } | |
4987 | if (ret) | |
4988 | break; | |
4989 | ||
4990 | ret = -1; | |
4991 | ||
4992 | pr_info(" read events: %ld\n", total_read); | |
4993 | pr_info(" lost events: %ld\n", total_lost); | |
4994 | pr_info(" total events: %ld\n", total_lost + total_read); | |
4995 | pr_info(" recorded len bytes: %ld\n", total_len); | |
4996 | pr_info(" recorded size bytes: %ld\n", total_size); | |
4997 | if (total_lost) | |
4998 | pr_info(" With dropped events, record len and size may not match\n" | |
4999 | " alloced and written from above\n"); | |
5000 | if (!total_lost) { | |
5001 | if (RB_WARN_ON(buffer, total_len != total_alloc || | |
5002 | total_size != total_written)) | |
5003 | break; | |
5004 | } | |
5005 | if (RB_WARN_ON(buffer, total_lost + total_read != total_events)) | |
5006 | break; | |
5007 | ||
5008 | ret = 0; | |
5009 | } | |
5010 | if (!ret) | |
5011 | pr_info("Ring buffer PASSED!\n"); | |
5012 | ||
5013 | ring_buffer_free(buffer); | |
5014 | return 0; | |
5015 | } | |
5016 | ||
5017 | late_initcall(test_ringbuffer); | |
5018 | #endif /* CONFIG_RING_BUFFER_STARTUP_TEST */ |